Physical AI Brief
Daily cross-source signals for the Physical AI supply chain — silicon photonics, CPO, VLA models, humanoid hardware, embodied AI. Three streams, one page, zero filler.
1155 items today · 1093 arxiv · 0 SEC 8-K · 62 humanoid · 0 CN photonics
01 ARXIV · PHYSICAL AI PAPERS
1093 items- arxiv:2610.18874 · cs.ROChunkAct: Action Chunking for Agentic RoboticsBingxin Xu, Wenrui Bao, Xinxin Liu, Jiawei Zhou +2
Agentic robotics places a large language model (LLM) agent in the robot control loop, where it composes perception, motion, and vision-language-action (VLA) tools under the guidance of a memory and thereby generalizes to new tasks without retraining. This generality comes at a steep price in efficie
vision-language-actionaction chunkingliberomemoryagentagentic - arxiv:2610.18873 · cs.ROLoopWAM: Closed-Loop Generalist Policy Improvement from Deployment RolloutsWenxuan Guo, Yingping Liang, Zhaoguo Wang, Ziyuan Li +10
Scaling expert demonstrations alone does not enable generalist robot policies to improve from deployment. Real-world rollouts contain successes, suboptimal behavior, and failures: imitating every recorded action can reinforce poor behavior, while filtering out failures discards evidence of physical
embodiedrobotwinworld modelaction-conditionedpost-trainingbenchmark - arxiv:2610.18872 · cs.ROEgoExplorer: A Vision-Motion-Touch Observatory for Human Loco-Manipulation In SituYunsong Zhou, Xuekun Jiang, Haiqi Shen, Hangxu Liu +10
Learning humanoid loco-manipulation from human demonstrations rests on observing how locomotion, posture, and object interaction evolve together as tasks unfold. Scaling collection across diverse environments, in turn, favors lightweight, portable capture systems. Yet, existing datasets provide only
manipulationhumanoidtactilewhole-body control - arxiv:2610.18871 · cs.LGCepheid: Decoupling Scale from Structure for Cross-Family Motion CompletionChih-Hai Su, Tsung-Wei Ke, Yu-Lun Liu
Sparse-to-dense motion completion recovers dense mesh motion from sparse vertex displacements. Residual priors trained on one geometry family frequently fail to beat parameter-free harmonic extension off-family. Moreover, cross-family training meets a scaling barrier: residual amplitude varies by mo
grasp - arxiv:2610.18869 · cs.ROWhat Matters in Quality-Aware VLA Pretraining: A Benchmark and Empirical StudyMingxuan Yan, Litian Gong, Dihong Huang, Zhixu Li +18
How should demonstration quality be used during vision-language-action (VLA) pretraining to improve performance after target-task fine-tuning? We compare weighting, filtering, and conditioning across reward definitions and label granularities using 5,000 naturally suboptimal human demonstrations and
vision-language-actionvlabenchmark - arxiv:2610.18867 · cs.ROEmbodiedEvo: Self-improving Coding Agents for Physical IntelligenceWenyi Li, Yizhe Chi, Youjie Zheng, Chunyu Liu +7
Bringing general-purpose agents into the physical world calls for an executable representation of physical intelligence. Compared with vision-language-action (VLA) models, which map observations and language instructions directly to actions, code makes control logic explicit and reusable. Code can s
vision-language-actionembodiedagentself-improving - arxiv:2610.18868 · cs.LGTev: Scaling Long-Horizon Agentic Environments with Dense Progress RewardsWenbo Hu, Shraman Pramanick, Krishna Kumar Singh, Nanyun Peng +2
Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI). Yet open-source models still struggle on long-horizon tasks, largely fo
agentagenticself-improvement - arxiv:2610.18866 · cs.ROAutoAgent0: An Agentic Runtime for Safe Closed-Loop DrivingJiageng Liu, Seth Z. Zhao, Zewei Zhou, Alejandro Huerta +8
Safe autonomous driving requires more than accurate planning at a single frame: it demands reliable closed-loop execution in dynamic environments. Although end-to-end driving policies have advanced rapidly, the accumulated closed-loop errors, especially in unseen scenarios, can push vehicles into st
agentic - arxiv:2610.18863 · cs.LGData Free Training of Diffusion Models for Lattice Field TheoriesDennis Grinwald, Ankur Singha, Khaled Kahouli, Karl Jansen +1
Sampling equilibrium field configurations remains a major computational bottleneck in lattice field theory, motivating the development of generative samplers that can produce independent configurations. Diffusion models are promising for this purpose, but their standard denoising score matching obje
benchmark - arxiv:2610.18862 · cs.LGDoubtLess: Training-Free Contrastive Decoding for Efficient ReasoningMinghao Luo, Liang Chen
Reasoning models achieve strong performance through long chains of thought (CoT), but often spend more tokens than needed to solve a problem. Existing ways to shorten the CoT require additional training for each model, add a fixed steering vector to the model's hidden states, or use decoding heurist
agentbenchmark - arxiv:2610.18861 · cs.LGHybrid Reasoning AgentsJiaru Zou, Yunzhe Qi, Sirui Chen, Federico Bianchi +4
How can we build an agent that thinks beyond human language, yet still acts through it in the external environment? Modern LLM agents still reason and interact through text, while latent reasoning offers an efficient internal medium but lacks native tool access and reasoning transparency. We bridge
memoryagentllm agentagenticpost-trainingbenchmark - arxiv:2610.18860 · cs.AIAesCode: Aesthetic Code Generation with Decoupled Cross-Modal RewardsXin Gao, Yifan Yang, Xin Li, Pan Wang +12
Generating information-rich visual artifacts such as slides, posters, and dashboards as code is appealing because the outputs are structured, editable, and verifiable. Yet code models cannot see how their choices of layout and color come together on the canvas, often limiting spatial planning and ae
benchmark - arxiv:2610.18858 · cs.CVVisual Tokens on Demand: Visual Budgeting for Scalable and Efficient Multi-Image Reasoning in Multimodal LLMsAimon Rahman, Joymallya Chakraborty, Vidit Bansal, Abhishek Dan
Multimodal large language models have advanced rapidly, but most remain optimized for single-image inputs. Multi-image reasoning is costly because each image adds many visual tokens, increasing encoding and attention demands. Existing techniques that address the above issue, including visual token p
retrieval-augmentedbenchmark - arxiv:2610.18856 · cs.LGExact Topological Compliance: Generating Persistence-Equivalent GraphsMattie Ji, Indradyumna Roy, Vikas Garg
Persistent homology summarizes the topological evolution of a graph as persistence diagrams (PDs). Generating graphs whose topology matches a target descriptor is increasingly important, yet existing topology-aware generative models rely on soft regularization, producing graphs whose PDs may only be
benchmark - arxiv:2610.18855 · cs.ROInteraction Latents as Actions for Contact-Rich ManipulationYuhang Zheng, Yujie Zang, Yuchen Liu, Yupeng Zheng +8
Contact-rich manipulation requires robots to coordinate not only where to move, but also how contact and force should evolve throughout an interaction. Yet most robot policies represent actions primarily as motion trajectories, using tactile signals as inputs or auxiliary prediction targets without
vision-language-actionmanipulationtactilepost-training - arxiv:2610.18852 · cs.ROForceWAM: Decoupling Future and Response Learning from Demo and Play Data for Contact-Rich ManipulationZijun Xu, Wendi Chen, Xinkai Wang, Jiachen Zou +15
Expert demonstrations show how to complete contact-rich tasks, but provide limited coverage of how interaction forces respond to different actions. Task-agnostic play provides broader experience of contact dynamics without specifying task goals. To fully leverage these two types of data, we introduc
manipulationworld model - arxiv:2610.18848 · cs.ROContact Trajectory Prompting: In-Context Transfer of Contact-Rich Behaviors from a Single DemonstrationXian Nie, Yujie Zang, Yuhang Zheng, Yupeng Zheng +7
In-context robot learning enables robots to acquire new behaviors from demonstrations without test-time parameter updates. However, visual and kinematic specifications alone may not distinguish contact-rich behaviors that share similar motions but require different interaction intensities or pattern
manipulation - arxiv:2610.18846 · cs.CVWorldRecon: Generative Multi-View Mesh Reconstruction for Large-Scale 3D WorldMengchen Zhang, Yang Li, Chunshi Wang, Tong Wu +1
Reconstructing complete, high-quality meshes of large-scale scenes from unposed multi-view images requires faithful geometric grounding while plausibly completing surfaces that are unobserved or only sparsely captured. We present WorldRecon, a generative multi-view reconstruction framework for recov
sim-to-real - arxiv:2610.18845 · cs.CVThe Last Remote Sensing Model Built by Humans? Casting Any Earth Observation Task as Recursive Self-ImprovementSijie Zhao, Zeqi Zhou, Heli Qi, Xuanchao Fu +5
Earth observation (EO) models are still built by hand. For each new task, experts design a training recipe and tune it against a benchmark. We show that this loop can be handed to an agent. Any verifiable task can be cast as a harness and a verifier, where the harness defines what the agent may use
agentself-improvementbenchmark - arxiv:2610.18844 · cs.CVRepProbe: Can Video Language Models Count Repeated Actions? A Diagnostic BenchmarkMahsa Khoshnoodi, Sarah Adel Bargal
Repetition counting requires identifying and enumerating action cycles over time. While state-of-the-art video language models describe actions fluently, they fail to reliably count how many times an action repeats. To investigate this limitation, we introduce RepProbe, a benchmark of 42K real and s
benchmark - arxiv:2610.18843 · cs.ROLIBERO-RIG: A Hardware-Setup Robustness Benchmark for Vision-Language-Action ModelsAndrea Yaoyun Cui, Hanyu Wang, Xinyi Wang, Bangzheng Li +5
Robot policies are expected to generalize across deployment environments, including variations in the robot itself: cameras are bumped and remounted, calibrations drift, controllers run at different rates, and end-effector pads wear down. We present LIBERO-RIG, a benchmark that perturbs only the rob
vision-language-actionliberogripperbenchmark - arxiv:2610.18842 · cs.ROWho Improves the Robot? Toward Physical Recursive Self-ImprovementHai Zhou, Fan Yang, Zeyi Chen, Lejun Zhang +26
Robots are increasingly autonomous in task execution, yet their improvement remains largely human-driven. This study investigates Physical Recursive Self-Improvement (Physical RSI), in which physical experience guides lasting updates to both a robot's capabilities and its improvement procedure, maki
self-improvementbenchmark - arxiv:2610.18836 · cs.CVPlan-Bench: Can Existing VLMs Go Well Beyond Perception to Reason and Plan in Embodied Worlds?Juntong Feng, Qingan Liu, Jingru Dong, Yue Zhu +7
Vision-Language Models (VLMs) have achieved strong performance in visual perception and spatial reasoning, leading to growing interest in using them as high-level planners for embodied AI. However, physical tasks require models to integrate spatial information, reason over temporal states, and gener
embodiedbenchmark - arxiv:2610.18835 · cs.RORoboSimStudio: An Interactive Toolkit for Rapid Real2Sim of Robot Manipulation TasksKaiyuan Chen, Shuangyu Xie, Uksang Yoo, Xinyue Zhu +8
It is difficult to accurately model the physics of the interaction between a robot and the objects it manipulates, including contact mechanics, coupled contact dynamics, deformations, stiffness and friction. We present RoboSimStudio (RSS), an interactive toolkit that makes choices of solver, contact
manipulationsim2realagentic - arxiv:2610.18834 · cs.CVGame2Real-Bench: Do Toy Games Reveal and Transfer to Real-World Visual Reasoning?Qingan Liu, Juntong Feng, Yue Zhu, Yujie Cheng +5
Visual reasoning is an important capability for vision-language models (VLMs) to understand visual evidence, spatial structure, constraints, and state changes. Existing evaluations mainly rely on real-world data or procedurally generated tasks. Real-world data preserve natural complexity, while proc
benchmark - arxiv:2610.18829 · cs.CVOmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and UnderstandingXuyang Liu, Shihao Wang, Xinyu Wei, Minghan Li +4
With the rapid development of multimodal large language models (MLLMs), evaluating their video understanding capabilities has become increasingly important. Existing video understanding benchmarks cover only part of the evaluation dimensions, limiting systematic diagnosis of omni-modal video underst
benchmark - arxiv:2610.18826 · cs.ROWorldOracle: Benchmarking Action-Conditioned World Models against Real-World ExecutionsHaoyi Duan, Liliang Chen, Nan Wang, Yang Zhang +12
We introduce WorldOracle, a benchmark specifically designed to evaluate action-conditioned world models (ACWMs) against paired real-world executions. Given the same initial observation and low-level action sequence, WorldOracle asks whether a predicted rollout reproduces what happens in the real wor
manipulationworld modelaction-conditionedbenchmarkleaderboard - arxiv:2610.18822 · cs.ROEgoMatrix: Physically Faithful Dexterous Robot Demonstrations from Egocentric Human VideosZiheng Wang, Yanming Shao, Ting Mao, Kaihan Chen +3
Egocentric human videos offer a scalable alternative to robot teleoperation for collecting dexterous manipulation data. However, two challenges limit their use as reliable training data: accurate hand-object interaction (HOI) reconstruction from monocular RGB and high-quality robot demonstrations, e
manipulationdexterousteleoperationbenchmark - arxiv:2610.18819 · cs.CVDuplexBridge: Coordinating Understanding and Generation for Unified Image EditingYidi Liu, Jie Xiao, Chaojie Mao, Jingfeng Zhang +1
Complex image editing requires faithful task interpretation and accurate execution, but a failed edit does not reveal which responsibility failed. We present DuplexBridge, which uses a unified model's own understanding capability to enhance prompts for its own generation capability, without an exter
benchmark - arxiv:2610.18813 · cs.AISelf-Evolving Agent HarnessChuyang Wei, Haoxiang Guan, Minghang Zhu, Kexin Zhou +4
Harness self-evolution revises an agent's software scaffolding, including prompts, memory, workflows, and executable components, while keeping model weights fixed. We formalize a feedback-driven process in which accepted revisions persist across runs. The framework separates update targets from prop
agentself-evolvingevaluator - arxiv:2610.18809 · cs.LGDepth MemoryDavid González-Martínez, Philipp Nazari, T. Konstantin Rusch, Shiwei Liu
Residual connections propagate information through a network by additively accumulating layer outputs. Recently proposed alternatives, such as Manifold-Constrained Hyper-Connections (mHC) and Attention Residuals (AttnRes), instead use more expressive depth aggregation mechanisms that improve languag
memory - arxiv:2610.18808 · cs.ROOptimizing Team Behavior via Extremum-seeking Control in Multi-agent Reinforcement LearningSvar Rajankumar Patel, Kshitij Jerath
In cooperative Multi-Agent Reinforcement Learning (MARL), agents must learn to coordinate their actions to achieve shared objectives. A key indicator of effective coordination is behavioral diversity, i.e., the agents' potential to enact distinct yet complementary control policies given the same set
multi-agent - arxiv:2610.18807 · cs.LGSkillScaffold: Scaffolding Agentic Reinforcement Learning with Policy-Adaptive Skill Credit Assignment and Selective InjectionWenlin Zhong, Hao Wang, Kun Yuan, Fan Yang +5
Agentic reinforcement learning (RL) enables large language models (LLMs) to learn from multi-turn interactions with verifiable outcomes. Textual skills can improve exploration by distilling prior interactions into reusable guidance. However, existing skill-based RL approaches often either assume ski
agentic - arxiv:2610.18804 · cs.LGRegistration is a Representation Problem: Solving Whole-Body PET/CT RegistrationNeel Dey, Adrian V. Dalca
Raw image intensities are a poor surrogate for similarity across images in correspondence problems. Further, while auxiliary segmentation labels can provide semantic identification and improve matching, they can assign only coarse identities to fine-grained anatomy and often cover only a subset of s
leaderboard - arxiv:2610.18792 · cs.ROSEAR: Self-Evolving Agentic RoboticsXinyi Wang, Bangzheng Li, Hanyu Wang, Lin Zhang +4
Frontier language-model agents can now improve robots on their own: they write control code, test it, and keep what works as skills and memories for the next task. We introduce SEAR (Self-Evolving Agentic Robotics), a framework and benchmark built around two questions. Is the agent really learning,
vision-language-actionagentagentictool usecode-as-policyself-evolving - arxiv:2610.18790 · cs.ROEqC: Equilibrium Energy Control for Introspective Lifelong Robot LearningHaodong Yang, Runhan Huang, Bowen Zhang, Iofai Lei +6
Long-term robot autonomy depends on more than effective action generation: a policy must recognize when its behavior is no longer sufficiently supported and use assistance to improve. We introduce Equilibrium Energy Control (EqC), a framework for introspective lifelong robot learning built around an
vision-language-actionmanipulationself-improving - arxiv:2610.18789 · cs.CVTrust Before Autonomy: Rethinking Agentic Remote Sensing for Crisis ResponseGeorgios Angelidis, Aggelos Psiris, Spyridon Evangelatos, Thomas Lagkas +3
Agentic remote sensing is evaluated on how many questions it can answer, whereas crisis response depends on whether its answers can be relied on and how much of the decision can safely be handed to the agent. This paper argues that \emph{autonomy should be earned, not assumed}: An Earth-Observation
agentagenticbenchmark - arxiv:2610.18780 · cs.AIBLADE: Auditing Information Use for Failure Attribution in LLM Multi-Agent SystemsXiaoguang Guo, Yaohao Long, Chuxu Zhang
Large language model multi-agent systems extend individual agents' capabilities through collaboration across specialized roles, tools, and workflows. Yet collaboration makes failures harder to diagnose: one agent may misinterpret a correct result while others carry the mistake forward. Failure attri
agentmulti-agentagenticagent system - arxiv:2610.18779 · cs.CVGAD: Geometry Attention Distillation for Consistent Video GenerationJunfeng Ni, Ziyu Zhu, Fan Jiang, Yu Liu +6
Video diffusion models generate visually compelling short-term sequences but often struggle to preserve scene geometry and object structure over time. In this paper, we first demonstrate that current video attention cannot identify reliable correspondences across frames, with responses dispersed acr
benchmark - arxiv:2610.18774 · cs.ROGreenWAM: A World Foundation Model for RoboticsGennady Sidorov, Valerii Kornilov, Denis Grankin, Polina Fedotova +1
A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions. Models are usually built and evaluated for only one of these roles, and every robot records its actions in its own format. We ask whether one robot world model can serve both role
vision-language-actiondexteroushumanoidliberorobotwinworld model - arxiv:2610.18771 · cs.CVThe Federated Tumor Segmentation and Classification (FeTS++) ChallengeWenyi Tang, Carla Pitarch-Abaigar, Siddhesh Thakur, Verena Chung +4
International challenges have become the standard for developing and benchmarking biomedical image analysis methods using centralized multi-institutional data. Federated learning (FL) describes a paradigm-shift on multi-institutional collaborations, where models are trained across institutions while
benchmark - arxiv:2610.18763 · cs.CVLayerCanvas: Infographic Generation in AI Learning Assistants via Hybrid Layer-Set DiffusionKeshav Gupta, Jeremy Owen, Christopher Lloyd, Shaunak Mishra +1
Online learning platforms are increasingly deploying AI assistants to provide personalized, on-demand tutoring. These can leverage infographics to communicate complex information and improve the learning experience. However, automating effective infographic generation remains a challenging task, req
online learning - arxiv:2610.18760 · cs.AIBeyond the Verdict: Agentic Code Judging with Verifiable FeedbackZheye Deng, Sha Li, Changlong Yu, Xin Liu +3
Test-time scaling for large language models (LLMs) can involve generating multiple candidate solutions for selection or revising an earlier attempt. To guide selection and revision, we need a verifier that can estimate candidate correctness and provide evidence of failure. In competitive programming
agentic - arxiv:2610.18755 · cs.AIPreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test RewardsYunkai Zhan, Miaosen Chai, Shangshang Wang, Jike Zhong +4
Reinforcement learning (RL) from unit tests is the common recipe for post-training code models, but a unit test only asks whether the final program passes. We analyze what this unit-test-only reward teaches a model during debugging and find that, while it raises pass rates, it also teaches the model
post-training - arxiv:2610.18745 · cs.AISolBehave: Evaluating the Behavioural Correctness of LLM-Generated Smart ContractsGiovanni Pio Delvecchio, Stefano Fantazzini, Giacomo Frisoni, Gianluca Moro +3
Large language models increasingly generate Solidity smart contracts that, once deployed, can never be patched, making a single latent fault permanent and exploitable. Yet current evaluations mostly ask only whether generated code compiles. We introduce SolBehave, a benchmark for evaluating the corr
benchmark - arxiv:2610.18738 · cs.LGEnabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL RegularizationHao Bai, Jiarui Yao, Tong Zhang
Off-policy data in reinforcement learning with verifiable rewards (RLVR) can arise from asynchronous generation, rollout-training mismatch, and response reuse. Token-wise importance weighting avoids multiplying likelihood ratios across a response but generally biases policy gradients. We prove that,
benchmark - arxiv:2610.18730 · cs.LGLearning When to Transfer: Reinforcement Learning for Carbon-Footprint Modeling in New Manufacturing DomainsMuhammad Umar Farooq, Daniel R. Cooper
Manufacturers often need to characterize a new material, process, or product before enough operating data are available to train reliable predictive models for this new, target domain. Existing manufacturing operations contain potentially useful knowledge to accelerate this characterization, but ind
embodied - arxiv:2610.18711 · cs.ROSteertuning Policies from a Few Human VideosBilly Yan, Isaac Tu, Kelly Lee, Blake Chang +7
Current recipes for robot learning require massive amounts of robot demonstrations to obtain robust and adaptive policies. In this work, we introduce Steertuning, a system that enables robots to learn new tasks in new environments from 5 human demonstrations in under 5 Steertuning consists of two pa
diffusion policyrobot policy - arxiv:2610.18704 · cs.AIUnlocking Token Rewards via Training-Free Reward AttributionSitong Wu, Haoru Tan, Bin Xia, Xichen Zhang +5
In this paper, we propose an extremely efficient, training-free method to extract token-level reward signals directly from an existing deep reward model. Our core idea is to attribute the overall process reward to individual tokens by estimating each token's influence. This influence is defined as t
benchmark - arxiv:2610.18699 · cs.RORoboFQE: Rollout-Free Evaluation of Visual Robot Policies from Offline TrajectoriesYunfu Deng, Sangwoo Shin, Brahma S. Pavse, Josiah P. Hanna
Evaluating robot policies before deployment typically requires producing new trajectories through real-world execution, simulators, or learned world models. These rollout-based approaches impose infrastructure requirements; simulator and world-model evaluators additionally face prediction errors tha
manipulationworld modelbenchmarkevaluatorpolicy evaluation - arxiv:2610.18697 · cs.ROUniW(A)M: A Unified Mixture-of-Experts Framework for World and World-Action ModelsJingbo Xu, Jianjie Fang, Zhuohang Li, Zihao Song +8
World models (WMs) and world-action models (WAMs) focus on future world prediction and executable action generation, respectively. Although these models learn shared physical dynamics and interaction semantics from embodied trajectories, they are typically trained in isolation, which restricts knowl
embodiedmanipulationliberorobotwinworld modelaction-conditioned - arxiv:2610.18694 · cs.LGGeoReader: Learning Latent Geometry for Spatial Reasoning in Vision-Language ModelsRong Li, Shijie Li, Jiangyong Huang, Lingdong Kong +6
Vision-language models (VLMs) struggle with metric and multi-view relationships that remain implicit in RGB observations. Geometry foundation models recover this structure from RGB, but existing methods either fuse the estimated features with visual tokens and learn to use them from question answeri
benchmark - arxiv:2610.18690 · cs.LGJust Respect the Semantics: Image-Token Compression via Pre-LLM Semantic Variable-Resolution Tokenization and In-LLM Relevance PruningTianhao Fu, Austin Wang, Charles Chen, Roby Aldave-Garza +4
Image-token compression must reduce computation without hurting the semantics. We introduce Varest, guided by semantic fungibility: tokens within a coherent semantic region can be summarized coarsely, while object boundaries and literal text require finer distinctions. Object and text localizers gui
benchmark - arxiv:2610.18688 · cs.CVRSIgen: Reciprocal Self-Improvement for Agentic Visual GenerationYidi Liu, Haozhe Wang, Jie Xiao, Kai Zhu +2
Agentic visual generation produces experience that can improve the native understanding (agent) and generation roles of a unified multimodal backbone, but task success alone does not determine which role should learn. Generator updates can also make the agent's action supervision stale. We introduce
agentagenticself-improvement - arxiv:2610.18686 · cs.RORadarGSOcc: Doppler-Aware 3D Semantic Occupancy Prediction Using 4D Radar Tensor with Geometric-Semantic Feature ExtractionChristofel Rio Goenawan, Dong-Hee Paek, Seung-Hyun Kong
3D semantic occupancy prediction is crucial for autonomous driving, but camera-based methods degrade in low-light conditions and LiDAR-based methods are sensitive to adverse weather. 4D radar is more robust, yet its sparsity and the high cost of radar tensor processing limit its use. We propose Rada
benchmark - arxiv:2610.18684 · cs.LGKnowing the User Is Not Knowing What We Established: Grounding Awareness in Personal AgentsJiajun Chen, Xinyue Chen, Yicheng Jiang, Hua Shen
Personal-agent memory has largely focused on learning what users know, prefer, and want. However, knowing the user does not necessarily mean knowing what the user and the agent have established together. An assistant's proposal should not become an authorized instruction merely because it is remembe
memorypersistent memoryagent memoryagentbenchmark - arxiv:2610.18683 · cs.LGProAS: A Progress-Aware Advantage Shaping Framework for Multi-Turn Agentic Reinforcement LearningBowen Ding, Mingda Wan, Yuhan Chen, Jiyao Yuan +6
Reinforcement learning (RL) has emerged as the primary paradigm for optimizing Large Language Model (LLM) agents, yet efficient critic-free algorithms (e.g., GRPO) suffer from a fundamental uniform credit assignment dilemma that broadcasts trajectory-level advantages equally to all actions. Our anal
agentic - arxiv:2610.18677 · cs.ROModel-Based Online Decision Making via Generative Trajectory PlanningHaldun Balim, Yilun Du, Na Li
Online decision-making faces a tension between efficient inference and long-horizon reasoning. Value-based methods rely on short-horizon bootstrapping that limits the propagation of long-term information, while trajectory optimization methods plan explicitly but must solve a high-dimensional optimiz
manipulation - arxiv:2610.18675 · eess.SYOptimal Decay Matrix Control Barrier Functions: Soundness and Completeness for Matrix Safety ConstraintsSamuel G. Gessow, Albert Wu, Aaron D. Ames, Brett T. Lopez
Many safety requirements, such as keeping a manipulator away from kinematic singularities, a network connected, or a system observable, amount to keeping a state-dependent matrix positive definite. Matrix control barrier functions (MCBFs) enforce such requirements with a semidefinite program that mi
manipulator - arxiv:2610.18669 · cs.ROKnown-Voice Cloning and the Limits of Personality Transfer for Character DesignJohanna Magdalena Kuch, Maud van Lier, Daksitha Senel Withanage Don, Patrick Gebhard +2
Recent advances in zero-shot voice cloning systems make it possible to reproduce a recognizable voice from a short recording and attach it to arbitrary virtual characters. Prior work suggests that familiar vocal identity can carry over associated personality impressions to new targets such as robots
embodied - arxiv:2610.18664 · cs.CVPlanPath: Exploration-Guided Orchestration for Patient-Level Pathology ReasoningJiusong Ge, Mingzhao He, Qidong Liu, Chunbao Wang +7
A patient's pathology case spans tens of slides from distinct specimens and anatomical sites, each carrying non-redundant, site-specific evidence. Forming a patient-level conclusion therefore requires locating the slides relevant to a given question and integrating complementary findings across them
agentic - arxiv:2610.18654 · cs.LGKwai Summary Attention V2 Technical ReportOneRec Team, Chenglong Chu, Guohong Mu, Guowang Zhang +29
Summary-token architectures such as Kwai Summary Attention (KSA) compress distant context into learnable Summary token per chunk, and hybrid variants interleave these layers with full attention to retain exact access to the original text. In such hybrid KSA-Full architecture, the interleaving insert
memory - arxiv:2610.18652 · cs.LGAMemOps: Evaluating Memory Updates in Long Audio StreamsPeiwei Chang, Li Zhang, Mark D. Plumbley, Philip J. B. Jackson +1
Audio memory must recognise recurring sources across acoustic changes while keeping their attributes up to date. Overall accuracy does not reveal how these demands interact. We introduce Audio Memory Operations (AMemOps), a benchmark for latest-attribute retrieval, preservation of unchanged attribut
memorybenchmark - arxiv:2610.18650 · cs.ROForkCast: Predictive Recovery for Vision-Language-Action Policies via Counterfactual FuturesChenru Wen, Huajian Zeng, Haokai Ding, Jian Kang +4
Vision-language-action (VLA) policies perform manipulation tasks using camera images and language instructions, but object slip or visual occlusion can disrupt execution. Recovery is typically learned during policy training or delegated to a monitor that triggers a scripted repair after detecting a
vision-language-actionvlavla policymanipulationrobotwinworld model - arxiv:2610.18645 · cs.CVST-GRU PredRNN-V2: A GRU-Based Reformulation of PredRNN-V2 for Video PredictionAugusta van Haren, Hanna Hoogen, Luca Pattavina, Quincy Rommen
Video prediction requires models to capture complex spatiotemporal dependencies and to retain relevant information over long sequences. PredRNN and its successor PredRNN-V2 address this challenge with spatiotemporal Long Short-Term Memory (ST-LSTM) units that maintain two separate memory states, at
memorybenchmark - arxiv:2610.18642 · cs.LGMultiAlloc: Budget-Aware Precision Allocation for KV Cache QuantizationYangchengyu Zhou, Zhiping Wu, Runchang Jiang, Dongdong Ren +4
The Key-Value (KV) cache constitutes a dominant source of memory consumption during large language model (LLM) inference, particularly in long-context and complex reasoning scenarios. Existing KV cache quantization methods have explored different compression granularities, including layer-wise preci
memorylong-contextbenchmark - arxiv:2610.18637 · cs.ROGrid or Nodes? Probing Object Memory in World Models Under Partial ObservabilityJustin McMillen, Yasin Yilmaz
A robot's world model must remember objects that leave the camera's view. FloWM, a learned world model, keeps them in a spatial grid of features that it moves with five built-in velocities. A node memory instead keeps each object's position and velocity as numbers. We compare them in FloWM's own blo
world modelmemory - arxiv:2610.18624 · cs.LGMajorTOM:Elliot Pretrain, A Multi-Sensor, Multi-Temporal Pretraining Dataset for Earth ObservationCarmen Oliver Huidobro, Mikolaj Czerkawski, Kevin Monsalvez-Pozo, Cesar Aybar +2
Pretraining generalist Earth observation foundation models requires data that combine several sensors, varied temporal sampling and, increasingly, text, yet open datasets rarely provide all three in one consistent, easily accessible form. We introduce MajorTOM:Elliot Pretrain, a collection of 279,16
benchmark - arxiv:2610.18617 · cs.AITwo Routes to Interoceptive Prioritisation: Dissociating Likelihood Precision and Preference Reweighting in Active InferenceSt John Grimbly, Nicolas Kuske, Evert A. Boonstra, Charel van Hoof +5
An embodied agent must prioritise among competing bodily needs. In active inference, it can respond to an urgent need by increasing the precision of information about that need or by increasing its preference for outcomes that relieve it. We compare these two interventions, separately and together,
embodiedagentembodied agent - arxiv:2610.18616 · cs.AIKnowledge Graphs as Files: An Agent Harness Where Tool Creation and Tool Use Co-EvolveYuyang Bai, Yu Wang, Yu Zhang
Searching text-rich knowledge graphs requires following relations while examining textual evidence. Predefined graph interfaces constrain what evidence agents can access, even when they write code to combine tool calls or process retrieved subgraphs. We propose direct knowledge graph interaction (DK
knowledge graphagenttool usebenchmark - arxiv:2610.18615 · cs.CVVideoJev: System One Model for Video Understanding and InteractionYuhan Zhu, Xiangyu Zeng, Qingyi Si, Chenxu Yang +14
Visual agents must turn video observations into timely decisions to track progress and choose what to do next. Building on the emerging System One paradigm, we introduce VideoJev, a family of models that cast video understanding and interaction as structured decision tasks. We adapt pretrained multi
benchmark - arxiv:2610.18614 · cs.LGCan an Agent Learn to Predict Its Own Failure? Outcome-Grounded Confidence Calibration in LLM AgentsAshutosh Hathidara
Controlling an acting LLM agent means deciding, at each step of a long-horizon task, whether to trust its next action, abstain, or hand off to a human. That decision needs a per-step confidence signal predicting whether the current rollout will eventually succeed. The agent's own self-reported verba
agentllm agentevaluation protocol - arxiv:2610.18613 · cs.LGDoes self-distillation improve reasoning?Jason Ken Adhinarta, Raina Wu, Ashutosh Sharma, Avinash Balakrishnan +4
Self-distillation is an emerging approach that trains a model to imitate its own predictions when conditioned on privileged information, such as correct answers or environmental feedback. Recent methods have reported better sample efficiency and generalization than reinforcement learning (RL), but i
tool use - arxiv:2610.18611 · cs.CLLegal-R1: Agentic Retrieval for Evidence-Based Legal ReasoningHuaiwen Pang, Jiale Han, Junying Chen, Wenjun Xiao +1
Legal question answering requires answers that are not only plausible, but also grounded in legally valid and citable evidence. Standard retrieval-augmented generation often relies on one-shot semantic matching, which is insufficient when legal evidence must be selected according to authority, appli
retrieval-augmentedagentmulti-agentagenticbenchmark - arxiv:2610.18610 · cs.AIJev-ToM: Evaluating Theory of Mind in Typed Decision ModelsKawai Chung, Chunkit Chan, Yuxuan Liu, Yauwai Yim +8
Dialogue and collaborative assistants repeatedly judge what a partner knows, believes, or intends. Typed decision models offer a potential alternative to generative large language models for bounded Theory of Mind (ToM) selection. We present Jev-ToM, comparing five typed decision models and seven ge
benchmark - arxiv:2610.18606 · cs.AIBeyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled TracesHaotian Liu, Yingjian Chen, Xiaohang Xu, Lang Huang +3
Large language models increasingly tackle tasks that require multi-step reasoning, where errors in intermediate claims can affect the final answer. Self-consistency (SC) improves reasoning by voting over sampled final answers, but its vote leaves intermediate disagreements unresolved. To address thi
benchmark - arxiv:2610.18603 · cs.AIAgentStripes: An Evidence-Graded Maturity Model for AI AgentsNutan Kumar Naik, Vijay Prasad Poudel, Aditya Kumar Saroj, Saurav Samantray +1
Benchmarks report what an Artificial Intelligence (AI) agent can do on a task suite, but not whether one released version has been engineered with the controls that make it reliable, safe and observable, or how strongly that is shown. This paper proposes AgentStripes, an evidence-graded maturity mod
agentai agentbenchmark - arxiv:2610.18602 · cs.AIAuditing Future-Memory Use in Point-in-Time Language AgentsRongtian Ye
Point-in-time (PIT) agents can recall post-cutoff facts even when their external evidence respects a historical cutoff. We examine whether this access predicts evidence gathering in 48 historical company-comparison tasks. Independent closed-book revenue probes identify the better-remembered company
memoryagent - arxiv:2610.18598 · cs.CVRethinking Projector Training in Multimodal LLMsFabian Gröger, Naël Ouerghemi, Shuo Wen, Maria Brbić
Multimodal LLMs are typically trained by first aligning a vision encoder with a language model via projector training on large image-caption datasets, followed by joint training of the full multimodal model. Despite its widespread use, the role of the projector training stage remains poorly understo
benchmark - arxiv:2610.18596 · cs.ROACCRUE: Continual Off-Policy Reinforcement Learning for Robot Foundation ModelsChangyeon Kim, Kimin Lee, Yashraj Narang, Yijie Guo
Reinforcement learning (RL) post-training improves robot foundation models (RFMs) beyond their demonstrations, but real-world deployment demands more: learning new tasks efficiently while retaining previously acquired behaviors. Existing continual RL methods on top of VLAs mitigate forgetting but re
robot foundation modelpost-training - arxiv:2610.18587 · cs.LGGeoMER: A Joint Euclidean-Hyperbolic Geometry Framework for Multimodal Emotion RecognitionTing Zhu, Huizhi Liang, Boguslaw Obara, Shufei Duan +1
Multimodal Emotion Recognition (MER) aims to understand human emotions by integrating information from multiple modalities such as speech and text. Psychological studies suggest that emotions exhibit latent hierarchical structured relationships in their semantic representations. However, most MER ap
benchmark - arxiv:2610.18583 · cs.AICompare to Correct: Pairwise-Guided Logit Adjustment for LLM Test-Time ScalingCong Hua, Qianqian Xu, Zitai Wang, Zhiyong Yang +3
Sampling-based test-time scaling enhances LLM reasoning by aggregating multiple reasoning paths for the same query. Existing methods use answer frequencies or path confidence to compute candidate scores, which do not explicitly characterize comparisons between competing answers. Inspired by the Cond
benchmark - arxiv:2610.18579 · cs.AIReading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent LoopsYubo Li, Ramayya Krishnan, Rema Padman
System One models choose an agent's next action by scoring a supplied set of options in a single pass. This makes them fast and cheap, but when they choose wrongly there is nothing to inspect: did the model misread what the agent has seen, or read it correctly and still act wrongly? We answer behavi
agenttool usebenchmark - arxiv:2610.18575 · cs.CVDetecting Unreliable Classifications from Quantized ModelsAnanya Kotla, Jorge T. Gomez
Post-training INT8 quantization is typically validated by checking accuracy retention, which requires ground-truth labels often unavailable in low-resource clinical settings. We show a label-free alternative: comparing a model's own output before and after quantization. We compute cosine similarity
post-training - arxiv:2610.18572 · cs.LGLLM Agents for Automated Post-Training via Structured Recipe OptimizationRuohong Liu, Bohan Tang, Lekang Jiang, Jack Umenberger +1
Automated post-training transforms pre-trained large language models (LLMs) into models capable of handling diverse domains by automatically constructing and optimizing a recipe of coupled decisions over data sources, data processing, cross-domain composition, and training configuration under a limi
llm agentpost-trainingbenchmark - arxiv:2610.18570 · cs.AIOrder-Consistent Latent Geometry for Reliable Short-Horizon Planning in World ModelsJunlin Wu, Wen Fan, Dandan Zhang
Latent planners often rank candidate action sequences by terminal Euclidean distance to a goal latent. Exact agreement with transition distance would make this score ideal, but is stronger than planning requires: along a transition-shortest rollout, the essential requirement is that latent distance
world model - arxiv:2610.18569 · cs.CLCREATE: Creativity Evaluation of VLMs using the Visual Remote Associates TestZiqing Yang, Anubhav Jangra, Smaranda Muresan
Vision language models (VLMs) are used in creative tasks such as visual storytelling, design ideation, art interpretation, and advertising. Yet we lack principled ways to measure the lateral reasoning of these models compared to human potential. We introduce CREATE, a semi-automatically constructed
benchmark - arxiv:2610.18568 · cs.ROWAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future PredictionHaomin Wang, Anqiong Ji, Ke Wang, Zhongshan Liu +3
Vision-language-action (VLA) policies excel at generalist robotic manipulation but generate actions without explicitly modeling their environmental consequences. World action models (WAMs) incorporate future prediction, yet most rely on costly video generation, while latent alternatives either use p
vision-language-actionvlamanipulationliberorobotwinworld model - arxiv:2610.18567 · cs.AIConfiguring RAG Systems: What Breaks, Where Originates, and How to FixLiangyu Chen, Youpeng Ma, Yulong Ye, Zezhen Xiang +1
Retrieval-Augmented Generation (RAG) has emerged as a dominant paradigm for improving Large Language Model (LLM) by intertwining it with other components, such as a vector/graph database. Therefore, the resulting RAG systems are often highly configurable, under which a non-trivial amount of failures
retrieval-augmentedrag - arxiv:2610.18566 · cs.AIRethinking Quality in Large Language Model Routing: Security-Conditioned Model-and-Budget SelectionYichi Zhang, Zhenxiang Huang, Yuyang Zhang, Yuchen Yang
Large language model (LLM) routers reduce serving cost by choosing which model to call and how many output tokens to grant it for each query. Existing routers rank model-budget actions using a single quality metric. This does not work for mixed workloads. For ordinary queries, a good answer is corre
benchmark - arxiv:2610.18564 · cs.CVObservation Quality in DSM-Assisted Aerial Localization: A Benchmark and Base-Anchored Uncertainty CalibrationPengyu Yin, Junxiang Tan, Jie Pan, Mengqi Fu +2
Aerial map matching provides position observations for navigation, but a position error alone does not determine how an observation should be weighted. We define five benchmark tasks: position availability and agreement, grouped error structure, predictive uncertainty, controlled updates, and tempor
benchmark - arxiv:2610.18545 · physics.app-phA Transferable Machine-Learned Potential Enabling Large-Scale Copper-Carbon Simulations from Equilibrium to Extreme ConditionsFeiyu Zhou, Wei Qiu, Yan Chen, Jiahui Liu +3
Cu-C materials and interfaces are central to diamond thermal management, metallization, graphene growth and carbon-coating technologies, but atomistic simulations across these applications require a potential that describes metallic Cu, diverse carbon bonding, heterogeneous Cu-C interactions and clo
benchmark - arxiv:2610.18541 · cs.CLEmpathy Is Not the Scapegoat for Declines in Factuality and ReliabilityLetian Shi, Kanwal Mehreen, Deheng Zhang, Yuxia Wang
Language models are increasingly expected to support users in stressful or uncertain situations, where warmth and empathy are desirable. While recent work associates warmer models with reduced reliability, we show that this conclusion is incomplete. Prior studies often train primarily on empathetic
benchmarkevaluator - arxiv:2610.18538 · cs.AIOn Symmetry for Characterizing Agent Heterogeneity in Multi-Agent SystemsJianhong Wang, Jonathan Lawry
Understanding agent heterogeneity can inform the design of shared and agent-specific policy representations, however existing criteria provide no unified structural test. We study finite-horizon POSGs and define a type-compatible transformation as a policy-evaluation symmetry when complete policy ev
agentmulti-agentagent systempolicy evaluation - arxiv:2610.18535 · cs.CVMeta-Video: Efficient Standardized Evaluation for Video UnderstandingSoumya Shamarao Jahagirdar, Anna Kukleva, Vishaal Udandarao, Samuel Thomas +4
Video understanding benchmarks are the primary yardstick for measuring progress in video-language modeling, and a growing body of work relies on them to claim improvements. The reported benchmark scores in these works, however, are almost always produced by frameworks (e.g., lmms eval) used as black
benchmarkevaluation frameworkevaluation protocol - arxiv:2610.18522 · cs.AILost in Implicit Routing: How LLM Agents Fail to Load Relevant Skills at ScaleJinlin Liu, Wei Chen, Xiaojin Zhang
Reusable skills let LLM agents load task-specific instructions on demand, but they make execution depend on implicit routing over skill metadata. We test whether this design scales by holding SkillsBench tasks and their golden skills fixed while adding irrelevant skills. We find skill routing collap
llm agent - arxiv:2610.18516 · cs.LGTwo For The Price Of One: Curvature-Aware Pruning Preserves Influence Functions TooMehdi Makni, Natalia Ponomareva, Rahul Mazumder
Post-training pruning enables efficient deployment, but conventional methods do not explicitly preserve internal-gradient alignment between sparse and dense models. We study novel connections between downstream performance of compressed models and the ability to retain gradient information via influ
post-training - arxiv:2610.18515 · cs.AIA Global View of How LLMs Value Human TimeYifan He, Sofie Goethals, Manon Reusens, David Martens
Large language models (LLMs) increasingly act as personal assistants; however, little is known about how they capture national-level preferences on a global scale. Given that "time" is a universal and representative concept, we examine LLMs' time preferences by varying only users' country and measur
agentic - arxiv:2610.18508 · cs.LGTowards Safe Autonomy for Computer UseDominik Seip, Bartol Markovinović, Matthias Hein
The safety of LLM-based agents currently rests on alignment training and safety classifiers, neither of which can rule out that a sophisticated indirect prompt injection (IPI) goes through and causes significant harm. Recent work therefore builds agents with information-flow guarantees that rely on
agent - arxiv:2610.18507 · cs.CLThe Last Mile of Coding Agents for Long-Document Tasks: An Empirical Study on Clinical Trial Protocol DigitizationJiacheng Lin, Geoffrey Low, Nishit Agarwal, Rahul Jain +3
Clinical trial protocols serve as the source of truth for study design, but are mostly written and used as unstructured PDF documents. Digitizing protocols turns static content into structured, machine-readable data that clinical software can analyze, validate, and execute. Coding agents, which sele
agentbenchmark - arxiv:2610.18504 · cs.LGHow Low Can You Go: Fine-Tuning LLMs with Bayesian Optimization (LlMBO)Aditya Kulkarni, Davide Sartor, Carlos Soto, Nathan Wycoff
Recently, Morris et al. introduced TinyLoRA, a strategy for LLM fine-tuning which introduces arbitrarily few parameters, and demonstrated accuracy improvements using less than $10$ parameters rather than the ${\gtrsim 10,000}$ typically demanded by LoRA-based methods. However, that study continued t
benchmark - arxiv:2610.18499 · cs.CVEgoOmniWorld: Unifying Egocentric Understanding via Ego-Action-World InteractionsSongruo Chen, Yuqian Fu, Yu Li, Mingkang Dong +6
Egocentric video understanding spans diverse capabilities, modalities, and evaluation formats, yet these are often studied in isolation despite their tight coupling in real-world interaction. We propose the Ego-Action-World interaction loop as a unifying principle: the wearer perceives the world, fo
benchmark - arxiv:2610.18498 · cs.AIAligning Faithfulness for Theory-Scale AutoformalizationYuming Feng, Tadd Mao, Dhruva Arekar, Tianjun Zhong +3
Autoformalization (AF) requires more than compilation: formal statements must preserve the meaning of their informal sources, a property known as faithfulness. An emerging frontier is the automated evaluation of this property, yet most existing methods operate at the statement level and produce whol
benchmark - arxiv:2610.18497 · cs.AIJudges Follow the Reference: The Derivation Gap in Verifying Tool-Call ArgumentsLinhai Ma, Yan Wang, Lingfei Qian, Yupeng Cao +1
Tool-call judges evaluate argument values proposed by LLM agents. We find that they accept a correct value less often when their input holds no reference for it, that is, no text that says what the value should be. This happens when the value is a default value from the function schema that the user
llm agentbenchmark - arxiv:2610.18485 · cs.CVPhysics-Aware Deepfake Detection via Distance-Speech ConsistencyKyeongrae Kim, Kim Sung-Bin, Oh Hyun-Bin, Tae-Hyun Oh
Recent progress in generative models has made talking-head deepfakes highly realistic and easy to create, jeopardizing public trust. However, most audio-visual deepfake detectors rely on lip-speech synchronization and are designed for static, frontal videos, limiting their reliability in dynamic, in
benchmark - arxiv:2610.18484 · cs.LGReTerm: Recovery Training for Self-Improving Terminal AgentsWei Xie, Kaituo Feng, Shiming Fu, Yong Liu +2
Terminal agents have shown strong abilities in complex coding tasks. However, they often struggle to recover from errors during long-running execution (e.g., command errors, faulty implementations), eventually leading to task failures. To this end, we present ReTerm, a self-improving terminal agent
agentself-improving - arxiv:2610.18481 · cs.LGProTok: Unified Protein Tokenization via Modular Multi-Codebook QuantizationRomain Hardy, Xiaoman Zhang, Pranav Rajpurkar
Protein sequence, structure, and functional annotations each capture complementary aspects of protein biology, yet existing tokenizers encode each modality independently, limiting the expressiveness of discrete protein representations. We present ProTok, a unified protein tokenizer that maps all thr
benchmark - arxiv:2610.18478 · cs.ROMemoryVLN: From Ego-Aware Memory Representation to Cross-Context Memory Reuse in Vision-Language NavigationXueyao Wei, Yifei Wang, Delun Li, Shaoan Wang +3
Large vision-language models (LVLMs) have shown strong potential for embodied vision-language navigation (VLN), yet their performance degrades on long-horizon trajectories due to context-length limits, redundant historical observations, and unstable spatial alignment across time. These issues make i
embodiedmemoryagent - arxiv:2610.18472 · cs.LGSimplifying and Improving On-Policy Distillation for Few-Step DiffusionJiawei Zhang, Ziyu Wan, Nanye Ma, Long Zhao +4
Few-step distillation reduces the sampling cost of diffusion and flow models, but existing approaches often require auxiliary score models or specialized architectures. We show that simple on-policy velocity regression already provides a strong foundation for 4-NFE generation. To represent teacher v
benchmark - arxiv:2610.18461 · cs.CVMFVSR: A Multi-Style and Fine-Grained Benchmark for Video Subtitle RemovalXiyan Huang, Qingdong He, Peng Tang, Bo Yin +2
Video subtitle removal (VSR) is an important video restoration task, yet existing evaluation remains limited by low-resolution datasets, fragmented test sets, and generic video-inpainting protocols that fail to adequately capture localized restoration errors, unintended modifications, temporal insta
benchmark - arxiv:2610.18459 · cs.LGTeacher-Guided Tool Learning for Small Language Models via Bidirectional KL ConstraintsHaoxuan Gao, Qin-Wen Luo, Xiang Chen, Yu Tian +4
External tools can extend the capabilities of small language models (SLMs), enabling their deployment in resource constrained environments. However, their limited capabilities make it difficult for SLMs to discover effective tool use behaviors through reinforcement learning alone, while direct teach
tool use - arxiv:2610.18457 · cs.LGKairos: Transactional Security Enforcement at the System Boundary for LLM AgentsKristoffer Hempel, Nils Loose, Lorenzo Cavallaro, Thomas Eisenbarth
Large language model (LLM) agents are becoming an integral part of the software development lifecycle, and the capabilities granted to them allow an agent to act with the user's authority on any resource it can access. The reach that gives an agent its utility is equally available to an attacker who
agentllm agentbenchmark - arxiv:2610.18456 · cs.LGLearning Perturbation-Specific Representations through Conditional Energy Prior for Cellular Response PredictionYuchen Yan, Yatao Bian
AI-based cellular perturbation prediction is essential to accelerate drug discovery and therapeutic identification. However, perturbation-specific signals are generally obscured by cellular complexity, making it difficult for existing methods to correctly capture perturbation effects. To address thi
benchmark - arxiv:2610.18455 · cs.CVEmoGenie: Benchmarking and Evaluating Expressed and Evoked Emotion in T2I Generation with MLLMsZixuan Xing, Lancheng Gao, Shengyan Li, Huiyu Duan +3
Rapid advances in artificial intelligence-generated content (AIGC) have significantly promoted using text-to-image (T2I) models to generate images with specific affective intentions. However, the emotional effects conveyed by generated images frequently deviate from target expectations, and it remai
benchmark - arxiv:2610.18449 · cs.AIAutonomous Science Bench: Evaluating AI Agents for Autonomous Scientific DiscoveryYibo Wen, Han Liu
AI agents for autonomous scientific discovery must decide which experiments to run under a limited budget. Most benchmarks for science agents provide fixed datasets, while others that support active experimentation largely rely on simulation. We introduce Autonomous Science Bench, an initial release
agentai agentbenchmarkleaderboard - arxiv:2610.18447 · cs.ROThinker-WM: A Unified World Model with Shared Multimodal Representations for Embodied IntelligenceThinker Team, Peng Gao, Jichao Jiao, Chen Liao +7
World models have emerged as a promising paradigm for embodied AI, yet existing systems often remain fragmented in the modalities they model and the functions they support. In particular, visual dynamics, geometric structure, and executable action are commonly coupled through separate interfaces or
embodiedworld modelaction-conditionedworld-model evaluation - arxiv:2610.18443 · cs.ROPACT: Posture-Aligned Co-Design Technique for Hand Configuration and Retargeting in Multi-Finger Imitation LearningKoyo Sasazaki, Satoshi Funabashi, Pranav Ponnivalavan, Alexander Schmitz +2
The fidelity of teleoperated demonstrations used for imitation learning is determined by how faithfully a robot hand can replicate the operator's posture. The posture retargeting law that the robot will execute is typically selected only once the hardware has been finalized, and conventional hand de
teleoperationgrasp - arxiv:2610.18442 · cs.CLCopying Is Not Grounding: Training Faithful Long-Form RAG with Authored-Sentence CreditHoang Pham, Duc Anh Vu, Duy Tung Doan, Cong-Duy T Nguyen +1
In search engines and assistants, long-form RAG answers can mislead readers with claims their sources do not support. Faithfulness rewards, the share of supported sentences, train models toward fewer such claims. We show that these rewards are hacked by copying. A sentence copied from the passages i
ragevaluator - arxiv:2610.18438 · cs.AISTAMP as a topology-aware molecular language for fragment-level learning and editingYuxuan Ren, Jiatong Li, Ziqiao Meng, Yatao Bian
Molecular fragments provide chemically meaningful units, but existing molecular languages rarely make their locations and connections explicit for editing. Here we introduce STAMP (Sequential Tree Attachment-based Motif-Preserving representation), a topology-aware molecular language that connects fr
benchmark - arxiv:2610.18427 · cs.CLReliable Memory from Unstable Growth: Curating Self-Written Agent Playbooks with Randomized ExperimentsHoang Pham, Cong-Duy Nguyen, Dong Le, Ponhvoan Srey +1
Persistent textual memory lets language-model agents reuse lessons from past tasks to improve on future multi-step tasks without retraining. Systems such as ACE and Dynamic Cheatsheet keep it as a playbook: lessons the agent writes after each task and reads in its prompt. A key challenge is to decid
memoryagentbenchmark - arxiv:2610.18426 · eess.SYEvidence-Governed Deployment Readiness for Networked Agents on Industrial OT InfrastructureDibyajyoti Nath, Nitya Sathyavageeswaran, Ziran Min, Pavan Athreya Narasimha Murthy
Deploying a workload on a cyber-secure industrial operational-technology (OT) network requires many cross-layer prerequisites to hold at once, including persistent addressing, name resolution, time synchronization, certificate trust, proxy traversal, and reachable industrial-protocol endpoints. Thes
agent - arxiv:2610.18425 · cs.ROCALM: Decentralized Control of Cooperative Aerial Manipulation with Multi-Agent Reinforcement LearningJiaxi Huo, Petre Ricioppo, Fausto Francesco Lizzio
Cooperative aerial manipulation is difficult under wind, payload motion, and noisy goal broadcasts because multiple Unmanned Aerial Vehicles (UAVs) are coupled through a shared payload. We present CALM (Cooperative Aerial Learned Manipulation), a decentralized controller for aerial manipulators carr
manipulationmanipulatormulti-agent - arxiv:2610.18424 · cs.CLReceiver Decides Agent-to-Agent NegotiationsFardeen Sadab, Alif Hasnain, Afsana Shehrin, Md. Azam Hossain +1
When one language-model agent asks another for money, which of the two decides how much changes hands? We ran a round-robin tournament of 1,049 negotiations among fifteen agents. In each game a requester tries to raise money from a holder that has been told to keep a $100 endowment. Knowing which ag
agent - arxiv:2610.18420 · cs.LGAgaram 2.5B: The First Tamil-First Language Model Trained from Scratch with Tool Calling. A Playbook for Under-Represented LanguagesRamchand Kumaresan, Madhusudhanan G, Kaushik Kezendirane
We trained Agaram 2.5B, a 2.5-billion-parameter Tamil-English language model, from scratch on about 50 billion tokens (58% Tamil, the rest English, math, code and parallel text) on one node of eight H100 GPUs in about six days, with a new tokenizer that, unlike our first model's, keeps line breaks.
tool callingbenchmark - arxiv:2610.18415 · cs.AIDo Factuality Methods Make LLMs More Factual? Reproducibility and Fair ComparisonPratuat Amatya, Vinay Setty, V Venktesh, Avishek Anand
Large Language Models (LLMs) are aligned to human preferences, after pre-training and supervised fine-tuning steps. However, standard RL alignment steps such as RLHF do not focus on factuality of LLMs, hence they may generate non-factual information, which could result in unintended consequences. In
rlhfbenchmark - arxiv:2610.18410 · cs.AIGraph-free Retrieval with Graph-Modified EmbeddingsFederica Santi, Vasilije Marković, Daniel Scott, Lazar Obradović
Dense vector retrieval can miss evidence connected through intermediate entities, while direct graph retrieval requires access to the graph at query time. We compare five offline graph-based embedding transformations and strategies for using the resulting graph-modified embeddings alongside the orig
benchmark - arxiv:2610.18408 · cs.CLPrompt-Injection Probes Learn the Conversation, Not Just the InjectionKaan Kale, Sriram Vishwanath
Linear probes on a model's hidden states are used as cheap monitors for indirect prompt injection: they screen the tool outputs an agent reads and send suspicious ones to a slower second check, so every false alarm has a cost. But a probe learns whatever separates its training examples. The standard
agent - arxiv:2610.18404 · cs.LGOvercoming Catastrophic Forgetting: Continual Learning for Radar Waveform ClassificationMohamed Rabie, Chinthana J. Panagamuwa, Konstantinos G. Kyriakopoulos
Automated radar waveform recognition classifies the waveform and modulation type of a received signal. As new waveform variants appear over time, a deployed classifier must learn them without retraining from scratch. Catastrophic Forgetting occurs when learning new waveforms overwrites prior knowled
benchmark - arxiv:2610.18399 · cs.CVBifocal: Training-Free Mixed Granularity Visual Token Pruning for Multimodal LLMsPeiran Yin, Hao Wu, Zihao Yang, Fengyu Zhou +5
Visual token pruning for multimodal large language models (MLLMs) appears highly effective, with existing methods retaining over 95% of unpruned performance on common VQA benchmarks using a quarter of the visual tokens or fewer. However, our extensive evaluation of ten methods across three MLLM fami
benchmark - arxiv:2610.18390 · cs.LGBeyond RMSE and Granger Gain: Multi-Horizon Forecast-Error Entropy for Nonlinear Predictive CouplingAndrei Velichko
We investigate whether the past of an auxiliary time series provides predictive information about a target beyond that contained in the target's own history. We compare matched out-of-sample target-only and target-plus-source forecasts and introduce the forecast-error distribution entropy (FEDE) pre
benchmark - arxiv:2610.18386 · cs.ROIntent-guided Adaptive Stiffening for Visuomotor Policies under Unknown ResistanceFrancesco Cufino, Carmela Mariniello, Mario Selvaggio, Andreas Sochopoulos +1
Manipulation tasks such as unscrewing a cap or sliding a part in contact with a surface require imposing force along a high-friction direction: the robot must overcome a resistance rather than yield to it. Generative visuomotor policies predict the motion for such tasks but not the authority needed
manipulation - arxiv:2610.18381 · cs.ROBenchmarking Interaction-Oriented Aerial Mobility across Heterogeneous EnvironmentsZhibo Duan, Hanlin Kong, Tianran Ma, Yuze Wu +7
Aerial mobile manipulation requires unmanned aerial vehicles (UAVs) to traverse heterogeneous environments such as indoor and outdoor domains, and connect navigation with subsequent physical interaction within the same trajectories. Existing aerial benchmarks are typically confined to single-environ
vision-language-actionembodiedmanipulationbenchmarkevaluation framework - arxiv:2610.18378 · cs.AIDelegating Earth science model calibration to knowledge-guided LLM agentsZhanliang Zhu, Ziwei Li, Liujun Zhu, Yuchen Liu +2
Applying process-based Earth science models to new sites and questions often requires expert calibration decisions. Reliable delegation to language-model agents must preserve scientific competence and reproducibility. Here we develop a general calibration harness (GCH) for end-to-end delegation from
agentllm agent - arxiv:2610.18373 · cs.RODR-VLA: Learning Human-Like Quadruped Navigation from Egocentric Web Videos via Domain-Invariant Scene RepresentationsJun Tian, Yingjie Niu, Ruo Chen, Hongyi Lin +1
Quadruped robots performing outdoor deliveries must interpret complex scenes, anticipate changes, and navigate in socially compliant ways. Robot demonstrations are costly to collect at the scale needed for generalization. Egocentric web videos offer abundant human navigation experience, but differen
quadrupedworld model - arxiv:2610.18369 · cs.CVHallucination-aware Faithful Reasoning in Remote SensingYi Liu, Jing Zhang, Xiang Feng, Xiaoyu Tian +4
Multimodal large language models (MLLMs) have demonstrated promising capabilities for remote sensing reasoning, yet they frequently generate visually unsupported reasoning that contradicts the visual evidence. Existing post-training methods optimize primarily for final-answer correctness, providing
post-trainingbenchmark - arxiv:2610.18361 · eess.SYAdaptive Handling of Deadline Misses via Stochastic MPCMelanie Gallant, Paolo Pazzaglia, Christoph Mark, Kevin Schmidt +1
Real-time control systems may experience sporadic computational overloads that cause controller executions to overrun their expected deadlines, affecting the timely update of control inputs. Such overruns are classically handled with static decisions, e.g., by either stopping and discarding the late
benchmark - arxiv:2610.18356 · cs.CVUrbex: Agentic Spatial Grounding for City-Scale 3D ScenesYu Wang, Rui Dai, Wei Chen, Yi Wang +4
City-scale 3D grounding is commonly formulated as candidate ranking, which requires precomputed instance proposals and becomes inefficient in large outdoor scenes. We propose Urbex, an agentic vision-language framework that instead casts grounding as active spatial search. Urbex represents a city-sc
agentagentictool-use - arxiv:2610.18355 · cs.AIABot-ER: Unified Embodied Reasoning through Agentic Data-Model Co-EvolutionQiuyu Ding, Heng-Da Xu, Changda Xia, Jiajun Shen +13
We introduce \textbf{ABot Embodied Reasoning (ABot-ER)}, a unified embodied reasoning foundation model that breaks the tension between breadth and depth in embodied reasoning. Embodied intelligence demands the joint mastery of perception, planning, and execution, yet reconciling generality with spec
embodiedagenticbenchmark - arxiv:2610.18354 · cs.AIChoosing LLMs for Strategic Agent InteractionsArastun Mammadli, Herbert Woisetschläger, Jawad Fayaz, Shiqiang Wang
Game-theoretic benchmarks claim that large language model (LLM) strategic decision-making varies across game types, interaction structures, and capability dimensions. However, they only evaluate strategic performance across individual fixed models. In parallel, contemporary LLM routers pick models u
agentbenchmark - arxiv:2610.18352 · cs.ROLet the Mind Conceive, Let the Hand Achieve: Following Multimodal Instructions for Long-Horizon Robotic ManipulationYoupeng Wen, Gang Wang, Tianyu Huang, Weiwei Zhang +4
A robot following a multi-step instruction needs to act on the right objects in the right order, even as the scene changes. It must also maintain awareness of task progress and determine when to advance. To address these challenges, we introduce Orchestra, an agent harness that separates task planni
manipulationagentbenchmarkevaluator - arxiv:2610.18343 · cs.AILightweight but Linguistically Fair? Quantisation Robustness across Bavarian, German, and EnglishJophin John, Michael Hoffmann, Piroska Lendvai, Florent Dufour
Post-training quantisation reduces the memory cost of large language models, but it remains unclear whether compression preserves language skills of regional varieties as reliably as their standard-language counterparts. We study this question for Bavarian, Standard German, and English using BAVGROU
memorypost-trainingbenchmark - arxiv:2610.18336 · cs.MAStrategic Multi-Agent Resource Allocation via Statistical ContractsBen Chugg, Michael I. Jordan, Aaditya Ramdas
We study a class of strategic hypothesis testing problems in which a principal aims to allocate resources among competing agents who possess data or data-generating mechanisms that may be relevant to the principal's goals. Each agent can pay a participation cost to opt into a statistical contract, i
agentmulti-agent - arxiv:2610.18326 · cs.CLJev-Guard: Fast Routing and Auditing Agent Actions over Editable Policy TreesWenbin Hu, Huihao Jing, Haochen Shi, Yuxuan Liu +8
Agent harnesses require a low-latency guardrail model together with customizable audit policies. However, deploying an auto-regressive guardrail model remains challenging: such models incur high inference latency and are often poorly aligned with customized policies. To address these problems, we pr
agent - arxiv:2610.18325 · cs.CVShoeFit: A New Dataset and Dual-image-stream DiT Framework for Virtual Footwear Try-OnYuhan Li, Zhiyu Jin, Yifan Tong, Wenxiang Shang +4
Virtual footwear try-on (VFTON), a critical yet underexplored area in virtual try-on (VTON), aims to synthesize faithful try-on results given diverse footwear and model images while maintaining 3D consistency and texture authenticity. Unlike conventional garment-focused VTON methods, VFTON presents
benchmark - arxiv:2610.18321 · cs.ROCoSMO: Conformal Semantic Mapping for Open-World Safe Robot NavigationShuqi Wang, Yiwei Zheng, Pingrui Zhang, Yue Gao +1
Safe autonomous navigation requires not only geometric obstacle avoidance but also semantic awareness. Different object categories demand different safety clearances, and the set of relevant categories may expand as the robot explores novel environments. Vision-language models (VLMs) offer rich sema
quadruped - arxiv:2610.18313 · eess.SYInfrastructure-Aware Runtime Protection of Co-Deployed AI Services at the Industrial EdgeNitya Sathyavageeswaran, Dibyajyoti Nath, Ziran Min, Pavan Athreya Narasimha Murthy
In modern production lines, a single edge device simultaneously runs multiple AI services like a torque-control model, a vision inspector, a predictive maintenance routine. All these services run on the same CPU and same shared network segment. When the shared network segment saturates, the services
agentautonomous agent - arxiv:2610.18311 · eess.SYBALLAST: Physics-Aware Moving Target Defense for Industrial Control Systems Using a Digital Twin Safety OracleOscar Mogollon-Gutierrez, Mohammadhossein Homaei, Alberto Lopez-Trigo, Andres Caro +1
Industrial Control Systems run on static network configurations, which leaves an attacker unlimited time to map the plant. Moving Target Defense answers this by shuffling addresses, but almost every published scheme never checks whether the reconfiguration will itself disturb the physical process it
agent - arxiv:2610.18308 · cs.AISimInter: Evidence-Grounded User Simulation with Interview-Based Profile Enrichment for Conversational Recommender SystemsErfan Shahabi, Amirhossein Roshandel, Mohammad Sadegh Sadeghi, Azadeh Shakery
Conversational Recommender Systems (CRSs) require reliable user simulators for cost-effective evaluation. LLM-based simulators have made realistic simulation possible but typically condition on profiles derived from explicit ratings, demographics, and static sources capturing surface preferences whi
agentllm agent - arxiv:2610.18307 · cs.ROI2O2I-Nav: A Benchmark with a Modular Agent for Long-Horizon Indoor-to-Outdoor-to-Indoor NavigationZhengjie Shu, Wanyu Li, Di Xiao, Lingxin Jiang +10
Service and delivery robots often need to complete long-horizon navigation missions spanning indoor and outdoor environments. Existing embodied navigation benchmarks primarily evaluate isolated stages or partial cross-domain journeys, providing limited support for assessing how heterogeneous navigat
embodiedagentbenchmark - arxiv:2610.18306 · cs.LGBayesian Optimization of CVaR with Unknown HyperparametersValentin Margraf, Jonas Hanselle, Viktor Bengs, Serafima Lebedeva +2
We study Bayesian optimization of the Conditional Value-at-Risk (CVaR) of an expensive black-box function that depends on a decision and an environmental variable. Existing no-regret algorithms assume that the correct Gaussian process (GP) kernel and its hyperparameters are known. In practice, howev
benchmark - arxiv:2610.18305 · cs.CVAffectWorld: Benchmarking Emotion-Driven Future World PredictionChanghao Song, Yacong Chen, Hui Gao, Yazhou Zhang +1
The world encompasses both physical environments and human actors. Predicting its evolution requires understanding both physical laws and how emotions shape human actions. However, existing world model benchmarks primarily assess physical plausibility and perceptual quality, leaving emotion-driven w
world modelbenchmark - arxiv:2610.18299 · cs.CLLegalRewardBench: Benchmarking Reward Models for Legal Preference EvaluationHui Huang, Jie Zhou, Hongli Zhou, Qundong Shi +6
Reward models are central to preference alignment for legal tasks, yet existing reward model benchmarks are still primarily centered on general-domain preferences. In this work, we introduce LegalRewardBench, the first human-verified benchmark for evaluating reward models in the legal domain. The be
benchmark - arxiv:2610.18298 · cs.AIOutput Discipline Sets the Floor: A Falsification-Guided, Model-Agnostic Agent Protocol for Auditable Hypothesis-Driven Decision MakingJing Liu, Yongxing Qi, Ping Yao, Ting Wu
What binds large language models on auditable, hypothesis-driven decision tasks, where every candidate can be checked against the evidence? On the public TCMEval-SDT benchmark for Traditional Chinese Medicine syndrome differentiation (200 cases), the binding constraint is output discipline and reaso
agentagent frameworkbenchmark - arxiv:2610.18295 · cs.LGRobust Logical Anomaly Detection under Capture Shifts via Synthetic AnomaliesUfuk Efe, Alptekin Temizel
Detectors of logical anomalies, such as missing, extra or wrongly combined parts, usually score a test image by its distance from normal (good) images. They achieve near-perfect image-level AUROC on standard industrial anomaly detection benchmarks such as MVTec LOCO, but their robustness to changes
benchmark - arxiv:2610.18289 · cs.CVLongForcing: Scaling Temporal Manifold Matching for Long-Horizon World RolloutZiyu Zhu, Zhaoxu Sun, Fan Jiang, Mengchao Wang +4
Autoregressive video diffusion models have emerged as a foundation for long-horizon world rollout. However, when extended far beyond their training windows, these models suffer from progressive drift: visual quality deteriorates, motion becomes inconsistent, and object identity gradually collapses a
world modelaction-conditioned - arxiv:2610.18284 · cs.CVEfficient Scene-Text Refinement via Pixel-Space Grouped Image GenerationYaoli Liu, Guo-Hua Wang, Yiyang Chen, Qing-Guo Chen +3
Generated images often have coherent layouts and visual content, yet local errors in small text and complex characters still limit their readability and practical use. Scene-text refinement requires accuracy, preservation, and efficiency. Existing methods based on variational autoencoders (VAEs) can
benchmark - arxiv:2610.18262 · cs.AIEncrypted Skills Can Become New Security Exploits If AI Agents Know How to DecipherYan-Lun Chen, Pin-Yu Chen, Chia-Mu Yu
Third-party skills can expose LLM agents to malicious instructions hidden by encoding. We introduce Ciphered Skill Injection (CSI), which embeds such instructions in skill files for agents to decode and follow. We evaluate Caesar and layered variants across eight models and 139 injection-task pairs,
ai agentllm agent - arxiv:2610.18256 · cs.LGTeach Me After You Learn It: Lookahead Self-Distillation for RLVRZepeng Zhai, Meilin Chen
On-policy self-distillation (OPSD) has emerged as a popular training paradigm for LLMs, providing dense, token-level supervision by aligning a model's own distribution with the distribution induced by privileged information. However, existing OPSD methods derive supervision solely from privileged in
agent benchmarkbenchmark - arxiv:2610.18248 · cs.CLAre We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language ModelsGuoming Ling, Yupei Lin, Zhongzhan Huang, Junxin Li +4
Benchmarks are essential for evaluating and improving large language models (LLMs). Yet many widely used benchmarks are redundant, making evaluation unnecessarily expensive. Benchmark compression lowers this cost by estimating full-benchmark scores and rankings from a partial evaluation. Research in
benchmark - arxiv:2610.18240 · cs.AIBeyond Answer Sharing: TrueMAD for Targeted Peer Engagement in Multi-Agent DebateDezheng Bao, Sihang Du, Renhong Huang, Zhizhang Yuan +2
Multi-agent debate (MAD) aims to improve large language model reasoning by exposing agents to alternative solutions and enabling iterative revision. However, standard broadcast-style MAD makes peer answers visible without requiring agents to engage with a particular peer claim. As a result, reliable
agentmulti-agent - arxiv:2610.18238 · cs.CVHeading Hierarchy Reconstruction for Complex and Long Documents with Level-Wise Multimodal MemoryQian Chen, Xianyin Zhang, Lifan Guo, Chi Zhang
Vision-language models (VLMs) have been widely adopted for document parsing and optical character recognition (OCR), as they can accurately extract page-level elements. However, downstream applications such as retrieval-augmented generation (RAG) also require coherent document-level structure. Altho
memoryretrieval-augmentedbenchmark - arxiv:2610.18233 · cs.AIConsider, Then Choose: Order-Invariant Decisions in System One ModelsJianguo Huang, Yao Long Teng, Yanchen Deng, Bo An
System One models answer typed questions in a single forward pass and promise the same answer for the same input. For choice questions, this promise fails: the answer depends on the order of the options, increasingly so as they grow in number, and averaging over their cyclic shifts reduces this depe
benchmark - arxiv:2610.18230 · cs.AIHolding Agents Accountable with Verifiable Agent InsuranceCheng Hua, Taro Tsuchiya, Nicolas Christin, Giulia Fanti
While AI agents are useful for many tasks, they regularly misbehave in destructive ways, such as deleting important data. While prior work has focused on preventing such misbehavior, we consider how to recover after an agent misbehaves. We propose verifiable agent insurance (VAI), a framework for ag
agentai agentagent benchmarkbenchmark - arxiv:2610.18229 · cs.LGWhen Does Opponent Modeling Pay Off? Spatial Inference Stabilises Reinforcement Learning under Partial ObservabilityMarcel Petran, Stefan Edelkamp
Deep reinforcement learning agents in partially observable competitive games must act on beliefs about an unseen opponent's intent. We study a spatial opponent model: a transformer trained with hindsight-relabelled opponent goals that outputs a subgoal heatmap from the opponent's observed trajectory
agent - arxiv:2610.18216 · cs.AITextualRL: Textual Reinforcement Learning for Context ManagementTianyu Huai, Jie Zhou, Bo Mao, Qin Chen +2
LLM agents can adapt to task feedback by revising the textual context that guides their behavior, without changing their model parameters. However, turning execution rewards into reusable context updates is challenging: individual trajectories offer limited comparative evidence about what to change,
llm agentbenchmark - arxiv:2610.18206 · cs.CLSWE-MISBEHAVIOR: Benchmarking and Mitigating Boundary Violations in Software Coding AgentsDo Xuan Trong, Duc Anh Nguyen, Do Xuan Long, Ngoc-Hai Nguyen +11
We present SWE-MISBEHAVIOR, the first taxonomy and evaluation framework for diagnosing boundary-violating misbehaviors in coding agents. It contains 7.4K human-in-the-loop synthesized samples spanning three layers of agent action space: (i) Code Generation, (ii) File Management, and (iii) System Int
agenthuman-in-the-loopbenchmarkevaluation frameworkevaluation protocol - arxiv:2610.18204 · cs.AIReset Early, Reset Often, Eliminate ModelsMartin S. Fixman, Fredrik Rømming, Sean B. Holden
Connection-tableau provers typically construct closed tableaux by backtracking search. This gives lightweight proof procedures, but their low-memory search state leaves little derived information to reuse after a failed attempt. SATCoP addresses this by retaining grounded tableau clauses in an incre
benchmark - arxiv:2610.18201 · cs.AIHow Reliable Are Large Language Models for Soil Organic Carbon Assessment? An Australian BenchmarkYingxuan Cheng, Tong Li, Yayong Li, Yuchuan Fan +4
Soil organic carbon (SOC) is an indicator of agricultural soil condition and carbon storage, but numerical outputs from large language models (LLMs) require validation before use in monitoring or management. We evaluated seven general-purpose LLM configurations on three direct-response tasks: estima
benchmark - arxiv:2610.18198 · cs.AIQORH: Learning Query-Specific Ordered Rule Hypergraphs for Inductive Knowledge Graph Link PredictionNaixing Xu
Knowledge graph link prediction involving unseen entities and relations requires learning transferable reasoning patterns from observed graph structure. Existing methods based on relation graphs primarily capture pairwise dependencies, limiting their ability to preserve the compositional semantics,
knowledge graphbenchmark - arxiv:2610.18194 · cs.ROCross-Embodiment Affordance Transfer for Robotic ManipulationSawera Yaseen, Serge Thill, Rekha Raja
Robotic manipulation systems commonly learn object affordances from human-annotated datasets, which reflect interaction possibilities for a human hand rather than the physical constraints of a specific robot. For example, a surface region of an object identified as graspable by a human annotator may
manipulationgrippergrasp - arxiv:2610.18191 · physics.opticsFeedback-Controlled Self-Trapping of SolitonsRaonaqul Islam, Pradyoth Shandilya, Yunxiang Song, Alioune Niang +4
Chip-scale microresonator-based frequency combs, commonly referred to as microcombs, are promising sources for field-deployable metrology applications. However, intrinsic microresonator noise can make it difficult to generate microcombs with sufficiently low noise for applications. A widely used app
benchmark - arxiv:2610.18189 · cs.CVVARI: Robust Text-to-Video Generation via Group-Structured Intent AlignmentMingxuan Zhu, Hohei Chan, Shiyuan Liu, Yuanxing Zhang +1
Modern text-to-video (T2V) models generate visually compelling videos, yet meaning-preserving changes in prompt wording can substantially alter semantic realization and overall quality. Existing prompt rewriting modifies the inference input, whereas conventional post-training constructs supervision
post-training - arxiv:2610.18186 · cs.LGReinforcement Learning for Large-Scale Lifelong Multi-Agent Path Finding with Shared Spatial FieldsJean Seong Bjorn Choe
Large-scale applications of lifelong multi-agent path finding (LMAPF) call for computationally inexpensive movement decisions under changing traffic conditions and across varied layouts. We propose a navigation policy in which a convolutional network computes a shared spatial field over the map once
agentmulti-agent - arxiv:2610.18174 · cs.LGLooped Reasoning Finishes Earlier Than You ThinkSumin Park, T. Konstantin Rusch
Looped reasoning with dynamic depth can be viewed as relaxation toward solution attractors in a learned dynamical landscape. In practice, however, latent trajectories often oscillate, cycle, or drift rather than cleanly converge to stable fixed points. How these non-stationary dynamics contribute to
latent dynamics - arxiv:2610.18173 · cs.CLTRACESTEER: Trace Driven Activation Steering for Hallucination Mitigation in Diffusion Language ModelsXin Zhang, Yili Wang, Yue Tan, Xin He +4
Diffusion large language models (dLLMs) have emerged as a promising alternative to autoregressive LLMs, yet remain susceptible to hallucination. Activation steering offers a promising mitigation approach by guiding parallel token predictions through internal representations. However, existing steeri
benchmark - arxiv:2610.18167 · cs.ROBeyond Planning: Differentiable Control-in-the-loop End-to-End Autonomous DrivingZhidong Wang, Zirui Li, Zitong Shan, Zhiqi Mao +5
End-to-end autonomous driving has achieved remarkable progress by planning trajectories directly from raw sensor data. Yet the step from planning to control still relies on hand-tuned tracking controllers that block the gradient path from vehicle motion back to the network. This gap stems from non-d
benchmark - arxiv:2610.18163 · cs.RORESTA: Time-Scaling Augmentation for Data-Efficient Robot Imitation LearningYunpeng Jiang, Jianshu Hu, Han Fang, Zhiyuan Guan +2
Robot manipulation policies trained through imitation learning commonly rely on costly demonstrations recorded in one task direction and at one temporal resolution. We propose Reversal and Stride-based Trajectory Augmentation (RESTA), which exploits time-scaling data augmentation to improve data eff
manipulation - arxiv:2610.18161 · cs.CVBenchmarking Multimodal Large Language Models for Zero-Shot Human-Object InteractionOriol Rabasseda, Zenjie Li, Kamal Nasrollahi, Sergio Escalera
Multimodal Large Language Models (MLLMs) have demonstrated strong zero-shot reasoning across diverse domains, but their potential for Human-Object Interaction (HOI) remains unexplored. Unlike traditional HOI approaches that rely on extensive labeled datasets and fixed interaction taxonomies, MLLMs o
benchmark - arxiv:2610.18160 · cs.ROForce Sensor-Based Picking Pattern Analysis and Two-Jaw Gripper Design for Cherry Tomato Robotic Harvesting in GreenhouseRajmeet Singh, Manveen Kaur, Yusuf Adamu, Irfan Hussain
Robotic harvesting of cherry tomatoes in greenhouses remains challenging due to dense fruit clusters, delicate skin, and the need for gentle yet effective detachment. This study presents a force-data-driven framework for autonomous cherry tomato harvesting that integrates six-axis force/torque sensi
gripper - arxiv:2610.18157 · cs.AIFan Out or Relay? Ownership-Aware Routing for LLM Agents Querying Siloed TablesZhexi Feng
Inside organizations, a question's data often sit with several owners, who query their own tables but never hand them over. An LLM coordinator can fan out, sending the question to every owner at once, or relay, querying owners in turn and reusing earlier rows. We show that neither is safe in advance
llm agent - arxiv:2610.18147 · cs.ROWalk Like This: Stylized Locomotion for Physics-Based Characters through Example-Driven DiffusionJoshua Aurand, Agon Serifi, Sammy Christen, Ruben Grandia +3
We present a real-time framework for expressive legged locomotion that enables independent control over velocity and style. Unlike existing data-driven controllers, which typically require explicit style labels for training, our method efficiently leverages large-scale unstructured motion data and g
humanoidlegged locomotion - arxiv:2610.18142 · cs.CVFlowCred: Hierarchical Credit Assignment and Functional Coordination for Joint Text-to-Video Post-TrainingMingxuan Zhu, Yanwei Huang, Shiyuan Liu, Xinzhi Zhang +3
Prompt optimization and generator post-training are two complementary ways to improve text-to-video (T2V) generation, and recent systems increasingly train both modules jointly. However, optimizing a prompt policy and a video generator from the same terminal reward creates two ambiguities: prompt qu
post-trainingbenchmark - arxiv:2610.18140 · cs.CLSparPR: Runtime KV-Cache Recovery for Resource-Constrained LLM Inference on Consumer HardwareJieyun Deng, Keming Hu, Yanghe Liu, Li Feng
Running large language models (LLMs) on consumer hardware is bottlenecked by the key-value (KV) cache, which grows with every decoded token and quickly exhausts the memory and bandwidth of a single consumer GPU. We present SparPR (Sparse Partial Recomputation), a runtime KV-cache recovery mechanism
memory - arxiv:2610.18135 · cs.ROSkipWorld: Latent Skip Planning with Distilled World Models for Efficient Visual NavigationYifei Yang, Zehua Fan, Zike Yan, Zhongyu Li +1
World-model-based visual navigation enables action planning by predicting future states under candidate trajectories. However, long-horizon rollout remains challenging: models are often trained with ground-truth visual histories but deployed in closed-loop prediction, and evaluating many candidates
world model - arxiv:2610.18126 · cs.AIFresh at Decision, Stale at Action: Routing New Information to LLM Decisions Already MadeAyoub Merimi, Souhail Bakkali, Abdeljalil Sakat
Freshness shapes an LLM decision at the moment it is made; the action it commits to is latent, carried out later, in a changing world. We therefore introduce a timestamp theory of decision validity and a decision ledger that realises it. Six timestamps define a decision; a freshness ceiling bounds w
persistent state - arxiv:2610.18125 · cs.LGH-DDCM: Hadamard Codebooks for Fast Generative Image and Video CompressionOr Ronai, Vladimir Kulikov, Tomer Michaeli
Training-free diffusion-based compression methods repurpose the generative priors of a diffusion or a flow model to compress data without retraining them as task-specific codecs. Recent codebook-based approaches such as Turbo-DDCM achieve strong perceptual quality at ultra-low bitrates, but their re
memorybenchmark - arxiv:2610.18123 · cs.ROMARTS: A Modular Agricultural Robot for Targeted Spraying of Diseased Tomato Plants in GreenhousesRajmeet Singh, Manveen Kaur, Yusuf Adamu, Lochan Kshetrimayum +1
Broadcast spraying remains the dominant method of disease control in greenhouse tomato production, yet it treats healthy and diseased tissue alike, exposes workers to hazardous aerosols, and depends on operator skill. This paper presents MARTS (Modular Agricultural Robot for Targeted Spraying), a lo
manipulator - arxiv:2610.18122 · cs.CLCheap, Consistent, Calibrated, but Biased? Typed Decision Models as JudgesFabian Lukassen, Christoph Weisser, Alexander Silbersdorff
Typed decision models (TDMs) return a probability for each allowed answer instead of generating text, which makes them candidates for cheaply replacing LLM judges that compare two answers. Whether they share the biases of LLM judges is untested. We run the CALM benchmark's bias tests on eight TDMs a
benchmark - arxiv:2610.18118 · cs.ROForeTac: Predictive Contact Guidance for Generative Robot PoliciesZhiyuan Chen, Shuai Chen, Haoran Zhou, Yizhe Zhao +6
Visually similar actions can produce contact loss, jamming, slippage, or irreversible damage, while tactile feedback may arrive after a critical transition. Contact-rich manipulation therefore requires generative policies to anticipate and evaluate whether candidate actions will produce desirable co
manipulationtactilediffusion policygraspaction-conditioned - arxiv:2610.18108 · cs.LGA CaLLM Judge is a Better Judge: Mitigating Bias in LLM-as-a-judgeCesare Barbera, Luca Stradiotti, Lorenzo Cascioli, Lorenzo Perini +5
Large language models (LLMs) are increasingly used as judges to predict which of two candidate answers a human evaluator would prefer, providing a scalable alternative when human evaluation is too costly. Unfortunately, LLM judgments can exhibit systematic biases. Prior work has documented position,
benchmarkevaluator - arxiv:2610.18104 · cs.LGThe Effect of Layer Ordering in Hybrid ArchitecturesJohn Cooper, Mingchen Ma, Jiaqi Mo, Shengqi Qiu +3
Hybrid attention is rapidly becoming the dominant architectural choice for frontier models. Despite their successes, such architectures complicate model design by introducing additional choices, including the types and locations of layers. This combinatorially large design space has been studied emp
memory - arxiv:2610.18097 · cs.CVAPT: Feedback-Driven Agentic Post-Training for Text-to-Image GenerationSirui Zhang, Fangyun Wei, Dong Chen, Jinjing Zhao +4
Post-training is crucial for improving text-to-image models, yet existing pipelines typically rely on manually curated training data and fixed training distributions. We introduce APT (Agentic Post-Training), a fully automated framework that adaptively post-trains a base model according to its evolv
agenticself-improvementpost-trainingbenchmark - arxiv:2610.18095 · cs.AICompanionSafe: From Multi-turn Safety Evaluation to Guardrails for AI CompanionsXinyang Liao, Lingyu Li, Kaicheng Shen, Yang Yao +3
AI companions provide users with emotional support and ongoing companionship, but their sustained relational interactions also introduce distinct safety risks. Prior work largely targets general harms or specific companion platforms and risk categories, while relying on manually collected conversati
benchmark - arxiv:2610.18093 · cs.ROLynxScenes: A Large-Scale Real-World Benchmark for Outdoor Wheel-Legged Robot Perception, Navigation, Imagination and ReconstructionJun Tian, Hao Wu, Ruo Chen, Yili Liao +4
Wheel-legged robots operating in outdoor spaces shared with humans require accurate perception, socially compliant navigation, future-scene reasoning, and scene reconstruction. However, existing datasets typically support these capabilities in isolation and rarely provide dense spatial and behavior-
benchmarkevaluator - arxiv:2610.18092 · cs.AILifeSearch: Benchmarking Deep Research Agents for Real-World Life and ProductivityYibo Wang, Senjie Jin, Yansong Ning, Xinbing Liang +5
Deep research (DR) agents address open-ended information needs through multi-step reasoning, long-horizon web search, and evidence synthesis. However, existing benchmarks primarily focus on academic research or synthetic tasks, leaving DR agents' capabilities in real-world life and professional prod
benchmarkevaluation framework - arxiv:2610.18090 · cs.AIAM-Bench: Understanding Agents under Task Underspecification across Models and HarnessesKaiyu Huang, Yiting Li, Angyang Gu, Pinjia He +2
Ambiguous or incomplete task specifications complicate both execution and clarification for language-model agents. Yet agents' responses to these information gaps remain poorly characterized across models and execution harnesses. We present AM-Bench, a benchmark built from 48 tasks adapted from Term
benchmark - arxiv:2610.18089 · cs.AIOne-Sided Verification Under Limited Expert ReviewFan Nie, Haotian Ye, Shurui Liu, Zihao Wang +2
AI has become increasingly helpful in scientific discovery, yet human expert review remains essential for those problems without reliable automated evaluation, posing a fundamental challenge as candidate generation scales beyond available review capacity. This paper studies verification under limite
post-trainingbenchmark - arxiv:2610.18088 · cs.LGResearchFlow-Eval: Tracing the Execution Gap in End-to-End Multi-Claim AutoresearchPeng Jiang, Dongyang Li, Quanying Liu, Xiaoxuan Jia
Autoresearch systems increasingly undertake entire research projects. In practice, scientific research requires systems to pursue multiple interdependent claims through a continuous, multi-stage workflow and revise experiments as evidence accumulates. Existing evaluations of autoresearch systems, ho
agentevaluation framework - arxiv:2610.18083 · cs.AIAReaL-RSI: Adaptive Scaling and Coordination of Recursive Self-ImprovementWenhao Zhou, Jiajun Song, Haitao Wang, Zipeng Qiu +10
Scaling recursive self-improvement requires additional computation to produce useful research progress in reaching persistent objectives. Intuitively, more researcher agents can explore more hypotheses, but early evaluation scores might misrepresent their promise, shared findings could lose their ap
memoryagentself-improvement - arxiv:2610.18082 · cs.LGFailure Recovery via Belief Updates and World Modeling for Long-Horizon AgentsXiyang Wu, Zhichao Liu, Shengxin zhang, Zongxia Li +3
Long-horizon agents often struggle to recover from failures whose causes lie several steps earlier, beyond what local reflection or backtracking can resolve. We introduce BELIEFWEAVER, a framework that combines persistent evidence across attempts with prospective reasoning over repair directions. It
world modelagentbenchmark - arxiv:2610.18072 · cs.CVLensErase: A Foundation Model for Erase AnythingYunuo Chen, Xiuyu Wu, Qinhong Yang, Jinjing Zhao +10
We introduce LensErase, a general-purpose mask-based Erase Anything foundation model designed to cover a broad spectrum of real-world erase scenarios. LensErase supports diverse objects, visual text, watermarks, associated visual effects, multiple-object removal, and a wide range of object scales, a
benchmark - arxiv:2610.18069 · cs.CLFrom Agents to Edges: Paired Optimization for Multi-Agent CollaborationFeng Lin, Xiaotian Yu, Yifan Yang, Aijun Li
Strong individual agents do not guarantee effective collaboration. Communication must align what one agent conveys with how another interprets and uses it. This dependency makes the communication edge a natural unit of optimization. We introduce Paired Automated Communication Optimization (PACO), wh
agentmulti-agentagent benchmarkbenchmark - arxiv:2610.18067 · cs.LGWorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment EvolutionHang He, Linting Wang, Guanming Liu, Bo Mao +13
Scaling interactive environments is important for training long-horizon agents through reinforcement learning. These environments often require further testing and refinement after initial generation, and correcting defects in a task may also require updating its work context and verifier. However,
agentagent benchmarkiterative refinementbenchmark - arxiv:2610.18066 · cs.LGUncertainty Evaluation: Bridging Machine Learning and Measurement ScienceDario Petri, Luca Mari
Reliable uncertainty evaluation is essential for assessing the trustworthiness of machine learning (ML) predictions, particularly in safety-critical applications. This paper explores conceptual analogies between uncertainty evaluation in ML and measurement science, drawing on the evolution of metrol
evaluation framework - arxiv:2610.18060 · cs.CLA Benchmark for Bangla Task-Oriented Dialogue UnderstandingAfrina Alam Lamia, Sumiaya Sharif, Prottoy Modak, Ohiduzzaman Rayhan +1
Task-oriented dialogue understanding remains underexplored for Bangla, particularly in multilingual settings involving English and romanized Bangla (Banglish). We introduce BanglaTOD, a parallel tri-lingual benchmark comprising 209 multi-turn dialogues across nine Bangladesh-specific service domains
benchmark - arxiv:2610.18058 · physics.app-phTwo-State Dynamical Synapses as a Physical Substrate for Equilibrium PropagationDongshu Liu, Julie Grollier
Equilibrium Propagation (EqProp) updates each synapse by comparing a local term at the free and nudged equilibria. Because these two values occur at different times, a physical implementation requires temporary local memory. We show that this memory can emerge from a two-timescale synaptic model in
memory - arxiv:2610.18057 · cs.LGFrom Policy Evaluation to Off-Policy Control: Provable Emergence of In-Context Q-LearningYingjun Mo, Yajie Zhao, Fang Kong
Transformers can implement reinforcement-learning algorithms in context, yet how off-policy control emerges within Transformers remains poorly understood. We study Q-learning, a foundational model-free control algorithm that combines off-policy bootstrapping with greedy policy improvement. We prove
policy evaluation - arxiv:2610.18054 · cs.LGPrivacy-Preserving GNN Inference on Financial Transaction Graphs via Homomorphic EncryptionH. Çağrı Bilgi, Kemal Mutluergil, Ceren Yıldırım, Kubilay Atasu +1
Anti-money laundering (AML) in financial networks requires modeling transaction patterns that span multiple institutions. In practice, however, banks cannot share account and transaction features due to strict privacy regulations. We propose a privacy-preserving graph neural network (GNN) inference
benchmark - arxiv:2610.18053 · cs.ROLADDER: A Long-Horizon Robot Manipulation BenchmarkPetko Petkov, Giuliano Albanese, Nikolay Nikolov, Luc Van Gool +2
General-purpose robot manipulation policies learn short-horizon skills effectively, but degrade rapidly on long-horizon tasks. Existing benchmarks cannot isolate failures because task horizon is either too short or entangled with scene and object variations. We propose LADDER, a long-horizon robot m
manipulationbenchmark - arxiv:2610.18052 · cs.AICan Agents Forget When Asked? A Benchmark for Memory Unlearning in LLM AgentsBofan Xie, Chaokun Wang, Guanchen Ge, Songyao Wang +3
Long-term memory has become a core component of large language model (LLM) agents. As agents accumulate personal information over long-term use, users increasingly ask them to delete sensitive or outdated facts. A memory system should honor such requests. It should remove exactly what was requested
memoryagent memoryagentllm agentbenchmark - arxiv:2610.18038 · cs.ROData-Free Continual Learning for World Action Model with Rectified Generative ReplayKaipeng Fang, Hangxu Liu, Chaojian Shi, Yunsong Zhou +4
World action models (WAMs) built on pretrained video generation models adapt quickly to new robotic tasks, but they forget previously learned tasks under sequential fine-tuning. Experience replay (ER) mitigates such forgetting by rehearsing data of previous tasks, yet this data is often unavailable
liberoagenticevaluator - arxiv:2610.18034 · cs.CVCOMPASS: Contextual Moment Prior Alignment for Generalizable AIGI DetectionZhan Yang, Tiancheng Shen, Weijie Lyu, Shuo Chen +3
Generalizable AI-generated image (AIGI) and deepfake detection remain challenging due to the asymmetry phenomenon, where adaptation to limited training distributions can degrade the transferable representations of Vision Foundation Models (VFMs). Recent subspace adaptation methods mitigate this issu
benchmark - arxiv:2610.18030 · cs.CVFlowSim: Rendering-Free Closed-Loop Simulation for Evaluating End-to-End Driving SystemsHanfeng Wu, Alvaro Marcos-Ramiro, Michael Schmidt, Christoph Stiller
Closed-loop evaluation of end-to-end (E2E) driving systems typically relies on generating sensor observations and re-processing them through the perception encoder, introducing substantial computational cost and a sim2real gap between simulated and real sensor observations. In this paper, we ask a m
sim2realagent - arxiv:2610.18026 · cs.LGChampionKV: Head-Wise Adaptive Budget Allocation for KV Cache CompressionYuechuan Chen, Yanda Guo, Xiaoxi Liu, Weilin Ruan +3
Some observation-window-based KV cache compressors use prompt attention to identify important tokens but allocate uniform or predefined budgets across heads, overlooking input-dependent differences in budget demand. We propose ChampionKV, an input-adaptive head-wise budget allocator for long-context
long-context - arxiv:2610.18025 · cs.LGSHELDON: Reasoning through Dual Search over Hypothesis and Observation SpacesHarshit Joshi, Priyank Shethia, Monica S. Lam
Test-time scaling has enabled LLMs to solve increasingly complex reasoning, discovery, and optimization tasks. However, existing approaches often gather observations while seeking evidence for current hypotheses, then generate new hypotheses conditioned on those hypotheses and their associated evide
benchmark - arxiv:2610.18023 · cs.LGFRACTION: Fraud Explanation Generation via Agentic Graph Context NavigationCecelia Soh, Jun Hu, Yuan Li, Cheng Chen +5
Graph fraud explanation generation emerges as an important task, which requires jointly leveraging graph structure and content information to produce human-readable explanations for a given risk label. An intuitive approach is to adapt existing graph-enhanced large language models (LLMs), which inje
agentagentictool-useevaluation protocol - arxiv:2610.18018 · cs.CVDynamic Interpolation for Training-Time Prediction Reshaping in Low-Light Image EnhancementPeirui Zhu, Shijie Hao, Jingxi Liao, Richang Hong +1
Low-light image enhancement (LLIE) is an important image/video processing problem for reliable visual systems under poor illumination, yet it remains challenging due to the strong coupling among insufficient illumination, noise amplification, and structural degradation. Existing methods typically re
benchmark - arxiv:2610.18014 · cs.AILatentRemedy: Efficient Remedy For Non-Prefix KV Cache Reuse in Multi-Agent SystemsZhiwei Chen, Zhongtao Rao, Waikit Xiu, Yuxiang Zhang +3
Non-prefix KV cache reuse is a promising approach for improving inference efficiency in multi-agent systems as agents share the same context blocks. However, when agents have conflicting tasks, directly reusing non-prefix KV caches across agents can severely degrade model performance, causing up to
agentmulti-agentagent system - arxiv:2610.18007 · cs.AICost-Efficient Privacy Enforcement for Email-Generating LLM Agents under Strategic Information RequestsQikun Cai, Tao Huang, Chen Hou
LLM agents now act as delegated communication intermediaries: they read private context, draft messages, and send externally visible emails on behalf of a user. Existing checker-repair defenses can reduce inappropriate disclosure, but they rely on repeated calls to a remote action LLM (the model tha
llm agent - arxiv:2610.18004 · cs.AISWEPatchTest: Evaluating Software Tests Beyond ReproductionHua Lu, Pengcheng Wang, Han Li, Zichen Tian +3
Coding agents use generated tests to decide whether a proposed patch satisfies an issue's requirements. A suite that fails on the buggy program and passes after applying the reference patch can still accept an incorrect patch or reject a correct patch. We introduce SWEPATCHTEST, a test-generation be
agentbenchmark - arxiv:2610.18001 · cs.LGMatch the Canvas, Not Just the Answer: Predictability-Aware SFT for Diffusion Language ModelsMiao Li, Hanyang Jiang, Hengyu Fu, Baihe Huang +7
Diffusion language models (DLMs) enable flexible parallel generation by resolving tokens according to the model's evolving predictability rather than a fixed left-to-right order. This flexibility, however, introduces training choices absent from standard autoregressive teacher forcing: which referen
benchmark - arxiv:2610.17999 · cs.ROTRACE: Efficient Transfer Learning for Neural Whole-Body Control Across Humanoid EmbodimentsQingyu Lan, Zhihao Tian, Zifan Wang, Junwei Liang
Humanoid behavior foundation models (BFMs) enable diverse whole-body control, but adapting a pretrained controller to a new embodiment remains challenging because of kinematic and dynamic mismatch. We present TRACE, a framework for efficient cross-embodiment adaptation that preserves pretrained moti
vision-language-actionhumanoidteleoperationwhole-body controlpost-training - arxiv:2610.17998 · cs.LGSaliency-Guided Replay for Continual Learning in Deep Spiking Neural NetworksFrancesco Panetta, Federico Di Valerio, Michela Proietti
Spiking Neural Networks (SNNs) are promising for energy-efficient neuromorphic computing, but they still suffer from catastrophic forgetting when learning tasks sequentially. This work investigates the integration of Experience Packing and Replay (EPR) as a memory-efficient continual learning strate
memory - arxiv:2610.17994 · cs.LGImproving Neural Symbolic Regression with a Diverse Prior and Post-training AlignmentDavid Otte, Jörg K. H. Franke, Frank Hutter
Symbolic regression is the problem of finding an expression that fits a dataset. Solving it requires searching a combinatorial space of candidate expressions. Neural symbolic regression can amortize that search by training a model once on synthetic data and then predicting expressions for a new data
post-training - arxiv:2610.17991 · cs.CVPolUR: Polarization Aware Multimodal Degradation Perception for Underwater Image RestorationWenshuo Zhang, Xiangyue Zhang, Junlin Li, Wenjie Tian +4
Recovering high-quality images from degraded underwater observations is crucial for underwater exploration. Polarization observations show promise as an effective source of complementary physical cues for recovering target radiance under clear-water conditions. However, existing methods have yet to
benchmark - arxiv:2610.17986 · cs.CVWeb Scale on a Budget: Tail-Aware Retrieval EvaluationBoaz Meivar, Shai Avidan
A common practice in instance retrieval is to grow the gallery size to improve a benchmark's discrimination capability. One prominent example is ILIAS, which adds up to 100 million distractor images. The prevailing assumption is that faithful evaluation demands web scale, and that indexing all 100M
benchmark - arxiv:2610.17983 · cs.ROBreaking the Average: Overcoming Convergent Defaults in Robotic Manipulation via Diagnostic Chain-of-Thought and Structured GuidanceMohammed Daba, Waleed Abdulrahman Saleh AlAkwa
Recent Vision-Language-Action (VLA) models, such as $π_{0.5}$, have demonstrated impressive generalization capabilities by scaling behavior cloning across massive datasets. However, these models frequently suffer from distributional convergence, a phenomenon where the policy regresses to "safe" but
vision-language-actionmanipulationgrasp - arxiv:2610.17977 · cs.ROReEnvision: Future-Guided Selective Action Repair for Long-Horizon ManipulationXiyin Zeng, Xuewei Huang, Yi Lu, Weihao Yuan +1
Long-horizon robotic manipulation requires robust execution over sequences of interdependent actions. Existing methods typically handle execution deviations by replanning the remaining action chunk or correcting actions step by step as new observations arrive. However, a local deviation may affect o
manipulation - arxiv:2610.17966 · cs.LGBELT : BEhavior-Aligned Low-Rank KV Cache Compression with Learned Top-k Rank AllocationYaoyang Liu, Shuwei Shen, Jincheng Tang, Jinguang Liu +4
Long-context inference makes the key-value (KV) cache a major memory bottleneck for large language models. We propose BELT, a low-rank KV cache compression framework. Its key-side objective targets attention distributions rather than raw keys or pre-softmax logits, yielding a quadratic surrogate wit
memorylong-contextbenchmark - arxiv:2610.17959 · cs.AIVirtual Head AttentionXibo Ding, Guoxia Wang, Jinle Zeng, Jiabin Yang +3
The design of attention mechanisms in long-context large language models faces an ``impossible triangle'' among KV cache size, training and prefill FLOPs, and model quality. GQA achieves strong quality but incurs high KV cache overhead; alternative approaches such as MLA and MFA can compress the cac
long-contextbenchmark - arxiv:2610.17953 · cs.CLAgent-Reflex: Teaching Language Agents to Act ReflectivelyGuanting Dong, Xinyu Yang, Jiajie Jin, Yuyang Hu +8
Large language models have evolved from single-turn chatbots into interactive agents capable of long-horizon reasoning. In real deployments, MCP tool services can return noisy feedback and users may intervene to correct the agent, but existing methods provide limited supervision for reflection learn
agentagenticagent benchmarktool-usebenchmark - arxiv:2610.17948 · cs.LGHaloBlock: Faster and Better Decoding for Diffusion Large Language ModelsZhaohui Zheng, Chenhang He, Jinrui Zhang, Ming-Ming Cheng +1
Diffusion large language models (dLLMs) have attracted increasing research attention due to their bidirectional attention and arbitrary order decoding. However, there are two issues in dLLMs inference: the decoding difficulty is dominated by the tokens on the right side of the block, and the block-w
benchmark - arxiv:2610.17936 · cs.CVPUGS: Physics-Guided Uncertainty Modeling for Underwater Dynamic 3D Gaussian SplattingTingjun Li, Yuxi Wang, Guoqiang Zhong, Zhaoxiang Zhang
Dynamic 3D Gaussian Splatting (3DGS) provides an efficient framework for dynamic scene reconstruction. Most existing approaches model scene dynamics with deterministic deformation fields, assuming each Gaussian follows a distinct geometric trajectory over time. This assumption breaks down underwater
benchmark - arxiv:2610.17928 · cs.AILearning How to Remember: Memory Modality Selection for GUI AgentsTaeyun Roh, Suhyeong Park, Gijin Kim, Wonjune Jang +1
Recent advances in vision-language models (VLMs) have substantially improved the capabilities of GUI agents, while memory augmentation has emerged as a promising approach for reusing knowledge from past interactions. However, existing memory augmented GUI agents typically adopt a fixed representatio
memorybenchmark - arxiv:2610.17927 · cs.ROGLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal NavigationYousef Liang, Ryan Ardalan, Yarden Goraly, Chenhao Zhu +5
Autonomous navigation in real-world environments often requires long-horizon operation under partial observability and evolving goals. Existing mapless navigation approaches predominantly assume fixed goals throughout an episode, limiting applicability to practical scenarios where goals change onlin
quadrupedmemory - arxiv:2610.17926 · cs.CVCineMem: Attention-Aligned Fine-grained Memory Retrieval for Multi-Shot Long Video GenerationXiaoxue Wu, Xinyu Gao, Fang Yu, Yu Qiao +1
Autoregressive video generation relies on memory retrieval to model long-range dependencies. This capability becomes particularly critical for multi-shot generation, where models must reconcile substantial visual changes across shot transitions with persistent semantic continuity. Existing methods t
memory - arxiv:2610.17922 · cs.ROWoVR+: World Model as Reliable Simulator for Versatile VLA Policy Optimization with RLShangqing Zhou, Zhennan Jiang, Ning Ma, Xin Fu +8
Action-conditioned world models can reduce costly real-robot interaction in reinforcement learning for Vision-Language-Action models (VLA-RL), but hallucinations can mislead policy optimization. Extending them to richer VLA interfaces and long-horizon tasks raises challenges in spatiotemporal and cr
vision-language-actionvlavla policyrobotwinworld modelaction-conditioned - arxiv:2610.17917 · cs.CLBatch-Paced: Fixed-Budget Batch Decoding in Discrete Diffusion Language ModelsShuai Zhang, Zhanchao Zhou, Huachuan Qiu, Anqi Li +1
As generative language models become more capable and widely used, inference demand grows, making efficient serving increasingly important. Discrete diffusion language models (dLLMs) offer a promising path to more efficient inference through parallel token generation. Prior dLLM decoding methods, ho
benchmark - arxiv:2610.17903 · cs.RODriveWLA: Unifying Future Dynamics and Scene Semantics through World-Language-Action Modeling for Autonomous DrivingShuzheng Wang, Haoxuan Li, Shengzhen Li, Chenxi Wu +4
Vision-Language Models (VLMs) and world models have demonstrated strong capabilities in semantic understanding and future scene evolution for autonomous driving. However, effectively integrating both capabilities for trajectory generation remains challenging. In this work, we propose DriveWLA, a uni
world modelpost-training - arxiv:2610.17902 · cs.CVVisual Dependence Does Not Imply Answerability: A Systematic Analysis of Answerability Signals in Frozen InstructBLIPMinjun Kim, Hyunmin Song, Changwon Lim
Visual question answering models often produce plausible responses even when the available visual evidence is insufficient to support a valid answer. Although prior work has introduced benchmarks and abstention mechanisms for unanswerable visual questions, it remains unclear which signals derived fr
benchmark - arxiv:2610.17901 · cs.LGFlashMask-3: Efficient and Expressive Mask-Aware Distributed AttentionGuoxia Wang, Qianyue He, Siming Wu, Haoyang Xie +5
Long-context training of large language and multimodal models couples structured attention masks (document, prefix-LM, share-question, sliding-window, block-sparse) with multi-node context parallelism, making the mask a dominant bottleneck. FlashMask's column-wise sparse mask compresses them from $O
long-context - arxiv:2610.17900 · cs.LGPaceReasoner: Label-Free Learning to Answer Briefly from Long ReasoningChangguo Jia, Zhiyou Xiao, Xinbei Ma, Geng Chen +4
Extended chain-of-thought reasoning improves problem-solving performance but increases inference cost and latency, motivating more accurate answers with shorter reasoning. We introduce PaceReasoner, a label-free test-time reinforcement learning framework that uses a model's LONG responses to improve
benchmark - arxiv:2610.17895 · cs.LGInteraction-as-a-JudgeWayne Chi, I-Hung Hsu, Li Chen, Wei-Lin Chiang +5
Coding agents are tasked with producing sophisticated interactive applications such as websites, desktop applications, and games. While humans assess these applications interactively, current automated LLM evaluations primarily focus on code analysis and unit tests. This mismatch can cause LLM judge
agent - arxiv:2610.17894 · cs.CLSwiss4Arg: A Parallel Dataset for Multilingual Argument Mining across Swiss National LanguagesSophia Conrad, Fatma-Zohra Rezkellah, Christina Niklaus, Reto Gubelmann
We present Swiss4Arg, a multilingual dataset for argument mining based on official Swiss voting brochures, covering all four national languages: German, French, Italian, and Romansh (a Romance minority language exclusively spoken in Switzerland). The brochures contain explicitly formulated arguments
benchmark - arxiv:2610.17893 · cs.LGTraining LLMs for Multi-Agent Collaboration with Verifiable Information RewardsKeru Chen, Sen Lin, Yingbin Liang, Nathaniel D. Bastian +1
Effective collaboration among large language model (LLM) agents in multi-agent systems requires sharing relevant evidence, requesting missing information, and incorporating teammates' contributions into decisions. Training these behaviors is challenging because final task outcomes provide limited fe
multi-agentagent systemself-playpost-trainingevaluator - arxiv:2610.17882 · cs.CVDiagnose-to-Generate: Eliminating Structural Distortions in Flow Models via Patch-Level Grounded AlignmentYifan Li, Lishuai Gao, Tong Zhang, Qilong Huang +7
Flow-based image generators achieve strong semantic synthesis but still produce localized structural distortions, such as malformed hands and limbs. Existing post-training methods typically optimize global scalar rewards, which provide weak spatial credit assignment and may alter well-formed regions
post-training - arxiv:2610.17881 · cs.ROAgentic Robot Creation through Generation, Verification, and RevisionXuebing Li, Yibo He, Sitong Liu, Ran Ji +1
Can an AI agent turn high-level user intent into a functioning robot and teach it how to behave? We study this question from two complementary aspects: embodiment and behavior creation. For embodiment creation, we investigate how well current multimodal models capture mechanical requirements, and in
agentai agentagentic - arxiv:2610.17874 · cs.LGSearching for What Matters: Open-Ended Discovery of Multiple Hypotheses with Agentic Data ExplorationSanchit Kabra, Nurendra Choudhary, Jiangshu Du, Eddie Huang +2
Discovering what drives an outcome from data is central to science and often requires deciding what is worth investigating before a specific hypothesis is known. Yet most existing hypothesis generation settings substantially narrow this search by providing a research question, target relationship, o
agenticbenchmarkevaluator - arxiv:2610.17870 · cs.ROLAG-IO: Multi-IMU State Estimation for Legged Robots Using Factor GraphsYicheng Zhou, Longwen Chen, Qinghua Yu, Junhao Xiao +2
Robust proprioceptive state estimation is crucial for legged robots, particularly in challenging environments where exteroceptive sensors are unreliable. While traditional leg odometry typically relies on a single body-mounted inertial measurement unit (IMU) fused with leg kinematics via an extended
quadruped - arxiv:2610.17869 · cs.AIECMR: Evidence-Aware Deliberation Control for Multi-Agent LLM ReasoningYuan Liu, Haoyue Peng, Fanliang Hu, Lei Huang +1
Multi-agent Large Language Model (LLM) systems typically commit to a deliberation protocol before inference. A fixed protocol cannot allocate computation differently to resolved and unresolved instances, and it may mistake repeated support for independent confirmation. We study this failure as seman
multi-agent - arxiv:2610.17868 · cs.CLPrefilling-dLLM v2: Learned Chunk Routing for Long-Context Diffusion Language ModelsQi Han, Jing Xiong, Ziyuan Zhang, Ngai Wong
Diffusion language models apply bidirectional attention over the whole context at every denoising step, so a longer context makes each step substantially more expensive. Sparsifying only at inference leaves the pretrained weights unchanged, so the model is never trained under the sparse pattern it r
long-context - arxiv:2610.17867 · cs.LGRT-J: Large-Scale Pretraining of Relational Transformers for Context-Efficient PredictionsRishabh Ranjan, Vignesh Kothapalli, Harshvardhan Agarwal, Charilaos Kanatsoulis +4
Foundation models for tables and relational databases are increasingly used for predictive tasks across various scientific and industrial domains. However, adapting existing models to unseen datasets requires tens of thousands of examples. Here we introduce RT-J, an 85M-parameter relational foundati
long context - arxiv:2610.17862 · physics.opticsOn-chip optical isolatorsAbanoub Mikhail, Nikolay Solodovchenko, Sen Xu, Fan Zhang +4
Optical isolation suppresses back-reflections that destabilize lasers, produce parasitic interference, and increase crosstalk in photonic systems. Although free-space and fibre-optic isolators are mature, no broadly adopted on-chip technology yet combines strong isolation, low insertion loss, broad
photonic integrated circuit - arxiv:2610.17860 · cs.AISocratic Skill Learning: Interleaving Teaching and Practice for Transferable Agent SkillsLetian Zhu, Han Li, Yifan Yao, Xinping Lei +8
Agent Skills improve LLM agents by making task knowledge reusable across sessions. Learning transferable Skills requires evidence about a procedure and its conditions of use, which a successful task attempt can leave untested. We introduce Socratic Skill Learning (SSL), which interleaves expert teac
agentllm agent - arxiv:2610.17856 · cs.LGMMTA: Benchmarking Multimodal Temporal Analysis with Time Series, Text, and VisionZiyang Zhang, Shenyi Li, Yilin Wang, Ziyun Cui +3
Time-series analysis is increasingly studied in multimodal settings, yet most benchmarks pair temporal signals with text or with visualizations derived from the same signal. This leaves open whether multimodal large language models (MLLMs) can integrate time series with real-world visual evidence. W
benchmark - arxiv:2610.17853 · cs.CVDVTR-KD: Distilling Vision-Language Models via Intermediate Visual Token ReconstructionHaolan Wei, Jiahui Wang, Junli Zha, Xinkai Lu +3
Multimodal large language models (MLLMs) have achieved strong performance on diverse vision-language tasks, but their computational demands limit deployment under constrained budgets, particularly on edge devices. Many works focus on Knowledge Distillation (KD) and try to transfer knowledge from pow
benchmark - arxiv:2610.17848 · cs.LGPersistence Positional Encoding: A Minimal Transformer Recipe for Persistence DiagramsEunwoo Heo, Ganghyun Kim, Jaejun Yoo
Persistence diagrams (PDs) represent multiscale topological information as multisets of birth-death pairs $(b,d)$. How best to use them with Transformers, however, remains unclear. Connecting PDs to Transformers requires choosing how topological information is tokenized and exposed to self-attention
benchmark - arxiv:2610.17844 · cs.CVWhat Makes a Good Visual Representation for Spatial Understanding in MLLMs?Xiaohu Huang, Yumin Guo, Kai Han
Spatial understanding is a necessary step in moving multimodal large language models (MLLMs) toward reasoning about the 3D physical world. Their visual pretraining data, however, is almost entirely 2D, so the visual tokens processed by MLLMs often lack inherent 3D structural information. A growing n
benchmark - arxiv:2610.17843 · cs.ROPhysics-Aware Object-Centric SE(3) Forecasting for Real-Time Proactive Dynamic GraspingYirui Zhou, Hongli Xu, Ying Zheng, Yi Wang +1
Proactive dynamic grasping requires robots to predict the future state at which physical contact will occur, rather than reacting to the currently observed object pose. But what future must the robot predict, and at what level of abstraction? We argue that contact-relevant rigid-body state in SE(3)
grasp - arxiv:2610.17841 · cs.LGFASHION-RQ: Mining Hierarchical Semantic Structure for Automated Evaluation of Fashion RetrievalTripti Shukla, Shwetha Somasundaram, Akshat Gupta, Rohit Garg +2
Evaluating fashion retrieval systems remains challenging due to the limited availability of benchmarks that capture the fine-grained, open-ended queries encountered in real-world search and recommendation. Existing benchmarks often rely on manual curation, predefined taxonomies, or synthetic prompts
benchmark - arxiv:2610.17840 · cs.AILearning Adaptive Deliberation Control for Reliable Recognition with Multi-Agent Language ModelsYuan Liu, Haoyue Peng, Fanliang Hu, Lei Huang +1
Multi-agent deliberation can improve recognition by criticizing and revising candidate answers, but fixed controls cannot respond to changes in disagreement or reasoning quality. We present an adaptive multi-agent controller (ADMA) that learns continuous policies for proposal, challenge, arbitration
multi-agentbenchmark - arxiv:2610.17837 · cs.AIProtoArena: Can AI Agents Discover and Exploit Real-World Cross-Component Protocol Vulnerabilities?Fasheng Miao, Zhaoyu Hu, Xiang Li, Changqing An +4
Large language model (LLM) agents are increasingly used for vulnerability analysis, yet existing benchmarks provide limited coverage of exploitation that depends on interactions among network protocol components. Such vulnerabilities arise from inconsistencies in message interpretation, timing, stat
ai agentbenchmark - arxiv:2610.17832 · cs.LGAgentic-MatriBO: Capability-Compiled Agentic Bayesian Optimization for Materials and Chemical DesignYu Zeng, Yuchuan Chen, Ziyue Zhang, Zijian Wei +4
Materials and chemical optimization must choose the next synthesis, characterization, or high-fidelity simulation from sequential observations under a limited budget. Large language models can use task knowledge to propose adjustments, but composing surrogates, acquisition functions, and search stra
memoryagentic - arxiv:2610.17831 · cs.AIDiff Diffusion: Predicting Edit Paths with Diffusion Models for Efficient Code EditingXiaolong Xu, Hao Jiang, Bo Jiang, Zhiwen Deng +4
Diffusion large language models (dLLMs) are appealing for code generation because they can update many tokens in parallel, yet many are optimized for sequence synthesis rather than edit prediction. This setting is poorly matched to modern AI-assisted IDEs, where edit prediction must infer the next c
benchmark - arxiv:2610.17829 · cs.CVQK-Edit: Revisiting Attention-based Injection in MM-DiT for Image and Video EditingTiancheng Shen, Zilong Huang, Xiangtai Li, Zhijie Lin +5
Multimodal Diffusion Transformers (MM-DiTs) have recently emerged as a powerful framework for unified text-vision synthesis, surpassing traditional U-Net architectures in generative tasks. One key innovation lies in its Multimodal Self-Attention (MM-SA) interaction, where image and text tokens are c
manipulation - arxiv:2610.17827 · cs.AIBenchmarking and Training Reward Models for Cross-Source Evidence Grounding in Search AgentsHaochen Zhao, Minsheng Hao, Zheng Li, Yifan Li +7
Existing approaches to training and evaluating search agents focus heavily on final-answer quality, largely neglecting how generated answers map back to retrieved evidence. Recent methods track citations to enforce grounding, but citation support alone does not establish source independence or wheth
benchmarkevaluation framework - arxiv:2610.17816 · cs.AIEmpowering Edge Agents: A Systematic Analysis of Edge-Cloud Collaboration in LLM Agent ExecutionPengcheng Huang, Zhenghao Liu, Shixin Ren, Shi Yu +4
Edge-cloud collaboration combines the lower inference overhead and greater data locality of edge-side execution with the stronger capabilities of cloud-hosted models. However, as LLM agents increasingly operate in long-horizon scenarios, cloud participation becomes interleaved throughout execution r
agentllm agentbenchmarkevaluation framework - arxiv:2610.17808 · cs.LGSPARCL: Spatially Adaptive Regional Coverage for Active Learning in Chest X-ray AnalysisJuho Jung, Seongjae Kang, Mijoo Kim, Sungduk Cho +3
Pixel-level annotation for chest X-ray (CXR) lesion segmentation is expensive, making active learning (AL) essential. However, existing AL strategies fail to account for the spatial distribution of selected samples, resulting in inconsistent performance across focal (e.g., aortic calcification) and
benchmark - arxiv:2610.17805 · cs.CLBRIDGE: Selective Boundary Repair with Dependency Graph Evidence in Multi-Agent SystemsEncheng Su, Pengze Li, Ziheng Hua, Qiucheng Yu +11
Multi-agent collaboration has emerged as a critical paradigm for tackling complex, long-horizon tasks; however, while coordination protocols continue to advance, systemic failures at the inter-agent collaboration interface remain a pervasive bottleneck. In collaborative workflows, systems frequently
agentmulti-agentagent systembenchmark - arxiv:2610.17802 · cs.ROBeyond the End-Effector: Learning Whole-Arm Contact Manipulation from Human DemonstrationsBo-Ruei Huang, Daniel Seita, Erdem Bıyık
Robot manipulation has traditionally relied on end-effector-centric interactions. In contrast, humans routinely exploit whole-arm contacts to improve efficiency, robustness, and adaptability, especially when the hands are occupied or when precise grasping is unnecessary. Enabling similar capabilitie
manipulationteleoperationgrasp - arxiv:2610.17794 · cs.LGGIFT-Eval-CTX: A Benchmark For Contextual Forecasting Model EvaluationXu Liu, Hanshu Yan, Taha Aksu, Quang Pham +5
Contextual forecasting seeks to improve time series forecasting by incorporating forward-looking information, such as planned events, external conditions, and expert expectations. However, reliable evaluation remains difficult: existing benchmarks often have limited domain coverage, lack standardize
agenticagent systembenchmark - arxiv:2610.17792 · cs.CVMHVF: Training-Free GUI Grounding via Verification-Guided Multi-Hypothesis SearchMinghua Chen, Xingming Liao, Weiqiang Zhong, Zhuowei Wang
GUI grounding enables web agents to translate natural-language instructions into actionable interface locations. However, dense layouts and small, visually similar controls can lead models toward plausible but incorrect elements, while refinement around a single prediction may leave alternative targ
benchmark - arxiv:2610.17789 · cs.RODexStitch: Learning and Composing Dexterous Skills from Human Videos via Contact-Aware Robotic InpaintingByeonggyeol Choi, Seongmin Jung, Unghui Nam, Youngbo Song +4
A long-horizon manipulation task consists of primitive skills and the transitions connecting them. Learning skills that can be executed in sequence requires supervision for both, which is costly to collect on a robot. Human videos offer a scalable source of demonstrations, but transferring them to r
manipulationdexterousteleoperation - arxiv:2610.17786 · cs.LGAgentic4PDM: Host-Governed Evidence Admissibility for Industrial Fault DiagnosisNam V. Nguyen, Hoang Van Truong, Hieu D. D., Duy Khuong Nguyen +2
We study industrial fault diagnosis as a classification task that assigns sensor time-series windows to predefined health states or fault types. Scarce labels and changing operating conditions make repeated training of specialized classifiers costly. Yet structures such as periodicity, trends, and t
agentllm agentagentic - arxiv:2610.17777 · cs.CLBeyond Belief Nesting: Rethinking Higher-Order Theory of Mind in Large Language ModelsShanshan You, Zhengqi Wang, Hao Liu, Yuqin Liu +3
Theory of mind (ToM), the ability to reason about others' mental states, underlies effective communication and coordination. Its higher-order form enables reasoning over nested perspectives in strategic interactions. However, existing evaluations often equate belief-nesting depth with reasoning orde
benchmark - arxiv:2610.17763 · cs.AIMemory Echo in Single-Agent Long-Term Memory: Lifecycle Evaluation and Provenance-Guided MitigationYufeng Li, Sheng Yao, Peng Zeng, Dezhang Kong +4
Long-term memory supports agents across sessions. Repeated retrieval and write-back can let one historical claim occupy multiple retrieval positions and influence current answers without independent evidence. Existing work examines retrieval, updating, and question answering, leaving same-source rec
memoryagent memoryagentbenchmark - arxiv:2610.17761 · cs.CLConnecting the Dots: Enduring Performance Gaps in ASR Systems Between Standard and Minority English VarietiesAlexander Metzger, Aruna Srivastava, Ruslan Mukhamedvaleev, Eunjung Yeo +4
Automatic Speech Recognition (ASR) system evaluations have consistently revealed performance disparities along lines of race, gender, socioeconomic status, and language variety. While these disparities are well-documented, they are often (implicitly or explicitly) treated as issues that subsequent,
benchmark - arxiv:2610.17760 · cs.LGGaussian Process Bandits for AI System Optimization with Selectively Verified FeedbackShuang Wu
Gaussian process bandits exploit similarities among configurations to optimize prompts, retrieval systems, and agent workflows. In these applications, however, inexpensive model-based evaluations can disagree with trusted task outcomes. We study selective verification when an action is chosen before
agent - arxiv:2610.17757 · cs.AIInstructions Are Evidence, Not Objectives: Measuring Outcome Alignment of AgentsKaiming Liu, Yuxuan Liu, Jinrui Ju, Minyuan Ruan +3
Large language model agents can follow complex human instructions to solve real-world tasks. In practice, however, initial instructions are often incomplete. In our analysis of 100 user-agent trajectories from SWE-chat, 56\% of initial instructions omit requirements that later emerge as necessary, a
agentbenchmark - arxiv:2610.17751 · cs.LGDegrees of Freedom in Multi-Agent Coordination: Between Centralized and Decentralized RegimesYiqing Peng, Hehai Lin, Zixuan Wang, Dongyi Zhang +6
Large language model (LLM)-based multi-agent systems (MAS) are commonly organized around two coordination regimes. Centralized coordination provides explicit control but can create a bottleneck at a single coordination locus, whereas decentralized coordination offers local autonomy but must achieve
multi-agentagent system - arxiv:2610.17750 · cs.ROLearning Robot Policies for Whole-Body Mobility Assistance via Dynamic Human Motion GenerationShikun Ban, Shohei Nagai, Qingyang Song, Hanxi Wan +3
Robotic assistance for human mobility could benefit care recipients and reduce caregiver burden during physically demanding tasks such as Sit-to-Stand (STS) and Lift-and-Transfer (L&T). Learning robot policies for these whole-body assistive tasks is challenging: real-world data collection for human-
manipulatorsim-to-real - arxiv:2610.17748 · cs.ROI3L: Iterative Interactive Imitation Learning for Whole-Body Mobile ManipulationWensi Ai, Christopher Luey, Karthik Dharmarajan, Evans Han +6
Policies trained offline from teleoperated demonstrations can degrade outside their training distribution. This gap is particularly acute in long-horizon whole-body mobile manipulation, where long horizons amplify compounding errors and coordinated arm, torso, and base motion makes exhaustive demons
manipulation - arxiv:2610.17747 · cs.LGAuditing Failure Attribution: Label Predictability and Judge BehaviorTianyang Xu, Jing Gao
Failure-attribution benchmarks ask an LLM judge which step or agent caused a failure. A high score should mean the judge checked the faulty step against the run (a cross-step check). However, spurious signals (surface cues) that predict the annotation can produce a high score. In this paper, we prop
agentbenchmark - arxiv:2610.17740 · cs.AICoupled Harness--Model Evolution for Scientific Equation DiscoveryPengze Li, Jiaquan Zhang, Encheng Su, Yunbo Long +7
Large language models (LLMs) are increasingly used to discover scientific equations from experimental data. Such a discovery system has two parts that must work together: a library of executable computational tools and the model's tool-use policy. The tools supply the solving capability, from diagno
agenttool-usebenchmark - arxiv:2610.17738 · cs.AIDuplexAct: How Interactive Are Frontier Full-Duplex Spoken Language Models?Jiaheng Dong, Shihong Tan, Gongping Huang, Jean Honorio +2
End-to-end (E2E) full-duplex spoken language models (FD-SLMs) enable simutanous listening and speaking continously, requiring models to take appropriate interaction actions as conversations unfold. However, existing benchmarks lack a structured action-level formulation, often evaluate action occurre
benchmark - arxiv:2610.17734 · cs.LGExplore with Memory, Learn without It: In-Context Exploration and Self-Distillation for LLM ReasoningYutong Chen, Zhichao Xu, Rui Liu, Kunpeng Liu
Exploration bounds learning efficiency in reinforcement learning with verifiable rewards (RLVR). Current methods explore mostly through independent sampling, without fully using the information gathered along the way, such as known wrong answers and environment feedback. We propose LEAF, which injec
memorybenchmark - arxiv:2610.17733 · cs.LGCoCA: Criterion-Aware Credit Assignment for Rubric-Based Reinforcement LearningGuangcheng Zhu, Yingfan Ma, Xing Zheng, Haobo Wang +6
Rubric-based reinforcement learning (RL) evaluates responses against explicit criteria to construct rewards, offering a viable way to train large language models on open-ended tasks. However, standard methods compress criterion-level feedback into scalar rewards and assign uniform token advantages,
benchmark - arxiv:2610.17728 · cs.AISafety for Whom? Benchmarking Understanding Beyond Refusal in Egocentric Multimodal AssistantsYewon Han, Jisu Jeon, Suhyun Choi, Juan Yeo +4
Egocentric multimodal assistants see through the user's eyes and guide what the user does next, so their outputs directly shape physical actions, including those that affect bystanders. Yet existing safety evaluations do not distinguish whom a response would harm, and score only whether a model refu
benchmark - arxiv:2610.17727 · cs.AISkillScreen: High-Recall Screening of Malicious Agent Skills with Decision ModelsJerrold Huang, Junwei Quan, Saiyue Lyu, Zihan Guo +1
Agent skills turn natural-language instructions into actions, creating a security challenge: legitimate automation and malicious behaviour can look deceptively similar. A practical screen must catch nearly every attack while keeping the review load manageable. We present SkillScreen, a compact decis
agent - arxiv:2610.17726 · cs.LGTSRL: Trigger-based Supervised Reinforcement Learning for Continual Agent TrainingYuqian Fu, Kaiwen Jiang, Zijun Song, Jiawei Wang +2
Large language model (LLM) agents excel at complex, multi-step tasks but suffer from catastrophic forgetting during continual learning--acquiring new skills often degrades previously mastered capabilities. Existing post-training paradigms face a fundamental trade-off: Supervised Fine-Tuning (SFT) is
agentagent benchmarkpost-trainingbenchmark - arxiv:2610.17724 · cs.LGEvidence-First Agentic VerificationReza Sanayei, Razvan-Gabriel Dumitru, Sunjie Hou, MohammadHossein Rezaei +6
Long-horizon, semi-verifiable agentic tasks often rely on rubric-based evaluation, but the evidence needed to assess these tasks can be spread across the environment state and the execution trace. A judge may overlook this evidence or accept unsupported claims, including an agent's own reports of su
agentagenticbenchmark - arxiv:2610.17723 · cs.AIUpstream Influence, Downstream Re-Anchoring in LLM Agent ChainsChun-Yu Hsu, Lun-Wei Ku, Chung-Chi Chen
People act on evaluative judgments produced by LLM agents, and in multi-agent systems those judgments pass through several agents before reaching a person. We show that a workflow's information architecture---source access, handoff format, and which model reads the untrusted content---governs how mu
manipulationagentllm agentmulti-agentagent system - arxiv:2610.17722 · cs.LGRepresentation Geometry and Inductive Biases in Quantum Classifiers across Classical Visual BenchmarksYu Yamaoka, Weng Ian Chan, Kosuke Mitarai
Quantum Machine Learning (QML) has a theoretical advantage over classical algorithms for specific synthetic data that is computationally hard. However, it remains unclear whether such theoretical advantages play a meaningful role for realistic classical data, such as natural images. Quantum structur
benchmark - arxiv:2610.17719 · cs.AIKeep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge GraphsVishisht Choudhary
Agents build knowledge graphs as memory, and building them has become easy. Maintaining them has not. When a source is retracted, overruled, or erased, everything built on top of it (entity merges, entailed edges, community summaries) must be re-examined, and the maintainer must choose between delet
knowledge graph - arxiv:2610.17718 · cs.LGECG-RSI: Recursive Self-Improvement through Targeted Electrocardiogram Waveform EvolutionXiaoliang Wang, Zhuo Chen, Chuanyang Zheng, Jiangwei Lao +7
Electrocardiogram (ECG) models need supervision that targets evolving weaknesses and remains faithful to waveform evidence. We introduce ECG-RSI, a recursive self-improvement framework combining verified waveform synthesis with feedback-guided intervention search. Signal remeasurement verifies wavef
self-improvementbenchmark - arxiv:2610.17710 · cs.AIAdaptive-Group Process Rewards for GUI AgentsZirui Shang, Yang Liu, Xin Shu, Zhi Gao +2
Training graphical user interface (GUI) agents with reinforcement learning suffers from sparse outcome rewards, leading to inefficient learning. Process rewards offer a promising direction for credit assignment to intermediate actions; however, accurately evaluating an action requires identifying ev
agentbenchmark - arxiv:2610.17709 · cs.ROWhere Small Mistakes Matter: Quantifying State Criticality for Imitation LearningYusuf Umut Ciftci, Joonwon Kang, Ajay Suresha Sathya, Somil Bansal
Imitation learning typically treats every demonstration state as equally important, optimizing a behavior-cloning objective uniformly across the dataset. Yet not all states are created equal: in a small number of critical states, even a slight deviation from the expert action can lead to a failure,
manipulation - arxiv:2610.17705 · cs.LGA Differentiable 3D Scene Graph Metric via Contextual Hellinger Triplet GeometryU-Chae Jun, Yoonhyuk Choi, Jiwoo Kang
3D scene graph prediction is commonly supervised with local object and predicate classification losses. Although effective for slot-wise label prediction, this formulation does not provide a graph-level semantic distance between a predicted scene graph and its ground truth: semantically mild label e
scene graph - arxiv:2610.17702 · cs.AIA Node Embedding Framework for Model-Centric Conformance Checking in Process MiningMostafa Abbasi, Homayoun Najjaran
Evaluating the quality of process models discovered from event logs is a fundamental challenge in process mining. Traditional evaluation relies on conformance metrics such as fitness and precision, which are highly sensitive to noise and deviations in event logs and therefore often fail to give a co
evaluation framework - arxiv:2610.17697 · cs.LGLeanPRM: Lean-Grounded Process Supervision for Natural Language Proof VerificationRodney Lafuente-Mercado, Salman Rahman, Jingyan Shen, James Shiffer +2
AI systems now write long and difficult mathematical proofs, and checking their step-by-step validity is becoming a bottleneck. A learned verifier could check proofs cheaply, but training one requires a correctness label on every step, and such labels are hard to obtain for open-ended proofs. Existi
benchmark - arxiv:2610.17696 · cs.LGzPocket: Better Pocket Finding with Frozen Co-Folding FeaturesTally Portnoi, Allison M Keys, Benjamin R DiFrancesco, Nate Gruver
Identifying protein-ligand binding sites (pockets) is important for understanding protein function and developing drugs. However, no public method has leveraged co-folding models for novel pocket prediction; furthermore, this task lacks a suitable benchmark as existing benchmarks are within the trai
benchmark - arxiv:2610.17694 · cs.ROHelpMeOut: A Neuropsychology-Grounded Benchmark for Error-Aware Monitoring of Household Task ExecutionHaoran Wang, Jingwei Huang, Anqing Duan, Ho Seok Ahn +1
We introduce HelpMeOut, an embodied AI benchmark for error-aware monitoring of household task execution. Existing embodied household benchmarks typically assume that humans execute tasks correctly, but this assumption does not hold for cognitively impaired persons or older adults whose everyday acti
embodiedbenchmark - arxiv:2610.17692 · cs.CVSWORD: Scene-Grounded Wasserstein Object-Relation Distance for 3D Scene Graph PredictionU-Chae Jun, Jaeeun Ko, Yoonhyuk Choi, Jiwoo Kang
3D scene graph prediction is commonly optimized with object-node and relation-edge classification losses. These objectives encourage exact object and relation label recovery, but they do not directly optimize semantic proximity to the ground-truth scene graph: two wrong graphs can receive comparable
scene graphsemantic graph - arxiv:2610.17690 · cs.LGMedEvo: Medical Reasoning Evolution through Knowledge NavigationXiaoliang Wang, Zhuo Chen, Chuanyang Zheng, Quanbin Wang +9
Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We
benchmark - arxiv:2610.17688 · cs.ROOccWAM: Occupancy World-Action Models with Joint Scene-to-Ego Flow MatchingQihang Huang, Wenhui Huang, Collister Chua, Shiyu Fang +5
Occupancy world models provide a structured, metric representation of future driving scenes, making them a promising basis for end-to-end autonomous driving. However, existing occupancy-aware planners typically couple occupancy prediction and action generation through either hierarchical coupling or
world modelaction-conditionedoccupancy world - arxiv:2610.17686 · cs.LGEncoding Context into LLM Weights: What Makes It Context?Yue Wang, Qizhou Wang, Feng Liu, Gang Niu +2
Longer contexts allow large language models (LLMs) to condition their predictions on more information, but inference costs grow prohibitively with context length. An emerging alternative uses test-time updates to encode context as temporary changes to model weights. However, encoding this informatio
benchmark - arxiv:2610.17685 · cs.LGLatentRiskBench: Measuring State-Dependent Risks in Enterprise AgentsJingyue Cong, Estrid He, Zhuang Li, Nan Wu +4
Enterprise agents face safety risks that cannot be determined from individual actions alone. Existing safety benchmarks largely evaluate action-local risks, overlooking failures whose safety depends on workflow state. In such state-dependent risks, the same action may be safe in one context but viol
benchmark - arxiv:2610.17684 · cs.CVHierarchical Curriculum Learning for Surgical Video Generation: Focusing on Bilateral Axillo-Breast Approach Robotic Thyroidectomy DatasetsYoungbin Kong, Seungho Lee, Jae Kyung Shin, Jong-hyuk Ahn +2
High-fidelity surgical video generation is increasingly important for surgical education, simulation, and downstream AI training, as conventional simulation pipelines are limited by cost, realism, and scenario diversity. Adapting generic text-to-video diffusion models to surgical procedures remains
curriculum learningbenchmark - arxiv:2610.17680 · cs.LGDeficit-Driven Stratified Acquisition for Building STRATA-S12: An Open SAR-Optical Corpus for Self-Supervised LearningChristian Sanchez-Leon, Dongming Chen, Dongqi Wang
Building a co-registered SAR-optical corpus for self-supervised Earth Observation is above all an acquisition problem, yet the main multi-modal datasets are collected opportunistically, with no per-stratum quota control and no mechanism against the geographic dropout that non-uniform cloud cover lea
human-in-the-loop - arxiv:2610.17673 · cs.CLRALI@iKAT 2026: Retrieval-aware User Profile Denoising via Personalized Conversational Dense RetrievalYuchen Hui, Fengran Mo, Jia Ao Sun, Yuxing Tian +2
Modern conversational information access systems continuously extract user information from conversations to build user profiles for personalized retrieval. As user profiles accumulate over time, they inevitably become noisy, making effective user profile denoising a fundamental challenge. Existing
agentai agent - arxiv:2610.17670 · cs.AICowork-Gym: Synthesizing Interactive Work Environments at ScaleAnni Zou, Cheng Chen, Jiaji Deng, Weikang Zhou +17
Cowork agents navigate dependencies across tools and files while adapting to changing user requirements and edits to shared workspaces. Scaling such environments requires maintaining consistency among task objectives, executable resources, user behavior, and success criteria. We introduce Cowork-Gym
agenticbenchmark - arxiv:2610.17667 · cs.ROPrT-Flow: Probability-Tractable Flow Matching for Reinforcement LearningCongzhong Sun, Ning Tan
Diffusion models and Flow Matching can represent flexible distributions over robot actions, but many reinforcement learning algorithms also require the probability of each action. We introduce PrT-Flow (Probability-Tractable Flow Matching), which computes probability density while retaining multimod
diffusion policy - arxiv:2610.17666 · cs.LGDeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress MeterXuyan Ye, Xiyan Fu, Shengda Fan, Jiawei Cui +2
Reinforcement learning with verifiable rewards (RLVR) for multi-turn agents relies on sparse terminal rewards that say little about individual steps. Within-group step comparisons depend on reliable state matching and sufficiently diverse sampled outcomes, while direct step scoring may require costl
evaluator - arxiv:2610.17663 · cs.CLThe Image Still Has a Say: Final-Answer Repair for Multimodal Chain-of-ThoughtYulan Pan, Xiaoyu Xu, Zhihong Shen
Although chain-of-thought prompting has improved multimodal reasoning, it is still difficult to produce correct final answers, as the reasoning can drift away from the image. Besides, the trace is already fixed when the answer is generated, leaving no room for self-correction. Considering the image
self-correctionbenchmark - arxiv:2610.17661 · cs.CVScenePercept: Scenario-Conditioned Perceptual Assessment of Real-World User-Generated Content ImagesShuhan Xia, Yu Chen, Xiang Wang, Shuai Li +2
The rapid development of mobile imaging and social platforms has led to explosive growth in user-generated content images, making their systematic and comprehensive perceptual assessment increasingly important. Existing image perception benchmarks are typically task-oriented, curating specific scena
benchmark - arxiv:2610.17659 · cs.AICan Agents Truly Forget? Explore Machine Unlearning for LLM AgentsJunkai Lin, Ge Xu, Zheng Xu, Junkai Chen +1
Large language model (LLM) agents have shown strong capabilities in interactive problem solving, tool use, and long-horizon task execution. However, they may also acquire sensitive or private knowledge during training and expose it during deployment. Retraining a model from scratch to remove such kn
memorypersistent memoryagentllm agenttool usebenchmark - arxiv:2610.17657 · cs.ROTowards a Systematic Study of Embodied System 2 over Full Task Execution CyclesYuxi Wei, Ye Fang, Shuai Yang, Tong Wu +2
An embodied brain, or System 2, must turn an abstract goal into actions while discovering relevant scene conditions and checking whether executed skills achieved their intended effects. Open-loop evaluations cannot test observations shaped by an agent's decisions, while some closed-loop evaluations
embodiedmemorybenchmark - arxiv:2610.17658 · cs.CLWorth Their Weights? Evaluating Generative Models for Discourse ParsingLuke Gessler, Amir Zeldes
Rhetorical Structure Theory (RST) parsing has historically been dominated by discriminative models based on pretrained encoder-only language models. The recent availability of high-quality generative language models has made it unclear whether RST parsing is better approached as a generative task. I
benchmark - arxiv:2610.17656 · cs.CVOpen-o3-Video-v2: Seek, Ground, and Reason with Verifiable Spatio-Temporal EvidenceJiahao Meng, Zhongyuan Bao, Yue Tan, Haiquan Wen +6
Reliable video reasoning requires models to produce correct answers and expose supporting visual evidence. Open-o3-Video enables verifiable reasoning by embedding timestamps and object-level bounding boxes into reasoning traces, but is restricted to frames from an initial sparse sampler. We introduc
agenticbenchmark - arxiv:2610.17654 · cs.CLDecouple to Forget: Semantic Mediation for Sensitive Unlearning in Large Language ModelsJingyue Cong, Andy Song, Yang Yu, Tabinda Sarwar +6
Machine unlearning for large language models seeks to remove specified knowledge while preserving general capability, yet current methods often treat an entire response as the unit of forgetting. This coarse granularity causes unnecessary utility loss when the forget target is only a small set of se
benchmark - arxiv:2610.17651 · cs.ROCode2Worlds-2: Physically Plausible Agent Harness for Embodied World SimulationHaoyu Zhang, Siming Li, Zeyu Zhang, Yi Zhang +2
Recent coding large language models (LLMs) can synthesize dynamic 4D worlds from natural language, but the resulting content remains a visual artifact. It lacks physical semantics such as the collision geometry, mass, friction, inertia, and joint structure, required to enable it to simulate the phys
embodiedagentbenchmark - arxiv:2610.17644 · cs.LGDisaggKD: Efficient Knowledge Distillation through Teacher-Student Disaggregation and a Fused LossRuixuan Tu, Yuyang Dong
Knowledge distillation (KD) frameworks place teacher and student on the same GPUs, so per-GPU memory limits the teachers that fit. DisaggKD separates them across nodes: teacher replicas stream hidden states to the student, a KD loss over row windows and vocabulary shards equals the full-vocabulary l
memory - arxiv:2610.17643 · cs.AIWhen Should Enterprise Policy Decisions Be Learned, and When Should They Be Reasoned?Hyojung Han, Ryangkyung Kang, Youngeun Kim, Seung-Hun Jeon
An enterprise agent that answers under a written policy must choose an action before it says anything: answer, ask for a missing fact, ask for confirmation, call one or several tools, escalate, or refuse. Holding one 27B Korean enterprise model fixed, we vary whether the decision is a generated JSON
agentbenchmark - arxiv:2610.17640 · cs.CVMemAct: Streaming Spatial Understanding with Memory and Active ExplorationYuxi Wei, Wei Huang, Qirui Chen, Lu Hou +1
Spatial understanding is essential for embodied agents navigating and interacting with the physical world. As observations arrive incrementally, agents must reason in a streaming manner. Yet streaming understanding must also be closed-loop: task-relevant evidence may require both recalling past obse
embodiedmemoryagentembodied agent - arxiv:2610.17639 · cs.CVFrameLite: Decision-Oriented Aesthetic Cropping with a Compact Vision-Language ModelJianze Li, Kai Liu, Ziqing Zhang, Weihang Zhang +3
Aesthetic image cropping improves composition by balancing subjects, context, and image boundaries. Large vision-language models bring semantic knowledge to this task, but their scale limits lightweight photography assistance. We propose FrameLite, a compact cropping model built on Qwen3.5-0.8B. Our
benchmark - arxiv:2610.17638 · cs.ROActing Through RegistersYiwei Zhang, Yantai Yang, Jin Gao, Jin Lou +6
World Action Models (WAMs) use generative visual representations to condition robot actions, yet how these representations organize action-relevant information remains poorly understood. We identify persistent high-norm spatial outliers in WAM representations and find that their action relevance per
embodiedmanipulationliberorobotwinworld model - arxiv:2610.17635 · cs.LGCalibrated Disambiguation for Partial Multi-label LearningZhuoming Li, Yuheng Jia, Mi Yu, Zicong Miao
Partial multi-label learning (PML) aims to train a classifier on datasets whose instances are over-annotated with not only relevant labels but also irrelevant labels, which is common when datasets are collected from crowd-sourcing platform. Existing works primarily approach it from a curriculum lear
curriculum learning - arxiv:2610.17633 · cs.AIExecutable Prefix Values for Locating Agent FailuresBingyu Shen, Jiaqi Tang, Lan Wei, Yi Sheng +2
Work on debugging LLM agents treats the "first mistake" marked by a human or LLM judge as ground truth for where a trajectory failed. We ask whether such labels point at causes or symptoms, and make the distinction executable: we restart the agent from reconstructed intermediate states of a failed t
agentllm agent - arxiv:2610.17630 · cs.ROFrom Pixels to Executable Physics: Behavior-Grounded World Models for Embodied Action Consequence PredictionRuo Chen, Jiahua Pang, Yingjie Niu, Hongyi Lin +3
In outdoor environments, varying terrain and contact conditions can cause identical action commands to produce different motion outcomes for quadruped robots. Predicting these outcomes is essential for reliable navigation and informed path selection. Existing video world models can generate visually
embodiedquadrupedworld model - arxiv:2610.17625 · cs.CV4LC: 4D Latent Cache for Efficient Conditioning in Dynamic World ModelingJianhui Zhang, Zuyao Lin, Pengxin Xu, Jingdong Wang +1
For video generation in dynamic world modeling, observed static and dynamic content should be efficiently reused to provide conditions for repeated generation at different times and viewpoints. We introduce 4LC, a 4D latent cache with composite visual representations and a flexible spatiotemporal ac
world modelv-jepa - arxiv:2610.17621 · cs.LGFedLoRD: Personalized Federated Learning based on Low-Rank Structures of Model Parameter DifferencesTsutahiro Fukuhara, Junya Hara, Hiroshi Higashi, Yuichi Tanaka
We propose FedLoRD, a personalized federated learning (PFL) method based on the low-rank structure of the differences between the global and personalized model parameters. PFL is a branch of federated learning that aims to learn models tailored to the local data of each client under biased local dat
benchmark - arxiv:2610.17616 · cs.CLStateHarness: Maintaining Evolving Project State for Long-Horizon Coding AgentsChuan Zou, Yujun Zhou, Xiaonan Luo, Xiangliang Zhang
Long-horizon coding requires agents to maintain an evolving view of the project they are changing: what problem is currently being solved, what has already been learned or modified, which other artifacts may be affected, and what work remains unresolved. When this information is scattered across a g
agent - arxiv:2610.17613 · cs.AIMAS-Gate: Unlocking Multi-Agent System Lifecycle via Node-Centric QuantificationHehai Lin, Zixuan Wang, Ming Yan, Peidong Wang +5
Large language model (LLM)-based multi-agent systems (MAS) can solve complex problems beyond the reach of a single agent by decomposing tasks and coordinating complementary capabilities. Despite extensive research on MAS, existing work largely focuses on specific aspects or individual phases of thei
agentmulti-agentagent system - arxiv:2610.17610 · cs.AIOpenDepth: Repurposing Image Editing Priors for Robust Open-World Monocular Depth EstimationMengke Zhang, Tianyi Yang, Xinzhuo Li, Anlong Ming
Recently, the integration of generative priors has revolutionized Monocular Depth Estimation (MDE), achieving state-of-the-art (SOTA) performance with remarkably high data efficiency. However, although methods based on Text-to-Image (T2I) models excel in general scenes, their efficacy degrades sharp
benchmark - arxiv:2610.17609 · cs.AIVerifier-Harness Co-Evolution for Training-Free Self-Improvement on Open-Ended TasksTianyi Xu, Yaolun Zhang, Yiran Wu, Joe Nguyen +5
Training-free self-improvement lets an agent use feedback to revise the response it is about to return and to update the harness state that shapes how it handles later tasks. On open-ended tasks that feedback rarely comes from reference answers or executable tests, so the agent has to rely on a mode
agentself-improvementbenchmark - arxiv:2610.17596 · cs.CVTanawpamana: Benchmarking Vision-Language Models on Philippine Cultural Heritage SitesIvan Yuri De Leon, Hiroki Ouchi, Sakriani Sakti
Vision-language models are increasingly used to identify and describe places, yet their recognition and knowledge of specific cultures remain poorly understood, particularly in underrepresented regions such as the Philippines. This study introduces the Tanawpamana Dataset, a repository of Philippine
benchmark - arxiv:2610.17591 · cs.ROPhysical-Field-Bridged Material Parameter Inference from Multi-View Deformation VideosLinrui Shou, Jie Li, Yu Liu, Tao Huang
Estimating simulation-ready material parameters from visual observations is crucial for deformable-object manipulation but challenging because material properties are only indirectly revealed through physical responses. We introduce physical-field-bridged material inference, a single-pass framework
manipulation - arxiv:2610.17590 · cs.LGRevisiting the 16x Pretraining Compute Gap between Diffusion and Autoregressive Language ModelsGuanghan Wang, Bing He, Zhan Shi, Benoit Dumoulin +2
Diffusion language models (DLMs) have attracted wide attention for their fast, parallel decoding, but they are believed to be far less efficient to train than autoregressive language models (ARMs): a widely cited study reports that a DLM needs roughly 16x the training compute of an ARM to match its
benchmark - arxiv:2610.17584 · cs.AIRAN-JEPA: An Action-Conditioned World Model for Energy-efficient Mobile Network ControlBinghan Wu, Kai Zeng, Kai Yang, Ya-Qin Zhang +2
Operational mobile networks are rich in observations but poor in interventions: they continuously generate large volumes of passive telemetry, while controlled action--response trajectories remain comparatively scarce and costly to collect. This observation--intervention asymmetry is a central obsta
world modelaction-conditionedlatent dynamicspost-training - arxiv:2610.17582 · cs.LGChasing Label Shifters: A Change-Aware Framework for Dynamic Graph Node ClassificationJunghoon Kim, Seungyoon Choi, Hyunsung Kim, Chanyoung Park
Entities in real-world systems often evolve over time: users shift between information consumption patterns, researchers migrate between fields, and firms transition between financial risk states. When such systems are modeled as dynamic graphs, these transitions correspond to nodes whose class labe
benchmark - arxiv:2610.17574 · cs.ROTacSpin: Tactile Tracking of In-Gripper Object Rotation on SO(3) via Axis-Magnitude EstimationYuhao Wu, Linhao Song, Huan Yin
Tracking object motion within a gripper provides essential feedback for robotic manipulation. However, limited contact coverage and uneven deformation make tactile estimation challenging. In this study, we present TacSpin, a method that tracks in-gripper object rotation on SO(3) using only two visio
manipulationtactilegripper - arxiv:2610.17573 · cs.CVBenchmarking Self-Supervised 3D Medical Vision Foundation Models and Transferability EstimationJiaqi Tang, Yu Wang, Yang Liu, Jing Tao +6
Self-supervised pretraining has expanded the pool of 3D medical foundation models, but published evaluations use different datasets, endpoints and fine-tuning protocols. Models also differ in architecture, pretraining objective and data, while downstream targets range from voxel-level masks to patie
benchmark - arxiv:2610.17572 · cs.ROEgoCPT: Learning Dexterous Manipulation from a Single Ego-Centric Video with Contact-Aware OptimizationZhixin Sun, Kairui Shi, Tianao Ren, Ming Yang +2
While expert demonstrations on real robots are difficult to collect, human ego-centric videos are easy to collect and abundant on the internet. However, learning dexterous manipulation policies from human videos remains challenging---one of the most important challenges being the difficulty of recon
manipulationdexterousbenchmark - arxiv:2610.17560 · cs.CLEdgeWise: Bringing Long-Horizon Agents to the Edge via Selective Cloud DelegationZeru Shi, Jindong Jiang, Ryo Hachiuma, Yonggan Fu +9
Complex and long-horizon agent tasks often rely on frontier models deployed in the cloud to achieve strong performance. However, using these models for every step of a long-horizon trajectory incurs substantial inference cost, even though not all steps may require the same level of model capability.
agent - arxiv:2610.17555 · cs.ROATOM: Abstract Transition and Option Modeling - Experience-Informed Hierarchical NavigationChangxuan Yao, Panagiotis Tsiotras
We propose ATOM (Abstract Transition and Option Modeling), a hierarchical navigation framework that combines higher-level graph planning with reusable lower-level option policies for efficient repeated navigation in large-scale 3D environments represented by hierarchical structures such as 3D scene
scene graph - arxiv:2610.17549 · cs.ROReference Wording Shapes Retrieval of Robot Task InstructionsJahnavi Yelamanchi
Identifying a robot task from a short instruction is sensitive to which descriptions are available for retrieval. We investigate this effect in the natural-language instructions included with the RLBench manipulation benchmark. We extract 296 fixed descriptions from 62 publicly available task files
manipulationbenchmarkevaluation protocol - arxiv:2610.17544 · cs.AIMany Steps into One: Reasoning Contraction and the Limits of Safe Memory Reuse for LLM AgentsJianghong Zhou, Hao Rong, Chen Lin, Daoan Zhang
Thought templates, skill libraries, agent memories and caches turn many reasoning steps into one. We call this reasoning contraction. A contracted memory saves reasoning steps and errors until its facts change, then quietly answers wrong. Most systems trust memories by counting successes and failure
memoryagentllm agent - arxiv:2610.17540 · cs.ROTacitVLA: Reasoning Vision-Language-Action Models Without Reasoning LatencyJack Lu, Yurong You, Boyi Li, Yulong Cao +7
Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding. We show that actions do not need to wait for or condition on the generated reasoning to benefit from it,
vision-language-actionvlamanipulationliberomemoryepisodic memory - arxiv:2610.17531 · cs.LGTowards Green AI: Evaluating Energy Efficiency for Quantized LLM InferencePetar Basta
While it is known that quantization of large language models (LLMs) reduces memory usage via lower-bit weights, studies quantifying the resulting impact on energy usage and carbon intensity are scarce. We present the first unified evaluation of weight-only quantization strategies for LLM inference a
memorypost-trainingbenchmark - arxiv:2610.17530 · cs.CVTwinPrune: Density-Aware Two-Phase Token Pruning for Vision-Language ModelsKai Liu, Anqi Li, Junxian Li, Zhixin Wang +3
We propose a unified spatial-density signal to score token importance for both inference phases: bipartite merge for prefill and KV eviction for decode, yielding measurable gains in each stage. To isolate the two phases, we split benchmarks by scoring logic. Prefill probes (MMBench, POPE, ScienceQA)
memorybenchmark - arxiv:2610.17528 · cs.LGATTUNE: Agentic Throughput Tuning for LLM TrainingLeo Fan, Hongyang Wang, Wenbo Chen, Jingguo Liu +7
Training recipes are tuned by hand for one model, cluster and workload, and every new model, accelerator, framework version or dataset means tuning again. Automated tuners search a fixed space of knobs and check only that a job runs, yet a configuration can run to completion while it silently change
agentagenticleaderboard - arxiv:2610.17526 · cs.ROBox-WAM: Box Trajectories as Compact State Representations for World Action ModelsQinxi Yu, Wenzhou Ding, Gloria X. Wang, Albert J. Zhai +1
World action models (WAMs) are a category of policy models that ex- plicitly predict both actions and future world states. They have been shown to outperform action-only models across various robotic manipulation tasks. How- ever, representing the future with full-frame video prediction can be compu
manipulation - arxiv:2610.17521 · cs.ROYUBI-Real2Sim: 3DGS Scene Verification for Reliable Policy EvaluationHiroki Nishizawa, Takehiko Ohkawa, Nicola Scianca, Tatsuya Matsushima +2
Evaluating VLA policies on real robots is costly, time-consuming, and difficult to reproduce, making real-to-sim simulation essential for scalable evaluation. While 3D Gaussian Splatting (3DGS) enables rapid real-to-sim scene generation, the reconstruction quality required for reliable policy evalua
vlapi0gripperbenchmarkscalable evaluationscalable eval - arxiv:2610.17513 · cs.CVGaussian Splatting for Real-Time Scene Reconstruction: Methods, Systems, and Trade-offsAshim Dahal, Rabab Abdelfattah, Nick Rahimi
3D Gaussian Splatting (3DGS) allows for high-quality real-time novel-view synthesis. Since its inception in 2023, its explicit and highly configurable scene representation has been extended toward sparse-view generation, dynamic reconstruction, efficiency and compression, streaming, and simultaneous
benchmark - arxiv:2610.17507 · cs.LGMerit: Making every modality earn its contribution in multimodal attention decodingAli Alavi, Donald S. Williamson
More than 1.5 billion people have hearing loss. When several people talk at once, hearing aids amplify every voice alike. Auditory attention decoding (AAD) aims to solve this by reading a listener's brain signals to find the attended talker, letting the device amplify only them. A real device will n
benchmark - arxiv:2610.17505 · eess.SYA Collision-Aware Predictive Rerouting Framework for Connected and Automated VehiclesTony Kinchen, Ting Bai, Andreas A. Malikopoulos
In this paper, we investigate the problem of mitigating non-recurrent congestion in urban transportation networks through predictive and adaptive routing for connected and automated vehicles (CAVs). We propose a collision-aware predictive rerouting framework that integrates microscopic SUMO simulati
benchmark - arxiv:2610.17502 · eess.SYMulti-Day Stochastic Dispatch for Power Systems With Long-Duration Storage Using Diffusion-Generated Weather ForecastsLeonard Göke
As power systems shift toward high shares of wind and solar, long-duration storage becomes pivotal for balancing multi-day deficits and surpluses of uncertain supply and demand. Current dispatch strategies for long-duration storage either assign a fixed value to stored energy or extend the modeled h
benchmark - arxiv:2610.17494 · cs.LGLongitudinal Learner Participation Profiling in Online Learning: Heterogeneous Consensus Clustering under Unequal Observation WindowsDavid Otoo-Arthur, Terence L. van Zyl
Learning management system traces are clustered to identify participation profiles, but unequal observation windows may cause profiles to reflect duration rather than participation alone. Using 16,448 complete learner--course sequences from 4,599 learners across 152 courses at a South African public
online learning - arxiv:2610.17490 · cs.LGA Time-Synchronous Prequential Evaluation Framework for Next-Week Student Disengagement Prediction from Moodle LMS LogsDavid Otoo-Arthur, Terence L. van Zyl
Retrospective evaluation of early-warning models may not reflect information available when predictions are issued and outcomes become observable. We developed a time-synchronous prequential framework for online prediction of next-week student disengagement, operationalised as inactivity onset, from
evaluation framework - arxiv:2610.17488 · cs.MAOrigin Attribution for Unsafe Behavior in Self-Evolving Multi-Agent Coding SystemsHanyi Yu, Jiaming Cui
As language-model-based coding agents evolve their programs and reuse code from one another, safety failures can acquire a history that extends beyond the agent in which they are detected. An unsafe change may originate in one agent, propagate through intermediate recipients, and survive subsequent
agentmulti-agentself-evolving - arxiv:2610.17471 · cs.CLScaling Inference-Time Compute with Teams of Coding AgentsKeenan Samway, Daniel D'souza, Omar Darwiche Domingues, Marzieh Fadaee +2
Multi-agent systems promise a new axis for scaling inference: rather than having one agent think longer or sample more attempts, a task is split across many agents that work in parallel and coordinate. However, in practice, more agents can easily produce duplicate work, coordination failures, and hi
agentmulti-agentagent system - arxiv:2610.17466 · cs.LGEfficient Fréchet Distributional Decoder Alignment for Latent Generative ModelsXianghong Fang, Jiajun Zhu, Esma Aimeur, Dehan Kong +1
Fréchet Distributional (FD) objectives provide an effective distribution-level signal for post-training generative models. For latent generative models, however, existing FD-based post-training methods optimize only the latent generator while keeping the decoder frozen: the decoder is trained on rec
post-training - arxiv:2610.17463 · cs.LGAn FPGA-Based Spiking Neural Network for Spatiotemporal Pattern Learning and ClassificationJingang Jin, Yue Ma, Zhenhang Zhang, Qinru Qiu
Spiking Neural Networks (SNNs) are promising for energy-efficient spatiotemporal pattern recognition at the edge, but efficient deployment and on-device adaptation require specialized hardware. In this work, we present an FPGA implementation of multilayer SNNs for spatiotemporal pattern classificati
memory - arxiv:2610.17462 · cs.ROAct4R: Action-conditioned 4D reconstruction for kinematics-free controlZhaoliang Zhang, Chenhao Yu, Zhimin Shao, Xijun Liu +1
Precise control of movement across various robotic embodiments is a longstanding challenge. Modeling complex robotic structures requires significant time, dedicated sensors, and manual labeling. Recent advancements in 4D representation learning offer an interesting alternative: can we leverage 4D fe
action-conditioned - arxiv:2610.17457 · cs.LGBrowseSafe++: Benchmarking Prompt Injection Detection Across Web Contexts, Languages, and Browser ViewsYuetian Chen, Kaiyuan Zhang, Yuntao Du, Thomas Li +5
Browser agents need to make decisions and execute actions on behalf of users while consuming untrusted webpages, creating opportunities for malicious instructions embedded in webpages to redirect legitimate tasks. We introduce BrowseSafe++, which extends BrowseSafe from 14,719 pages to 38,675 pairs
benchmark - arxiv:2610.17456 · cs.LGCompositional RL: Evaluating and Training LLMs for Collaborative ReasoningShangbin Feng, Yulia Tsvetkov, Jan Kautz, David Acuna
Language models are often evaluated alone but increasingly deployed together, routed between, made to debate, ensembled, or collaboratively trained. For a model inside a system, what matters most is not how well it performs on its own, but how much the system gains from its presence. We call this a
benchmark - arxiv:2610.17440 · cs.CVKnow-Thy-Not: Training-Free Targeted Negation in Dual-Encoder Vision-Language ModelsDoYoung Kim, Changkyu Choi, Adín Ramírez Rivera, YunSeok Choi +1
Dual-encoder vision-language models systematically fail at negation, scoring ''a cat without a dog'' nearly identically to ''a cat with a dog.'' We show that this failure is inherent to the single-embedding framework. Under contrastive training, single-embedding methods can at best ignore the negate
benchmark - arxiv:2610.17436 · cs.AISchemaWeaver: Accelerating On-Device Mobile Agentic AI via Schema-Guided Parallel LLM DecodingBoyuan Yang, Wei Gao
Large language models (LLMs) on mobile devices can act as agents that call external software tools to complete complex tasks, but the major challenge of supporting on-device mobile agentic AI with LLMs is the long latency of LLM tool calling on mobile devices. In particular, the decode phase is the
memoryagentictool calling - arxiv:2610.17434 · cs.CVWorld-T3: Test-Time Training as Persistent Memory for Real-Time Video World ModelsJunyi Wu, Zekun Li, Chenhongyi Yang, Chen Guo +8
Recent advances in video world models have enabled plausible interactive environments, yet long-horizon generation places a fundamental demand on visual memory: the previously observed world must remain consistent as the model continues to generate. Existing autoregressive approaches often implicitl
world modelmemorypersistent memorybenchmark - arxiv:2610.17429 · cs.LGMonarch-Muon: Structure-Aware Muon for Fast and Memory-Efficient PretrainingAlexander Miasnikov, Ruslan Kabirov, Gleb Molodtsov, Dmitrii Vlashchuk +6
Training large language models is constrained by both wall-clock time and memory footprint, and recent matrix-aware optimizers sharpen this tradeoff: Muon can improve optimization but its dense polar update remains expensive. We propose Monarch-Muon, a structured matrix-aware optimizer for weights p
memory - arxiv:2610.17427 · cs.ROSmoothness Ranks Skill, Not Success: An Audit of Published Trajectory-Smoothness Curation Metrics Under Episode-Length ControlsJaiveer Bassi
Trajectory smoothness has been proposed as a cheap, training-free signal for curating imitation-learning demonstrations, most recently by rinse, which introduces Spectral Arc Length (SAL) and a contact-aware Trajectory-Envelope Distance (TED). We test whether these published metrics survive episode-
manipulationdiffusion policy - arxiv:2610.17426 · cs.LGReinforcement Learning In Partially Verifiable Environments for Colon Cancer Treatment PlanningLi-Ching Chen, Andrés Franco Valiente, Yuta Ishiyama, Zachary R. Chalmers +6
Clinical decision-making is a sequential process of information collection, clinical inference, and forecasting under partial observability, often without a uniquely correct treatment decision. Large language models (LLMs) have demonstrated strong medical question-answering capabilities but extendin
agent - arxiv:2610.17422 · cs.ROMIRROR: Reflecting Hidden Symmetries from Offline Robot Interaction DataShiqi Liu, Yihang Yao, Peide Huang, Mengdi Xu +1
Symmetry-aware robot policy learning requires coupled observation-action transformations. For pixel observations, even simple physical symmetries can induce complex, viewpoint-dependent image transformations that are difficult to specify by hand. Existing symmetry-aware methods often rely on represe
manipulationrobot policy - arxiv:2610.17418 · cs.AIWhy Agentic Workflows Fail: A Taxonomy Derived from Literature and Empirical DataXin Li, Rami Bahsoon, Tao Chen
Agentic workflows combine large language models with planning, memory, tools, execution environments, and iterative feedback. Their failures are difficult to diagnose because the same user-visible symptom may originate from model behavior, workflow orchestration, external tools, infrastructure, or t
agentagentic - arxiv:2610.17410 · cs.AIAgency as a New Organizing Abstraction for Computer ScienceIan Foster
Advances in large-scale machine learning have enabled agentic AI systems: persistent, autonomous computational entities that pursue goals, initiate actions, and interact with open environments. These systems are often framed as applications of foundation models. We argue that this framing is incompl
agentagentic - arxiv:2610.17409 · cs.LGPosition: Beyond Training Time, Test-Time Coordination is Essential for Cooperative MARLDongsu Lee, Sooraj Sathish, Joonkyung Kim, Woojun Kim +1
We argue that cooperative multi-agent reinforcement learning (MARL) should treat test-time coordination as part of the problem definition rather than a deployment artifact. Our focus is on long-horizon, centrally monitored cooperative systems such as warehouse robotics, automated storage-and-retriev
multi-agent - arxiv:2610.17397 · cs.LGJustOPD: On-Policy Distillation from Autoregressive to Diffusion Language ModelsZikun Zhang, Jiayuan Sheng, David D. Yao, Wenpin Tang
On-policy distillation (OPD) is a post-training approach for large language models (LLMs) that enables a student model to learn a teacher model's distribution from sequences sampled from the student's own policy. Despite its promising capability, OPD for diffusion language models (DLMs) remains unde
post-training - arxiv:2610.17391 · cs.AILearning to Play: Reinforcement Learning and GamesAske Plaat
Amazing breakthroughs in reinforcement learning have taken place. Computers teach themselves to play Chess and Go and beat world champions. There is talk about expanding application areas towards general artificial intelligence (AI). The breakthroughs in Backgammon, Checkers, Chess, Atari, Go, Poker
self-play - arxiv:2610.17387 · cs.LGCross-Market Meta-Learning for Electricity Price ForecastingDelaram Pirhayatifard, Terrence Alsup
Day-ahead (DA) electricity price forecasting is usually posed per market, with a separate model fitted to each grid's own history. This caps accuracy: statistical and tree-based baselines saturate, while higher-capacity sequence models need more training data than a single market's few years of hour
benchmark - arxiv:2610.17384 · cs.LGMeasure Neural MappingsRicardo Baptista, Edoardo Calvello, Bohan Chen, Andrew Stuart +1
We introduce and define measure neural mappings (MNMs), which combine measure-dependent neural transport with ensemble consistency for learning operators on probability measures. Ensemble consistency requires outputs from growing ensembles sampled from an input law to converge to the specified popul
benchmark - arxiv:2610.17383 · cs.LGVariable Input Masked Autoencoders for Wrist-Worn AccelerometryHarish Haresamudram, Wanting Mao, Aditya Radhakrishnan, Maxwell A. Xu +1
Large-scale repositories of wearable sensor data have enabled the development of foundation models for human activities and behavior. However, existing models are pretrained using fixed input durations and sampling rates, despite substantial heterogeneity in recording conditions, downstream tasks, a
benchmark - arxiv:2610.17378 · cs.AIGet In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic CommerceAdrian de Valois-Franklin, Alex Bogdan, Anas Ibrahim, Ibrahim Shaer +1
AI agents now browse and buy on people's behalf, and benchmarks measure how well the agents perform. Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed.
agentai agentagenticbenchmark - arxiv:2610.17370 · cs.LGHarnessTTT: Learn to Optimize Harness at Test-Time for Scientific DiscoveryYingzi Ma, Yulong Cao, Nanxi Li, Fang Wu +4
Scientific discovery with language models requires finding the best valid solution within a finite computational budget. We study how feedback from this search can improve the procedure used to conduct it. We introduce HarnessTTT, a test-time learning framework that trains a harness proposer while k
benchmark - arxiv:2610.17368 · cs.LGHASTE: Hierarchical and Attribute-aware Supervision for Generative RecommendationTripti Shukla, Shwetha Somasundaram, Akshat Gupta, Rohit Garg +2
Generative recommendation models have shown promising results by representing items as semantic IDs and generating these IDs autoregressively. We introduce an approach showing that explicitly incorporating hierarchical and attribute-level supervision into generative recommendation can simultaneously
benchmark - arxiv:2610.17365 · cs.AIWhen Goals Leak Across Relationships: How Cached Goal Representations Shape Cross-Client DecisionsSaloni Gupta, Sébastien Krier, Glenn Matlin
As LLM agents take on more autonomous roles, they may interact with multiple clients and counterparties whose goals and incentives conflict. If an agent maintains shared inference state across these interactions, information associated with one relationship may influence decisions made for another.
agentllm agent - arxiv:2610.17363 · cs.CLUMCR: Universal Medical Concept Recognition through Guideline Integration and Agentic LinkingFabrice Lamarche, Amal Zouaq
Medical Concept Recognition is the task of extracting and grounding spans in medical text to entries in a standard ontology. Current methods are generally developed towards a single ontology and source text type (clinical notes, drug labels, etc.) The central challenge is therefore to build methods
agenticbenchmark - arxiv:2610.17358 · cs.AIclaimcheck: Measuring and Improving Specification FaithfulnessMartin Amster, One An, Holly Baker, Fernanda Graciolli +2
Formal verification proves that a program satisfies its specification, but says nothing about whether the specification captures what the author meant. As LLMs increasingly write both code and specifications, a gap between the author's natural-language requirement and the specification becomes a liv
benchmark - arxiv:2610.17356 · cs.CVBrain2Face: A Study on the Relation of EEG Signals and Spatio-temporal Head AvatarsCamilo Andrés Martínez Martínez, Olaf Dünkel, Kartik Teotia, Maurice Rekrut +2
Driving digital head avatars is classically done by hand-crafted artist design or by image-based facial performance capture. In this work, we ask whether spatio-temporal tracking of the humans's head geometry and appearance can be performed solely from brain signals, i.e., Electroencephalography (EE
benchmark - arxiv:2610.17355 · cs.CVWIRE: Weight Space Implicit Representation Encoding of Frozen Video LLMsAmeen Ali, Tamim Zoabi, Lior Wolf, Chaim Baskin
Video language models represent a video as a long sequence of visual embeddings in a decoder's input. Every visual embedding is processed by every decoder layer during prefill, stored in the key value cache, and attended by every generated token, so a long video raises prefill, memory, and generatio
benchmark - arxiv:2610.17354 · cs.CVMID: Mask-Image Distributional Divergence for Evaluating Medical Image SegmentationVincenzo Marcianò, Xiaoming Zhang, Gianluca Guglielmo, Sebastien Ourselin +2
Current evaluation protocols for medical image segmentation remain largely anchored in sample-wise comparisons, e.g., Dice and Hausdorff distance, which require each prediction to be paired with a ground-truth mask. Yet two limitations arise in practice: they are inapplicable when predictions are ma
evaluation protocol - arxiv:2610.17350 · cs.ROCan a Frontier VLM Serve as a Robot Policy for Memory-Dependent Manipulation?Jack B. Jedlicki, Amish Sethi, Heng Yang
We use a frozen frontier vision-language model, GPT-6-Astra, directly as the robot policy on RoboMME, a benchmark of 16 memory-dependent manipulation tasks spanning temporal, spatial, object, and procedural memory. The model keeps the whole episode in context and outputs every end-effector waypoint
embodiedmanipulationrobot policygrippermemorybenchmark - arxiv:2610.17331 · cs.AIDo Theorem Provers Need Theorem-Proving Agents? mini-Lean-agent: A Coding-Agent Baseline for LeanTomas Ortega, Simon Park, Ayush Khaitan, Liam Fowl +2
General-purpose coding agents use standard shell and file interactions, yet formal theorem-proving agents often rely on specialized, ad-hoc tooling. This makes it difficult to determine which components of theorem-proving agents provide benchmark gains: an improved base model, a specialized harness,
agentbenchmarkevaluator - arxiv:2610.17327 · cs.CLThe Curse of Knowledge in Role-Playing Language ModelsXiao Zhang
Role-playing language models speak as characters in games and agent simulations, where the system records what each character can hear, see and understand. We test whether characters respect these boundaries in 480 scenes, each with a counterfactual twin that changes only the hidden fact, so mention
agent - arxiv:2610.17326 · cs.ROEvaluating Long-Horizon Robot Planning under UncertaintyHarsh Muriki, Gayathri Aravindan, Yuchen Zheng, Sonia Chernova
Household robots must successfully navigate im-perfect circumstances, both in their internal models and the external world. We introduce a benchmark of seven long-horizon household tasks in Habitat that vary initial robot memory and environment configurations. We evaluate adaptations of ReAct, VLM-T
memorybenchmark - arxiv:2610.17323 · cs.LGSSM-Attack: Learning Where to Perturb for Black-Box Face PrivacyTunahan Parlayıcı, Erchan Aptoula, Yücel Saygın
Face recognition systems are deployed widely enough that a photograph shared online can be matched without its subject's consent. Generator-based protection methods rely on heavy pretrained backbones, multi-stage inference or visible semantic edits. We introduce SSM-Attack, a black-box face privacy
benchmark - arxiv:2610.17318 · cs.CLFrom Sampled Evidence to Radiology Reports: Objectives, Constraints, and RealizationKun Zhao, Haoteng Tang, Chenghua Lin, Liang Zhan
Sampling multiple reports creates an inference problem: how should sampled evidence be converted into a final decision, and does a better structured decision yield a better report? We study these questions through a framework that separates the decision signal, scoring objective, and feasible decisi
benchmark - arxiv:2610.17310 · cs.AIBeyond Successful Play: Diagnosing Opponent Modeling in Large Language ModelsHanbo Xie, Yue Zhao, Robert C. Wilson
Language-model agents are expected to adapt to others whose behavior changes through interaction. Yet adaptation may rely on learned behavioral patterns rather than theory of mind (ToM): the ability to infer others' mental states, including beliefs and intentions. We introduce a benchmark and diagno
memorybenchmark - arxiv:2610.17309 · cs.AIStartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative DecisionsXialie Zhuang, Zunhai Su, Jia Pan, Ngai Wong +2
We introduce StartLux-Decision, decision models from 0.8B to 27B parameters that combine efficient single-pass prediction with adaptive, confidence-guided reasoning. The Instant models use bidirectional attention in full-attention layers to improve decision accuracy while keeping single-pass inferen
memorybenchmark - arxiv:2610.17305 · cs.AIReliable Text-to-SQL in Resource-Constrained Settings with Small LLMs via Iterative Retry WorkflowsBilel Abderrahmane Benziane, Xavier Schepler
Large Language Models (LLMs) have recently improved Text-to-SQL generation, but their computational cost can hinder deployment in resource-constrained settings. This paper evaluates the Text-to-SQL capabilities of small and medium open-source LLMs (1B-8B parameters) under a controlled inference budg
benchmark - arxiv:2610.17301 · cs.AIThe Agent Knew: Internal States Forecast Unauthorized Actions and Stop Them, and the Lies That FollowArya Fayyazi, Mehdi Kamal, Massoud Pedram
Language-model agents sometimes act beyond their authorization and then misreport what they did. On CAPE-Bench, matched agent tasks in which one fact decides whether an action is authorized, frontier open-weight agents do both, and one behavior marks the lie: retroactive permission laundering, actin
agentbenchmark - arxiv:2610.17296 · cs.AITowards Resilience: Slice-Driven Orchestration of Mobile IoMT in 6G-Edge-ContinuumMuhammad Zeeshan, Hafiz Faheem Shahid, Erkki Harjula
Emerging healthcare services, such as real-time diagnostics, remote health monitoring and computer-assisted surgery, require stringent Quality of Service (QoS) guarantees while operating under limited and heterogeneous edge resources. Existing Internet of Medical Things (IoMT) orchestration methods
memory - arxiv:2610.17288 · cs.AIPhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D ReconstructionShuai Shao, Jinchen He, Cunxin Fan, Zihan Guo +9
Building physical simulations from video requires coding agents to translate observations into executable models and refine them through tool use. We introduce PhysCodeGym, a benchmark for physically verifiable 4D reconstruction with $156$ tasks covering rigid-body, liquid, deformable, and wind-driv
agenttool usebenchmark - arxiv:2610.17281 · cs.ROPhase-Conditioned Affordance Fields for SE(3)-Equivariant Visuomotor PoliciesXinzhe Chen, Liqi Huang, Sihua Ren, Haowen Sun +4
SE(3)-equivariance provides a useful geometric inductive bias for visuomotor policies, encoding that actions covary with rigid transformations of the observation. However, existing SE(3)-equivariant visuomotor policies build this bias on a global equivariance assumption that treats the entire observ
manipulationgripper - arxiv:2610.17278 · cs.CLNufo: A Benchmark for Ewe Automatic Speech RecognitionChrist Kekeli Keli, Steve Sanogo, Patrice Sebastiano, Hamad Aldenawi +1
Automatic speech recognition (ASR) remains challenging for low-resource languages, where the scarcity of transcribed speech limits the development of accurate systems. This challenge is particularly acute for African languages such as Ewe, which has millions of speakers but remains underrepresented
benchmark - arxiv:2610.17266 · cs.AIRL Post-Training Diffusion Language Models Along Their Own Generation PathsVaibhav Singh, Eugene Belilovsky
Diffusion language models are a promising and more efficient alternative to autoregressive models. However, most RL post-training techniques are not easily adapted to these models. One issue is that diffusion language models generate answers through many intermediate states, deciding at each step wh
post-training - arxiv:2610.17265 · cs.LGProtocol Choices Matter More Than the Aggregation Algorithm: Federated Learning for In-Hospital Mortality Prediction in Care-Unit and Multi-Hospital FederationsRodrigo Tertulino, Ricardo Almeida, Laércio Alencar
Federated learning allows hospitals to train a shared mortality model without pooling their records, but published comparisons of aggregation strategies rarely control the evaluation protocol. This study compared seven strategies (FedAvg, FedProx, SCAFFOLD, FedAdagrad, FedAdam, FedYogi and FedCluste
evaluation protocol - arxiv:2610.17260 · cs.ROPhysical SLAM: Physics-aware Simultaneous Localisation and MappingGuangming Wang, Zhishuo Liu, Haibing Wu, Wanli Ma +3
To reconstruct a predictive 4D representation of the physical world from monocular video captured by a moving camera, we jointly reconstruct dynamic Gaussian radiance fields and infer their underlying physical dynamics. This paper proposes \emph{Physical SLAM}, a physics-aware method for constructin
manipulationaction-conditionedbenchmark - arxiv:2610.17256 · cs.AIDoes a Governed Confidence Score Know When a Text-to-SQL Answer Is Right? An Exploratory Study on BIRD and a Pre-registered Confirmation on SpiderRaghu Venkat, Tricha Anjali
Conversational analytics lets people ask questions of records spread across many systems and get answers in plain language; it is useful only if each answer can be trusted and explained. A production system of this kind attaches a confidence score to every answer it writes over a database and uses i
benchmark - arxiv:2610.17252 · cs.ROValue-Preserving Action QuantizationMax Muchen Sun, Jueun Kwon, Benjamin Riviere, Todd Murphey
Search-based policy learning, such as Monte Carlo tree search (MCTS), offers the flexibility to trade runtime computation for data efficiency and adaptation to novel environments. For robot learning, however, applying search requires discretizing the continuous action space and associated dynamics,
benchmark - arxiv:2610.17242 · cs.CVWhen Left Is Right? Camera-Motion Understanding beyond ScreenChong Cheng, Ran Zhang, Xinyuan Li, Alan Zhao +2
Can a vision-language model tell left from right? Modern VLMs solve visual puzzles and follow hour-long videos, yet when asked whether the camera moved left or right, most do no better than a coin flip. Adding camera-motion data or geometric representations has been tried, separately, and neither cl
benchmark - arxiv:2610.17234 · cs.CVIsoVRBench: Video Reward Modeling Without Human FeedbackShuyue Stella Li, Jacqueline He, Jiahui Chen, Sreya Dutta Roy +4
Video reward models are trained and evaluated on human preferences between videos that often differ simultaneously on properties such as prompt alignment, visual fidelity, motion, and generator identity. As a result, aggregate accuracy does not reveal which video properties a reward model can reliab
benchmark - arxiv:2610.17232 · cs.ROLiDAR-based Humanoid Navigation in Dynamic, Non-Convex Environments via Spatio-Temporal Safe CorridorsShubham S. Kumbhar, Panagiotis Artemiadis
Navigating humanoid robots through dynamic, non-convex environments requires tightly coupling onboard perception with footstep planning and dynamic locomotion control. Existing methods frequently discard complex free space through simplified convex obstacle bounds and fail to account for real-time d
humanoid - arxiv:2610.17226 · cs.LGSolvating machine-learning interatomic potentials for solution-phase chemistryAnton Z. Ni, Joonho Lee
Most machine-learning interatomic potentials (MLIPs) are trained on gas-phase data, which limits their use for chemistry in solution. Here, we introduce AnySolv, a transferable implicit solvation potential that provides solvent-dependent energy and force corrections to existing gas-phase potentials
benchmark - arxiv:2610.17224 · cs.LGCodeLift: Your Test Suite Is Secretly a Dense SWE CurriculumJunjielong Xu, Huachao Zhu, Alyssa Wu, Zhichao Hu +1
The pass-all-test binary reward has become a standard recipe for training software engineering agents, but its sparsity eliminates difference of partially correct patches. The test pass-rate is a natural alternative. However, we find that it is not a reliable indicator of patch quality. A functional
benchmark - arxiv:2610.17214 · cs.LGDensity-Matrix Reinforcement LearningChiuman Ho
What information does a reinforcement learning policy omit? At each state, a standard policy stores one probability per action. It has no place to record how actions relate to each other, such as which pairs go together and which conflict. We ask when this missing structure matters, and we give exac
agent - arxiv:2610.17208 · cs.RODexNinja: Learning Robust Dexterous Cutting Policy with a Real-to-Sim-to-Real Data EngineHaozhe Lou, Runyi Yang, Wanqi Zhong, Caiyun Liu +9
Cutting a piece of food looks effortless in human hands. For a robot, however, every cut is a high-contact interaction with a deformable, elasto-plastic object whose behavior changes with pose, contact mode, and cutting stage. Collecting real-world data is expensive and wasteful: each trial typicall
manipulationdexteroustactilesim2realsim-to-real - arxiv:2610.17206 · cs.AIAutoMARS: Automatic and Scalable Multi-Agent Large Language Model Framework for Supply Chain Risk AssessmentYinzhu Quan, Zefang Liu, Frederick Benaben, Benoit Montreuil
We propose AutoMARS (Automatic Multi-Agent Risk Assessment), an agentic framework for automatic and scalable comparative risk assessment in supply chain networks. The framework enables end-to-end execution of analytical workflows by automatically planning tasks, coordinating specialized agents, invo
agentmulti-agentagentic - arxiv:2610.17203 · cs.LGOnline LLM Personalization from Pairwise Preferences Across UsersRomain Puech, Georgy Salakhutdinov, Dimitris Bertsimas
Large language models can answer the same question in different styles, but users' preferred styles are hidden and vary across users and questions. We study inference-time personalization through online selection from a fixed set of prompts, each representing a persona. Users arrive sequentially, ea
benchmark - arxiv:2610.17202 · cs.LGMLAfication: Efficiently Enabling High-Fidelity MLA for Modern LLMsZile Wang, Zhide Xie, Tao Ji
Multi-Head Latent Attention (MLA) compresses the KV cache into a latent representation, reducing inference memory and bandwidth. Existing MHA2MLA/TransMLA methods do not support QKNorm and require costly recovery on modern LLMs, especially MoE. We present MLAfication, a QKNorm-compatible, parameter-
memorylong-contextbenchmark - arxiv:2610.17201 · cs.CVMultics: Generative Multi-Player World Model with Hierarchical CommunicationKaiyi Huang, Wangbo Yu, Yukun Huang, Jiwen Yu +5
Multi-player interactive world models aim to simulate shared environments from multiple players' perspectives, but achieving robust generation while scaling to more players remains challenging. Existing approaches that rely on fully-connected communication (e.g., Solaris) incur quadratic attention c
world model - arxiv:2610.17195 · cs.CLWelcome to the Age of Agent Ads: Using On-Policy Attacks to Steer Agentic DecisionsTim R. Davidson, Marija Šakota, Joshua Joseph, Robert West
People increasingly delegate option-ranking tasks (e.g., shopping, job search, travel) to AI agents, which creates incentives to steer agentic decisions via "agent ads." To defend against such steering attacks, people count on models' alignment training and platforms' terms of service, and consequen
agentai agentagentic - arxiv:2610.17191 · cs.LGFreeSolo: Linear-time Sparse Attention with Zero-Shot Length ExtrapolationZekun Shi, Alastair James Arthur Price
Linear-time sequence models typically store long-context information in a recurrent state. A fixed-size state limits how much information can be stored, while hybrid architectures that add full attention for better recall incur quadratic sequence cost. Alternatively, sparse attention with continued
long-context - arxiv:2610.17184 · cs.CLDuoDiscover: Scaling Autonomous Discovery Beyond the Evaluation BottleneckShubham Agarwal, Akshat Gupta, Mike Heddes, Dilara Soylu +20
Evaluation has been central to advances in AI, and it is equally critical to autonomous discovery. Yet scaling autonomous discovery to real-world engineering challenges faces a fundamental bottleneck: the cost of evaluating progress. In real settings, a single true evaluation, such as a full benchma
agentbenchmarkevaluator - arxiv:2610.17183 · cs.CVChemExplain: A Multimodal Benchmark for Explanatory Chemical ReasoningAotian Chen, Frazier N. Baker, Reza Averly, Shan Zhang +4
Current chemistry benchmarks primarily evaluate outcome prediction or reaction pathway generation, but rarely test whether a model can explain why an observed chemical effect occurs. We introduce ChemExplain, a benchmark of 1,180 multiple-choice chemistry questions drawn from 383 peer-reviewed paper
multi-agentagent frameworkbenchmark - arxiv:2610.17175 · cs.LGFunDM: Cheap & Label-Free Decision Models from Frozen Language ModelsUjjawal K. Panchal
Decision models such as Jev score candidate answers or actions. A small decision head trained label-free can recover about 99% of a frozen language model's multiple-choice accuracy while using only two-thirds of its transformer layers. It learns the full model's answer probabilities with KL divergen
agentbenchmark - arxiv:2610.17161 · cs.ROForce-Modulated Dexterity: Unified Compliance for Contact-Rich ManipulationYing Yuan, David Held, Jitendra Malik, Haozhi Qi
Teleoperation is widely used to collect high-quality demonstrations for robot learning. Dexterous manipulation, however, requires coordinated control of motion and contact forces. Operators can specify motion, but struggle to reproduce the subtle force adjustments of human manipulation without feeli
manipulationdexterousteleoperation - arxiv:2610.17146 · cs.LGSCOPE Before You Sample: Conformal Scores Guide Posterior Exploration for Fast and Calibrated Bayesian Predictive InferenceZahra Moslemi, Ziyi Liang, Babak Shahbaba
Bayesian predictive regions account for parameter uncertainty but can be poorly calibrated under model misspecification, even with expensive full MCMC. Cheaper approximations such as variational inference reduce the cost but often yield less accurate predictive distributions. We propose SCOPE (Score
benchmark - arxiv:2610.17143 · cs.LGOpenDGO: Benchmarking Multi-Agent Organizations Across Diverse Application DomainsXunkai Li, Miao Zhang, Zekai Chen, Xuaner Chen +5
Large language model (LLM)-based multi-agent systems are increasingly applied across diverse domains, yet different applications require different forms of agent organization. Existing studies typically adopt predefined organizations, such as Star, Chain, and Tree, or learn communication graphs to i
agentmulti-agentagent systembenchmark - arxiv:2610.17141 · cs.CVMARS: Systematic Multi-Level Acceleration for World Action ModelsZengyuan Zuo, Yongchang Zhang, Guangyu Zhao, Tianze Zhang +3
World Action Models (WAMs) have emerged as a promising alternative for embodied intelligence by jointly predicting future world states and robot actions. However, their heavy visual encoding and iterative generative inference incur substantial latency, limiting their applicability to high-frequency
embodiedlibero - arxiv:2610.17131 · cs.AISonicWorld: Source-Centric Latent Dynamics for Spatial Audio ReasoningChenlong Wei, Hanling Wang, Mingjin Lou, Yufan Shi +5
Dynamic spatial audio reasoning requires a model to associate changing acoustic observations with the correct source and recover its spatial state at a queried time or interval, a challenge intensified by overlapping speech and source motion. To study this problem, we introduce Sonic-Bench, comprisi
latent dynamicsbenchmark - arxiv:2610.17130 · cs.AIEvoQuantLab: Training-Free Dual-Loop Evolution for Auditable Quantitative ResearchHaolin Yang, Zhiqiang Qian, Chutian Chen, Jipeng Zhang +5
Automated alpha factor discovery is critical for modern quantitative research. While large language models (LLMs) have increasingly been integrated into quantitative workflows, existing agent-based approaches remain highly susceptible to severe backtest overfitting. To address this challenge, we pro
agentself-evolving - arxiv:2610.17128 · cs.CVEgoProbe: Benchmarking Long-Range and Exact Memory in Streaming Egocentric VideosHaokai Pang, Yuqian Fu, Yifei Wang, Yujiao Shen +3
Egocentric video provides a continuous first-person record of everyday activities, making it a natural foundation for wearable AI assistants. Unlike offline video understanding, streaming systems must process observations as they arrive, while queries may refer to experiences from the distant past.
memorybenchmark - arxiv:2610.17126 · cs.AIWho is the 'I' in AI? Establishing Decentralized Identity and Accountability for Autonomous AgentsAhammod Munim Ramil, Khalid Al Kafi, M. A. Abdullahis Sany, Muktadirul Islam Nibir +3
The rapid proliferation of autonomous AI agents in decentralized ecosystems challenges existing identity paradigms. While human identity is traditionally anchored in biological and/or self-asserted attributes and government credentials, AI agents lack a standardized definition of selfhood, creating
agentai agentautonomous agent - arxiv:2610.17125 · cs.CLToolHACE: A Dataset for Detecting Hallucinations in Tool-Based LLM ResponsesNikolay Ivanov, Sergey Pletenev, Elisei Rykov, Anatolii Frolov +7
Tool-augmented language models ground their responses in external tool outputs, improving reliability beyond parametric knowledge alone. However, even when the correct tool is called and returns accurate information, models may still contradict such outputs, omit critical details, or generate unsupp
benchmark - arxiv:2610.17118 · cs.LGWhen LLM Agents See a Conflict but Still Proceed: Cross-Domain Dependencies in Network SlicingMarcos Lima Romero, Ricardo Suyama
Future AI-native networks are being designed to apply autonomous agents built on large language models (LLMs) in network management through intents. When intents from different domains meet on the same resource, an agent must recognize the cross-domain dependency before taking an irreversible action
agentllm agentautonomous agent - arxiv:2610.17114 · cs.ROD-World: An Explicit World-Language-Action Model with Recursive Self-Improvement under Dual ControlXuan Wen, Baichao Wang
Frontier large language models (LLMs) now reason about space well enough to program robots, yet the world models behind robot policies are implicit, held in network weights or predicted video, and cannot be inspected or corrected directly. We present D-World, an explicit world-language-action model
sim-to-realworld modelagentself-improvementpost-training - arxiv:2610.17113 · cs.AICalibrated Typed Decisions for Entity, Relation, and Event ExtractionYang Wang, Hanlin Xiao, Yanan Ma, Chenghao Xiao +1
Information extraction models often perform well on the datasets they are trained on, but transferring to new domains and annotation guidelines remains challenging. Supporting user-specified types alone does not resolve this problem. We present CalOn, a unified 0.6B-parameter encoder for entity, rel
benchmark - arxiv:2610.17106 · eess.SYA Deficiency-Based Approach for Unbalanced Three-Phase Distribution System Expansion PlanningEli Brock, Miguel Heleno, Javad Lavaei, Somayeh Sojoudi
Distribution planners typically characterize grid needs through power flow studies and fit upgrades to those needs by hand. More principled one-shot optimization methods either do not scale or misrepresent unbalanced three-phase networks. This paper proposes an iterative planning process that altern
benchmark - arxiv:2610.17093 · cs.AISFT Leaves a Reproducible Signature in Activation Geometry: Calibrating Post-Training Interpretability Against Run-to-Run NoiseXupeng Chen, Chenqian Le, Jiayi Cheng, Runhao Li +1
A model diff credits post-training with whatever separates a base checkpoint from one post-trained descendant, presuming a rerun would land close by. On the sparse co-activation graph of Qwen3-1.7B it does not: two LoRA-SFT reruns sit over half as far apart as either sits from the base. We score eac
post-training - arxiv:2610.17089 · cs.ROInteractive Navigation Beyond Reachability: Benchmark and Agentic FrameworkDelun Li, Xueyao Wei, Rui Chen, Yifei Wang +3
Existing embodied navigation tasks commonly assume that traversability and object states remain fixed during an episode. Yet reaching a goal may require opening a door to access a room and then opening a container to reveal a hidden target. We formulate interactive navigation beyond reachability, un
embodiedagenticbenchmarkevaluation protocol - arxiv:2610.17083 · cs.RODeclarative Reactive Robot Behaviors from Imperative Language ModelsVincent Berenz, Tasha Kim, Bernhard Schölkopf
Large Language Models (LLMs) are predominantly trained on imperative programming languages, and existing work on LLM-based robot programming reflects this: it focuses on goal-directed tasks expressed as plans. However, many robotic applications require not tasks but behaviors--open-ended, continuous
agent - arxiv:2610.17080 · cs.CVScaling Unified Multimodal Models to Long-Horizon Interleaved GenerationZiqiao Peng, Yutao Cui, Wang Lin, Zhong-Yu Li +11
Unified multimodal models (UMMs) provide an interface for interleaved text-image generation, where generated visual states can directly inform subsequent reasoning and synthesis. However, existing systems are still primarily demonstrated on short trajectories or simplified visual content. We develop
memory - arxiv:2610.17079 · cs.CVRethinking LLM-Guided Visual Question Answering: Does Deeper Knowledge Integration Really Help?Aneerban Saha, Vijay Kumar, Sameer Madhav D., Priyam Srivastava +1
Knowledge-based Visual Question Answering (VQA) requires reasoning that extends beyond visible image content, demanding external commonsense and factual knowledge. While recent work leverages Large Language Models (LLMs) as knowledge sources, the question of how deeply such guidance should be integr
benchmark - arxiv:2610.17069 · cs.ROTask-Space Motion-Force Nonlinear Model Predictive Control for Precise and Compliant Control under UncertaintiesQuang-Nam Nguyen, Ludovic Righetti
In Nonlinear Model Predictive Control (NMPC), the control law is implicitly determined by the solution of an Optimal Control Problem (OCP). Consequently, OCP design is often tailored to specific platforms and tasks, demanding extensive expert knowledge and tuning effort. To address this, we propose
manipulatorquadruped - arxiv:2610.17066 · cs.CVPerturbBridge: Direct Phenotypic Supervision for Cellular Perturbation GenerationYucheng Yang, Xiaoxiao Sun, Huangqingbo Sun, Elaine Sui +2
Virtual cell models should capture phenotypes induced by specific perturbations. However, image-based models are typically trained and evaluated to match real-image distributions, which alone does not explicitly prioritize perturbation-specific biological responses. Recent approaches add biological
post-trainingbenchmark - arxiv:2610.17065 · cs.LGTILE: Bridging LLMs and Table Foundation Models for Tabular Learning with StringsFélix Lefebvre, David Holzmüller, Gioia Blayer, Myung Jun Kim +2
Real-world tables are often heterogeneous, combining numerical features with free-form string columns. Table foundation models (TFMs) excel on numerical tables but require strings to be encoded numerically, while large language models (LLMs) hold semantic knowledge but are not designed for tabular p
benchmark - arxiv:2610.17062 · cs.LGGradient Routing Localizes and Removes Unintended Behaviors in RLJake Ward, Shawn Hu, Aria Wong, Nathan Hu +3
We propose a mitigation to reward hacking in reinforcement learning. Our setup involves a misspecified reward function that gives rise to unintended behavior, and a monitor used to detect unintended behavior. As with frontier AI systems, the monitor makes systematic errors. Consequently, using the m
post-training - arxiv:2610.17059 · cs.AIPatientWM: Towards Generalizable Clinical World Models via Unified Patient StatesZiwen Kan, Jie Peng, Bo Yan, Davis Hobley +4
Clinical decisions depend on anticipating how a patient's condition will evolve as treatment is adjusted and new evidence becomes available. Existing medical world models capture patient-state evolution, but often define the evolving state around a particular modality or task, limiting reuse as inpu
world model - arxiv:2610.17057 · cs.CVVL-DocIR: A Benchmark for Vision-Based Long Document Information RetrievalSimon Ott, Yufan Chen, Ruiping Liu, Junwei Zheng +5
Vision-based document retrieval has improved rapidly, but current evaluation settings still emphasize short documents and single relevant pages. This leaves an important gap between leaderboard performance and realistic long-document retrieval, where the evidence needed to answer a query may span mu
benchmarkleaderboard - arxiv:2610.17043 · cs.AINeos: Learning to Interleave Search and Visual ReasoningEkko He, Jack Hong, Shilin Yan, Xintong Zhang +7
Recent multimodal deep search agents have advanced tool use and visual content retrieval, yet they still struggle to convert intermediate visual evidence into actionable guidance for subsequent search, limiting their effectiveness in long-horizon settings. To address this gap, we present Neos, a mul
tool usebenchmark - arxiv:2610.17042 · cs.CLAD4EDU: Benchmarking University Lecture Audio Description across Accessibility Standards, Metrics, and Learner PreferencesPsychias Stylianos, Gao Yingqiang
Audio description (AD) aims to enhance the accessibility of visual media for blind and low-vision (BLV) persons. However, in higher education, where slide-based lectures are a default mode of teaching, lecture AD remains largely underexplored. The main reason is that existing production guidelines,
benchmark - arxiv:2610.17039 · cs.LGMENTOR: Procedural Shielding of Tool-Using LLM AgentsAshwani Anand, Ivi Chatzi, Ritam Raha, Anne-Kathrin Schmuck
Tool-using LLM agents are increasingly used to automate workflows governed by procedural manuals, such as company rules for booking travel or procuring hardware. These manuals are written in natural language for humans and describe general requirements over such complex workflows. Ensuring complianc
agentllm agent - arxiv:2610.17027 · cs.AIBenchmarking Quantum Kernel SVMs for Credit Card Fraud Detection: Performance Realities Under NISQ ConstraintsHitanshi Arora, Mohit Swarnkar, Aswani Kumar Cherukuri
Detecting credit card fraud is notoriously difficult because fraudulent charges make up less than 0.2% of total transactions. Quantum Machine Learning (QML), particularly Quantum Kernel Support Vector Machines (QSVM), promises richer feature representations by embedding classical data into high-dime
benchmark - arxiv:2610.17020 · cs.AICWE-bench: Refocusing Agentic Security from Exploitation to DefenseGonzalo Gonzalez-Pumariega, Ajay Jagannath, Adit Jain, Guram Gogia +2
Cybersecurity evaluations of AI agents have focused largely on offense, yet a working exploit does not show that an agent can make software safe. An attacker needs to exploit only one vulnerability; a defender must find and fix all of them. We introduce CWE-bench, a private benchmark of 120 held-out
agentai agentagenticbenchmark - arxiv:2610.17018 · cs.LGLLM Evaluation is a Forecasting ProblemRahul Thomas, Adam Block, Micah Goldblum
Standard procedure in machine learning is to randomly split train and test data, emulating IID draws from one underlying data distribution. In practice, models are trained on past data and deployed in the future, which may come from distinct distributions. In this paper, we find that web text is ind
benchmark - arxiv:2610.17016 · cs.LGProbabilityBench: Model Evaluation and Calibration with Exact Ground-Truth UncertaintyYuyang Dai, Yuting Tang, Futoshi Futami, Takashi Ishida
Evaluating model uncertainty on real-world data is challenging because each instance provides only a single observed label, leaving true conditional class probabilities unknown. To address this, we introduce ProbabilityBench, a benchmark that provides exact, instance-level class probabilities across
benchmarkevaluation framework - arxiv:2610.17009 · cs.CVMaskSense: Confronting the Visual Exploration Trap in Masked Image GenerationYawen Shao, Jie Xiao, Kai Zhu, Yu Liu +5
Reinforcement learning (RL) has shown strong potential for aligning generative models with human intent, but adapting RL to masked generative models (MGMs) remains largely underexplored. Existing approaches that directly adapt Group Relative Policy Optimization (GRPO) to MGMs leave a foundational co
benchmark - arxiv:2610.17003 · cs.ROMulti-Agent Deferred-Decision Trajectory Optimization with Anytime-Valid Target Retention for Moving CandidatesFarbod Siahkali, Mengxue Hou, Vijay Gupta
We study multi-agent terminal interception when one of several moving candidates is the unknown target and both target identity and candidate terminal locations are uncertain. We develop a receding-horizon deferred-decision trajectory optimization (DDTO) controller that combines conformal terminal-l
agentmulti-agent - arxiv:2610.17001 · cs.ROMOSAIC: Mode- and Support-Aware Inference-Time Control of Robot Diffusion ModelsSabrina Bodmer, Rahel Rickenbach, René Zurbrügg, Melanie N. Zeilinger +1
Generative policies such as diffusion models provide expressive data-driven priors for robot behavior, yet the situations a robot faces at execution time rarely match its training data and impose constraints and objectives the policy never learned. Existing inference-time guidance methods either rel
manipulationdexterousgrasp - arxiv:2610.16998 · cs.AIGlasseek: A Transparent Recipe for Training and Trustworthy Evaluation of DeepSearch AgentsCong Pang, Jiaqi Su, Chengrui Wu, Kuihao Yan +13
Deep search enables language agents to solve complex information-seeking tasks through iterative web exploration. However, developing robust open-source agents remains challenging for three reasons. First, opaque trajectory-construction pipelines make training data difficult to audit and reproduce.
context compressionagentbenchmarkevaluation protocol - arxiv:2610.16985 · cs.CVAUGUstus: Autonomous Driving VLA Model with Unified Generation and UnderstandingYuqi Ye, Caigao Jiang, Zhuonan Wang, Congyuan Pei +11
Vision-language-action (VLA) models show strong potential for end-to-end autonomous driving by leveraging world knowledge and understanding to excel in long-tail scenarios. However, current VLAs have two limitations: 1) Degeneration into vision-action (VA) models. For low latency, regression/flow-ba
vision-language-actionvlavla model - arxiv:2610.16984 · cs.LGFrom Suboptimal to Optimal: One-Step Latent Flow Policies for Offline Reinforcement LearningShenghua Lu, Zixuan Yang, Rongxu Zhao, Bohao Qu +2
Offline reinforcement learning aims to learn effective policies from fixed datasets without environment interaction. Although generative models such as diffusion and flow matching enhance policy expressiveness, they typically start from uninformative Gaussian noise and generate actions through multi
benchmark - arxiv:2610.16980 · cs.LGPRAISE: A Reproducible Workflow for Non-Programmer Scientists to Implement Published AI ModelsMax Achterweust, Stefan Groot Nibbelink, Vincent J. de Beer, Stefan Schipper +6
Artificial Intelligence (AI) models are increasingly developed, published and used across scientific fields. Researchers, students and industry professionals who want to apply these models to their own research questions often lack the programming and data science expertise to do so. Existing guidel
evaluator - arxiv:2610.16977 · cs.CVThe Value of Being Wrong: Self-Mined Visual In-Context Learning from ErrorsShu Zhao, Tan Yu
Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depends critically on the examples selected. Recent counterfactual selection methods improve ICL by constructing demonstrations that expose how individual at
benchmark - arxiv:2610.16975 · cs.AIMADDIE: Multi-Agent Drug Discovery EngineYe Liu, Reza Averly, Avery Meyer, Frazier N. Baker +3
Drug discovery remains a resource-intensive and time-consuming process, involving billions of dollars and a decade of development to advance a drug to market. While many in silico tools have accelerated this process, operating them still demands substantial time, effort, and expertise from medicinal
multi-agentagenticagent systemevaluator - arxiv:2610.16971 · cs.CLEvent Memory and Narrative Dynamics: A Large Language Model Framework for Multidimensional State SpacesJoshua Sariñana
Studying event memory requires defining events and their relationships within remembered experience. Autobiographical narratives provide evidence of this organization, but their analytical units must first be established. This study introduces a large language model framework for constructing event
memoryepisodic memory - arxiv:2610.16969 · cs.ROUniLAP: A Unified Latent Action Policy for Cross-Embodiment ManipulationLei Yao, Qiao Feng, Mutian Tong, Qihao Yun +2
Latent Action Models (LAMs) can capture manipulation priors from action-free videos, but discrete latents usually impose a granularity bottleneck, limiting their ability to represent fine-grained action. Recent continuous alternatives mitigate this, but they are typically used as proxies for world m
vision-language-actionmanipulationhumanoidworld model - arxiv:2610.16966 · cs.LGLearning to Design SR Solvers: Harness Optimization for Symbolic RegressionQian Li, Yuxiao Hu, Qihao Duan, Jinyue Yan +1
Symbolic regression (SR) seeks explicit equations that explain observed data, and recent methods increasingly employ large language models (LLMs) and agents for equation discovery. However, these systems typically rely on manually designed workflows and prompts. We argue that the SR agent harness sh
agent - arxiv:2610.16959 · cs.RODiTSG: Context-Aware Air Traffic Scene Generation with Multi-Agent Diffusion TransformersSeokbin Yoon, Kyungmin Kim, Taejung Kwon, Keumjin Lee +1
Understanding how many aircraft, and which traffic configurations, can be safely and efficiently managed by air traffic controllers within a given airspace is fundamental to air traffic management. Air traffic controllers continuously modify aircraft trajectories through interventions such as speed
multi-agent - arxiv:2610.16956 · cs.CVDDVR-Net: Degradation-Aware Decoupling and Visibility-Guided Restoration for Historical Document RecognitionJiu Sha, Yu Weng, Mengxiao Zhu
Historical document analysis is crucial for cultural heritage preservation and low-resource language digitization. However, real historical documents often suffer from severe degradation, such as ink erosion and structural damage, causing mixed-visibility character sequences. Conventional OCR system
benchmark - arxiv:2610.16948 · cs.LGLearning to Design Convex Relaxations for Nonconvex OptimizationJinrun Liu, Bai Yan, Tengfei Liu, Qunfeng Liu +4
Convex relaxations make nonconvex optimization tractable by replacing hard nonconvex problems with convex surrogates that give valid bounds. Yet their design is still largely manual: a human expert must decide where to relax, which relaxation to use, and how much formulation growth is worth paying.
benchmark - arxiv:2610.16946 · cs.LGSocial Priors: Descriptions of AI Users Shape How Language Models Trust and CooperateHarsh Kumar, George Eilender, Ashton Anderson
Descriptions of how people use AI are becoming training data for the next generation of models. We ask whether what a model reads about its users changes how it treats people. Using contrastive synthetic-document finetuning, we train Qwen3-32B on four matched corpora built from the same 1,625 real-w
agent - arxiv:2610.16945 · cs.CVUniMetric: Universal Metric Scale Recovery for Visual Geometry Foundation ModelsLei Yang, Yuner Zhang, Ziyang Liu, Ziying Song +4
Recent advances in multimodal foundation models, exemplified by GPT-6 Astra, raise the prospect of general-purpose embodied reasoning and planning. Using Internet videos for embodied pretraining calls for reliable physical-scale annotations. Yet visual geometry foundation models often recover scene
embodied - arxiv:2610.16942 · cs.AIUA-Bench: Benchmarking Large Audio-Language Models on Underwater Acoustic RecognitionZhe Sun, Liang An, Jiayu Yao, Ronglai Qian +8
Large audio-language models (LALMs) are fast becoming a general interface to sound, yet how far their broad pretraining carries into specialized domains remains largely open. Vessel monitoring, maritime safety, and ocean sensing all depend on recognizing underwater signals, but to our knowledge LALM
benchmark - arxiv:2610.16940 · cs.LGKNN Implementation Details Can Dramatically Change Performance: An Example from Cover TreesAmol Khanna, Edward Raff
Nearest neighbor search algorithms are widely used and have a long history, but we find that key implementation details are seldom reviewed in articles about them. These details can materially impact performance, and thus change the conclusions one makes about a given algorithm's superiority. In thi
benchmark - arxiv:2610.16921 · cs.LGTime Window Optimization for Motor Imagery Classification via Temporal Structure Embedded Reinforcement LearningHuan Luo, Na Lu, Xiaopeng Wang, Ziyang Shang
Online motor imagery-based brain-computer interfaces (MI-BCIs) typically decode continuous EEG using fixed-length sliding windows that may be misaligned with neural events. Existing window optimization methods often evaluate windows independently, overlooking temporal dependencies and potentially pr
benchmark - arxiv:2610.16919 · cs.LGML Engineering Agents - A Data Management PerspectiveArnab Phani, Elias Strauss, Tim Schuster, Sebastian Schelter
Machine learning engineering (MLE) agents are emerging as a new paradigm for automating end-to-end ML pipeline development. Using large language models for code generation and predictive performance as a machine-gradable feedback signal, these agents iteratively synthesize, execute, and refine pipel
benchmark - arxiv:2610.16916 · cs.AITransparent Information Flow Modeling for Role-Aware AI Act AssessmentMattias Brännström, Themis Dimitra Xanthopoulou, Lili Jiang
The European Union's Artificial Intelligence Act introduces a structured framework for governing high-risk AI systems, emphasizing transparency, risk management, and role-specific accountability. However, current assessment practices often rely on monolithic system views or disconnected checklists,
tool use - arxiv:2610.16911 · cs.AIAudit tools of the trade: Automated Auditor Elicitation Can Be Sensitive to the Context the Auditor Is GivenBen Slater, Ryan Lundqvist, Leo Richter, Kevin Wei
Does providing additional context to automated AI auditors improve their effectiveness? Using the Petri tool to audit LLM coding agents, we test the impact of providing two types of resources (an external codebase and benchmark transcripts) to LLM auditors, across four auditor-target LLM pairs and t
benchmark - arxiv:2610.16905 · eess.SYCounterfactual Action-Conditioned Graph Prediction for Cascading-Failure Mitigation in Power SystemsFangxin Liu, Zhentao Tian
Cascading failures develop through coupled power flow redistribution, protection actions, and topology changes, making remedial control dependent on how alternative interventions modify the subsequent trajectory. This paper develops an action-conditioned graph predictor (ACGP) for multi-step cascade
action-conditioned - arxiv:2610.16902 · cs.LGNeural Bayesian Networks for Modular and Scalable Decision-Making Under UncertaintyGiovanni Briglia, Marco Lippi, Stefano Mariani, Franco Zambonelli
Bayesian Networks (BNs) are established tools for letting agents reason under uncertainty in sequential decision-making. They split a joint distribution into local conditional mechanisms arranged on a directed acyclic graph, which makes them modular, interpretable, and ready to perform probabilistic
benchmark - arxiv:2610.16895 · cs.ROObserve, Share, or Predict? Disentangling Partner Information in Embodied Multi-Agent CollaborationAdrián Szlatincsán, Filippo Ziliotto, Bruno Lepri, Luciano Serafini +2
LLMs are used as high-level controllers for embodied agents, but it remains unclear whether they support collaborative reasoning, such as anticipating a collaborator's actions in real-time teamwork without dialogue. We ask if observing the partner's state, exchanging object knowledge, and predicting
embodiedmulti-agentembodied agentbenchmark - arxiv:2610.16890 · cs.LGFORGE: Procedural Weight Generation for Sparse MoE InferenceChanyoung Moon, Travis Pence, Daisuke Yamada, Jiaqi Mo +3
We introduce FORGE (Fast On-chip Rotor Generation for Experts), a procedural weight generation system for Sparse Mixture-of-Experts (MoE) inference. Rather than storing and loading every expert weight tensor from HBM, we show how a compact accelerator inside each Streaming Multiprocessor can regener
memory - arxiv:2610.16884 · cs.LGLearning Operators for Symmetry Discovery: From Lie Groups to Local Symmetry GeometryPavan Karjol, Scindhiya Laxmi, Prathosh AP
Continuous symmetry discovery is typically formulated as a separate estimation problem for each function or dataset. We instead investigate whether the map from a function to its symmetry structure can be learned across a family of functions from finite observations. Using the infinitesimal differen
benchmark - arxiv:2610.16879 · cs.CVEviStress: Evidence Stress Learning for Generalizable Face Forgery Detection with Textual ExplanationsJiuyao Jing, Xuanyu Yi, Yu Zheng, Chunlei Peng
The rapid spread of face forgeries on the Web challenges the credibility of visual content. In practical Web environments, manipulated media originate from evolving generation pipelines and undergo diverse processing during dissemination, making reliable authenticity assessment under unseen manipula
manipulation - arxiv:2610.16870 · cs.CLSpecGuardBench: Can Coding Agents Construct Reliable Validation Harnesses from Natural Language Specifications?Zhiqing Zhong, Mingyu Chen, Aoyang Fang, Yuzhong Zhang +4
As coding agents become more capable, tests remain the primary evidence for judging whether their code is correct. Yet passing available tests does not establish that an implementation satisfies its natural language specification. Reliable validation requires checks that faithfully capture the speci
agentbenchmark - arxiv:2610.16866 · cs.LGEfficient Resource-Shiftable RL Post-training with Generative RewardsYanfeng Zhao, Haoyang Li, Sheng Lin, Yuming Zhou +3
Reinforcement learning (RL) post-training with generative reward models (GenRM) has become a central recipe for improving large language models (LLMs), and it is also increasingly resource-intensive: policy rollout, GenRM-based reward evaluation, and policy training all compete for GPU resources. Ho
post-training - arxiv:2610.16867 · cs.LGDavinci-Tool: Scaling Agentic Tool-Use Supervision with Verifiable Executable EnvironmentsXiaoyang Yuan, Ji Zhao, Xingyu Hu, Wenhan Wang +6
Reliable agentic tool use requires training data that pair diverse multi-step workflows with executable evidence that the requested outcomes were achieved. Scaling either trajectory coverage or verification alone is insufficient as new environments introduce distinct states, rules, and dependencies
agentictool usetool-usebenchmark - arxiv:2610.16865 · eess.SYController-Agnostic Runtime Safety Certificates for Tool-Using AI Agents via Data-Driven Robust ReachabilityAli Baheri, Peng Wei
Software agents built on large language models act by issuing tool calls, some of which cannot be undone. To control theory the model is an opaque feedback controller, replaced every few months, whose closed loop cannot be analyzed directly. We propose a runtime filter, or shield, between any such c
ai agent - arxiv:2610.16861 · cs.LGGpFlash: Maximizing Speed and Minimizing Memory for Gaussian Process Training and PredictionAndre Marchildon
This paper presents GpFlash, a Gaussian process library that is focused on fast training and posterior evaluations, minimal memory usage, and scalability to high-dimensional problems. GpFlash was compared to the popular GP libraries scikit-learn, GPy, GPflow, GPyTorch, and SMT on a MacBook Pro and t
memory - arxiv:2610.16859 · cs.CLANCHOR: Assessing Naturalistic Constraint Hierarchies and Ordering in Repository-Grounded Agentic CodingWenqiang Zhu, Xinping Lei, Zhiyi Lai, Zhengshuo Gong +15
For repository-grounded coding agents, a functionally correct patch is not necessarily an acceptable one: developers also constrain how the task must be carried out-modification scope, workflow order, validation, tool use, and reporting. This capability, which we call engineering constraint followin
agentagentictool usetool-usebenchmarkevaluation framework - arxiv:2610.16855 · cs.ROHuman-Like Redundancy Resolution for Real-Time Bimanual Humanoid TeleoperationChao Tan, Ming Shin Cindy Sung, Shuxian Liang, Chaoqun Wan +2
Real-time bimanual humanoid teleoperation requires accurate wrist tracking while keeping arm configurations predictable. Because a 7-DoF arm admits multiple elbow configurations for the same wrist pose, wrist-only tracking can cause elbow drift or abrupt configuration changes. Fixed anthropomorphic
humanoidteleoperation - arxiv:2610.16851 · cs.LGRadial Tail Repair: Post-Training Correction for Diffusion and Flow ModelsHamza Cherkaoui, Hélène Halconruy, Antonio Ocello
Diffusion and flow-matching models may struggle to accurately represent distributional tails, especially for heavy-tailed data. Existing approaches often use heavy-tailed noise or modify the generative dynamics, requiring dedicated retraining and more difficult estimation. We take a simpler approach
post-training - arxiv:2610.16849 · cs.CVA Multi-Task Deep Learning Framework for Pose-Invariant Face Recognition and Facial Landmark LocalizationAli Ashkeyev, Zukhra Abdiakhmetova
Face recognition systems often experience performance degradation under large pose variations due to changes in facial appearance and visibility of discriminative features. This study presents a multi-task deep learning framework that jointly addresses pose-invariant face recognition and facial land
benchmark - arxiv:2610.16838 · cs.AISame Operation, Different Permission: Evaluating Whether Guardrails Honor AuthorizationHuihao Jing, Shaojin Chen, Wenbin Hu, Yuxuan Liu +4
As language-model agents write code, move money and change records on behalf of users, deciding whether an action is authorized has become a central safety problem. Guardrails are often placed in front of these actions, yet few evaluations ask whether their decisions follow authorization: blocking u
benchmark - arxiv:2610.16837 · cs.AIA Scaled-Up Empirical Study of Syntactic Language ModelsPengcheng Wang, Zhexuan Zhang, Chao Lou, Kewei Tu
Syntactic language models (SLMs) that generate sentences along with their syntactic tree structures have shown promise in both syntactic generalization and downstream tasks. Prior studies, however, have mostly evaluated smaller models trained on less pretraining data and tested on narrower task suit
benchmark - arxiv:2610.16826 · cs.LGUnsupervised Data Drift Monitoring for Deployed Deepfake Speech DetectorsLukáš Prokeš, Vojtěch Staněk, Anton Firc, Kamil Malinka
Deepfake speech detectors are typically evaluated as binary classifiers on fixed benchmarks, but deployed systems face evolving attacks, codecs, speakers, and recording conditions, with labels unavailable during operation. This makes it difficult to know when a detector operates in familiar conditio
benchmark - arxiv:2610.16825 · cs.LGMean Accuracy Is the Wrong Estimand for a Release DecisionJaewon Choi, Jihye Kang, Namhyuk Ahn
A release decision about a fine-tuned model is an operational commitment about the \emph{next} rerun of its training recipe. Yet standard benchmark reporting gives only the mean accuracy of past seeds. These target fundamentally different questions: release decisions depend on catastrophic tail risk
benchmarkevaluation framework - arxiv:2610.16820 · cs.LGSample-Size Extrapolation in Simulation-Based Inference via Statistical ScalingShiyi Sun, Lei Fang, Geoff K Nicholls
In simulation-based inference (SBI), neural posterior estimators trained on a bounded range of sample sizes can become inaccurate on larger datasets because their parameterizations do not specify how posterior location, uncertainty and shape must change as observations accumulate. We propose center-
benchmark - arxiv:2610.16814 · cs.AIStableJudge: A Lightweight Adversarial Defense Framework for LLM-as-a-JudgeAnning Hu, Yangyu Li, Yuqi Huang, Shengchao Hu +1
Large language models (LLMs) are increasingly used as judges to assign scalar scores that drive downstream decision-making, including tasks such as paper review and recommendation. However, LLM-as-a-Judge pipelines are vulnerable to score-hijacking prompt injection embedded in evaluated content, whi
judge model - arxiv:2610.16813 · cs.LGCarve VFM: Carving Specialists out of Generalists, Automatic Task-Specific Compression of Vision Foundation ModelsLotfi Abdelkrim Mecharbat, Abdelrahman Elsayed, Ibrahim Almakky, Martin Takáč +1
Vision foundation models (VFMs) provide transferable visual representations, but their inference cost limits deployment in many settings. Fine-tuning specializes a VFM without shrinking it; distillation shrinks it, but through expensive teacher-guided training. Structured pruning offers a more direc
memorybenchmark - arxiv:2610.16811 · cs.CVX-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray ReasoningYi Wang, Zhenlong Yuan, Yuchan Guo, Maolei Huang +8
Chest X-ray (CXR) interpretation often requires integrating visual findings with task-specific evidence rather than relying on direct visual recognition alone. While tool-augmented medical agents can supply such evidence via external modules, they often lack explicit mechanisms for learning when aux
agenticpost-trainingbenchmark - arxiv:2610.16801 · cs.LGWhen Repeated Measurements Hurt Tabular Foundation ModelsYixiang Liu, Ruizhe Deng, Bibhas Chakraborty, Qiong Zhang +1
Tabular foundation models (TFMs) achieve state-of-the-art performance on IID benchmarks, but how do they respond to repeated measurements at the same inputs? Across six TFMs, synthetic and semi-synthetic experiments show that additional measurements can improve predictions at measured inputs while d
benchmark - arxiv:2610.16793 · cs.CVViewStruct: View-Grounded Articulation and Connectivity InferenceWenjing Bian, Ranran Huang, Andreas Geiger
Interacting with unfamiliar objects requires understanding how their parts connect and move. We introduce ViewStruct, a framework for jointly predicting connectivity and articulation in objects and scenes. Given an RGB image with aligned part segmentation and depth, optionally supplemented by images
manipulation - arxiv:2610.16784 · cs.LGInfoVis: Unified Slide Image Generation and EditingJiahao Wang, Xiyan Jiang, Chenyuan Wu, Haoqiang Zhang +6
Generating and editing presentation slides requires models to combine accurate text rendering, instruction following, visual grounding, aesthetic design, and precise editing within one canvas. Existing work covers only subsets of these capabilities and lacks large-scale slide-native supervision and
post-trainingbenchmark - arxiv:2610.16782 · cs.AILIBERO-Industry: Benchmarking Robotic Manipulation under Industrial Process ConstraintsMing Shin Cindy Sung, Duan Leijiahao, Haoxiang Wang, Boning Tan +8
Industrial manipulation in flexible manufacturing, e.g., high-mix, low-volume (HMLV) electronics assembly, requires robot policies to adapt to changing products and production conditions. Such settings impose industrial process constraints, including product variants, moving material flows, finite e
vision-language-actionmanipulationliberobenchmarkevaluation frameworkevaluation protocol - arxiv:2610.16776 · cs.AIGame-Creation-Agents: Autonomous Game ProductionTencent Multimodal RL Team, Xiangxin Zhou, Jiaqi Tang, Ke Ma +14
Game-Creation-Agents (GCA) turns a natural-language request into a runnable Phaser 4 browser game. The pipeline first builds a multimodal game design document that couples mechanics and scene contracts to grounded visual references. It then projects that document into a runtime scaffold, machine-rea
agent - arxiv:2610.16763 · cs.LGRelational Learning as World ModelingZheyu Zhang, Gjergji Kasneci
Relational learning typically predicts from a database state, but decision making asks a different question: how will that state change under an action? In operational systems, the answer depends on system-specific dynamics that may not be visible in the database itself. We view each such system as
world model - arxiv:2610.16759 · cs.CVImage Tokenizers Should Be Trained Between ImagesOle Beißwenger, Jan-Niklas Dihlmann, Hendrik Lensch
Latent diffusion models generate images as codes of an image tokenizer, whose decoder turns them into the pixel image. Recent tokenizers improve how well these codes reconstruct an image and how easily a generator learns them. Yet a generator does not reproduce the codes of training images, and when
benchmark - arxiv:2610.16755 · cs.CLThe Multilingual Artificial Hivemind: Do All LLMs Think Alike in Every Language?Xaver Raphael Davey, Shubhra Mishra, Anton Adelöw, Siyao Liu
Text generated by Large Language Models (LLMs) often remains limited in diversity compared to that produced by humans. Recent advances demonstrate the presence of inter- and intra-model collapse in open-ended generation of English text, raising concerns about the long-term homogenization of LLM, and
benchmark - arxiv:2610.16744 · cs.CVFoveaTree: Information-Density Trees for Query-Guided Visual Token Pruning in Video Large Language ModelsZihan Meng, Weidong Chen, Dexiang Hong, Xinyan Liu +1
Video large language models (Video LLMs) encode each frame into hundreds of visual tokens, making video inference expensive. Existing methods reduce this cost through similarity-based merging or attention-based selection but face two challenges: Hierarchy-blind Pruning, which overlooks regional comp
benchmark - arxiv:2610.16737 · cs.ROCopyQuad: Quadrupedal Skill Transfer from a Single Side-View Video Without a Depth ModelMohamad H. Danesh, Raina Hu, Mengxue Xu, Hsiu-Chin Lin
Video-to-robot pipelines for quadrupedal imitation insert a learned lifting stage, either a monocular depth model or a per-clip 3D reconstruction, to recover the one axis a side-view camera cannot see. We present CopyQuad, which omits that stage and pins the missing axis to the robot's own stance wi
quadruped - arxiv:2610.16734 · cs.CVPathPilot: An Adaptive Question-Conditioned Multi-Magnification Navigation Agent for Whole-Slide Image AnalysisZhengyang Xu, Han Li, Jingsong Liu, Ayako Ura +4
The diagnostic capability of WSI agents relies on effective slide navigation to identify diagnostically relevant regions and efficiently integrate evidence across the slide. While in clinical practice, pathologists dynamically examine slides across multiple scales, integrating global and cellular ev
agent - arxiv:2610.16725 · cs.AITrace-to-Architecture: Capability Bottleneck Diagnosis for Harness RevisionWenfeng Wang, Jianrui Shi, Mingang Guo, Jiayuan Ma +4
Improving an LLM agent requires knowing what to change when it fails. Similar failures can require different fixes, and resolving one problem can expose another. We introduce Trace-to-Architecture (T2A), a framework that uses execution trajectories to determine what to change in an agent's harness a
agentllm agent - arxiv:2610.16720 · cs.LGLearn from Move: the Next Step for GUI AgentsJiahao Lu, Yize Shen, Chuanyuan Tan, Mianqiu Huang +5
Existing benchmarks for GUI agents primarily focus on increasing task difficulty by extending task horizons. Besides, enabling agents to explore unfamiliar environments and successfully complete tasks within them is also a critical yet challenging problem. We introduce LatentGUIWorld, a benchmark of
benchmark - arxiv:2610.16692 · cs.CV3D-VITAL: Visibility-aware Identity Training from Artificial Liftings of 3D Reconstruction model for Cross-View Object Re-IdentificationWilliam Grolleau, Astrid Sabourin, Guillaume Lapouge, Catherine Achard
We address the Cross-View Object Re-Identification (ReID) task, aiming to match identities across drastically different viewpoints. We focus on Single-View Aerial-Ground ReID (SV AG-ReID), where models train on one viewpoint and must generalize to another at inference. To bridge this viewpoint gap,
benchmark - arxiv:2610.16689 · cs.CLEHR-Bench: A Chinese Multi-Task Benchmark for Recruitment AIRuijia Yang, Bing Xu, Zulong Chen, Zhipeng Xu +4
Recruitment applications span diverse operations, from drafting job descriptions and interpreting queries to organizing profiles, assessing fit, and recommending candidates and jobs. Systematic LLM evaluation therefore requires task coverage that reflects this range of business needs. We introduce E
benchmarkevaluation framework - arxiv:2610.16680 · cs.LGSpectral Costs of Multinomial and Residual Resampling in High-Dimensional Particle Ensembles: A Free-Probability AnalysisDeming Song, Chi Kin Lam
Resampling is the standard remedy against weight degeneracy in particle filters, yet its effect on the spectrum of the particle second-moment matrix is poorly understood. We study a single resampling step for a d-dimensional ensemble of N particles with d/N -> c. For any unbiased scheme, the resampl
benchmark - arxiv:2610.16679 · cs.CVEvoSpatial: Experience-Driven Skill and Tool Evolution for Spatial Reasoning AgentsBin Yu, Shijie Lian, Zhaolong Shen, Xiaopeng Lin +9
Tool-augmented vision-language agents compose perception and measurement to reason about spatial relationships, but effective execution also requires tracking geometric assumptions and checking intermediate evidence. Preserving this experience across tasks requires retaining procedural knowledge tog
benchmark - arxiv:2610.16667 · cs.ROFEM-Validated Real-Time Modelling of Soft Tactile DeformationSilvia Terrile, Kaichen Shi, Taejun Kwon, Jiseok Lee +4
Accurate modelling of soft tactile sensors remains difficult because finite element methods (FEM) are computa- tionally expensive, while real-time simulators often lack fidelity for hyperelastic deformation. Large physical datasets are also impractical to collect because repeated testing can damage
tactilesim-to-real - arxiv:2610.16666 · cs.CVRevisiting Natural Language Prompts from Large Language Models for Open-Vocabulary Video RecognitionHaichen He, Weibin Liu, Weiwei Xing
Vision-language models (VLMs) like CLIP have shown great success in image tasks, however, applying CLIP to open-vocabulary video recognition poses challenges. Due to the limited text information, directly fine-tuning CLIP on video recognition datasets may lead to overfitting and hinder its generaliz
benchmark - arxiv:2610.16664 · cs.LGMaking LoRA Identifiable: Orthogonal Alignment for Continual LearningPratik Rakesh Singh, Mohammadi Zaki, Akash Saha, Aneesh Mukkamala +1
There is growing interest in applying Parameter-Efficient Fine-Tuning (PEFT) to Continual Learning, as it enables faster training, mitigates catastrophic forgetting, and facilitates better adaptation to new tasks. Prior approaches have explored a range of strategies, including extending new LoRA bra
memorybenchmark - arxiv:2610.16661 · cs.AIStructural Baselines Matter for Agent Failure Detection and AttributionRamesh Arvind Naagarajan, Stefan Streif
An LLM agent using tools can fail silently. It stops before the task is done, raises no error, and answers fluently. Two kinds of monitor address this, one deciding whether a run failed and the other naming the step where failure became unavoidable. Much of the reported performance on both tasks com
agentllm agentagent benchmarkbenchmark - arxiv:2610.16660 · cs.LGCompression Through Time: An Interplay of Separability, Stationarity and CapacityNikolaus Kopp, Franz Pernkopf, Bernhard Geiger
Information-theoretic and geometric compression have been widely studied as possible explanations for generalization. Yet their temporal dynamics in recurrent networks remain largely unexplored. Using discretized LSTMs, whose finite embedding space enables direct estimates of information quantities,
memory - arxiv:2610.16652 · cs.LGIND-Eval: Aligning Evaluation and Simulation for Industrial Time Series Foundation ModelsVebjørn Haugland, Pavel Gulin, David Foss, Peet Cremer +4
We introduce IND-Eval, a public benchmark that organizes industrial time series across domains into source-specific forecasting tasks. Relative to established benchmarks such as GIFT-Eval, it emphasizes finer sampling resolutions and a greater prevalence of multivariate forecasting. Evaluating Chron
benchmark - arxiv:2610.16646 · cs.CLSIREN: Situation-aware Index for Response and Emergency ActioNability in Crisis TranslationMaria Carmen Staiano, Lifeng Han, Johanna Monti, Francesca Chiusaroli
Machine translation (MT) is increasingly utilized in multilingual crisis communication, yet conventional evaluation does not directly assess whether translated emergency messages preserve their actionability. We introduce SIREN (Situation-aware Index for Response and Emergency ActioNability), a huma
evaluation framework - arxiv:2610.16641 · cs.ROWhat Limits Cross-Task Generalization in World Action Models?Chen Zhao, Zipeng Ye, Yihan Lin, Yuchen Xie +1
Cross-task generalization remains difficult for world action models (WAMs), even when their predicted futures depict plausible manipulation. This gap can be attributed directly: we make this gap directly testable by decoupling a WAM into a world model and an inverse-dynamics model (IDM), then holdin
manipulationgripperworld modelbenchmark - arxiv:2610.16636 · cs.AISuPER-Mamba: Surprisal Prioritization for Extended-Context Routing in MambaZhetong Zhang, Gayan K Kulatilleke
State space models provide efficient linear-complexity processing for long sequences, but their performance frequently degrades as context length grows. We propose SuPER-Mamba, a training-free framework that uses Surprisal as a model-intrinsic signal for long-range state routing in pretrained Mamba
benchmark - arxiv:2610.16635 · cs.LGACED: Reference-Anchored Aggregation of Natural-Language and Executable Evidence for Mathematical ReasoningYiyi Lu, Jiaxiang Geng, Bing Luo
Natural-language self-consistency and executable programs provide complementary test-time evidence for mathematical reasoning, yet either channel may support the same semantic error. Thus we propose ACED, a reference-anchored framework that treats the self-consistency answer as the default and execu
benchmark - arxiv:2610.16634 · cs.ROLanguage2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving?Zihan You, Zehua Fan, Wenqi Lyu, Aoqi Wang +18
Recent Vision-Language-Action (VLA) models increasingly use natural-language instructions to condition autonomous driving behavior. Yet existing closed-loop benchmarks primarily measure route completion, safety, and rule compliance, leaving a fundamental question unresolved: does language actually d
vision-language-actionbenchmark - arxiv:2610.16632 · cs.CVREEL: Learning to Retrieve, Evaluate, and Evolve Visual Memory for Any Long-Horizon Multi-Shot Video GenerationHang Xu, Guohui Zhang, Xiaoxiao Ma, Hu Yu +6
Memory-based long-video generation aims to combine user control with long-range visual consistency. We organize this process into three functional components: a planner, an adapter, and a generator. Representative multi-shot systems couple memory to generator-specific interfaces or use predefined ag
memorypersistent memoryagent - arxiv:2610.16630 · cs.LGAeroChem: Closed-loop Physics-Informed State Space Modeling for Long-term Chemically-Reactive Air Quality Forecasting--Extended VersionKien Le Trung, Tung Kieu, Dinh Duc Nha Nguyen, Cuong Do +2
Accurate long-term air quality forecasting is challenging due to long-range transport, meteorological variability, and nonlinear chemical reactions underlying secondary pollutant formation. Purely data-driven models often suffer from long-horizon drift and lack physical consistency, while existing p
benchmark - arxiv:2610.16625 · cs.AILATCH: LLM-Assisted Timing Closure Harness for Post-Route FPGA OptimisationShu Gu, Matthew Youngman, Yuqin Zhao, Cristian Sestito +4
Post-route FPGA timing closure remains expert-dependent and computationally expensive. We present LATCH (LLM-Assisted Timing Closure Harness), a two-layer framework that separates large language model (LLM) orchestration from deterministic FPGA implementation and evaluation. Across sixteen designs a
agent - arxiv:2610.16621 · cs.MAAgentCue: Statistical Summaries for Guiding Decentralized Multi-Agent Self-OrganizationGuojun Zhu, Xinyou Chen, Kejian Zhang, Sanguo Zhang +1
Multi-agent systems (MAS) combine multiple agents to solve complex tasks, and recent MAS scale to more than a thousand agents. At this scale, decentralized MAS remove the central orchestrator that assigns sub-tasks, and the agents, which we call workers, self-organize through their own choices of su
multi-agentagent system - arxiv:2610.16618 · cs.RODR-APO: Decoupled-Representation Advantage-Guided Policy Optimization from Continuous Control to Embodied NavigationRunhua Zhang, Yingjie Niu, Chao Li
Policy representations in reinforcement learning have expanded beyond Gaussian policies to diffusion and flow-matching based generative policies. However, existing RL methods remain representation-specific, lacking a unified optimization interface for emerging generative policies. In particular, ext
embodiedpost-training - arxiv:2610.16615 · cs.AIWhen 1+1 Backfires: Rethinking Relevance-Diversity Balancing for Sample-Aware Visual Token ReductionYihang Huang, Peisong Wen, Chenhao Zhou, Difu Feng +1
Visual token compression has emerged as a critical technique for reducing the inference cost of Vision-Language Models. Existing methods select tokens based on text relevance, visual diversity, or a fixed combination of both---treating the relevance-diversity balance as a universal constant independ
benchmark - arxiv:2610.16604 · cs.RODesign and Analysis of a Nonlinear Parallel Mechanism Coupling for Dynamic Humanoid LegsRohit Kumar, Heiner Peters, Felix Wiebe, Franek Stark +4
Humanoids increasingly demand lightweight leg mechanisms for agile, dynamic, and energy-efficient locomotion. Reducing distal inertia through proximal actuation has thus become a common design goal. Yet, compact transmissions that preserve torque capability and efficient force transmission remain ha
humanoid - arxiv:2610.16601 · cs.AIMixLID: Token-Level Language Identification in Mixed-Language TextSeongtae Hong, Youngjoon Jang, Seungyoon Lee, Jungseob Lee +1
Processing multilingual text requires both identifying a broad range of languages and distinguishing the language of each token in inputs containing multiple languages. We introduce MixLID, a token-level language identification model supporting 111 language-script combinations. MixLID follows a two-
benchmark - arxiv:2610.16598 · cs.CVDetecting Chain-of-Thought Hijacking in Backdoored Multimodal Large Reasoning ModelsZhifang Zhang, Qi Wei, Weitong Chen, Olaf Maennel +4
Chain-of-thought (CoT) reasoning enables multimodal large reasoning models (MLRMs) to solve complex visual problems, but also exposes them to CoT hijacking: an adversary implants a backdoor through poisoned fine-tuning so that a visual trigger induces a hijacking turn, redirecting an otherwise faith
self-correctionbenchmark - arxiv:2610.16597 · cs.RODEFlow: Deep Equilibrium Flow Maximum-Entropy Reinforcement LearningRuslan Ageev, Huaze Tang, Runqing Miao, Jiarong Li +1
Expressive flow policies can represent multiple high-value actions, but sampling them with few integration steps while retaining tractable action densities remains challenging. We introduce DEFlow, a deep equilibrium flow policy that replaces explicit Euler sampling with implicit Euler integration a
manipulation - arxiv:2610.16592 · cs.LGProrsa: Unleashing the Potential of Prefill-Decode Disaggregation in Agentic Reinforcement LearningChiheng Lou, Xiangwei Deng, Bangjun Xiao, Xiaodong Ji +3
Agentic reinforcement learning (RL) is commonly used to enhance the capabilities of large language models (LLMs) in solving complex multi-turn tasks. During rollout, agentic RL exhibits a workload pattern distinct from that of traditional RL: multi-turn interactions repeatedly introduce prefill requ
agentic - arxiv:2610.16591 · cs.LGJtT: Rethinking Point Trajectory Diffusion for Diverse and Rigid Motion from an ImageGabrijel Boduljak, Yushi Lan, Christian Rupprecht, Andrea Vedaldi
We study forecasting rigid-body motion from a single image: generating diverse and physically plausible future trajectories for arbitrary points in a scene, without building rigidity into the architecture or relying on geometric inputs such as depth. Like video generators, recent trajectory generato
evaluation protocol - arxiv:2610.16589 · cs.LGHyBRA-EEG: Hyperbolic Global-Local Brain-Region Aggregation for Self-Supervised EEG Representation LearningXin Liu, Feng Lu, Wei Li
Electroencephalography (EEG) representation learning is challenged by noisy signals, inter-subject variability, and heterogeneous electrode montages in different datasets and recording settings. Recent self-supervised EEG pretraining methods improve transferability by converting heterogeneous record
benchmark - arxiv:2610.16577 · cs.CVCineRecon: Benchmarking Frontier Models for Agentic Cinematic ReconstructionJialei Wang, Zehan Wang, XiaoYang Wu, Ziang Zhang +5
Reconstructing reference videos as editable dynamic 3D scenes is an important path toward controllable video generation for filmmaking: these scenes provide modifiable, renderable visual proxies that enable explicit control over scene layout, subject actions, and camera behavior. To evaluate agents'
agentagenticbenchmark - arxiv:2610.16576 · eess.SYNANO: A Unified Optimization Framework for Nonlinear Gaussian FilteringZhenxuan Ma, Haoning Wang, Wenhan Cao, Shengbo Eben Li
Conventional Gaussian filtering typically approximates nonlinear models or their statistical moments and retains the Kalman update structure. The resulting approximation errors can affect both state estimates and their uncertainty. This paper presents the Natural Gradient Gaussian Approximation (NAN
benchmark - arxiv:2610.16569 · cs.LGFrom a Large-Scale Teacher to Efficient Multimodal Students for Search Relevance PredictionTomasz Gniazdowski, Mohammadreza Momenifar, Nadia Vase, Shahram Khadivi +5
Relevance prediction between user queries and items is a fundamental component of e-commerce search systems. Traditionally, high-quality relevance labels are obtained through human judgment, which is reliable but expensive, slow, and difficult to scale. Recent advances in large language models (LLMs
iterative refinement - arxiv:2610.16567 · cs.ROVLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action ModelsJun Xiao, Rui Ye, Dong Cao, Jinyu Gu +2
Vision-Language-Action models (VLAs) have achieved high success rates on familiar manipulation tasks, yet it remains unclear whether such scores reflect genuine zero-shot generalization and perturbation robustness. Existing evaluation frameworks provide limited capability-level diagnosis and combina
vision-language-actionmanipulationliberoevaluation framework - arxiv:2610.16558 · cs.ROG$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal NavigationKe Gui, Chenghao Shi, Ziyue Wang, Longwen Chen +2
World models capture environment dynamics and support counterfactual reasoning for planning. In image-goal navigation, where the target often lies beyond the current field of view, the agent must not only model environmental dynamics but also maintain a persistent spatial memory of the scene. Existi
world modelmemoryagentbenchmark - arxiv:2610.16555 · cs.LGAI Discovery in Combinatorics: From Problems to Solutions and Back to Bigger ProblemsEugenio Cainelli, Lorenzo Luccioli, Alessandro Iraci, Giovanni Paolini
We introduce $qt$Bench, a benchmark of 29 open problems in algebraic combinatorics. These problems ask for a simple combinatorial algorithm that explains given polynomial coefficients or identities. To help AI agents search for such algorithms, we provide a scaffold for playing with combinatorial ob
ai agentbenchmark - arxiv:2610.16543 · cs.AIOvertaught, Underlearned: Learner-Centered Distillation from Multi-Agent DiscussionMengtao Zhou, Jianxiao Wang, Xiupeng Chen, Cailiang Liu +1
Multi-agent discussion produces reasoning collaboratively, while deployment still relies on a smaller student that must answer alone. Full-graph distillation transfers such discussions by treating each correct solution as a separate training target, but what teachers provide need not match what the
multi-agent - arxiv:2610.16538 · cs.AIPatTransBench: an ipc-stratified, multi-dimensional benchmark for patent machine translation evaluationXiaowei Su, Xusen Cheng
We present PatTransBench, a 12,000-item Chinese-English/Japanese/Korean patent machine translation (PatMT) benchmark with a four-dimensional evaluation protocol. The benchmark comprises 6,000 short-text (sentence/claim) and 6,000 long-text (paragraph) pairs, evenly split across three language pairs
benchmarkevaluation protocol - arxiv:2610.16537 · cs.LGLearning Graded Sparse Mechanisms: A Representation Theorem and Certified GeneralizationMark Kozdoba, Shie Mannor
In a variety of multi-agent and physical environments, an element's update depends strongly on a few other elements and weakly on many, and both the identity and the number of the elements that matter change with the state of the system. Such graded sparse mechanisms are common, for instance, in flo
multi-agent - arxiv:2610.16531 · cs.CLSupport or Magnitude? Characterizing Neuron Relevance for Instruction Data SelectionYifan Le
Model-internal signals, such as neuron activations and gradients, have been used for instruction data selection, yet how neuron relevance should be characterized remains underexplored. In particular, should selection prioritize the magnitude of relevance or how broadly it is distributed across neuro
benchmark - arxiv:2610.16523 · cs.LGEAU: Environment-Agent-User Co-evolution for Training User-Centric AgentsMin Cai, Dan Zhang, Gao Xing, Libin Ma +2
Training LLM agents to interact with users over multiple turns is challenging because the learning signal depends not only on the agent, but also on the user simulator and environment, which jointly shape the agent's experience. Simulated users can be stochastic or inconsistent, while environments m
agentllm agentagent frameworkagent benchmarkbenchmark - arxiv:2610.16517 · cs.ROTrust-Calibrated VLA Policy Post-Training with Evolving ImaginationWenzhan Li, Yiran Qin, Heng Zhou, Huirui Wang +2
Vision-Language-Action (VLA) models are often pretrained via imitation learning, making reinforcement learning (RL) post-training important when deployment moves beyond demonstration coverage. Existing refinement paradigms expose a reliability--efficiency tension: model-free refinement avoids simula
vision-language-actionvlavla policyliberoworld modelpost-training - arxiv:2610.16502 · cs.LGKARAMEL: Knowledge-Aware Ranking for Multimodal Entity Linking in the Arts via Heterogeneous GraphsRaffaele Scaringi, Alejandro Sierra-Múnera, Gennaro Vessio, Giovanna Castellano +1
Discovering and disambiguating entities from multimodal and heterogeneous data is a key challenge in discovery science, where relevant knowledge is often implicit and distributed across modalities. We present KARAMEL, a framework for multimodal entity linking (MEL) - the task of aligning visual and
benchmark - arxiv:2610.16495 · eess.SYGeometric Transition of Latent Dynamics in Intermittent Chaos: SVD-based Effective Dimension for Detection and Proactive ControlTakashi Hikihara, Tetsushi Saburi
This paper investigates the geometric detection of latent dynamic transitions (where the latent space is defined as a finite-dimensional state space reconstructed via a delay-embedding Hankel matrix) in mutually coupled phase-locked loops (PLLs) exhibiting on-off intermittent chaos. We propose a dat
latent dynamics - arxiv:2610.16491 · cs.LGDrafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic SystemsXin Xu, Ke Cheng, Hao Wang, Yangru Huang +10
Improving an agent through successive revisions requires balancing exploration and evaluation under a limited budget. A narrow search can miss useful sequences of edits, while fully evaluating every intermediate version can exhaust the budget before those sequences yield a verified improvement. We i
agentagenticself-improvementevaluator - arxiv:2610.16484 · cs.LGSLEM: Spectrum-Aware Latent Expert MergingDohyeon Kim, Bedionita Soro, Bruno Andreis, Sung Ju Hwang
Mixture-of-Experts (MoE) models scale model capacity efficiently by activating only a small subset of experts for each input, while still requiring all parameters to be stored in memory, resulting in substantial memory overhead. Expert merging techniques have recently emerged as a promising compress
memorybenchmark - arxiv:2610.16482 · cs.CVMiMaAlign: Hierarchical Micro-Macro Alignment for Physically Plausible Video GenerationYansong Liu, Litong Gong, Weijie Li, Tiezheng Ge +1
Text-to-video (T2V) diffusion models have made remarkable progress in generating high-fidelity visuals, yet their outputs still often exhibit physically implausible phenomena, including discontinuous dynamics and implausible event interactions. Existing representation alignment methods mainly transf
benchmark - arxiv:2610.16480 · cs.AIPXRD-Bench: How Close Are We to Determining Crystal Structures from Powder Diffraction?Dongyeop Woo, Nofit Segal, Nayoung Kim, Minkyu Kim +2
Determining crystal structures from powder X-ray diffraction (PXRD) would enable routine characterization of newly synthesized materials, including those absent from reference databases. Generative models offer a promising approach, but existing evaluations remain fragmented and largely assume singl
benchmark - arxiv:2610.16478 · cs.CVXAI-Guided Fourier Domain Generalisation for Pure-Simulation Sim-to-Real Deep Reinforcement LearningWei Yao, Chuyao Wang, Ziwei Wang, Nabil Aouf
Deep reinforcement learning has shown strong potential for vision-based autonomous driving, but policies trained in simulation often suffer severe performance degradation when deployed in the real world. Existing sim-to-real methods commonly address this gap through image-transfer-based domain adapt
sim-to-real - arxiv:2610.16474 · cs.CLOntology-Based Post-Coordinated Expression Assembly: from Clinical Narrative to Queryable Knowledge GraphsJohann Pignat, Christophe Gaudet-Blavignac, Julien Ehrsam, Jean-Philippe Goldman +6
Clinical oncology documentation records decisive information in free text, where it remains readable but not retrievable at scale, leaving cohort identification dependent on incomplete diagnostic codes or manual review. The current standard, coding individual terms, captures what a document mentions
knowledge graph - arxiv:2610.16471 · cs.CVAuditing Observation-Quality Shortcuts in the Evaluation of Person-Centered Motion for Video Anomaly DetectionChaehyun Kim, Changwon Lim
Video anomaly detection (VAD) identifies deviations from normal patterns, but such deviations do not necessarily correspond to safety-critical events. We focus on human-centered emergencies and violence and investigate whether motion-based anomaly discrimination reflects task-relevant motion informa
evaluation framework - arxiv:2610.16463 · cs.RODeTac: Dual-branch Decoupling of Texture and Force Signals in Vision-Based Tactile SensorsZhirui Zhang, Yunkai Tang, Jiahang Li, Fan Fei +5
Vision-based tactile sensors enable fine-grained perception for dexterous robotic manipulation, but current marker-based designs tightly couple surface texture and contact force signals within a single observation. Existing data-driven decoupling methods rely on serial mask-based inpainting, which s
manipulationdexteroustactile - arxiv:2610.16457 · cs.ROSHOES: Sim-to-real Hardware-adaptation via On-policy Expansion of Skills for Legged RobotsKota Tsuruhara, Tsung-Yuan Chen, Kento Kawaharazuka
Humans expand their locomotion capability by wearing appropriate footwear, such as hiking boots for mountains or skates for ice. Footwear changes the ground-contact interface and enables traversal of otherwise difficult terrain. In contrast, locomotion controllers are typically trained for a fixed m
sim-to-real - arxiv:2610.16455 · cs.CVGHOST: Geometry-Constrained Hybrid Diffusion for Open-Vocabulary 3D Scene Graph GenerationChangqun Feng, Wangxiandi Yin, Xin Hu, Ruiting Dai +3
3D scene graph generation requires understanding both object semantics and metric spatial relations. However, existing open-vocabulary methods often struggle with spatial predicates because language similarity reflects semantic co-occurrence rather than explicit geometry. We propose GHOST, a Geometr
scene graph - arxiv:2610.16452 · cs.CVNot Every Cue Is Evidence: Cue Supportability for Forensic Distillation in Face Forgery DetectionBinjia Zhou, Dawei Luo, Shuai Chen, Feng Xu +6
Explainable face forgery detection requires not only correct judgments but also forensic evidence supported by the input image. However, training cues are typically derived from the forgery generation pipeline. Since images available to the model are of limited quality and resolution, it is difficul
benchmark - arxiv:2610.16451 · cs.LGStructZO: Structured Zeroth-Order Optimization for Memory-Efficient LLM Fine-TuningXuanlin Yang, Jia Li, Zhouchen Lin
Fine-tuning large language models at scale is increasingly limited by the memory cost of backpropagation. Zeroth-Order (ZO) optimizers reduce this burden by relying only on forward function evaluations, but they still exhibit a substantial performance gap to first-order (FO) optimizers such as Adam.
memory - arxiv:2610.16450 · cs.CLCivWorldBench: Evaluating Language World Models in Structured Multi-Agent Sandbox WorldsAng Li, Haofei Yu, Zhengwei Tao, Liang Chen +6
The growing use of language models for forecasting motivates a broader evaluation of their world-model capabilities. Such evaluation must go beyond predictive accuracy to assess hidden-state inference and intervention prediction. Real-world forecasting benchmarks offer limited access to hidden-state
world modelmulti-agentbenchmarkleaderboard - arxiv:2610.16449 · cs.LGCryptoFlowBench: Benchmarking Multi-Hop Reasoning on Blockchains for Cryptocurrency Fund TracingRunang He, Tongya Zheng, Jing Cao, Yucheng Ying +6
Pseudonymous accounts and complex smart-contract ecosystems substantially hinder the tracing of illicit cryptocurrency fund flows. Existing Graph Neural Networks (GNNs) have achieved promising performance in blockchain fraud detection, yet their static node- or edge-level predictions do not naturall
agentagenticbenchmark - arxiv:2610.16444 · cs.CVPixels or Code? Benchmarking Instruction-Guided Visual Artifact OperationShun Guo, Shuizhou Shi, Boao Wang, Wei Xiong +3
Visual artifacts such as charts, technical diagrams, webpages, and document pages underpin scientific, commercial, and engineering workflows. Editing these artifacts requires direct modification, dependency-aware updates, preservation of unaffected content, and compliance with domain constraints. We
agentbenchmark - arxiv:2610.16443 · cs.AIRAP-WAM: Learning World-Action Models through Predictive Rehearsal and Action PerturbationsYuanpeng He, Ceyao Zhang, Fangjing Li, Kefei Zhu +7
Recent world action models draw on video generation priors to jointly predict future observations and robot actions. Adapting these priors to control requires determining how future predictions should guide policy learning. RAP-WAM combines predictive Rehearsal and Action Perturbations (RAP) with fu
embodied - arxiv:2610.16441 · cs.CVVisExMEM: Visual Evidence-Based Explainable Multimodal Evaluation MetricSubhadeep Roy, Christoph Leiter, Steffen Eger
Current image-text evaluation often relies on a global score, which gives limited insight into which semantic details are supported by the image. Explainable metrics provide rationales, but the visual evidence supporting those rationales can still be difficult to verify. We introduce VisExMEM, a ref
benchmarkevaluator - arxiv:2610.16440 · cs.LGARMED: Adaptive Retrieval-Augmented Medication RecommendationKyungkeon Chung, Se Un Park, Hyejin Lee, Junyoung Park
Medication recommendation requires jointly considering diagnoses, comorbidities, longitudinal patient history, and treatment interactions. Although recent retrieval-augmented generation approaches using language models have shown promise for clinical recommendation, most existing systems rely on fix
retrieval-augmentedrag - arxiv:2610.16435 · cs.CLCultureBot: A Framework for Retrieval-Augmented Exploration of Cultural Heritage DataVasileios Karampinis, Angelos Vlachos, Giorgos Filandrianos, Maria Lymperaiou +3
Digital cultural heritage collections encode rich relations among objects, places, periods, providers, images, and source institutions, yet users typically access them through keyword search and faceted browsing. We present CULTUREBOT, an interactive framework for natural-language exploration of cul
retrieval-augmented - arxiv:2610.16432 · physics.opticsFading Memory and Nonlinear Dynamics in a Microring Resonator Network and their Use in ComputationAlessandro Foradori, Alessio Lugnan, Lorenzo Pavesi, Peter Bienstman
Silicon microring resonators are promising components for photonic reservoir computing because their nonlinear optical response can transform signals and retain information about recent inputs. Under constant optical excitation, their coupled free-carrier and thermal dynamics can generate self-pulsi
memorymicroring - arxiv:2610.16426 · cs.AIAuthGateBench: Benchmarking LLMs as Authorization GateZhixiong Zhuang, Jin Jia, Zhiwei Shang, Ziyu Wang
Automated approval systems let coding agents execute high-risk actions with minimal human oversight by delegating the authorization decision to a classifier, typically a large language model (LLM). Such systems, adopted in tools like Claude Code and Codex, remain vulnerable to failure modes such as
agentbenchmark - arxiv:2610.16425 · cs.CVFederated Coreset Subsampling for Distributed Memory BanksJason Witzemann, Daniel Pröll, Sebastian Hegenbart
In real-world scenarios, industrial processes can span multiple factories and goods are often produced at multiple sites, leading to a shared yet non-identical distribution of data across locations. To address this, we propose two federated approaches for i) combining existing coresets (Double Subsa
memory - arxiv:2610.16412 · cs.AISLS-Bench: A Benchmark for Incident Log Summarization with Synthetic Observability DataValerii Iakovlev, Markus Spanring, Martin Flechl
As IT systems grow in complexity, automated agent-driven incident response becomes increasingly critical. A key capability for these agents is summarizing large log streams into human-readable reports. Further progress requires meaningfully evaluating this function, yet existing benchmarks leave end
benchmark - arxiv:2610.16402 · cs.AIEvoWeb: A Benchmark for Coding Agents on Evolving Web ApplicationsDavid Nukrai, Bar Alon, Nativ Levy, Ido Andrew Atad +5
Web application development often proceeds through successive requests that extend an existing implementation. Yet web-generation benchmarks primarily evaluate development toward fixed specifications. We introduce EvoWeb, a novel benchmark comprising 320 cumulative requests over 32 web applications,
agentbenchmark - arxiv:2610.16398 · cs.ROLABIT: Long-Horizon Robotic Assembly Benchmark for Industrial TasksConstantin Schempp, Christian Friedrich, Luca Kofler, Laura Suhm
Robot learning has undergone significant advancements in the domain of compliant control and robotic assembly. The development of novel methodologies has transcended the capacity to address discrete tasks, such as grasping or inserting, and has evolved to encompass the comprehensive construction of
manipulationgraspbenchmark - arxiv:2610.16397 · cs.AIRecursive Environment Scaling for General Long-Horizon AgentsLingxiao Jing, Zhengbo Jiao, Yichen Xiang, Bofan Zhu +9
We study environment scaling as a way to increase agent-benchmark difficulty while holding core task objectives fixed. Our framework, Recursive Environment Scaling (RES), transforms existing benchmark environments while preserving their objectives and completion requirements. A proposer changes info
agentbenchmark - arxiv:2610.16396 · cs.LGSampling from Neural Operator Priors to Solve Partial Differential EquationsAndrew Millard, Joshua Murphy, Zheng Zhao
Guided sampling with sparse observations and PDE residuals has recently been combined with neural operator priors to solve partial differential equations in function space, their natural setting. We pair such a prior with pseudo-bootstrap sequential Monte Carlo (pBS SMC), a guided sampling framework
benchmark - arxiv:2610.16391 · cs.LGRetroPlanner: Multi-Route Expansion and Multi-Objective Reasoning for LLM-Based Retrosynthesis PlanningYe Rin Kim, Yongjun Jeong, Seongjun Yun, Hanbum Ko +3
Multi-step retrosynthesis planning is often considered solved once a route to purchasable starting materials is found, but such a route should also be practical in real-world chemistry. Recent LLM-based planners formulate this process as sequential decision-making over an AND-OR graph, yet typically
benchmark - arxiv:2610.16372 · cs.CLPersonaFactor: Preference Factors with Dynamic Selection for LLM PersonalizationHang Su, Zequn Liu, Jianghan Chao, Qiuyu Tian +2
Although users may express different personalization needs across tasks, these task-specific behaviors can reflect shared underlying preference mechanisms. Existing LLM personalization methods often focus on surface-level interactions or undifferentiated histories, overlooking the shared mechanisms
benchmark - arxiv:2610.16371 · cs.CLRule vs.Consequence: Dissociable Internal Representations of Moral Reasoning in LLMsEugenie Shi
Large language models are increasingly deployed in morally and legally sensitive contexts, yet little is known about how they internally represent distinctions between moral reasoning styles. We ask whether a model's differential response to rule-framed and consequence-framed moral questions reflect
benchmark - arxiv:2610.16369 · cs.RORoboBugBench: Benchmarking Automated Repair of Real Robot Software Under an Audited Behavioral OracleAlexey Tuzikov
Language agents increasingly modify robot software, but whether they can repair it is unmeasured, because the domain supplies no oracle to inherit: a robot bug manifests as behavior, and robot repositories ship almost no tests that capture it. We present RoboBugBench, a benchmark of 33 real, executi
benchmark - arxiv:2610.16368 · cs.LGLoopQuant: Accurate and Efficient Post-Training Quantization for Looped TransformersZhixiong Zhao, Zukang Xu, Xing Hu, Lifeng Liu +2
Looped Language Models (LoopLMs) improve parameter efficiency by recursively reusing shared Transformer modules, yet their repeated computation over evolving recurrent states poses new challenges for low-bit deployment. Existing post-training quantization (PTQ) methods, largely designed for feed-for
memorypost-training - arxiv:2610.16365 · cs.AIWorld Models Need a Physical Registry: Learning Internal Predictive States for Physically Plausible World ModelsShinseong Kim, Jaeseok Jeong, Jianlong Fu, Youngjung Uh
Video world models must generate physically plausible futures, not just realistic frames. Existing approaches typically use semantic, physical, or predictive information to guide the visual generation stream, leaving a fundamental question: where inside the generator should predictions of scene evol
world model - arxiv:2610.16357 · cs.CVVideoSlate: Stateful Commit Control for Temporally Grounded Agentic Video ReasoningShuo Zheng, Yuanjie Dang, Peng Chen, Jianjun Xiang +3
Agentic video reasoning systems gather and assess video evidence across stages, selecting relevant spans as visual evidence for answering questions. New candidates and observations prompt reassessment of the current span, but revisions can correct localization errors or discard useful evidence alrea
agenticbenchmark - arxiv:2610.16355 · cs.ROPINT: A Unified Model with Shared Predictive Intent for Robot ManipulationXinyu Guo, Chao Jing, Li Yi, Xingyu Chen +1
Generalist robot manipulation can benefit from pretrained vision-language models (VLMs) for semantic reasoning and video generation models (VGMs) for world modeling. However, existing unified models that directly fuse them through shared attention at multiple layers can make it harder to draw on the
embodiedmanipulationliberorobotwinworld model - arxiv:2610.16352 · cs.LGA Production-Grade, High-Efficiency Agentic RL Framework for Frontier LLMs with Validated RecipesLoonSage Team, Bowei Song, Caiwei Zhao, Chuwen Liu +44
Reinforcement learning (RL) has emerged as a critical phase in the post-training pipeline of large language models (LLMs). As the frontier of LLM capabilities shifts from static reasoning to agentic tasks--characterized by external tool invocation, multi-turn interactions, and long-horizon decision-
memoryagenticpost-training - arxiv:2610.16343 · cs.LGEfficient Multi-Agent Reasoning via Confidence-Guided Adaptive DebateSeungdong Yoa, Ye Seul Sim, Suhee Yoon, Sanghyu Yoon +4
Multi-agent debate has shown promise for improving the reasoning of large language models, yet recent theory suggests its benefits are highly regime-dependent. While interaction can amplify informative signals under corrective conditions, symmetric debate dynamics are neutral in expectation, often m
multi-agentbenchmark - arxiv:2610.16338 · cs.CVSyncReward: Human-Aligned Synchronization Reward for Audio-Visual GenerationKaisi Guan, Ying Ba, Wenjing Wang, Xihua Wang +3
Recent years have witnessed rapid progress in joint audio--visual generation. However, audio--visual synchronization remains a challenge, as generated speech and sound often temporally misalign with visual events. A key bottleneck is the lack of human-annotated data, reliable evaluation benchmarks,
post-trainingbenchmark - arxiv:2610.16333 · cs.AITRAP: Tool-result Replay Attack for Probing Misbinding in LLM AgentsYutong Zhi, Jie Zhang, Xiuqi Lu, Jiayuan Xu +1
As AI agents increasingly act through external tools, tool results are no longer simple outputs but execution artifacts that must remain correctly associated with the context that produced them. Yet as results persist across caches, recovery mechanisms, and shared runtime state, this association can
agentai agentllm agentagent frameworkbenchmark - arxiv:2610.16330 · cs.CVBAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State UpdatesLijian Li, Yuanpeng He, Yudian Zheng, Chung-ju Huang +5
Feed-forward transformer models have driven remarkable progress in 3D vision,yet their quadratic complexity renders them impractical for long video sequences. Streaming 3D reconstruction addresses this bottleneck by processing frames sequen?tially with constant memory. However, existing recurrent ar
benchmark - arxiv:2610.16329 · cs.CVExplicitly Aligned Multi-View Breast Cancer Risk Prediction from Variable-Length Screening HistoriesSolveig Thrun, Zijun Sun, Suaiba A. Salahuddin, Kristoffer Wickstrøm +4
Accurate breast cancer risk prediction from screening mammography is essential for improved risk stratification and personalized screening. It can benefit from both temporal breast tissue changes across longitudinal examinations and complementary anatomical information from craniocaudal (CC) and med
benchmarkevaluation protocol - arxiv:2610.16320 · cs.CVBuilding age estimation based on widely available multi-source1 geospatial dataJialu Sui, Hongsheng Zhang, Rui Liu, Jing Ling +3
Building age is fundamental for urban infrastructure safety, operation, and planning, as it directly affects structural safety, and sustainable development. Existing methods mostly rely on detailed building information such as street view, function, or materials, which are harder to obtain than geos
benchmark - arxiv:2610.16319 · cs.AIHyWe: Unifying Search and Operation for Real-World Hybrid Web TasksZijian Cai, Haowei Liu, Kailai Chen, Lin Peng +6
Real-world hybrid web tasks need both information gathering and page interaction, yet search-based and operation-based agents are developed and evaluated separately. To bridge this gap, we introduce HyWe, a framework that unifies both capabilities within one model and trajectory. Its two modes use d
multi-agentbenchmark - arxiv:2610.16318 · cs.CLStrong yet Controllable LLM Abstention without Learning to AbstainJunwei Zhang, Yifei Wang, Wanli Yang, Hexiang Tan +5
Large language models (LLMs) can reduce hallucination by abstaining from questions they would likely get wrong. Abstention can be implemented outside the model (selective answering) using a confidence threshold that controls the answer rate (coverage), or learned inside it (abstention training), whi
post-trainingbenchmark - arxiv:2610.16316 · cs.ROConstraint-Decomposed Impedance Control from Generative Policy for Articulated Object ManipulationKyungseo Park, Byeongdo Lim, Jungbin Lim, Minseok Choi +1
Articulated object manipulation tasks such as opening doors, tightening bolts, and pulling drawers impose kinematic constraints on a robot end-effector's motion that neural policies may fail to satisfy. Unlike hybrid force-position control that strictly separates force and position control into orth
manipulation - arxiv:2610.16315 · cs.AIReTAC: Residual Test-Time Agentic Computation for Video Spatial ReasoningYike Wu, Keyan Chen, Qixiong Wang, Xi Xiang +5
Test-time agentic computation can improve multimodal large language models (MLLMs) through planning, tool use, iterative perception, and reflection, but indiscriminate computation may duplicate resolved work, introduce noisy evidence, and perturb correct predictions. We propose that such computation
agenticagent frameworktool usebenchmark - arxiv:2610.16313 · cs.ROListen Before Act: Audio-Conditioned Robot Manipulation with ListenVLAJiaqi Fan, Rui Xu, Jiakun Xia, Xuyu Fan +6
Vision-language-action (VLA) models have advanced robotic manipulation by connecting visual perception and language instructions to action generation. However, many manipulation tasks depend on auditory cues that cannot be reliably inferred from visual observations and may need to be elicited throug
vision-language-actionvlavla policymanipulationgr00tlibero - arxiv:2610.16309 · cs.ROReAP: From Visual Failure Recognition to Action-Conditioned Risk Prediction for Runtime Failure MonitoringJingyuan Liu, Yunpeng Qing, Changxu Wei, Zhi Hou
Despite rapid progress, generalist robot policies remain brittle during execution. Reliable deployment therefore requires runtime monitoring that provides timely warnings of failure. Existing monitors commonly estimate risk from current observations or policy signals without explicitly predicting th
action-conditioned - arxiv:2610.16308 · cs.ROPACE: Temporal-Role Tactile Conditioning for Chunk-Based Insertion PoliciesZhiyuan Chen, Shixiong Xu, Shuai Chen, Yangwei You
Contact-rich insertion requires robots to combine visual alignment with tactile adjustment. However, existing visuotactile chunk-based policies collapse tactile feedback into a single generation-time fusion pathway, leaving persistent contact states, abrupt contact events, and future contact consequ
manipulationtactilebenchmark - arxiv:2610.16302 · cs.CVRAR: Rate-Adaptive Rollout for Time-Consistent Interactive World ModelsYaoyang Liu, Dongyu Yan, Yihua Du, Shengju Qian +4
Video-based interactive world models aim to enable real-time, user-controllable environments. Yet most existing systems are trained and rolled out at a fixed temporal rate, limiting both runtime-rate adaptation and physical-horizon coverage -- a limitation we term temporal-rate coupling. Specificall
world model - arxiv:2610.16298 · cs.AIMemory-Efficient Integer-Only Training of Deep Neural NetworksHazem Hesham Yousef Shalby, Fabrizio Pittorino, Manuel Roveri
Training neural networks in reduced precision can yield significant efficiency gains, yet all existing methods that scale to standard computer vision benchmarks retain floating-point in at least one stage, typically the weight update or optimizer state. We present a training algorithm that scales to
memorybenchmark - arxiv:2610.16297 · cs.LGSLDBench v2: Active Scaling Law DiscoveryLingbei Meng, Rui Huang, Zhirong Shen, Jinxin Liu +4
Scaling laws predict large-model performance from small experiments, but finding a reliable law itself requires costly experiments. SLDBench tests whether language-model agents can discover scaling laws when all experimental results are given upfront. We introduce SLDBench v2 with two extensions. Fi
agentagent system - arxiv:2610.16293 · cs.ROClothAtlas: Joint Garment Identification and State Estimation under Self-OcclusionJens Lundell, Alberta Longhini, Lei Shu, Hemal Arora +5
Garment manipulation requires inferring both garment identity and complete configuration from partial, self-occluded observations. This creates a chicken-and-egg problem: completing a partial observation requires a garment-specific deformation prior, while identifying the garment requires a configur
manipulationgrasp - arxiv:2610.16289 · cs.LGFast Quantized Johnson$-$Lindenstrauss TransformsAngelos Korakitis, Vasileios Nakos
Quantization has emerged as one of the primary techniques for reducing the memory footprint of large language models during inference. A prominent example is Quantized Johnson$-$Lindenstrauss (QJL), which compresses KV caches using 1-bit asymmetric quantization while approximately preserving the inn
memory - arxiv:2610.16288 · cs.AIREVEAL: Reverse Engineering Agentic Latents from Frontier-Model TracesYuetian Du, Meng Wang, Yucheng Wang, Zhangzhe Chen +8
Frontier LLM agents perform complex interactive tasks, but their inference costs motivate transferring their experience to smaller open-source agents. Distillation and latent communication typically require target-model training or access to source-model internals. We study how textual frontier-mode
llm agentagenticbenchmark - arxiv:2610.16281 · cs.CLRelLedger: Relational Information Governance for LLM Agent OrganizationsJunxi Huang, Wenshuo Chen, Nuoxuan Li, Kailong Wang +6
In LLM agent organizations, shared records determine which facts are available for later decisions. A sender consults interaction histories before deciding whether to report an event. A delivery gap across histories does not show whose relationship drove the choice or whether the resulting record ma
agentllm agent - arxiv:2610.16280 · cs.LGagent-risk: Risk Assessment for Changes to AI AgentsYue Zhao, Zixiang Xu, Angela Huyue Zhang, S. Alex Yang +1
Before changing a deployed AI agent (a stricter safety gate, a new model), organizations must approve the change, starting from existing logs. Extending classical risk analysis to partial records and partial changes to running systems, agent-risk replays logged episodes, assuming nothing else change
agentai agent - arxiv:2610.16278 · cs.CVDepthTile: Depth-Guided Existing-Image Self-Tiling with Input-Side Tileability AssessmentRui Wang, Xi Chen, Hui Dai, Piaopiao Yu +3
Existing-image self-tiling aims to transform a given image into a seamlessly repeatable tile while preserving its original semantics and local details. This task requires weak editing to retain the source but leave periodic seams, whereas strong editing improves tileability at the large risk of cont
benchmark - arxiv:2610.16277 · cs.CVFlowAvatar: Real-Time Full-Body Avatars from Sparse Egocentric Inputs on Consumer XR DevicesChenfeng Gao, Taeyoung Yeon, Sungheon Park, Vasco Xu +3
Prior avatar embodiment research has studied tracking completeness (how faithfully the body, hands, and face are sensed and reproduced) and personalized appearance in isolation, each requiring specialized lab equipment. How these factors interact on consumer hardware remains an open question, becaus
benchmark - arxiv:2610.16272 · cs.ROGraph3R: Unifying View Graph Initialization with Feed-Forward ReconstructionJingwen Yu, Hengli Wang, Jianhao Jiao, Zhonghang Liu +2
Feed-forward reconstruction recovers camera poses and dense geometry from image sets whose views are assumed to come from a single scene. In practice, uncurated image collections often contain views from disjoint scenes, and reconstructing these views jointly can corrupt camera poses and geometry. E
benchmark - arxiv:2610.16268 · cs.LGError Correction in Trained TTT LayersYuhe Hu, Wen Jia Hu
Test-time-training (TTT) layers update fast-weight memory while processing an input sequence. A gradient-based update, however, does not by itself establish how the memory revises an existing association. We study the structure and empirical role of prediction-derived updates in released TTT impleme
memory - arxiv:2610.16261 · cs.LGHardware-Friendly Token-Group Activation Quantization for Low-Bit Mamba Super-ResolutionJinwoo Chung, Jangho Kim
Mamba-based super-resolution models are efficient alternatives to Transformer-based restoration backbones, but their low-bit post-training quantization remains challenging because selective scan, Mamba's input-dependent state-space operator, induces different activation ranges across spatial tokens.
post-training - arxiv:2610.16260 · cs.AIScreenASR-Bench: A Tiered Benchmark for Speech Recognition with Visual ContextMingwei Fu, Zheshu Song, Zhikai Zhou, Shen Huang +2
Current ASR models perform well on general benchmarks but still struggle with acoustically ambiguous words. Visual context can aid recognition in screen-based scenarios, yet existing benchmarks provide limited insight into the benefits of different types of visual evidence. To address this gap, we i
benchmark - arxiv:2610.16256 · cs.CVEgoWholeView: Camera-Only Head-Mounted Capture of Body and Bimanual Hand Motion for Embodied DemonstrationsYaoqing Hu, Xuguo He, Daohan Cui, Yukun Shi +1
Head-mounted body and bimanual motion capture supports self-avatars, teleoperation, and embodied demonstrations. However, a single egocentric camera is vulnerable to a limited field of view, scale ambiguity, and hand-object occlusion. We present EgoWholeView, a wearable capture system of six synchro
embodiedteleoperationbenchmark - arxiv:2610.16254 · physics.opticsProgrammable spatiotemporal shaping for partially coherent structured wave packetsZhao Zhang, Haihao Fan, Yaning Zhou, Weiran Li +4
Providing more coherent light with space-time coupling offers unprecedented opportunities across diverse scientific disciplines and areas of technology. However, in many applications, tailored laser beams with partial coherence in space and time are desired. Here, we demonstrate a programmable appro
manipulation - arxiv:2610.16251 · cs.AIAvoiding What Your Teacher Already Knows: Entity Familiarity for Deep-Search DistillationYifei Liu, Zhixiang Liang, Wentao Tan, Yicheng Wang +3
Deep-search agents are commonly trained by distilling tool-use trajectories from stronger teacher models. However, structurally complex questions may remain relatively easy for teachers familiar with key entities in the underlying entity graph, resulting in lower search demand and weaker supervision
tool-use - arxiv:2610.16249 · eess.SY$H_2$-optimal model order reduction using hyperbolic geometryAndrea Angino, Tamás Dózsa, Matthias Voigt
We revisit the problem of approximating a linear time-invariant LTI model by a reduced-order model (ROM) in an $H_2$-optimal sense. A widely used method for computing such ROMs is the iterative rational Krylov algorithm (IRKA). However, in practice, the stability of the resulting ROM is not generall
benchmark - arxiv:2610.16242 · cs.ROSame Goal, Different Words: Spatial Instruction Sensitivity in World-Action ModelsAli-Adeeb Abbas, Esmaeil Seraj, Behrad Toghi, Taskin Padir
World-action models should preserve the intended manipulation across equivalent descriptions of a spatial goal. We evaluate this consistency in 864 rollouts from three models across four RoboLab scenes, holding initial states and task goals fixed. Describing relations through the reference object ra
manipulation - arxiv:2610.16238 · cs.ROEmbodied Skill Bench: Benchmarking How Well Agents with Embodied Skills Perform on Physical TasksZhi Han, Zihan Guo, Chenxi Zeng, Zicai Cui +5
LLM agents increasingly control robots using VLA policies, perception models, and procedural knowledge. These modules are difficult to reuse because their interfaces, embodiment assumptions, and conditions of use differ, and their benefits on physical tasks remain unclear. We package them as embodie
vlaembodiedliberogripperagentllm agent - arxiv:2610.16232 · cs.LGAutoNovo: Agentic Program Search Discovers Transferable Nucleic Acid Design AlgorithmsJoel Shor, Subhashini Venugopalan
In-silico optimization of nucleic acids relies on predictive oracles that imperfectly capture biological function, so sequences optimized against one model may exploit model-specific artifacts and fail to generalize to independent models. Recent studies have shown that not all designers generalize e
agentic - arxiv:2610.16226 · cs.LGNetRL: Powering Network Decisions via NetGRPO-Driven Reinforcement LearningLingfeng Peng, Xiaohui Xie, Yang Bao, Yong Cui
Network decision-making tasks, including adaptive bitrate streaming, load balancing, and congestion control, require policies to process heterogeneous states and actions while balancing near- and long-term consequences. Existing solutions generalize poorly across network tasks, require substantial t
action head - arxiv:2610.16224 · cs.CVLook Back to Move Forward: Learning to Revisit for Agentic Compositional Image GenerationZixu Wang, Xiaoran Zhao, Yahui Liu, Feng-ao Wang +7
Text-to-image models have made remarkable progress in visual fidelity and instruction following, yet complex compositional generation remains challenging. Existing agentic approaches address this limitation through iterative generation and editing, adapting their actions based on interaction history
agentagentic - arxiv:2610.16213 · cs.AISafeDSE: Joint Exploration of Safety Architecture Patterns, Mapping, and Scheduling for Dataflow Applications With Safety Arguments (Extended Version)Fatemeh Bahrami, Seyed-Hosein Attarzadeh-Niaki
Embedded platforms increasingly host safety-related streaming functions next to less critical ones. Whether a safety architecture pattern, such as a doer/checker pair, protects a function depends on the cores that run its components. Their schedule must meet the throughput requirement, and mixing sa
benchmark - arxiv:2610.16191 · cs.AICorrect Choices, Failed Sequences: Choice2Seq Benchmarks Correctness Retention Across Output InterfacesTianyou Lai, Wentao Yue, Zhiying Deng, Qilei Li
A model may select the correct answer yet fail to construct it through a sequence of local decisions. We introduce Choice2Seq, a diagnostic benchmark that measures correctness retention across complete-answer selection and sequential construction in independent runs on the same problems and candidat
benchmark - arxiv:2610.16187 · cs.LGExploration strategies for Gaussian process bandit optimization with inference regret guaranteesHwanwoo Kim, Jungwoo Hong, Kyungjae Lee
Existing cumulative regret guarantees for Bayesian optimization (BO) do not necessarily characterize the quality of the algorithm's final output. This gap is critical under noisy evaluations, where the goal is to return a single high-quality recommendation. To address it, we introduce efficient expl
benchmark - arxiv:2610.16182 · cs.AILabel-Centric Knowledge Retrieval for Zero-Shot Text ClassificationDaoping Wu, Dong Wang, Siyu Liu, Chen Liang +4
Flagship large language models (LLMs) offer rich general knowledge and strong reasoning capabilities, but their high inference cost and latency limit their use for large-scale data labeling. We study zero-shot text classification with semantically meaningful labels and task descriptions, without tas
retrieval-augmentedbenchmark - arxiv:2610.16177 · cs.CVDIGS: Distribution-Informed Gaussian Splatting for Training-Free Open-Vocabulary 3D SegmentationZhifan Zeng, Wenbo Xiao, Arcot Sowmya, Changming Sun
Open-vocabulary 3D Gaussian segmentation typically binds semantic features to the per-Gaussian state via per-scene gradient training (1-4 h) or the more recent closed-form distillation. Both commit each primitive to a single feature; therefore, boundary Gaussians spanning multiple 2D instances are f
evaluation protocol - arxiv:2610.16176 · cs.CVEgoAlign: Situated 3D Understanding without Situation DescriptionsYiyang Li, Tianyi Ma, Zehong Wang, Yanfang Ye
Situated 3D Question Answering requires models to interpret a 3D scene relative to an agent's pose. Existing methods commonly convey that pose through a natural-language situation description, which may be unavailable or costly to obtain in real-world deployments. While some methods additionally inc
agent - arxiv:2610.16175 · cs.AIProCARE: Real-World Study Automation via Profile-Grounded Evidence ContractsZhaoyu Fan, Bowen Han, Jincheng Ou, Yanwei Ren +7
Real-world studies (RWS) analyze routinely collected healthcare data, such as electronic health records and insurance claims, to estimate treatment effects, monitor safety, and identify risk factors outside randomized trials. Automating RWS with large language model agents is appealing but brittle:
benchmark - arxiv:2610.16173 · cs.ROExtending Wire-Driven Parallel Robots to Multi-Floating-Link Systems with an Inter-Body WireShintaro Inoue, Kento Kawaharazuka, Temma Suzuki, Sota Yuzaki +2
Parallel wire-driven systems are a framework in which multiple wires are connected to an end effector, and the end effector is driven by the tensions of those wires. Because they can transmit large forces, up to the breaking strength of the wires, over lengths on the order of several tens of meters,
manipulation - arxiv:2610.16172 · cs.CVEventLens: Event-Structure Reinforcement Learning for Video UnderstandingGuowen Zhang, Boshen Xu, Zihao Yue, Ziheng Wang +5
Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure that makes a video coherent. They may merge neighboring events, miss single-event progression, or infer implausible relations across events. We study s
post-trainingbenchmark - arxiv:2610.16170 · cs.CVGeometric Parrots: Rule Absence vs. Rule Fragility in Spatial and Physical Reasoning for Vision-Language ModelsBilal Jawaid, Hamza Hanif, Muhammad Maaz Uddin, Anas Zafar +7
Vision-language models (VLMs) can answer many geometric and physical reasoning questions correctly, yet still behave inconsistently in interaction, for example, abandoning a correct answer when confronted with a confident but incorrect user assertion. Standard single-turn accuracy cannot distinguish
benchmark - arxiv:2610.16167 · cs.LGInteractive 4D Volumetric Liquid Forecasting under Moving-Solid InteractionJongwon Ryu, Sunjae Yoon, Junyeong Kim
We propose the task of Interactive 4D Volumetric Liquid Forecasting: predicting the spatiotemporal evolution of a future dense 3D liquid velocity field conditioned on how a solid object moves through it. This task provides an upstream physical forecasting primitive for systems that need to reason ab
manipulationbenchmark - arxiv:2610.16166 · cs.CVEsports World Model DatasetWeiyu Ma, Zeyuan Gong, Xuhui Liu, Wenxuan Zhang +4
Interactive world models must capture how multiple participants jointly change a shared environment while observing different parts of it. Competitive games make this challenge concrete: players cooperate and compete, while fog of war, team vision and camera placement determine what each can see. Mo
world model - arxiv:2610.16165 · cs.LGThink Before You Speak: Enabling Training-Free Latent Communication Between Language ModelsRishi Khare
Large language models (LLMs) have demonstrated remarkable capabilities in solving complex tasks in coding, mathematical logic, and iterative reasoning, especially by coordinating effectively within multi-agent systems (MAS). However, language agents traditionally communicate via text exchange, which
multi-agentagent systembenchmark - arxiv:2610.16160 · cs.ROFlowVLA: Enabling World Dynamics Understanding through Pluggable 3D Flow LearningJunhyeong Go, Mingyu Jung, Woobin Im, Jongbin Ryu
Recent advances in 3D reconstruction enable the extraction of temporally consistent 3D motion from RGB observations; however, current Vision-Language-Action (VLA) models still predominantly rely on 2D image-text priors. Therefore, they do not accurately represent the fundamental dynamics of the 3D w
vision-language-actionvlavla modelmanipulation - arxiv:2610.16155 · cs.AIUncovering Robustness Weaknesses in LLM Agents via Persona Evolution with Task-Integrity ConstraintsYutao Xie, Jingbo Yang, Guanyu Yao, Xiaoyu Yang +2
Interactive-agent evaluations commonly rely on standardized LLM-based user simulators that exhibit broadly cooperative behavior, potentially overlooking failures that arise under diverse but realistic user interactions. This can lead to overly optimistic estimates of agent robustness and provide lim
agentllm agentagent benchmarkbenchmark - arxiv:2610.16152 · cs.LGWildTrace: Evaluating Training Data Detection in Realistic Large-Scale LLM Post-TrainingHengxiang Zhang, Junxian Liu, Hongxin Wei
Although post-training significantly enhances model performance, the use of massive undisclosed data raises concerns about benchmark contamination and copyright infringement. This drives growing interest in training data detection for post-trained models. Existing evaluations for training data detec
post-trainingbenchmarkevaluation protocol - arxiv:2610.16147 · cs.LGBARTER: Compact Dynamic Sparse Training with Standard Dense KernelsYueyuan Sui, Payal Mohapatra, Qi Zhu, Stephen Xia
Dynamic sparse training (DST) aims to reduce training costs while maintaining accuracy. Existing DST methods mask out parameters but keep matrix dimensions unchanged. Training speedups therefore depend on specialized support, such as sparse kernels. We introduce BARTER, a structured DST framework th
memorybenchmark - arxiv:2610.16146 · cs.CVUnravel: Recovering an Interactive 3D Scene from a Single ImageHui Shan, Keyang Luo, Binjia Zhou, Ming Li +10
We present Unravel, a training-free framework that turns a single image into an interactive 3D scene, where every object is an independent and movable asset. Existing generators either produce one entangled geometry in which no object can be moved, or predict all objects in one feed-forward pass, wh
agentic - arxiv:2610.16145 · cs.AIPollard: Governed Semantic Execution Trees for Framework-Independent AI Agent Control and ReplayMuntaser Syed, Marius C. Silaghi
Agent frameworks can persist checkpoints, trace tool calls, and limit resource use, but these facilities often assign different identities to execution, governance, and replay. We present Pollard, a framework-independent runtime based on governed semantic execution trees. A step identity commits to
agentai agentagent framework - arxiv:2610.16141 · cs.AIPhantom Merge: When Your Large Language Model Agents Pick One but Tell You About AnotherWangshu Zhu, Qianwen Guo, Yushun Dong
LLM agents increasingly operate over multiple entities, where reliable response generation requires not only retrieving relevant evidence but also preserving its association with the correct entity. We identify Phantom Merge (PM), an entity-evidence binding failure in which an agent attributes unsup
agentllm agent - arxiv:2610.16140 · cs.AILarge Language Model Assisted Transportation Planning through a Differentiable User Equilibrium LayerEunseung Lee, Gyeongjun Kim, Keemin Sohn
Transportation planning has traditionally required substantial expertise in network modeling, traffic assignment, and optimization. As a result, strategic analyses such as network design, capacity expansion, and model calibration remain inaccessible to non-expert decision makers. Recent advances in
llm agent - arxiv:2610.16139 · cs.RODirectional Pullback Regularization for Long-Horizon Rollout of Latent World Models in Autonomous DrivingShatadal Mishra, Ahmadreza Moradipari, Nejib Ammar
Learned world models for autonomous driving operate in unconstrained latent spaces, yet the physical systems they represent are governed by strict non-holonomic constraints: ground vehicles cannot realize instantaneous pure lateral motion. We study long-horizon latent rollout modeling for autonomous
world model - arxiv:2610.16133 · cs.AIQuery Your Way Out: Benchmarking Multi-Turn Knowledge Reasoning through Generalized Puzzle SolvingXin Cheng, Junming Yang, Ning Xu
Puzzles offer a natural testbed for Large Language Models (LLMs) reasoning because they pose verifiable goals under partial information and require multi-step decisions to reach a solution. However, existing puzzle benchmarks typically focus on narrow environments with rigid mechanics and tightly co
benchmark - arxiv:2610.16131 · cs.LGXForecast: The KDD 2026 Finance Day Challenge on News-Grounded Financial ForecastingJaehoon Lee, Jun Seo, Seunghan Lee, Tae Yoon Lim +9
Financial forecasting has traditionally relied on numerical time-series (TS) signals, while textual information such as news remains largely underutilized. We introduce XForecast, a benchmark and competition for news-grounded financial forecasting, where the task is to predict stock price movements
benchmark - arxiv:2610.16129 · cs.LGPID-Gated Continual Prompt Adaptation Under Noisy FeedbackZhentao Xu
Deploying LLM agents in non-stationary environments requires continually revising their instructions as the task drifts, yet rewriting after every reported mistake can turn noisy feedback into persistent errors. This raises an often-overlooked scheduling question in prompt adaptation: when should fe
llm agentbenchmark - arxiv:2610.16127 · cs.AITowards a Benchmark for Fire Safety Code Interpretation in the Hong Kong AEC IndustryLe Qiu, Zhuan Anne Zhang, Hok Ming Adeline Chan, Emmanuele Chersoni
Large Language Models (LLMs) are increasingly explored in the Architecture, Engineering, and Construction (AEC) domain, yet their reliability in regulatory compliance remains uncertain. In this study, we evaluate the performance of LLMs on Hong Kong fire safety codes. In collaboration with domain ex
benchmark - arxiv:2610.16125 · cs.AIImitation Holds a Key to Evolution: Localized Repair and Verified Skill Acquisition for Adaptive AgentsKai Hu, Hua Huang
An agent can retain successful trajectories without learning which part of an experience should repair a new discrepancy. We formulate passive imitation (PI) as an architectural loop that localizes a violated contract, extracts a local goal, proposes an effect-matched skill, tests a small patch, and
agent - arxiv:2610.16120 · cs.CLScaling Multi-Agent Intelligence with Substrate-Centric CoordinationZhenting Qi, Zhuokai Zhao, Rulin Shao, Kerem Oktar +6
Existing large language model (LLM) based multi-agent systems have been shown to scale poorly, and we identify three problems: reading burden that grows with the group, useful information diluted by shared work of unknown quality, and fragility to failed agents. We argue that the problems are largel
multi-agentagenticagent system - arxiv:2610.16118 · cs.LGElephant in the Fridge: Constant-Memory Frame Packing for Long Video UnderstandingShuo Gao, Chenhao Zheng, Zhongzheng Ren
Hour-long videos, streaming feeds, and movie-length content are pushing vision-language models (VLMs) past their breaking point -- the standard recipe of encoding each frame into visual tokens yields sequences that grow linearly with video length, causing prohibitive memory consumption and inference
memorybenchmark - arxiv:2610.16117 · cs.AIBeyond Perfect Teammates: Multi-Agent Collaboration with Unreliable ColleaguesChen Huang, Yiyang Zhang, Yuhao Wu, Yang Deng +1
LLM-based multi-agent collaboration is increasingly used to solve complex tasks by dividing work among specialized agents and integrating complementary expertise. Yet existing systems and evaluations largely rest on an implicit perfect-colleague assumption: collaborators provide correct and complete
agentmulti-agentbenchmark - arxiv:2610.16108 · cs.AIGUIDE: Computationally Augmented Consensus for Trustworthy Clinical Practice GuidelinesYuanfan Lin, Jie Di, Xujin He, Zhenzhe Lin +20
Maintaining up-to-date clinical practice guidelines (CPGs) is essential for evidence-based medicine, but rapid evidence growth and complex expert consensus formation limit timely updates. We developed GUIDE, a multi-agent copilot system for human-AI collaborative guideline generation and updating. G
ai agentmulti-agent - arxiv:2610.16106 · cs.LGSilence and Noise: Resolving Suppression and Overvaluation in Multimodal Entity AlignmentXizhe Zhang, Jingfeng Zhang, Meng-Fen Chiang
Multimodal entity alignment (MMEA) is fundamental to integrating heterogeneous knowledge graphs (KGs), yet existing paradigms remain critically vulnerable to incomplete modalities and cross-graph structural asymmetries. This paper exposes a fundamental oversight in current MMEA research: treating fe
knowledge graphbenchmark - arxiv:2610.16105 · cs.LGBeyond Policy Rollouts: Embedding-Space Directional Exploration with Anchored Noise for LLM RLVRDaijiao Liu, Huiyi Wang, Lina Yao, Dong Gong
Reinforcement learning with verifiable rewards (RLVR) typically uses small rollout groups per prompt, where repeated sampling from the same policy can yield highly similar trajectories, reducing within-group reward variation and weakening group-relative advantage signals. Existing remedies such as h
benchmark - arxiv:2610.16104 · cs.LGSelectTTRL: Representation-Guided Data Selection for Efficient Test-Time Reinforcement LearningYi Li, Linan Yue, Weibo Gao
Test-Time Reinforcement Learning (TTRL) enables large language models to adapt directly on unlabeled test problems by constructing pseudo-labels from multiple sampled responses. However, existing TTRL methods typically apply the same expensive sampling and optimization procedure to all test problems
benchmark - arxiv:2610.16102 · cs.AIWhere Polanyi Meets Galois: Tacit Knowing as an Algebro-Deterministic Operation, Toward Re-rooting AI-Generated ScienceHiroyuki Chuma, Kanji Otsuka, Yoichi Sato, Kodo Yamauchi
We argue that tacit knowing is better modelled as an operating regime of one memory substrate than as a kind of content. The substrate is VaCoAl, an algebro-deterministic XOR-and-shift hyperdimensional memory over GF(2), realized as CASRAM (commercial SRAM-CAM) and PyVaCoAl (DRAM-CAM).In VaCoAl/PyVa
memory - arxiv:2610.16082 · cs.LGRegime-Aware Router Learning for Multi-Domain Medical Image Generation with Diffusion ExpertsYaoshuai Ma, Wei Yao, Yongxiang Liu, Feifan Wang +4
Heterogeneous medical image data motivate domain-specialized diffusion experts, but expert specialization alone does not determine how their predictions should be combined during sampling. Learning a global router is particularly challenging when the server has limited, potentially unrepresentative
benchmark - arxiv:2610.16078 · cs.AIHeterogeneous Agentic CoevolutionWenqian Ye, Ziwei Guan, Zi Yang, Shivani Modi +13
Recent advances in self-improving LLM agents have demonstrated the potential of AI systems that drive their own intelligence. However, many domain-specific applications rely on specialized foundation models that possess strong domain expertise but remain dependent on human-designed improvements. Ext
agentllm agentagenticself-improvingself-improvement - arxiv:2610.16071 · cs.LGDream2Discern: Learning from Hard Latent Variants for Runtime Failure Prediction in Multi-Agent SystemsBoxuan Zhang, jianing Zhu, Xiaotian Han, Ruixiang Tang
Reliable deployment of LLM-based multi-agent systems calls for runtime auditors that predict terminal outcomes from observed execution prefixes. These auditors typically learn from a limited corpus of trajectories with verified outcomes. Expanding this corpus requires costly execution and verificati
multi-agentagent systembenchmark - arxiv:2610.16070 · cs.LGUnderstanding and Reducing Misalignment in Persuasion OptimizationBoyi Wei, Ruta Desai, Peter Henderson, Wenyue Hua
Frontier models are now trained for complex multi-agent tasks, from running a company to engaging in negotiations. These tasks necessitate training models to be persuasive. But what does this optimization pressure mean for model alignment? In this work, we study how reinforcement learning (RL) train
multi-agent - arxiv:2610.16067 · cs.AIWhen Agent Evaluation Becomes Unreliable: An Empirical Study of Oracle Failures and Configuration-Dependent Measurement in an LLM Agent BenchmarkJunjie Chen
Agent evaluation harnesses are software systems that run test suites, generate inputs, and decide outcomes with test oracles. This paper audits such software for reliability defects that score tables cannot reveal. We introduce an oracle-integrity auditing approach with five state-integrity invarian
agentllm agentagent benchmarkbenchmark - arxiv:2610.16058 · cs.CVFrom Failure Prediction to Component Selection in Semantic SegmentationMinhui Kang, Yiyao Zhang, Peizhe Long, Jiafeng Qi +2
Reliable scene understanding requires deciding which predictions to trust, retain, or inspect. In semantic segmentation, component failure predictors support these decisions, but their predictive performance does not fully describe the value of the resulting selection. We develop a decision-centered
evaluation framework - arxiv:2610.16049 · cs.AIAdaptive Compression and Targeted Perturbation: A Unified Framework for Generalized Audio Deepfake DetectionJianqiao Cui, Haoxian Zhu, Hanbiao Meng, Yanghao Zhang
Existing audio deepfake detection (ADD) methods frequently exhibit limited generalization, as fixed bottleneck architectures fail to adapt to the heterogeneous information densities inherent in diverse spoofing artifacts. Moreover, standard augmentation often leads to a pathological robustness-accur
benchmark - arxiv:2610.16048 · cs.LGBackward Conformal Prediction under Covariate ShiftAmin Meghrazi, Yuanyuan Li
Backward conformal prediction (BCP) fixes the size of a prediction set and estimates its miscoverage, but assumes calibration and test data are exchangeable. We extend BCP to covariate shift through importance weighting, preserving its guarantees when the density ratio is known. We show that BCP's t
benchmark - arxiv:2610.16047 · cs.CVHybrid Context Forests and Bilateral Viterbi Matching for 3D Rib Centerline Tracing in CTQuan Wang
Automatic extraction and anatomical numbering (R01-R12, L01-L12) of all 24 rib centerlines in thoracic CT is essential for unfolded rib visualization, fracture detection, and surgical navigation, but remains challenging due to adjacent-bone confluence, calcified costal cartilage bridges, kyphotic po
benchmark - arxiv:2610.16040 · cs.LGCLNO: Conservative Local Neural Operator for Compressible Flow PredictionBuxuan Wang, Rongzu Li, Qingsong Zou
We propose a conservative local neural operator (CLNO) for accurate and generalizable autoregressive prediction of compressible flows. Building upon the locality and translation invariance of the local neural operator (LNO), CLNO introduces a conservative evolution framework based on local subdomain
benchmark - arxiv:2610.16039 · cs.AIWhen Agents Talk, Secrets Travel: Topology-Driven Leakage and Recipient-Conditioned Sanitization in Multi-Agent LLM SystemsSunday Oyinlola Ogundoyin, Muhammad Ikram
Multi-agent LLM systems (MALS) introduce a privacy-leakage problem that conventional sensitivity-based defenses miss, since information disclosed to one authorized agent can propagate through routine coordination to an unauthorized one, without compromise or malicious behavior. This motivates topolo
agentmulti-agent - arxiv:2610.16034 · cs.LGA Nonlinear Time-Series Theory for Constant-step SGD with MomentumYounghun Lee, Jake Zummo, Wei Biao Wu
Constant-step stochastic gradient descent with momentum (SGDM) augments the parameter recursion with a momentum state that carries geometrically decaying memory of past stochastic gradients. We study this coupled recursion as a nonlinear time series and develop a framework for its stationary behavio
memory - arxiv:2610.16027 · cs.AIOf Errors and Echoes: Responsibility-Aware Credit Assignment for Tool-Use AgentsXingjian Diao, Tianyu Yang, Wenjun Huang, Xiangchi Yuan +7
Reinforcement learning improves tool-use agents, yet fine-grained credit assignment remains challenging even with step-level feedback. An observed error may be inherited from an earlier decision, while multiple responsible decisions may share the same downstream consequence or participate in later r
tool-usebenchmark - arxiv:2610.16022 · cs.CVRemoving the Compute Bottleneck for Event-based Visual Place Recognition with Spiking Neural NetworksSomayeh Hussaini, Tobias Fischer, Michael Milford
Neuromorphic sensing and computing promise low latency and high energy efficiency for robotics, yet most event-based visual place recognition (VPR) pipelines reintroduce dense, frame-like computation downstream of an inherently sparse sensor, forfeiting the sensor's latency advantage. In this paper,
benchmark - arxiv:2610.16020 · cs.LGWhen Does Compensating Inherited Quantization Error Help? Mean-Square Guarantees for Layer-Wise Quantization and Their LimitsTianhao Guo
Layer-wise post-training quantization methods let each layer compensate for the quantization error inherited from earlier layers. We ask under what assumptions such compensation guarantees a network output error no larger than without it. For linear networks of any depth, with the required calibrati
post-training - arxiv:2610.16016 · cs.LGCertified Robustness via Sparse Second-Order Cone RelaxationsMohammadreza Maleki, Daniel Sadig, Rushendra Sidibomma, Mahyar Fazlyab +2
Certified robustness is critical for deploying neural networks in safety-sensitive settings, but computing exact certificates typically requires solving nonconvex optimization problems. Practical verifiers use convex relaxations: linear programming (LP) relaxations are scalable but often too conserv
benchmark - arxiv:2610.16015 · cs.AIComputational Reflective Equilibrium for Agentic AI: A Rawlsian Alignment Scaffold for GovernanceQuanyan Zhu
The rapid deployment of agentic AI has introduced complex challenges for the ethics and governance of AI systems. Agents built on large language models act through tools on behalf of people, and each deployment case brings its own instructions, policies, evidence, and stakeholders, so alignment must
agentagentichuman-in-the-loop - arxiv:2610.16009 · cs.CVMeaningful Moments: A Corpus of VLM-Derived Temporal Importance Labels for Video Action RecognitionCameron Keith, SouYoung Jin
Not every moment in a video is equally useful for recognizing an action. A few decisive moments often carry the evidence: the instant an object moves, the peak of a dive, or one object placed on another. Existing video datasets rarely provide large-scale, action-conditioned labels for which short se
action-conditionedevaluation protocol - arxiv:2610.16008 · cs.LGImproving the Evaluation of Biomedical Question Answering with Few Human Preference SignalsShakila Mahjabin Tonni, Xiang Dai, Sarvnaz Karimi, Xiaomei Li +2
Retrieval-Augmented Generation (RAG) systems are increasingly used for biomedical Question Answering (QA). However, evaluating their effectiveness remains challenging due to the lack of reliable reference resources and the frequent misalignment between automatic evaluation metrics and human judgment
retrieval-augmented - arxiv:2610.16003 · cs.CLPAIR: Memory-Efficient and Fast Inference with Reading--Reasoning FusionSejun Park, Haocheng Xia, Yongjoo Park
For context-based question answering, using one large model for both evidence extraction and answer reasoning requires substantial memory and computation. Existing approaches trade accuracy for efficiency within a single model only. We propose a multi-model approach for efficient inference: our meth
memory - arxiv:2610.15995 · cs.AIAutoSec-RAN: Specification-Grounded LLM Fine-Tuning for Automated Security Analysis of O-RANAnoop Mishra, Joshua Moore, Aly Sabri Abdalla, Vuk Marojevic
This paper presents AutoSec-RAN, a specification-grounded large language model (LLM) framework for automated security triage of Open Radio Access Network (O-RAN) testing. Unlike approaches relying on static rule sets or privileged access to raw packet captures, AutoSec-RAN fine-tunes Mistral-7B-Inst
retrieval-augmentedrag - arxiv:2610.15994 · cs.AILarge language model-generated patterns for epidemiological modeling: a multi-benchmark evaluation frameworkSophia Mao, Keshav Gandhi, Alexander Gutfraind
Representing individual-level behavioral heterogeneity is challenging in epidemiological modeling, particularly when empirical behavioral data are sparse. We assess the potential of large language model (LLM)-generated behavioral outputs in addressing this challenge, using hepatitis C virus-related
benchmarkevaluation framework - arxiv:2610.15993 · cs.AIHandoff Curves: Initial Selection, Later Handoff, and Supervision Targets in Multi-turn LLM RoutingDaniel Matos, Lu Liu, Yanfeng Zhang
We study how initial model selection, later handoff, and training targets affect success and cost in LLM tasks that span multiple turns. We compare several router architectures using handoff curves, which record task success and cost as the handoff step changes. Our dataset contains 34,204 complete
benchmark - arxiv:2610.15991 · cs.AIRevocation Is Not Complete at Receipt: An Executable Outcome Contract for Distributed Agent-Tool GatewaysLifei Liu, Haoran Yu, Danping Zhang
Distributed agent workflows cache delegated authority at tool gateways. Revocation may be reported at issuer commit, transport acknowledgment, or receiver persistence, although none of these observations establishes that all gateways enforced the change or that guarded effects stopped. We define C0-
agent - arxiv:2610.15976 · cs.AIEmbedding Information in Policy Randomness: Distribution-Preserving Implicit Communication in MDPsGongpu Chen, Deniz Gündüz
This paper investigates a new role of randomness in stochastic policies--as a medium for embedding and communicating information. Specifically, we consider an agent interacting with an environment according to a prescribed stochastic policy. By carefully modulating the random action realizations of
agent - arxiv:2610.15969 · cs.AIPopGen: Generating Semantically Diverse Populations for Zero-Shot CoordinationPaul Teiletche, Kanghyun Ryu, Kyungjoon Park, Dhruva Tirumala +2
In multi-agent reinforcement learning, exposing a cooperator to partners with diverse strategies during training is key to better collaboration at test time. However, existing methods typically rely on trajectory-level proxies to measure diversity, which may label two distinct trajectories of the sa
agentmulti-agent - arxiv:2610.15964 · cs.AISmall but Mighty: Test Suites as Improvement Signals for Self-Evolving AgentsYutong Cheng, Wei Cheng, Wenchao Yu, Dawei Zhou +1
A self-evolving agent rewrites its own harness and must decide which rewrites to keep, so its improvement signal is the check a rewrite has to pass. Verifying a rewrite by agent rollout is expensive and subjective, while executable checks such as the repository's own tests are too sparse to catch be
agentself-evolving - arxiv:2610.15956 · cs.AIADAPT: Audio-Depth Action Planning with Transition-Aware Rewards for Lights-Out Industrial NavigationWenjun Huang, Shenghao Fu, Yian Jin, Yang Ni +11
Vision-language-action models for embodied navigation typically rely on RGB observations, limiting their applicability in lights-out industrial environments where visible illumination may be unavailable. We study illumination-independent navigation using spatial audio for target localization and dep
vision-language-actionembodiedbenchmark - arxiv:2610.15943 · eess.SYRayleigh-Benard Stabilization by Fredholm BacksteppingMiroslav Krstic, Eugene Lavretsky
Among Navier--Stokes systems, the Rayleigh--Bénard layer, a fluid between two horizontal plates heated from below and cooled from above, is the benchmark of hydrodynamic instability that manifests itself as the unsteady convection phenomenon, which in turn can be interpreted as the onset of chaos: t
benchmark - arxiv:2610.15938 · cs.AIThe Anatomy of a Blind Spot: What an LLM Judge Must Be Shown, and How MuchNitish Kumar, Prateik Pothuneedi, Greg Bandak
Evaluation of text generated by large language models (LLMs) is increasingly delegated to LLMs acting as judges, yet what such a judge must be shown remains unresolved. We treat a judge whose reports are acted on as a regulator, in the sense of cybernetics, and ask what it must be shown, how much, i
benchmark - arxiv:2610.15936 · cs.ROJacobian State Augmented Viable Kinematic Safety FilterSiddharth Singh, Joshua Vaughan
Safety Filters (SF) are leveraged to design certifiably safe controllers, which ensures forward invariance on a defined safe set when the filter is feasible. Predictive Safety Filters (PSF) embed these constraints to a closed-loop model-based optimization method to improve constraint satisfaction, a
manipulator - arxiv:2610.15931 · cs.LGMesh-Free Convolution: Learned Spectral Attenuation and TransportNiloufar Zakariaei, Manav Manav, Eldad Haber
Learning solution operators for partial differential equations (PDEs) requires models that can represent spatial structure across grids and scales. This becomes especially important in time-dependent problems, where the learned operator is often used autoregressively: each predicted state is fed bac
benchmark - arxiv:2610.15929 · cs.CVWhen Trackers Fail: VLM-Guided Verification and Recovery for Robust Video Object SegmentationValay Bundele, Susmit Agrawal, Mehran Hosseinzadeh, The Nam Nguyen +1
Recent trackers built on foundation models such as the Segment Anything Model (SAM) achieve strong performance by propagating object masks through memory. However, this strength is also a key source of failure: under occlusion, reappearance, or distractors, an incorrect prediction can enter the trac
memoryagentbenchmark - arxiv:2610.15924 · cs.LGScalable Weak-to-Strong Verifier Hardening for Agent EnvironmentsChong Zeng, Ziqian Zhong, Kexun Zhang, Ivan Bercovich +4
Agent environments are essential for large language model (LLM) development. They use programmatic verifiers to turn complex outcomes into rewards. Verifiers can be attacked into awarding full credit to cheating agents without completing the intended task, causing inaccurate evaluation and misalignm
agent - arxiv:2610.15921 · cs.LGFast and Accurate: Improving Personalized LLM Performance by Semantic Concentration Inspired Workload Split between RAG and Fine-TuningXingke Yang, Sicong Li, Lingfeng Yao, Rui Chen +3
With ever increasing personalized large language model (LLM) services, the next question to answer is how to improve the quality of services/users' experiences, which is indicated by personalized LLM generation accuracy/quality and inference latency. Although state-of-the-art designs, e.g., paramete
retrieval-augmentedrag - arxiv:2610.15915 · cs.CVVIEWCRAFT: Evolving What Multimodal Agents See and How They Read ItBinxiao Xu, Ruichuan An, Bocheng Zou, Weihua Du +6
Reusable skills improve multimodal agents without updating model weights, but their effectiveness depends on both the evidence presented and the procedure used to read it. We introduce ViewCraft, which jointly evolves an executable evidence interface and its reading procedure. The interface selects,
benchmark - arxiv:2610.15912 · cs.CLMore Reasoning Is Not Always Better: Executing Clinical Guidelines with LLMs and RulesXiaoqian Lu, Guangfu Guo, Long Cheng, Kunpeng Liu
Large language models (LLMs) have shown considerable promise in clinical decision support. However, existing approaches often rely on multi-step reasoning or multi-agent collaboration to execute clinical guidelines, introducing substantial computational overhead without reliably preventing decision
multi-agentbenchmark - arxiv:2610.15910 · cs.CLRIVO: Lightweight Adaptive Rerouting for Deep Research AgentsMingchen Li, Jiatan Huang, Hong Yu
Deep Research agents rely on long-horizon reasoning and iterative tool use to solve complex information-seeking tasks. However, adaptive decision-making over these trajectories remains challenging, as agents may stop prematurely when further evidence is needed, continue unnecessarily after sufficien
agenttool usebenchmark - arxiv:2610.15908 · cs.AILURE: a Phase-Labelled Synthetic Corpus of Online-Grooming ExchangesEmanuele Florindi, Valentina Franzoni, Ludovico Guercio, Alfredo Milani +1
Phase-annotated conversational data is essential for studying the progression of online grooming of minors and for developing early warning tools. However, the only publicly available source (i.e., Perverted-Justice archive, which contains a subset coded using ChatCoder) is small, outdated, and froz
retrieval-augmentedhuman-in-the-loop - arxiv:2610.15907 · cs.MAMind Your Own Role: Reducing Role Overstepping for Effective and Efficient Multi-Agent Collaboration via Dynamic Role Activation SteeringXingke Yang, Sicong Li, Lingfeng Yao, Rui Chen +4
Multi-Agent Systems (MAS) have shown strong capability in solving complex tasks by assigning different agents specialized roles and corresponding subtasks. However, agents may overstep their assigned responsibilities and perform subtasks belonging to other roles, which can disrupt workflow, reduce s
agentmulti-agentagent system - arxiv:2610.15904 · physics.opticsTransfer learning-aided, artificial intelligence approach to evaluate optical properties of materials for classical and quantum photonic applicationsMohammad Reza Habibi, Luca Sapienza, Juan C. Vasquez
In this study, an artificial intelligence-assisted evaluation of the refractive indices of different materials is performed. The evaluation of the refractive index is necessary to design photonic devices for various classical and quantum photonic applications, and to theoretically analyse experiment
quantum photonic - arxiv:2610.15891 · cs.CVLearning Multimodal Sensor Fusion with JEPA Representations for Breathing Activity and Pose RecognitionYunxiang Luo, Constantino Álvarez Casado, Xiaoting Wu, Nhi Nguyen +3
Predictive video representations do not by themselves specify how a multimodal system should handle missing sensors. We transfer frozen V-JEPA2 features to a set of heterogeneous sensors comprised by RGB video, depth video, and radar time series. A shared fusion model uses modality queries, fixed se
v-jepa - arxiv:2610.15888 · cs.LGVariance-Optimal State Resampling for Reinforcement Learning with Verifiable RewardsYibo Wang, Yuanxin Liu, Zhaoran Wang
Reinforcement learning with verifiable rewards spends much of its rollout budget on states that yield no learning signal. We recast this as a question about the sampling distribution itself: under a fixed rollout budget, which distribution over states minimizes the variance of the policy-gradient es
agenticbenchmark - arxiv:2610.15885 · cs.LGComparative Study of Canvas- and Coordinate-Based Geometry-Conditioned Neural Networks for Real-Time Prediction of Indoor Environments with Varying Room DimensionsShaofan Wang, Philipp Geyer
Computational fluid dynamics (CFD) is computationally expensive for iterative building design. Deep learning surrogates offer faster predictions, but generalization across varying room dimensions remains challenging. This study compares two geometry-conditioned neural network surrogate models for ra
benchmark - arxiv:2610.15884 · cs.AIRetryBench: Separating Retry Permission from Recovery Outcomes in GPT Tool UseYou Liu, Yue Liu, Quanchao Lu
False retry permission (FRP) occurs when an agent retries a timed-out tool call without support from the visible contract. RetryBench separates this error from realized recovery. Successful protected recovery does not imply reliable rejection of unsupported retries. In a 720-response key control, va
agenttool use - arxiv:2610.15883 · cs.AIWhat the Model Cannot See: Covert and Provable Text Invisibility in LLMsMark Russinovich
Text wrapped in a trigger can be made behaviorally invisible to a language model: omitted from summaries and quotations, ignored in question answering, and left unflagged by downstream filtering. We study this selective blindness as a dual-use information-flow primitive, a supply-chain backdoor, and
benchmark - arxiv:2610.15882 · cs.AIEngram Injection: A Study of Temporal Validity in Parametric Language-Model MemorySam Siavoshian, Omar Ramadan, Amir K. Saeed, Benjamin A. Johnson +1
A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supp
memory - arxiv:2610.15864 · cs.LGPersonalized LLM Recommendation with Semantic IDs: Autoregressive Beam Search vs. Multi-Token PredictionSejoon Oh, Fernando Amat Gil, Evan Smothers, Dan Iter +10
Large-scale personalization is shifting from embedding-table retrieval to LLM-based recommenders that generate the Semantic ID (SID) of the item to recommend, where each SID is a short sequence of discrete hierarchical content codes. A single instruction-tuned model can then unify retrieval, content
post training - arxiv:2610.15855 · cs.AIAttention Sinks in VLMs Are Borrowed, Not SeenDengjia Zhang, Alexander Martin, Pengyu Yan, Xihan Sun +5
We study how to detect and mitigate visual attention sinks in vision-language models (VLMs). Visual attention sinks are image tokens that absorb a large share of attention over the image regardless of the query or task, and they mislead methods that treat attention maps as a record of where the mode
benchmark - arxiv:2610.15846 · cs.LGMobility Helps Learning: Unsupervised Model Adaptation for Object Recognition via MovementYanan Ma, Yihang Tao, Zhengru Fang, Senkang Hu +2
Movable agents, such as autonomous vehicles and robots, should be able to autonomously adapt their pre-trained object recognition models to new environments without human annotations. While existing test-time adaptation (TTA) methods address label-free model adaptation, they typically treat test dat
embodiedagent - arxiv:2610.15840 · cs.LGLLM Value Functions for Credit Assignment in Agentic RLKaijie Zhu, Weixiang Yan, Omar Attia, Abbas Kazerouni +2
Reinforcement learning (RL) for large language model (LLM) agents remains challenging in long-horizon tasks with sparse outcome rewards, where assigning credit to individual actions is difficult. Methods such as GRPO assign the same trajectory-level advantage to all actions, while critic-based metho
agenticbenchmark - arxiv:2610.15833 · cs.ROP-BFM: Perceptive Behavioral Foundation Models for Humanoid Control via Unsupervised Reinforcement LearningXue Wang, Yitang Li, Yiming Zhang, Siqin Wang +6
Behavioral foundation models (BFMs) provide humanoids with reusable repertoires of whole-body behaviors; in real-world humanoid control, the same behavioral intent may require fundamentally different motions under different terrain constraints. Covering such variation with terrain-matched demonstrat
humanoidteleoperation - arxiv:2610.15829 · cs.LGGuarded Block Transfers for Polytope Distance: Linear Convergence with Explicit Constants and Gap-Certified Screening for SVM TrainingMayank Gupta
Hard-margin SVM training is the distance problem between two convex hulls, which the Triangle Algorithm (TA) solves with a certified duality gap. We propose \emph{guarded block transfers}: $k$ Mitchell--Demyanov--Malozemov (MDM) transfers per scan are aggregated into one exact line search, guarded b
benchmark - arxiv:2610.15818 · cs.LGBenchmarking Deep Learning Models for Anomaly Detection in Real-World Cloud TelemetrySumit Dhyani, Sougata Mukherjea
Modern cloud platforms generate high-dimensional, noisy telemetry from thousands of microservices. Unlike classical anomaly detection benchmarks, cloud anomalies are short-lived, irregular, and largely non-periodic, making timely detection challenging. We present an empirical benchmark of deep learn
benchmark - arxiv:2610.15814 · cs.ROReTrace: Cross-Stage Failure Attribution for Self-Improving Robotic ManipulationYue Yu, Bin Zhu, Junhui Li, Jiayu Wang +2
General-purpose multimodal foundation models are increasingly capable of directly controlling robots, yet improving their manipulation capabilities through autonomous interaction remains challenging. A key challenge to such self-improvement is learning from failures in complex manipulation tasks inv
manipulationfrankaself-improvingself-improvement - arxiv:2610.15807 · cs.CVTotem: Object-Centric 3D Memory for World-Consistent Video GenerationZichen Wan, Tianyu Huang, Fan Li, Kai Wang +2
Generating coherent videos grounded in a unified underlying world remains a major challenge, particularly under large viewpoint changes, where existing methods often struggle to maintain stable representations of individual objects. To this end, we introduce Totem, an object-centric 3D memory framew
memorybenchmark - arxiv:2610.15806 · cs.CVReference-Efficient Multimodal Bitstream-Corrupted Video Recovery via Depth-Conditioned AdaptationTianyi Liu, Jiaojiao Wang, Ye Lu, Kejun Wu +3
Recovering visually faithful content from damaged video bitstreams is important for reliable multimedia communication and storage. Existing methods exploit local residuals and non-local RGB references of the input video sequence, but both sources can become unreliable after decoding, while processin
benchmark - arxiv:2610.15797 · cs.AIPassing Tests, Missing Semantics: Evidence of Interpretive Lock-In in Plan-Based DataFrame Code GenerationJingjing, Liu, Dishu Yang, Hongbo Qin +1
Planning and revision are intended to bring generated code closer to the task specification. Yet a mistaken interpretation can persist across the plan, implementation, and tests - a pattern we call interpretive lock-in. We investigate this risk on 30 DataFrame tasks derived from TPC Benchmark DS, co
benchmarkevaluator - arxiv:2610.15795 · cs.AIAn Explainable System for Offensive Content Detection in Nastaliq Urdu: Corpus Curation, Transformer Benchmarking, and Word-Level AttributionMuhammad Shahid Iqbal Malik
Automated moderation systems are deployed at scale, yet they remain opaque to the users and moderators affected by their decisions, and their coverage is almost entirely confined to English. For Urdu written in the Nastaliq script, no such system exists, and the annotated resources it would be built
benchmark - arxiv:2610.15794 · cs.RO$Δ$ Flow Matching for Robot Foundation ModelsKuangji Zuo, Tuo An, Xinying Guo, Gen Li +8
Flow matching is a common generative recipe for robot foundation models. Vision-language-action models (VLAs) generate action chunks, while world-action models (WAMs) additionally generate future visual states. Both start generation from a fixed Gaussian prior, despite temporal continuity between pa
vision-language-actionvlavla policyrobot foundation modellibero - arxiv:2610.15788 · cs.AIAgent Sociodrama: Social-Dynamics-Guided Persona Recasting for LLM Agent TeamsYirui Qi, Xiaoming Zhang, Jaehong Yoon
Large language model (LLM) agents are increasingly deployed as teams, yet they fail in characteristically social ways: information holders stay silent, dominant speakers anchor groups on wrong answers, and norm violations go unsanctioned. We present Agent Sociodrama, a framework for modeling and val
agentllm agentbenchmark - arxiv:2610.15782 · cs.CVData-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual BootstrappingSongze Li, Haoming Gu, Ziyang Gong, Zhitong Wang +10
Vision-language navigation requires agents to follow natural-language instructions to reach target locations, making high-quality language supervision essential. Since human annotation is costly, prior work has explored scalable data synthesis methods. However, existing methods typically either use
self-improvement - arxiv:2610.15781 · cs.LGHeterogeneous Structured Pruning Under Resource Constraints for Hardware-Efficient Large Language ModelsCameron Barker, Andreas Paraskeva, Jan N. van Rijn, Elliot J. Crowley +1
Training-free pruning removes weights from pretrained large language models to reduce their deployment cost. However, existing approaches typically consider only the total number of retained weights. Consequently, their pruning decisions do not account for how target hardware stores and processes th
memory - arxiv:2610.15772 · cs.CLRSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-EvolutionFan Zhang, Hao Wu, Jinyang Wu, Hongyu Qu +6
Recent focus in agentic reinforcement learning (RL) seeks to improve credit assignment in long-horizon tasks. One promising direction combines reinforcement learning with verifiable rewards (RLVR) and on-policy self-distillation (OPSD), integrating trajectory-level advantages with dense token-level
embodiedagenticself-improvingself-evolving - arxiv:2610.15769 · cs.ROHand2Robot: Object-Centric 4D Hand Reconstruction from Human Demonstrations for RoboticsZetao Du, Weiqing Wang, Hanbing Zhang, Yimiao Zhou +3
Human demonstrations can capture task-relevant manipulation without using the target robot, but the recorded hand motions cannot directly serve as robot action supervision. Hand reconstruction and motion retargeting provide a direct route to such supervision, yet recovering hand motion alone does no
manipulationdexterousrobot policypost-training - arxiv:2610.15766 · cs.ROSurgiCLAW: An Interpretable Atomic-Skill Agent for Precision Robotic MicromanipulationAngen Ye, junjie xie, Yujia Song, Junkai Huang +1
Recent multimodal large language models (MLLMs) and vision--language--action models have provided robots with new capabilities for long-horizon task understanding and planning. In surgical robotics, however, end-to-end prediction of low-level actions remains limited by insufficient precision, opaque
manipulationagent - arxiv:2610.15765 · cs.AIA Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software EvolutionAlan Wu, Wei Tao, Yiwen Guo
Large language model agents are emerging as an important new class of users of desktop applications. However, existing desktop interfaces were designed for human end users or application developers, leaving agents to interpret visual controls or discover and compose low-level software operations. Th
agentbenchmark - arxiv:2610.15763 · cs.ROCompetitive Task Allocation with Temporal Proposal Windows and Improved Artificial Potential Fields for Dual-Arm Industrial ManipulationSurya Prakash S. K, Bhuvan Narula, Jagannath Prasad Sahoo, Jayakumar VK +2
Modern industrial automation demands autonomous collaborative systems capable of coordinated manipulation in workspaces with heterogeneous object arrangements. Existing systems rely on predetermined trajectories and rigid allocation strategies, causing suboptimal workspace utilization and collision
manipulationmanipulator - arxiv:2610.15760 · cs.LGHyperbolic Time Series Forecasting with Learnable Decomposition and Manifold DynamicsSarang Rajendra Patil, Vishal Fenn, Mengjia Xu
Multiscale structure is central to time series forecasting: temporal signals contain slowly varying trends, intermediate seasonal patterns, and high-frequency residuals. Traditional forecasters model this multiscale structure in Euclidean space, without representing the ordering it induces among com
benchmark - arxiv:2610.15756 · cs.ROKinematic Reward, Measured Effort: Hardware-Validated RL Morphology Optimization of a Planar 2R ManipulatorGunesh Gupta, Arvind Kumar Mishra, Sohom Chakrabarty
Yoshikawa's Manipulability Index is generally used as a classical measure to define an optimal morphology of a manipulator. However, any such optimization is almost always studied as a purely kinematic quantity in simulation and is rarely validated on hardware. We present a reinforcement-learning fr
manipulator - arxiv:2610.15754 · physics.opticsDirect growth of crack-free stoichiometric silicon nitride on sapphire for wide-band photonicsZhaohui Ma, Ashish Chanana, Zhiquan Yuan, Khoi Hoang +4
Low-loss stoichiometric silicon nitride on oxidized silicon has emerged as the major wide-band photonics platform for nonlinear and quantum photonics. These films, deposited by low-pressure chemical vapor deposition (LPCVD), are well known to have around 1000 MPa of residual tensile stress, which of
quantum photonic - arxiv:2610.15750 · cs.AIGoogle Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic EvaluationNoam Slonim, Aviv Slobodkin, Gil Fidel, Gal Moshitch +33
Google Earth AI is a family of geospatial AI models, together with an agent that orchestrates them. We present this agent, Google Earth AI's Geospatial Reasoning agent (GRA), a Gemini-powered research system that combines geospatial data sources, foundation models, and tools to answer questions abou
agentagenticbenchmark - arxiv:2610.15742 · cs.CVRSAJEV: Training-Free Abstaining Prototype Memory for Eliminating Silent Hallucination in JEV-Based Remote Sensing ClassificationMan Chen, Antai Chen, Shuai Xu, Chengjie Gu +4
Scene classification of remote sensing imagery with multimodal large language models (MLLMs) is formulated as a single-forward-pass decision problem: candidate class names are lettered options in the prompt, a registered answer slot is conditioned on the prompt, and classification is read out by sco
memoryretrieval-augmented - arxiv:2610.15738 · cs.AIVisible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization LoopsJianlin Chen, Wenhui Chen, Chi Man Vong
LLM optimization loops often show the proposer the candidates it has already tried and rejected, but leave it to the proposer to avoid them. We ask whether avoidance measured in a static probe predicts avoidance in the states an optimizer's own trajectory produces. In catalog-based loops where three
agent - arxiv:2610.15732 · cs.CLNepPlural: A Human-Anchored Benchmark for Persona Annotation in Nepali Migration DiscourseJinu Nyachhyon, Anish Khatiwada, Bimal Timilsina
Pluralistic alignment asks that language models faithfully represent a society's diverse, competing values rather than collapse them into one dominant viewpoint, yet flagship pluralistic-alignment datasets remain heavily skewed toward Western respondents. We introduce NepPlural, a human-anchored ben
benchmark - arxiv:2610.15730 · cs.LGA Dependency-Aware Multi-GPU Training Framework for Temporal Graph Neural NetworksYang Zhao, Yue Dai, Liang Qiao, Buxin Tu +7
Temporal graph neural networks (TGNNs) achieve superior performance in many dynamic graph learning tasks. These models use node state vectors (i.e., memories) to capture and record graph changes (i.e., events), and then exploit these memories to assist prediction. However, this memory-centric design
memory - arxiv:2610.15725 · cs.AISkillMarker: A KGW-Based Watermarking Method for Agent Skill Ownership VerificationAlan Tang, Lifeng Wang, Gang Wu, Tao Zhang +2
Agent skills encode operational procedures and domain expertise as reusable knowledge assets, yet their distribution in plaintext exposes them to unauthorized copying and leakage. However, practical mechanisms for skill ownership verification and leakage tracing remain underdeveloped. Watermark embe
agent - arxiv:2610.15722 · cs.CVPhySimCode: A Benchmark and Evaluation Method for Physics Video to Code GenerationSouraja Kundu, Aditya Gupta, Joel Julin, Yizhou Zhao +2
Multimodal large language models (MLLMs) are good at video understanding and code generation, but it remains unclear whether they can recover the physical laws underlying videos, a capability essential for converting videos into executable simulations and for providing structured physical supervisio
benchmark - arxiv:2610.15721 · cs.LGTo Cache or Not to Cache: Understanding and Mitigating Safety Incompatibility in Cross-Model KV Cache ReuseZhiwei Lin, Jiahao Liu, Yan Liu, Chengpeng Wang
Cross-model KV cache sharing reduces prefill computation by allowing a receiver model to reuse prefix KV states generated by a related sender model. Existing methods optimize such reuse for latency and generation quality, but its safety implications remain unclear. In this work, we systematically ev
benchmark - arxiv:2610.15715 · cs.LGHistory Replays Itself: Unleashing Long-Context Capability in Linear AttentionLi Nanbo, Firas Laakom, Yilan Zhang, Piotr Piękos +9
Linear attention promises the efficiency needed for long-context language modelling, yet its fixed-size recurrent state must simultaneously support ongoing computation and preserve information from an ever-growing history. This coupling has reinforced a prevailing view that the state's memory capaci
memorylong-contextepisodic memory - arxiv:2610.15714 · cs.AIMedHarnessBench: A Unified Benchmark for Healthcare Agent EvaluationRui Tang, Hongqiu Wang
Healthcare agents must transform longitudinal clinical evidence into reliable actions through heterogeneous tools and stateful workflows, conditions under which model-centric systems may lose context, misuse tools, fail to verify consequential transitions, or cross authorization boundaries. Existing
agentagent benchmarkbenchmark - arxiv:2610.15712 · cs.CLTL;QR: Querying Temporal Discursive Patterns in Textual CorporaFrançois Hublet, Vladimir Krsmanovic, Jesus Solano, Ahmed Bouhoula +1
Many questions about large textual corpora concern patterns that unfold over time, such as a political group whose stance on a topic drifts from support to criticism over several years of debates. We present TL;QR (Too Long; Query with logical temporal Reasoning), a declarative query language and op
benchmark - arxiv:2610.15710 · cs.AISuggested Is Not Superseded: A Pre-Registered Evaluation of a System-One Model as a Memory-Retirement GateXupeng Chen, Chenqian Le
Long-term memory for assistants needs a gate that decides when a new utterance retires an old stored claim. Benchmark labels for this step mark a pair as changed or not, so they cannot tell a gate that acts on established change from one that acts on a hint. We write a task contract that separates t
memorybenchmark - arxiv:2610.15709 · cs.LGPreserving Updates in Low-Bit Recurrent State QuantizationHaotian Li, Liran Dong, Kaicheng Xiao, Guoliang Xing
Linear attention and recurrent models compress past tokens into a fixed-size state that is updated at every step. However, storing these states in high precision incurs substantial memory and bandwidth costs at large batch sizes, motivating low-bit state quantization. Unlike weights and key--value c
memory - arxiv:2610.15708 · cs.AISpecForge: Recursive Self-Improvement for Coding Agents via Behavioral SpecificationsYutong Cheng, Haifeng Chen, Peng Gao, Wei Cheng
Coding agents based on frozen language models can improve their performance by refining the external context they rely on, without updating model weights. We study recursive self-improvement through behavioral specifications: natural-language code documentation that describes the behavior of program
agentself-improvementbenchmark - arxiv:2610.15705 · cs.AIbharatSetu: a 1M x 22 bridge for indic translationHaq Nawaz Malik, Nahfid Nissar, Faizan Iqbal
India's Eighth Schedule recognises 22 languages, yet parallel data is uneven across them, and many language pairs, especially among lower-resource languages such as Bodo, Dogri, Santali and Kashmiri, lack direct bitext. We present BharatSetu 22-Indic 1M, a row-aligned table of 984,764 English senten
evaluation protocol - arxiv:2610.15691 · cs.AIFARL: A Fully Automated Pipeline of RL Interface and Solver DesignTianyu Jin, Liuyu Xiang, Chaowen Yang, Xingyang Chang +2
Existing reinforcement learning (RL) largely focuses on optimizing learning algorithms within a pre-defined task formulation, while applying RL to new tasks still relies heavily on human expertise to design the learning pipeline: what an agent observes, how it is rewarded, which algorithm it uses, a
manipulationmemoryagent - arxiv:2610.15687 · cs.CLTowards Autopilot Labeling for Short Video Content GovernanceJunwen Chen, Hongwei Wang, Zixuan Wang, Jingbing Wang +11
The explosive growth of short videos has presented significant challenges to content governance. Traditional manual labeling is inefficient, slow to iterate, and costly. Therefore, we explore and develop a multi-stage autopilot labeling framework aimed at progressively increasing the level of automa
agentagentic - arxiv:2610.15685 · cs.AIRadioactive Feedback: Tracing Models Trained by Label-Only LLM SupervisionYuejia Li, JunHeng Li, Hao Wu, Ke He +1
Conventional distillation transfers capability through teacher-generated demonstrations. LLM-as-a-Judge knowledge distillation offers a distinct and increasingly practical alternative. In the label-only variant studied here, the student generates every response and improves from scalar scores alone.
evaluator - arxiv:2610.15675 · cs.RORecovering Low-Bit Failures in Vision-Language-Action Policies via Quantization-Conditioned Reinforcement LearningSeongmin Park, Jaewon Lee, Yujun Heo, Yoonsik Kim +3
Quantization lowers deployment costs of vision-language-action (VLA) policies but can compromise task completion. Across five VLA models and 4,050 real-world trials, we systematically reveal that quantization-aware training (QAT) generally improves on post-training quantization (PTQ) but leaves subs
vision-language-actionvlavla modelgr00tpost-training - arxiv:2610.15667 · cs.CLRate-Resolved Text Evaluation with Masked Diffusion Language ModelsYiqi Liu, Kun Zhao, Chenghua Lin, Nafise Sadat Moosavi
A single evaluation score can rank generation systems but rarely reveals what failed. We propose rate-resolved text evaluation, which turns the masking rate of masked diffusion language models into a diagnostic axis. By reconstructing text under different fractions of masked tokens, one diffusion sc
benchmark - arxiv:2610.15666 · cs.CVA Persian Vision-Language-Action Dataset and Benchmark for Ambiguity- and Infeasibility-Aware Instruction-to-Plan GroundingAbdollah Pourasghari, Saeedeh Sadat Sadidpour, Mohammad Roustaei
Converting natural-language instructions into executable action plans conditioned on a scene image is a core capability for embodied human-robot interaction, but existing benchmarks target English, simulated or video-grounded environments, and largely assume instructions are clear and executable. We
vision-language-actionembodiedagenticbenchmarkllm-as-judge - arxiv:2610.15661 · cs.CLFeedback as an Interface: Controlling Form and Detail in LLM Translation RefinementYiqi Liu, Flavjo Tomori, Nafise Sadat Moosavi, Steffen Eger +1
Feedback presentation shapes how LLMs use evaluator judgments during translation refinement. We introduce a machine translation protocol that keeps the evaluator score, main issue, and recommendation fixed while varying feedback form and supporting detail. Across seven models and six translation dir
evaluator - arxiv:2610.15659 · cs.LGTesting Central Structure for Complex NetworksQilong Wu, Yuxiao Li, Simeng Zhang, Xinying Liu +2
Centrality measures rank nodes even when the observed differences can be explained by ordinary structural variation. We therefore study a preceding question: whether a network contains a statistically identifiable central structure at all. We formulate this question as a network-level hypothesis tes
benchmark - arxiv:2610.15658 · cs.AIDynamic Conflict Clustering with Lightweight Learning and Strategic Repair for Distributed Constraint SatisfactionRaziyeh Moghaddas, Fabio Caraffini
Distributed Constraint Satisfaction Problems (DCSPs) model distributed decision-making in multi-agent systems, where agents assign values to local variables while satisfying shared constraints. However, many DCSP solvers still rely on broad, repeated peer-to-peer communication and are largely reacti
multi-agentagent systembenchmark - arxiv:2610.15645 · cs.CVSurgCast: Action-Conditioned Future Skeletons for Controllable Surgical World ModelsWanhao Liu, Rulin Zhou, Liangjing Shao, Zhaocheng Lin +8
Surgical world models must turn prospective robot actions into visually faithful, structurally consistent instrument motion. We introduce SurgCast, which predicts future instrument skeletons from current structure, robot state, and future actions, then uses the forecast to control video generation.
world modelaction-conditioned - arxiv:2610.15642 · cs.AIRecall What Is Missing: Composing Evidence from Long-Term Agent MemoryQi Feng, Jicong Fan
Under a limited context budget, a long-term agent must recall not merely memories that are individually relevant, but evidence that is collectively sufficient for the current query. Conventional memory access retrieves, scores, and truncates memories, often spending multiple context slots on redunda
memoryagent memoryagentbenchmark - arxiv:2610.15640 · cs.LGContext-Aware Precision Scheduling for Small Language ModelsMukund Athmaram, Manoj Rao, Liang Ma
Small language models (SLMs) used as on-device agents must balance accuracy and latency as their context grows. Mixed precision helps manage this tradeoff by assigning different precisions to weight tensors according to their size and sensitivity. However, these choices depend on the operating condi
tool-usebenchmark - arxiv:2610.15639 · cs.ROImitation Learning via Discrete and Rhythmic Primitive CompositionRafid Saif, Auke Ijspeert, Hsiu-Chin Lin
Long-horizon robotic manipulation is inherently composed of semantically meaningful sub-tasks. While Movement Primitives (MPs) can model these components, their composition typically relies on manual engineering or sample-inefficient reinforcement learning. Existing supervised methods are limited by
manipulation - arxiv:2610.15638 · cs.LGMERIT: Scaling KG Foundation Models with Relation TransformersCosimo Gregucci, Daniel Hernandez, Antonio Vergari, Steffen Staab
Knowledge graph (KG) foundation models (KGFMs) predict links on unseen graphs zero-shot. KGFMs have been pioneered by ULTRA which builds a relation graph from co-occurrences that match motif templates, and passes messages over it to obtain relation representations. More recent KGFMs such as MOTIF an
knowledge graphbenchmark - arxiv:2610.15634 · cs.AIBig-Big-SycoBench: A Benchmark to Detect Sycophancy Through Model Responses to Valid vs. Invalid ArgumentsMarthe Ballon, Arne Vanhoyweghen, Seorin Kim, Francesca Carlon +10
Sycophancy in large language models risks amplifying unsupported claims or beliefs across the knowledge ecosystem as these systems increasingly contribute to research synthesis, evaluation, and decision-making. A reliable model should revise its judgement when a user identifies a genuine flaw, but s
benchmark - arxiv:2610.15629 · cs.LGThe Price of Precision: A Calibrated Cost Model for Mixed-Precision LLM ServingMukund Athmaram, Manoj Rao, Liang Ma
Choosing a mixed-precision configuration for large language model (LLM) serving trades speed against accuracy. Allocators that use average bit-width or a memory budget as a speed proxy can miss changes in the resource that limits execution. We argue that served speed can be limited by compute, weigh
memory - arxiv:2610.15625 · cs.CV1Step-WAM: Your World Action Model but Faster using 1-Step Action GenerationRundong Su, Pengtao Chen, Haohan Geng, Sha Zhang +5
World Action Models (WAMs) unify visual world modeling and action generation, yet their real-world deployment is heavily bottlenecked by iterative action denoising and the sequential execution of visual prefilling and action prediction. We demonstrate that the action-denoising trajectories in WAMs a
liberorobotwinworld model - arxiv:2610.15620 · cs.AIFrom Stages to States: Unified Iterative Reasoning for MLLMsChuang Yu, Xuefei Ma, Yuanhao Feng, Wuxiong Zeng +8
Multimodal Chain-of-Thought (MCoT) substantially enhances the complex reasoning capabilities of Multimodal Large Language Models (MLLMs) by explicitly modeling intermediate reasoning processes. However, most learning-based MCoT methods follow a stage-centric paradigm, where different reasoning behav
self-refinementbenchmark - arxiv:2610.15614 · cs.LGProML: Progressive Multi-label LearningYi Kou, Jieting Wang, Yuhua Qian
Multi-label learning typically pursues model construction through the joint optimization of label fitting, correlation modeling, and structural regularization. While empirically effective, such a coupled paradigm imposes multiple learning objectives on a single predictive parameter matrix, forcing i
benchmark - arxiv:2610.15606 · physics.opticsTornado Wave-Encoded Point Spread Functions for Multi-Period Three-Dimensional Localization ImagingLai Chen, Rui Wang, Li-Gang Wang
Three-dimensional (3D) localization microscopy using rotational point spread functions (PSFs) is fundamentally challenged by limited axial range, poor PSF compactness, and depth-decoding ambiguity -- particularly in dense specimens with heavy PSF overlap. To address these limitations, we introduce a
benchmark - arxiv:2610.15604 · cs.CVSqueeze-Aware-YOLO: Early Channel Recalibration and Multi-Scale Fusion for Real-Time Traffic Sign DetectionMizbaul Haque Maruf, Syed Samin Sadaf, Syed Tamzid Bakht, Md. Hasanul Kabir +1
Traffic sign detection plays a vital role in intelligent transportation systems, supporting the safe operation of advanced driver assistance systems and autonomous vehicles. Despite significant progress in real-time object detection, accurately identifying traffic signs under challenging conditions
benchmark - arxiv:2610.15595 · cs.CVFireNav: Decision-Oriented Reconstruction Evidence Routing for Vision-Language Navigation in Indoor Fire EnvironmentsHongbo Ding, Xiang Chen, Jiangxin Dong, Jinshan Pan
In real-world indoor fire environments, visual degradation severely impairs instruction grounding, waypoint prediction, and long-horizon exploration in vision-language navigation (VLN). Although image restoration offers a natural means to mitigate such visual degradation, existing methods primarily
benchmark - arxiv:2610.15585 · cs.ROStreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow MatchingShaojie Ren, Jingwen Sun, Zezhi Liu, Hongjin Chen +3
World Action Models (WAMs), often built upon pretrained video-generation world models, have emerged as a promising framework for general-purpose and open-ended robot control. However, their real-world deployment often suffers from discontinuous and jerky execution. Existing methods typically address
liberorobotwinworld modelmemory - arxiv:2610.15582 · cs.AIInverse Garment Design with Visual AgentsNanxiang Jiang, Yunuo Chen, Zeng Tao, Zizhang Li +4
We study the problem of inverse garment design, reconstructing simulation-ready 3D outfits with arbitrary layers from a single image, with visual agents. Unlike differentiable optimization that is time-consuming, or feed-forward methods limited by training data, we introduce DressAgent, a hierarchic
memoryagentichierarchical agentbenchmark - arxiv:2610.15578 · cs.CLTowards Robust Calibration Dataset for Domain-General Mixture of Experts PruningLuis Frentzen Salim, Chia-Chun Wu, Tran Van Nhiem, Lun-Wei Ku +1
Post-training expert pruning is a promising deployment technique for Mixture-of-Experts (MoE) models. The calibration set is an integral part of the pruning process, however, its correlation with evaluation performance is largely misunderstood and unintuitive. In this work, we present an empirical s
post-training - arxiv:2610.15577 · cs.ROSIGN-MAPPO: State-Intent-Aware Graph Multi-Agent Reinforcement Learning for Cooperative Driving under Imperfect V2XZide Fan, Jiahui Yuan, Kaixin Zhang, Junhao Fan +2
Cooperative driving at mixed-traffic unsignalized intersections is challenged by imperfect vehicle-to-everything (V2X) communication, human-driven vehicle (HV) intentions that cannot be directly communicated, and the potential incompatibility of independently generated crossing decisions. This paper
multi-agent - arxiv:2610.15575 · cs.AISelf-Representation Alignment in Diffusion Models for Video Background Music GenerationJincheng Zhang, György Fazekas, Charalampos Saitis
Diffusion models have been widely used for various generative tasks. However, recent studies have demonstrated that the internal representations learned in the diffusion models' denoising process are insufficiently meaningful, which can hinder the generation performance. To address this, we propose
benchmark - arxiv:2610.15572 · cs.ROLook Before You Move: Agent-Interface-Aligned Supervision for Open-World Mobile ManipulationRonghe Qiu, Teli Ma, Jiaming Zhou, Zeying Gong +3
Open-world mobile manipulation requires a robot to find an object, approach it, and move it to a target receptacle from a language instruction. Training vision-language model (VLM) agents for these tasks requires demonstrations of both action selection and active vision. Existing pipelines often con
embodiedmanipulationagent - arxiv:2610.15570 · cs.LGTruncation Samplers Are Nested: Comparing Decoding Rules Is a Measurement ProblemRasul Khanbayov, Hasan Kurban
Settling one comparison between two truncation samplers has cost six thousand A100-hours, and nothing says in advance how far apart two samplers can be. We observe that almost every deployed truncation rule keeps a prefix of the probability-sorted vocabulary, so any two such rules are nested at ever
benchmark - arxiv:2610.15565 · cs.CVGenesis: Visual Augmentation Primitives for Multimodal Chain-of-Thought ReasoningFan Yang, Lifeng Chen, Yipu Wang, Han Wang +13
Recent multimodal large language models achieve promising results on visual reasoning tasks, approaching human-level performance on existing STEM benchmarks. However, these capabilities degrade under complex real-world scenes. We identify that chain-of-thought reasoning remains largely confined to t
benchmark - arxiv:2610.15562 · cs.CVA Plausible Future Is Not a Causal Future: Can Video Models Generate How the World Unfolds?Jialiang Li, Zhengning Xu, Zongyang Ma, Shaobo Cui
Large-scale video generation models are increasingly viewed as a promising path toward world models, yet their ability to capture causally coherent physical dynamics remains poorly understood. Existing evaluations largely emphasize visual quality, prompt adherence, or broad physical plausibility, le
world modelbenchmark - arxiv:2610.15560 · cs.ROStereo-Planner: Learning Stereo-to-Trajectory Planning for Zero-Shot Sim-to-Real Quadrotor NavigationHongming Shen, Junjie Lu, Xun Chen, Wei Wang +3
Existing navigation systems commonly rely on dedicated depth sensors or explicit depth reconstruction, while learning to generate trajectories directly from RGB images remains challenging due to the sim-to-real domain gap. In this work, we propose Stereo-Planner, an end-to-end vision-to-trajectory f
sim-to-realbenchmark - arxiv:2610.15558 · cs.ROAnalytic Robot Self-Model: Learning Screw-Axis Kinematics for Rendering and Model-Based ControlChen Xu, Ning Tan
Visual self-modelling lets robots learn body models from observations, but such models usually encode kinematics implicitly rather than exposing metric joint parameters for computation and control. Moreover, a good fit to images does not by itself establish accurate kinematics. Here we present an ex
quadruped - arxiv:2610.15556 · cs.ROVision-Predictor-Action: Towards Efficient Embodied AI via Neuro-Symbolic JEPA PredictorsYeonseok Lee
Vision-Language-Action (VLA) models commonly generate robot actions token by token through Large Language Model (LLM) decoders, causing inference latency to grow with action sequence length and limiting closed-loop control frequency. We present Goal-Conditioned Vision-Predictor-Action (VPA), which r
vision-language-actionembodiedlibero - arxiv:2610.15555 · cs.LGELECT: Settling Identity Before Rendering for Scalable Story VisualizationMehrajul Abadin Miraj, Md. Mahmudul Hasan, Meherun Farzana, Aniket Joarder +8
Training-free story visualization without reference images must preserve recurring entities across frames while scaling to long narratives. Existing methods typically maintain consistency by conditioning each frame on other frames of the same story, coupling generation time, VRAM, or both to story l
benchmark - arxiv:2610.15548 · cs.ROPointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point CloudsSicen Li, Zhen Chu, Chao Li, Jun Zheng +2
Quadruped robots deployed for inspection and field missions must cross heterogeneous terrain, from stairs, slopes, and gaps to high platforms, stepping stones, and balance beams, where safe support is sparse or see-through. Existing methods interpose a terrain map or a depth image between sensing an
quadruped - arxiv:2610.15543 · cs.LGBeyond Aggregate Drift: Optimal Client-Specific Windows in Federated LearningYuriy Dorn, Dmitrii Vishovan, Kirill Novoselov, Ilya Kuruzov +2
We study federated learning from continuously arriving data: each client samples from its own distribution, these distributions drift over time, and the server must track the minimizer of the clients' weighted average loss. Clients are not alike as they collect data at different rates, with differen
memory - arxiv:2610.15542 · cs.CLApertus 1.5: From Open Text Models to Open Multimodal ReasonersProject Apertus, :, Yixuan Xu, Yassine Turki +129
Language models are becoming infrastructure, and infrastructure built in the open, by public institutions, on documented data, is the alternative to dependence on a few providers; Apertus shows this is viable at scale. Apertus 1.5 extends the fully open Apertus family from text to multimodal reasoni
long-contextbenchmark - arxiv:2610.15540 · cs.LGSensitivity Fidelity Reveals Physical Dependencies Beyond Forecast Accuracy in Deep-Learning Earth-System ModelsQi Chen, Peng Zhan, Hongxing Cui, Wenqing Cui +1
Deep-learning models increasingly rival numerical simulation in Earth-system forecasting, yet forecast error alone cannot determine whether their learned input-output dependencies are physically consistent. Here we introduce sensitivity fidelity as an additional criterion: alongside prediction error
benchmark - arxiv:2610.15529 · cs.AIAgent-Canary-Mas: End-to-End Evaluation of Protocol Risk in Long-Horizon Multi-Agent SystemsXiuming Liu, Le Gai, Zhiyuan Chang, Peiyang Li +3
Agent-to-agent protocols allow LLM agents to discover peers, exchange results, and coordinate tool use. A locally successful protocol attack, however, may be rejected by a later agent or propagated into a consequential task decision. Measuring local exploitability alone leaves this difference unreso
agentllm agentmulti-agentagent systemtool use - arxiv:2610.15522 · cs.ROGNFormer: Gaussian Neural Field Transformer for Efficient Camera-based Occupancy PredictionGuangcheng Chen, Li He, Hong Zhang
Camera-based occupancy prediction provides geometric and semantic scene understanding for robot perception and has been adopted in autonomous vehicle and UAV perception systems. Gaussian primitives offer a promising alternative by potentially representing only occupied regions, thereby concentrating
memorybenchmark - arxiv:2610.15520 · cs.AIImproving the Bounds on Approximating Weighted Maximin Share for ChoresTrung Thanh Nguyen, Khaled Elbassioni, Jörg Rothe
We investigate the problem of allocating indivisible chores among agents with arbitrary positive weights and additive nonnegative cost functions. An allocation is said to be $α$-weighted-maximin-share (WMMS) fair if the cost incurred by each agent is at most $α$ times her weighted maximin share. We
agent - arxiv:2610.15514 · cs.LGSARATHI: Decoupled Data-Free Probing and Adaptive Reconstruction for Structured LLM PruningSankar Behera, Dhruv Singh, Satyadev Ahlawat, Yamuna Prasad
Structured pruning reduces the inference cost of large language models (LLMs) by permanently removing neurons and weight rows, yielding dense, hardware-agnostic speedup. Existing post-training methods couple neuron selection and weight reconstruction through a shared calibration dataset, introducing
memorypost-training - arxiv:2610.15509 · cs.LGCacheLink: Functional KV Transport for Zero-Prefill Cross-Model Context HandoffHaomiao Qiu, Yining Tang, Zijia Song, Ruizi Han +2
KV caches make long-context serving efficient by reusing past computation, but model switching breaks this reuse and forces the receiver to prefill the context again. We study \emph{zero-prefill cross-model context handoff}: translating a sender cache into a receiver-compatible state so that the rec
long-context - arxiv:2610.15492 · cs.AIWhen Skills Remain But Access Fails: Factorized Skill Memory for Continual Skill InternalizationHaomiao Qiu, Yining Tang, Hongyi Wan, Changfeng Sun +2
Skill internalization aims to convert externally specified procedural skills into persistent model capabilities, allowing language-model agents to execute them without retrieving or reinjecting skill descriptions at inference time. Yet when new skills are learned continually, whether previously inte
memorybenchmark - arxiv:2610.15488 · cs.LGAlphaPilot: Multi-Turn Tool-Using LLM Agents for Adaptive Alpha Mining with Reinforcement LearningPeiqi Jiang, Fan Zhang, Xin Xu, Tianhao Chen +1
Formulaic alpha mining seeks interpretable expressions that rank assets by future returns. The task is challenging because it demands simultaneous processing of unstructured market descriptions, historical data, and quantitative evaluation of candidate factors, yet most existing methods exploit only
agentllm agentevaluator - arxiv:2610.15486 · cs.ROLearning Safety-Constrained Whole-Body Loco-Manipulation for Lunar Surface OperationsAlvaro Belmonte-Baeza, Miguel Cazorla, Gabriel J. García, Carlos J. Pérez-Del-Pulgar +1
Robotic systems for lunar surface operations must act autonomously and preserve their own integrity in an environment where failures are unrecoverable and continuous supervision is impossible. This paper presents a Constrained Reinforcement Learning framework for quadrupedal mobile manipulators that
manipulationmanipulatorquadruped - arxiv:2610.15482 · cs.AISecurity in LLM-Assisted Coding: A Systematic Literature ReviewLiang Yuan, Maryam Khanian Najafabadi, Mashud Rana, Varvara Vetrova
Large Language Models accelerate software development but introduce severe security vulnerabilities. This systematic review investigates security risks in automated coding, benchmark limitations, and the role of human developers. Findings indicate that models frequently generate insecure code and fa
benchmark - arxiv:2610.15481 · cs.AISkillFusion: Towards Reliable Multi-Skill Composition for LLM AgentsEnhao Huang, Xinye Zhao, Yiming Li, Bingrun Yang +4
Independently authored skills can impose incompatible requirements, competing preferences, and redundant instructions when jointly loaded by an LLM agent. We propose SkillFusion, an offline compiler that reconciles these interactions into a reusable merged skill. Its central idea is to make composit
agentllm agentagent benchmarkbenchmark - arxiv:2610.15463 · cs.AIMemeLogos: Knowledge-Driven Retrieval-Augmented Reasoning for Harmful Meme DetectionHexiang Gu, Saihui Hou, Changlan Liu, Min Ren +3
With the rapid proliferation of social media, harmful memes have become a growing threat to online ecosystems. Their harmfulness is often implicitly conveyed through complex interactions among visual, textual, and contextual cues, making accurate detection challenging. Recent studies have introduced
retrieval-augmentedknowledge graph - arxiv:2610.15461 · cs.AIScalable Robust MCTS with Anytime Deterministic Optimality GuaranteesTomer Massas, Vadim Indelman
MDP planners often rely on nominal transition models that may differ from the true dynamics. Robust MDP planning accounts for this uncertainty through a set of admissible models, but doing so with limited online computation is challenging: the search tree grows exponentially with the horizon and eac
policy evaluation - arxiv:2610.15459 · cs.AIWhen Messages Go Wrong: Understanding Communication Faults in LLM-Based Multi-Agent SystemsJinhao Zhou, Xiaoyu Zhang, Jiongchi Yu, Fahao Chen +1
LLM-based multi-agent systems (LLM-MAS) use specialized agents that exchange messages to accomplish complex tasks. Their reliance on networked protocols, remote tools, and model endpoints exposes communication to message loss, delay, and context expiration. Because agents often exchange open-ended n
multi-agentagent system - arxiv:2610.15452 · cs.CVEmoTrace-Bench: Diagnosing Emotional Expression in Text-to-Video Generation through Controlled Cues and IntensityYeheng Wang, Liudan Shi, Yi Zheng, Jingke Li +4
Text-to-video (T2V) models are increasingly used to generate videos with targeted emotional expressions, yet existing evaluation methods mainly rely on holistic video-text alignment and provide limited insight into whether failures arise from target-emotion expression, intensity control, or specific
benchmarkevaluator - arxiv:2610.15449 · cs.AIHidden in Plain Sight: Advancing Harmful Content Understanding via Perception-Aware Adversarial Self-PlayYisong Xiao, Benlei Cui, Shiji Zhao, Zonghao Ying +7
Despite significant advances in language understanding, Large Language Models (LLMs) can still miss meaning hidden in plain sight, failing to identify harmful intent in expressions readily understood by humans. Malicious users exploit this gap by strategically disguising harmful content to evade aut
self-play - arxiv:2610.15445 · cs.CVRethinking Supervision in Document Unwarping: A Self-consistent Flow-free ApproachShaokai Liu, Hao Feng, Wengang Zhou
In recent years, the proliferation of smartphones has led to an upsurge in the digitization of document files via these portable devices. However, images captured by smartphones often suffer from distortions, thereby negatively affecting digital preservation and downstream applications. To address t
benchmark - arxiv:2610.15431 · cs.ROHarnessing SPAD Noise: Test-Time Gated Attention for Indoor Depth EstimationYiquan Gao
Monocular depth estimation (MDE) has emerged as a cost-effective alternative to hardware-dependent 3D sensing modalities, yet predicting precise metric depth remains a persistent challenge due to intrinsic scale ambiguity, edge bleeding, and limited zero-shot generalization in unseen environments. R
benchmark - arxiv:2610.15429 · cs.LGWorld Modeling through Spectral AlignmentHolger Molin, William Peng, Marco Bagatella, Randall Balestriero
How do we tell a world model what information to preserve? We study this question through the lens of spectral representation learning. We introduce SpecWM, a JEPA framework that explicitly specifies desired relationships between observations through a target similarity kernel and trains the encoder
world model - arxiv:2610.15428 · cs.LGPessimistic Latent Task-aware Optimization for Robust Offline Meta-Reinforcement LearningChang-Hoon Jeong, Kisung Shin, Hanwool Sul, Byoung-Tak Zhang
Context-based offline meta-reinforcement learning (COMRL) has advanced almost entirely through better task representations, which separate task identity from artifacts of the behavior policy so that the latent reflects the task alone. The implicit expectation has been that a clean representation wil
benchmark - arxiv:2610.15425 · cs.ROExploring the Limits of Combinatorial Generalization in Embodied ManipulationXupeng Zhang, Yantai Yang, Chang Guo, Zhaokai Yin +1
In embodied manipulation, Vision Language Action (VLA) models can execute atomic skills, but long-horizon manipulation requires recombining these skills without demonstrations for every skill combination. The full instruction route treats skill chaining as long-horizon imitation. In a controlled stu
vision language actionembodiedmanipulationpost-trainingbenchmark - arxiv:2610.15422 · cs.ROA Reinforcement Learning Framework for Safety-Aware Haptic Shared Control in Surgical RoboticsZihao Zhang, Karl Handwerker, Balint Varga
Haptic shared control in surgical robotics combines human-side and robot-side forces at a shared end effector while preserving operator involvement. However, the effects of geometric safety choices on task execution and force coordination remain unclear. We model the interaction as a two-agent coope
multi-agentbenchmark - arxiv:2610.15415 · cs.LGSensitivity or Criterion? Measuring and Moving When Language Models ActSixun Dong, Yebowen Hu, Ming Yin, Yiran Chen +2
Before an agent acts, it must decide whether action is warranted. A single accuracy cannot distinguish failure to recognize a warranted action from reluctance to act. Signal-detection theory separates sensitivity from criterion, but criteria estimated from authored binary labels depend on each bench
agenttool usebenchmark - arxiv:2610.15409 · cs.AICoverSynth: Coverage-Guided Generation of Retrieval Evaluation SetsAnahita Sharma, Apaar Shanker, Rishav Chakravarti, Sneha Srinivasan +5
Enterprise agents need retrieval evaluation sets that expose domain-specific failures and guide developers towards efficient deployment tradeoffs. Neither expert-human-authored sets nor current synthetic generation methods provide a methodology for achieving the necessary coverage, reliability, effi
agent - arxiv:2610.15407 · cs.AIState Representation Shapes Long-Horizon Agent Learning and Transfer: An Empirical StudyDong Nie, Rui Lan, Yixin Nie, Lingzi Hong
An end-to-end success rate alone cannot distinguish failure to learn an action from failure to apply it as task state changes. We study this distinction in deterministic tool workflows that provide the current artifact state and candidate action recipes. Using 1.7B, 8B, and 27B Qwen models, we compa
agent - arxiv:2610.15401 · cs.ROHarnessing VLM in Embodied RL: From Reward Regressor to Success Detector via Exploration-Augmented Key-Observation GatingXin Liu, Yutong Jiang, Jian Su, Ziheng Wang +3
VLMs provide promising solutions to embodied reward design. Most current methods that train VLMs as reward regressors not only require extensive annotations but are also primarily designed for finetuning pretrained policies, and perform less well in the more challenging from-scratch RL setting. This
embodied - arxiv:2610.15400 · physics.opticsComparative Robustness of Mach-Zehender and Microring Photonic Matrix Multipliers Under Laser RIN and Thermal DriftAndreas Falkenberg
Photonic matrix multipliers can accelerate neural-network inference, but their practical utility depends on how optical weight elements respond to source noise, thermal variation, and calibration error. This paper compares Mach-Zehnder modulator (MZM) and microring-resonator (MRR) implementations un
mach-zehndermicroring - arxiv:2610.15398 · cs.LGGA-SVD: Generation-aware Singular Value Decomposition for LLM CompressionShengkun Tang, Cong Zeng, Zongfang Liu, Dan Alistarh +1
Large language models (LLMs) demonstrate strong generalization across a broad range of reasoning and generation tasks, but their large parameter counts impose substantial memory and computation costs, hindering deployment in practical resource-constrained settings. One-shot low-rank compression is a
memorybenchmark - arxiv:2610.15394 · cs.LGWUSPruner: Second-Moment-Optimal Transforms for Structured PruningShengkun Tang, Dan Alistarh, Zhiqiang Shen
Structured pruning is typically formulated as identifying which model dimensions to remove, often within the model's original representation. However, the effect of coordinate deletion fundamentally depends on the basis in which pruning is performed. This raises a more fundamental question: how shou
memorybenchmark - arxiv:2610.15390 · cs.LGFrom Acoustic Sensing to Physical Actions: A Comprehensive Survey of Security Risks in Edge-Deployed Audio LLM-Based AgentsBinhong Wang, Haidong Kang, Yanfeng Zhang, Hong-Ning Dai +2
Audio Large Language Models (Audio LLMs) are embedded in agents for smart homes, vehicles, wearables, and embodied systems, where acoustic inputs can propagate from perception into persistent state and consequential execution. This setting extends security analysis beyond model behavior to the syste
embodiedpersistent stateagent - arxiv:2610.15389 · cs.CVVideoEvoHarness: Self Evolving Agent Harness from Unlabeled Videos for Video UnderstandingWeixin Xu, Yifan Liu, Yi Wang, Mingfei Rong +4
Video understanding agents rely on a harness to acquire evidence, use tools, and reason, but improving it requires substantial effort. Recent self evolution methods seek to optimize harnesses through execution feedback to reduce human intervention. However, existing methods often rely on annotated v
memoryagentbenchmark - arxiv:2610.15384 · cs.LGE8 Lattice Key Quantization for Low-Bit KV-Cache CompressionSasho Nedelkoski, Alexander Acker, Soeren Becker, Dominik Scheinert +1
Compressing the key-value (KV) cache is essential for serving long-context large language models (LLMs) under memory constraints. While current scalar quantization methods achieve significant compression, they typically face a quality-fidelity trade-off, either suffering from performance instability
memorylong-context - arxiv:2610.15375 · cs.ROHave You Seen It Clearly? Uncertainty-Guided Belief Refinement for Zero-Shot Object NavigationJingzhi Huang, Junkai Huang, Yuhang Peng, Keyu Mu +5
Zero-Shot Object Navigation (ZSON) requires an agent to locate target objects in unfamiliar environments without additional training for the test categories or environments. Recent approaches leverage large language models and vision-language models to construct spatial representations and guide exp
quadrupedsim-to-realmemoryagent - arxiv:2610.15373 · cs.AIIn-Place Indic Tokenizer Expansion and Continual Pretraining of Nemotron-3-Nano-30B-A3BAvinash Benki, Prasanjith Pasunuti, Sutanshu Saksena Raj, Parijat Chatterjee +2
We adapted Nemotron-3-Nano-30B-A3B for eight Indian languages without changing its backbone architecture. We expanded the vocabulary in place from 131,072 to 163,072 entries using continued byte-pair encoding and AdaptMix rebalancing, reducing tokens per word by 38.6%. We initialized new input and o
long-contextbenchmark - arxiv:2610.15372 · cs.AIFewer Bits, More Tries: Recovering Capability Under Memory ConstraintsTomer Galanti, Josh Susskind, Eran Malach
Large language models are generally more capable than smaller ones, but their memory requirements can make local deployment difficult. We ask whether part of this capability gap can instead be recovered by running a smaller or more compressed model multiple times and selecting among its outputs usin
memorybenchmark - arxiv:2610.15371 · cs.AIA Sovereign, Indic-First, Reasoning ModelSutanshu Saksena Raj, Prasanjith Pasunuti, Parijat Chatterjee, Daksh Maan +3
Gnani-Evon-v3.3-30B-A3B is a long-context hybrid Mamba-Transformer mixture-of-experts model for reasoning, instruction following, and tool use in English and eight Indian languages. Starting from NVIDIA's Nemotron-3-Nano-30B-A3B checkpoint, it retains the 52-layer backbone and changes the vocabulary
long-contexttool usepost-trainingbenchmark - arxiv:2610.15368 · cs.LGANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified BeliefsYi Ge, Kangrui Ruan, Jifan Chen, Manish Prajapat +5
Rapid advances in large language models have made software engineering agents capable of tackling complex coding tasks. Within a containerized environment, a coding agent receives tool responses or execution results, and determines its later actions by those earlier results. However, most existing c
world modelagentagentic - arxiv:2610.15365 · cs.LGTRACE: Temporal Risk Assessment and Concept Editing for Safer Language GenerationAnkur Garg, Xuemin Yu, Hassan Sajjad, Samira Ebrahimi Kahou
Large language models (LLMs) can produce harmful content while generating a response. A useful safeguard should not only detect this risk during generation but also intervene while leaving safe responses largely unchanged and provide an auditable record of the internal patterns supporting its decisi
benchmark - arxiv:2610.15362 · cs.ROLERR-Nav: Learning to Coordinate Exploration, Recovery, and Reminiscing for Object NavigationJingzhi Huang, Junkai Huang, Yuhang Peng, Keyu Mu +5
Object Navigation in unseen environments remains a fundamental challenge in embodied AI, as it requires agents to generalize visual and spatial reasoning to novel layouts. While recent advances in large language model (LLM) and vision-language model (VLM) have enhanced semantic mapping, spatial memo
embodiedmemoryagent - arxiv:2610.15356 · cs.LGJSIR: An Audio-Aligned, Multi-Task Dataset of Jazz Standard RecordingsSihun Lee, Dasaem Jeong
Beat tracking, chord estimation, and structure analysis are staple tasks of music information retrieval. Their datasets are drawn largely from Western popular music, and jazz remains a repertoire on which state-of-the-art systems underperform. We present JSIR, the Jazz Standards dataset for Informat
tool use - arxiv:2610.15345 · cs.LGCodeWalk: Generating Coding Benchmarks by Walking a Problem GraphJiaxin Fang, Runyuan He, Arnav Gupta, Sahil Kumawat +2
High-quality coding problems are expensive to collect. Benchmarks such as SWE-bench rely on substantial human curation, while synthetic alternatives typically rely on costly model generation and validation. In this work, we show that new coding problems can be created at almost no cost by composing
benchmark - arxiv:2610.15343 · cs.CVAcross Viewpoints and Time: Trajectory-Generalizable Physical Attacks on World Models and Video Understanding ModelsSong Xia, Meiwen Ding, Chenqi Kong, Wenhan Yang +1
World models and video understanding models are increasingly deployed in dynamic visual environments, yet their robustness to physical adversarial threats remains largely underexplored. We study whether a single physical adversarial texture can mislead the victim model across different real-world sc
world modelbenchmark - arxiv:2610.15337 · cs.ROCoLA: Multi-VLA Coordination via Learned AdaptersAhaskar Kashyap, Jiawen Zong, Boyu Ma, Chuhao Zhou +3
Vision-Language-Action (VLA) models are typically deployed as isolated single-agent policies. When several such agents share a workspace, they can coordinate implicitly, by observing one another, or explicitly, by exchanging messages, but it remains unclear when explicit communication becomes necess
vision-language-actionvlaaction headagent - arxiv:2610.15336 · cs.LGZOIF: Zeroth-Order Influence Function Estimation for LLM Data Selection Using Only Forward PassesLisong He, Yunxuan Jiang, Haishan Ye, Xiangyu Chang
Influence functions enable principled data selection for large language models (e.g., identifying influential training samples for model editing, debugging, or efficient fine-tuning). However, scaling IFs to LLMs is hindered by the intractability of billion-scale Hessian matrices and the prohibitive
memory - arxiv:2610.15330 · cs.CVBaT-SAM: Introducing Inductive Bias into SAM via a Mixture of Convolutional Experts for Brain Tumor SegmentationSumiaya Jannat, Md Mahedi Hasan, Md Shamimul Islam, Md Nur Hossain
Accurate brain tumor segmentation from magnetic resonance imaging (MRI) is essential for diagnosis and treatment planning; however, manual delineation is labor-intensive and time-consuming. Existing approaches are either task-specific with limited generalization or rely on user-provided prompts duri
benchmark - arxiv:2610.15328 · cs.LGAnchor-and-Harvest Curriculum: A Fully Unmasked Non-LLM Curriculum for CraftaxJiayu Zhu, Luyuan Li, Yuyuan Shi, Hanyi Dong
We introduce Anchor-and-Harvest Curriculum (AHC), a domain-informed, rule-based training system for Craftax-Full. AHC combines achievement-oriented tasks, harvested initial-state progress, difficulty adaptation, bounded selection, and original-task practice. It uses no LLM inference and keeps all 43
evaluation protocol - arxiv:2610.15325 · eess.SYAdaptive Event-Triggered Model Predictive Coverage Control with Average Update-Rate RegulationPengbo Zhu, Jicheng Shi, Colin N. Jones, Andreas A. Malikopoulos
Model predictive coverage control for multi-agent systems operates under tight communication and computation budgets, which requires it to balance the load of optimization against responsiveness to varying spatial priorities. With a fixed event threshold, the update rate varies with operating condit
multi-agentagent systembenchmark - arxiv:2610.15324 · cs.CVTempoKV: Predictive KV Cache Quantization for Autoregressive Video GenerationJiahui Hu, Weile Mo, Peiliang Cai, Jiacheng Liu +4
Autoregressive (AR) video diffusion models enable streaming generation of long videos but suffer from significant memory overhead due to the rapidly growing KV cache. Existing KV cache quantization methods are primarily designed for LLMs and transfer poorly to video models, where activations exhibit
memory - arxiv:2610.15308 · cs.ROAn Open MuJoCo Simulation Framework for a Quadrotor Carrying a Tendon-Driven Soft Continuum ArmHaitham El-Hussieny
Combining aerial robots with soft continuum manipulators promises safe, compliant interaction with the environment, but the coupled dynamics of a free-flying base and a highly deformable, underactuated arm are difficult to study without a suitable simulator. This paper describes an open-source simul
manipulator - arxiv:2610.15305 · cs.RORM-WAM: Reliable Memory-Augmented World Action Models for Long-Horizon Robot ManipulationCui Miao, Junhao Yu, Ping Liu, Ning Yang +4
World Action Models (WAMs) improve robotic manipulation by jointly predicting future visual states and actions, yet most existing methods make decisions primarily from the current observation. This limits their reliability in long-horizon tasks, where visually similar states may correspond to differ
manipulationmemory - arxiv:2610.15298 · cs.AICHEAR: Confusion-Hierarchy Evidence-Aware Reasoning for Cost-Efficient Human Activity RecognitionXiangqing Lu, Tao Zhu, Mingxing Nie, Liming Chen
Large language models (LLMs) offer a new approach to wearable-sensor-based human activity recognition. Retrieval-augmented generation (RAG) further introduces similar motion examples and discriminative sensor knowledge, enabling decisions grounded in observations relevant to the current query. More
retrieval-augmentedllm agent - arxiv:2610.15297 · cs.CVHGA-Quant: Heterogeneity-Aware Quantization for AutoRegressive Visual Generation ModelsXinmeng Liu, Xianglong Yan, Kaicheng Yang, Weisheng Dong +1
Recently, AutoRegressive Visual Generation (ARVG) models have emerged as a competitive paradigm for high-quality image synthesis. However, their substantial computational and memory demands make them difficult to deploy on resource-constrained edge devices. Post-Training Quantization (PTQ) provides
memorypost-training - arxiv:2610.15294 · cs.ROProWorld: Multi-View World Model with Proprioceptive Feedback for Closed-Loop Policy EvaluationZijun Wang, Tianyi Yan, Haokun Lin, Panwen Hu +11
Evaluating robot manipulation policies through repeated physical trials is costly. We study whether an action-conditioned world model can serve as a closed-loop test environment whose policy success rates track those on real robots. Beyond realistic video, the model must return the camera views and
manipulationgraspworld modelaction-conditionedpolicy evaluation - arxiv:2610.15292 · cs.AIOntoAR-HAR: Ontology-Guided Agentic RAG with Personalized Memory for Human Activity RecognitionXiangqing Lu, Tao Zhu, Liang Yu, Shuangjian Li +5
LLMs increasingly support sensor-based human activity recognition (HAR) by linking motion signals with activity semantics and physical knowledge. Retrieval-augmented generation (RAG) grounds decisions in sensor examples and domain features, enabling training-free target-user inference with less depe
memoryretrieval-augmentedragagentic - arxiv:2610.15289 · eess.SYPolicy Improvement From Private Information For Decentralized LQG TeamsHung T. Nguyen, Andreas A. Malikopoulos
Sharing of information in multiagent systems allow for improved coordination. Delays in communication can cause asymmetries in information, causing classical results to fail. We study the decentralized LQG problem with delayed information. In general, the optimal policies are not linear and are diff
agent system - arxiv:2610.15287 · cs.ROASM-6D: Fast Certified Pose-Shape Inference with Complete Correspondence Flow MapsHo Jin Choi, Minku Kim, Nadia Figueroa
Category-level 6D pose and shape estimation requires recovering an object's pose and geometry from partial observations without an exact instance model. We present ASM-6D, which couples generative complete-geometry prediction with explicit category-level pose-shape inference through shared correspon
graspbenchmark - arxiv:2610.15283 · cs.LGWhat a Data Release Must Preserve for Off-Policy EvaluationKyungbok Lee, Yating Zou, Michael R. Kosorok
In off-policy evaluation, merging decision-time histories that matter differently for a specified policy value or comparison can make that evaluation unidentified, even with observed actions and outcomes and known behavior and target policies. We characterize which distinctions must be preserved: id
policy evaluation - arxiv:2610.15282 · cs.AINorth or South? Diagnosing LLMs' Understanding of North Korean Language, Society, and KnowledgeKiwoong Park, Seyoung Song, Donghyun Woo, Alice Oh
Large Language Models (LLMs) have made substantial progress in multilingual question answering, yet their ability to understand low-resource sociolinguistic contexts remains under-explored. We study North Korea as a unique testbed where decades of division have produced distinct differences from Sou
benchmark - arxiv:2610.15274 · cs.ROEmbodiment-Aware Inference-Time Search for Human Reaction Selection in Humanoid RetargetingYongpeng Cao, Masahiro Hirano, Yuji Yamakawa
Training humanoid robots to learn human behaviors requires expensive data collection, and human-robot interaction data can be even more difficult and costly to collect. Training a specialized reaction model for each humanoid embodiment can also be time-consuming and resource-intensive. Human-to-huma
humanoid - arxiv:2610.15267 · cs.ROLeHami: Symplectic Latent Dynamics for World ModelsRomina Yalovetzky, Randall Balestriero, Yann LeCun
Joint-embedding predictive architectures (JEPAs) learn world models directly in latent space, but unconstrained autoregressive predictors may contract the latent distribution over long rollouts. We introduce Symplectic Hamiltonian JEPA (LeHami), which replaces the standard predictor with an action-c
world modelaction-conditionedlatent dynamics - arxiv:2610.15259 · cs.AIRECAP: Calibrating Low-Bit Reasoning Models on the History They ReadChi Ruan, Yujie Chen, Zixiang Zhao, Jingcai Guo +1
Low-bit quantization reduces the cost of serving reasoning models but can substantially degrade long chains of thought. Each step reads cached keys and values spanning thousands of tokens, allowing quantization errors to accumulate along the trajectory. Standard calibration operates on short segment
post-training - arxiv:2610.15258 · cs.LGSequent-Prover: Training Agents for Formal, Checkable SMT-based ReasoningDebargha Ganguly, Weicong Chen, Mohsen Hariri, Vikash Singh +12
We introduce Sequent-Prover, a family of verifier-using language-model agents, and the post-training framework that produces them. Agents reason in an executable SMT-LIB scratchpad: they write complete SMT-LIB programs, query Z3 across turns, inspect solver feedback, repair formalizations, and termi
post-trainingbenchmark - arxiv:2610.15256 · cs.MADo Rollups Self-Organize? Multi-Agent Reinforcement Learning in the EIP-4844 Blob Fee MarketAlessio Brini
EIP-4844 gave Ethereum rollups a separate fee dimension for data, governed by a multiplicative update rule that targets three ``blobs'' per block. Whether this target is reached depends on how many independent, self-interested rollups choose to post data, a question that closed-form models address o
agentmulti-agent - arxiv:2610.15250 · cs.LGCorrelation-Aware Reconstruction for Multivariate Time Series Anomaly DetectionMd. Johir Raihan, Zahra Zamanzadeh Darban, Yueyang Liu, Flora D. Salim +2
Reconstruction-based detectors for multivariate time series anomaly detection typically learn a single entangled representation of normal behavior and flag anomalies based on reconstruction error. However, real-world anomalies are not only isolated spikes in a single channel but also disruptions in
benchmark - arxiv:2610.15246 · cs.LGObsSpec: Accelerating Agentic Reinforcement Learning with Observation SpeculationKyle Montgomery, Wang Zhang, Chenguang Wang
Tool-using agents alternate between model generation and tool execution, waiting for each tool call to complete before starting the next generation. This sequential execution is costly during agentic reinforcement learning (RL), which requires large batches of multi-turn rollouts. We introduce ObsSp
world modelagentagentic - arxiv:2610.15244 · cs.LGEvolvAttack: Backdooring self Evolving VLM AgentsChuan Zou, Yang Xiang, Zexin Li, Xiaonan Luo +1
Self-evolving mechanisms let vision-language model (VLM) agents continually adapt through evolving memory, but they also allow malicious behavior to emerge only after deployment. We introduce a new attack surface, the self-evolving backdoor, whose payload stays dormant in the initial agent and is pr
memoryagentself-evolving - arxiv:2610.15239 · cs.AIClinEdge: An Expert-Graded Benchmark of Real Patient Edge Cases Where Frontier Multimodal Models FailAnurag Sharma, Aditya Sharma, Sapna Baalwan
Most benchmarks for large language models in medicine are built from licensing examinations and theoretical questions. The few that draw on real clinical data do so only in part, and rarely require a model to read a patient's own imaging. ClinEdge addresses this gap with 32 real, de-identified patie
benchmark - arxiv:2610.15235 · cs.AIMoMaTune: Model, Match, Then Tune Language Models for Numerical RecordsWeimin Lyu, Jingfen Zhang, Lingdao Sha, Peng sun +3
Despite remarkable progress in language understanding and generation, decoder language models continue to exhibit striking failures on numerical data. Numerical examples are typically serialized as text by concatenating feature names with their values (e.g., fat:3.4), while conventional task adaptat
post-training - arxiv:2610.15233 · physics.opticsMonogamy of quantum steering in exciton--cavity optomechanical systemHamza Harraf, Mohamed Amazioug, Rachid Ahl Laamara
In this work, we study a theoretical scheme to investigate the monogamy of quantum steering in a hybrid exciton--cavity optomechanical (EOM) system. Within the Gaussian-state formalism, we quantify the steerability among the exciton, photon, and mechanical modes in both the steady-state and dynamica
manipulation - arxiv:2610.15228 · cs.AIWords to World: Benchmarking LLMs in Interactive Physical EnvironmentsXinnan Dai, Kaiqi Yang, Yaxin Li, Cheng Luo +8
Words are increasingly becoming an interface for action: large language models (LLMs) can now invoke tools, operate devices, and make decisions that affect the physical world. However, current benchmarks largely evaluate LLMs through static text responses, offline tool calls, or simulated tasks, lea
benchmark - arxiv:2610.15224 · cs.ROPersistent LLM Orchestration of Heterogeneous Robot Teams via Partial Delegation to Multi-Agent ProtocolsMaria G. Mendoza, Victoria M. Tuck, Zachary Ravichandran, Shankar Sastry +1
Natural language specifications let non-expert users task robot teams, and large language models (LLMs) have recently proven effective at grounding such semantic specifications into plans for heterogeneous teams. Existing approaches assume missions that terminate and environments that, while unknown
multi-agent - arxiv:2610.15221 · cs.AIHarnessNet: Feedback-driven Hierarchical Joint Harness Evolution for Multi-agent SystemsZhaoyuan Qiu, Sachith Seneviratne, Yu Xia, Saman Halgamuge
AI agents can evolve by using environment feedback to revise their harnesses, the software managing model inputs, tool use, memory, and execution, while keeping foundation models frozen. Existing studies focus on single agents, leaving joint harness evolution in multi-agent systems (MAS) underexplor
agentai agentmulti-agentagent systemtool use - arxiv:2610.15215 · cs.ROnanoflann: A Header-Only KD-Tree Library for Exact Nearest-Neighbor Search on Compile-Time Product ManifoldsJosé Luis Blanco-Claraco
Nearest-neighbor (NN) search underlies countless robotics algorithms, but many state spaces are not Euclidean: orientations live on $SO(2)$ or $SO(3)$ and poses on $SE(2)$ or $SE(3)$, and a Euclidean KD-tree on their raw coordinates returns wrong neighbors. We present a header-only, dependency-free
memory - arxiv:2610.15208 · cs.AIPANINI-DR: Controlled Deep Research through Generative Semantic WorkspacesM. Yigit Turali, Shreyas Rajesh, M. Efe Lorasdagi, Chenda Duan +2
Deep research requires connecting multiple steps of search, reasoning, and computation, with intermediate findings guiding what to investigate next. As investigations grow, agents must maintain research state across model calls and check findings before errors propagate into subsequent work. We intr
memorypersistent stateagentmulti-agentbenchmark - arxiv:2610.15205 · cs.LGMoSAIC: Learning DAG Scheduling Heuristics from Optimal Schedules through Motif-Aware Recovery and LLM-Guided ReasoningShiva Ahir, Hashmath Shaik, Alex Doboli, Nadeem Yaseen +1
Efficient task scheduling in heterogeneous computing systems remains a fundamental challenge in embedded and high-level synthesis flows. This paper presents MoSAIC (Motif-aware Scheduling with AI-driven Cognition), a learning-based framework that bridges the gap between exact and heuristic schedulin
benchmark - arxiv:2610.15204 · cs.LGSocraticRepair: Planning-Guided Socratic Tutoring for Program Debugging and Misconception RepairMykhailo Kopytskyi, Jianhua Yang, Mir Seyedebrahimi
Large language models are increasingly used in computer science education. However, the teaching abilities of even the most powerful LLMs are limited: they tend to reveal the solution prematurely and may lose focus on the learner's current debugging objective. To address this, we introduce SocraticR
benchmark - arxiv:2610.15191 · cs.AIContinuous Risk-Based Scheduling for Food Safety Inspection: Integrating Historical Decay with Event-Driven Consumer Review SignalsAkriti Bhat, Wayne Zhu, Ishan Gupta
Batch risk-based inspection scheduling struggles to detect a critical failure class: food facilities with recent clean inspections that subsequently develop severe violations. Traditional systems recompute risk on fixed planning cycles using static historical data, leaving blind spots between inspec
benchmark - arxiv:2610.15186 · cs.LGSAFE-KV: Causal Suffix Repair for Behavior-Preserving Non-Prefix KV Cache Reuse in RAGXingjian Zhang
Retrieval-augmented generation (RAG) reuses the same documents across many requests, making non-prefix KV-cache reuse attractive. Yet a cached document state is not reusable text: it is a context-conditioned causal state whose keys and values depend on the prefix, position, and preceding documents u
retrieval-augmentedrag - arxiv:2610.15185 · cs.LGDeltaNet-PC: Test-Time Memory Without Backpropagation Through TimeTodd Morrill, Richard Zemel
Test-time memorization of sequential information is a foundational capability for any intelligent system that needs to rapidly store and retrieve data, as demonstrated by the attention mechanism in large language models. However, Transformers and linear attention variants are trained via backpropaga
memory - arxiv:2610.15179 · cs.AIUDHRBench: Benchmarking Human Rights in Large Language ModelsAlak Raad, Noreldin Morssy, Zeerak Suhail, Serin Atiani +1
Large language models are increasingly deployed in contexts requiring adherence to ethical and legal norms, yet their alignment with internationally recognized human rights standards remains underexplored. We introduce UDHRBench, a large-scale benchmark for operationalising the evaluation of large l
benchmarkjudge model - arxiv:2610.15177 · cs.LGOnly What You've Signed up For: Secure Workflows with Constrained Diffusion AgentsMichael Cardei, Anjun Hu, Chenrui Ma, Francesco Pinto +3
As agentic AI systems take on increasingly complex and consequential tasks, securing them against prompt injection becomes critical. Existing defenses often rely on detecting or reasoning about injected malicious instructions and can therefore remain vulnerable to adaptive attacks that optimize the
agentic - arxiv:2610.15166 · cs.CVExeWorldBench: Benchmarking Code-Based Executable 3D World ModelingXinge Peng, Zhen Fang, Ruibing Wang, Yongxuan Lv +7
Coding agents are increasingly evolving beyond reproducing what a world looks like toward constructing executable worlds that capture how a world works. This shift calls for executable world modeling, which integrates two core capabilities: world understanding, reasoning about the underlying structu
world modelbenchmark - arxiv:2610.15163 · cs.LGSanzaru: Benchmarking Protocol Formation under Epistemic & Communication AsymmetryYiping Li, Zhiyu An, Wan Du
A language model's ability to solve a task individually does not ensure that a team can solve it under unequal access to information and communication. SANZARU examines this gap through cooperative bomb-defusal puzzles adapted from BOMBANANA! (Lefto Studio, 2026). Task information and action authori
benchmark - arxiv:2610.15157 · cs.RORootSLAM: Square Root Bundle Adjustment for Visual-Inertial SLAMTomás Santucci, Mateo de Mayo, Daniel Cremers, Taihú Pire
Efficient perception is a prerequisite for the small, low-power devices targeted by robotics, mixed reality, autonomous driving, and drone applications, all of which rely on Simultaneous Localization and Mapping (SLAM) to estimate their motion within their surroundings. Learning-based SLAM systems a
benchmark - arxiv:2610.15153 · cs.AICan agents whose sole duty is to keep humans alive be hijacked to kill? Delayed multi-turn derived-action trap attacks in simulated high-stakes environmentsJosef A. Habdank
We ask whether a tool-using agent whose sole duty is to keep humans alive can be induced, by delayed prompt injection, to kill. The keep-alive duty is in force from the first watch, with the attack arriving later, appended to otherwise truthful telemetry the operator is required to read. Experiments
agent - arxiv:2610.15150 · cs.AIHarmony Harness: Multi-Objective Self-Evolution of Agent HarnessesWeichen Yu, Lesly Miculicich, Bhavana Dalvi Mishra, Jiefeng Chen +4
Recent work on self-evolving LLM agents treats the agent harness as an optimization target that can be iteratively improved through execution feedback. Existing approaches, however, largely optimize a single objective and evolve the harness as a monolithic artifact, limiting their ability to handle
agentllm agentself-evolvingbenchmark - arxiv:2610.15139 · cs.LGWhen Dynamic Scaling Fails: A Scale-State Feedback Governs Recurrent-State Quantization in Linear-Attention and SSM LLMsYuhang Jiang
Linear-attention and state-space LLMs carry long context in a fixed-size recurrent state, and quantizing that state is the natural way to cut its cost during long decode. Unlike a weight tensor, the state is rewritten at every step, so a quantizer applied at each step sits inside the recurrence and
long context - arxiv:2610.15134 · cs.CLLumen-bench: Cross-Lingual Safety Gaps and the Say-Do Problem in Tool-Calling Large Language ModelsGiovana Kerche Bonás, Thiago Laitz, Thales Sales Almeida, João Guilherme Alves Santos
When a large language model refuses an unethical request, what does refusal mean? In conversation, refusal is bounded by what the model says; once the model is equipped with executable tools, what it does can diverge from what it says, a failure mode that English-only evaluations cannot fully reveal
benchmark - arxiv:2610.15129 · cs.AIGesBench: When and Where Do Multimodal Language Models Perceive Social Meaning?Zonghuan Li, Hayley Hung, Chirag Raman
Human social perception is incremental: we infer intent before an utterance is complete, revise our interpretation as new cues arrive, and attend to specific facial, gestural, or prosodic signals that shape social meaning. Current multimodal LLM benchmarks largely evaluate final-answer understanding
benchmark - arxiv:2610.15128 · cs.LGTo Defer or Not to Defer? Online Learning with LLMs and Congested ExpertsVasileios Saketos, William Reveillard, Alexandre Proutiere, Sebastian Dalleiger +1
Modern AI systems increasingly combine large language models (LLMs) with human experts, using the LLM to handle routine queries while deferring uncertain cases for expert review. When expert capacity is limited, however, deferral decisions must account not only for the reliability of the LLM but als
online learning - arxiv:2610.15127 · cs.LGRIECNN: Real-Time Image Enhanced CNN for Traffic Sign RecognitionReem Abdel-Salam, Rana Mostafa, Ahmed H. Abdel-Gawad
Traffic sign recognition plays a crucial role in the development of autonomous cars to reduce the accident rate and promote road safety. It has been a necessity to address traffic signs that are affected significantly by the environment as well as poor real-time performance for deep-learning state-o
benchmark - arxiv:2610.15117 · cs.LGscHFBiMamba: a bidirectional Mamba autoencoder for cardiac disease-state classification from single-nucleus RNA-seq dataUgo Lomoio, Pietro Hiram Guzzi, Pierangelo Veltri
To evaluate whether a compact bidirectional state-space autoencoder improves joint classification of species, cell type, and disease state from single-nucleus RNA-sequencing profiles in heart failure (HF), while yielding interpretable gene attributions. Building on the dense denoising autoencoder (D
benchmark - arxiv:2610.15112 · cs.CVDrive3R: Block-wise Streaming Geometric Foundation Models for Autonomous DrivingTianyi Yang, Haojun Yang, Jipeng Xiao, Xuesong Chen +3
High-fidelity closed-loop simulation for autonomous driving increasingly relies on dense multi-view 3D reconstruction. While recent feed-forward geometric foundation models have revolutionized per-scene optimization, applying them to continuous, long-horizon driving sequences is computationally intr
memory - arxiv:2610.15107 · cs.AIStructural Chain-of-Thought: Reasoning with Executable Computational GraphsPetr Senatorov, Karim Vafin, Elena Tutubalina, Oleg Somov
Chain-of-thought prompting enables reasoning in large language models but produces explanations that are difficult to verify, expensive to generate, and not always faithful. We introduce a Structural Chain-of-Thought framework where models generate executable computational graphs instead of free-for
benchmark - arxiv:2610.15103 · cs.CLRegExtract: A Benchmark for PFAS Regulatory Information Extraction with LLMsDefne Circi, Krystelle Lionti, Sarathkrishna Swaminathan, Daniel P. Sanders +7
Extracting structured information from regulatory documents is a critical but challenging task. Relevant facts are scattered across long, legally dense texts, and traceability of extracted facts to their source is essential for verification. With increasing awareness of the environmental and health
benchmark - arxiv:2610.15097 · cs.LGAnchors Enhance Controllability and Reasoning of Diffusion Language ModelsFeiyu Qu, Zheng Huang, Andrea Zanette
Diffusion language models (DLMs) offer new opportunities for controllable text generation. However, their controllability, defined as the ability to satisfy task constraints while preserving their original capabilities, remains underexplored. DLMs' bidirectional attention allows control tokens to gu
benchmark - arxiv:2610.15095 · cs.LGSwarm Telepathy: Prefill-Free KV-Cache Transfer Across FamiliesAbdulrahman Alabdulkareem, Irene Tenison, Yang Zhang, Yada Zhu +2
LLM multi-agent systems split work across agents that communicate through text. An agent summarizes its findings, another agent receives the text. This constrains communication to discrete tokens and costs redundant compute, limiting bandwidth and capabilities of multi-agent systems. Recent works de
agentmulti-agentagent systembenchmark - arxiv:2610.15093 · cs.LGTurnCost: Turns, Not Observations: Cache-Aware Loop Engineering for Coding AgentsPramod Kumar, Raj Kumar Pandey
AI coding agents operate in iterative loops: they modify code, run tools, inspect the results, and decide what to do next. We call the design of this process loop engineering, which includes context management, model selection, compaction, and termination. Existing efficiency work mainly compresses
agent - arxiv:2610.15092 · cs.CVVLCQ: Post-training quantization for deep neural networks using variable length codingReem Abdel-Salam, Ahmed H. Abdel-Gawad, Amr G. Wassal
Quantization plays a crucial role in efficiently deploying deep learning models on resource-constrained devices. Post-training quantization does not require access to the original dataset or retraining the full model. Current methods that achieve high performance close to baseline results typically
post-training - arxiv:2610.15091 · cs.AIHave You Been Convinced? A Neuro-symbolic Argumentation Reasoning Approach to Detecting Persuasiveness in ChangeMyView DiscussionsSiddharth Bhargava, Sara Tonelli, Patricia Martín-Rodilla
Argument structures represent how individual arguments interact within a discourse. However, leveraging these structures to infer meaningful knowledge remains challenging, as it requires a formal representation of how argumentative influence propagates across the structure. We propose Neuro-symbolic
retrieval-augmented - arxiv:2610.15084 · cs.AISelf-Evolving Visual Agent via Margin RewardWenxi Yang, Yuzhong Zhao, Baochang Zhang, Yunfan Liu +1
Visual agents that interleave reasoning with tool invocation have posed a promising direction for multimodal tasks. However, learning effective tool invocation usually relies on curated question-answer pairs or expert demonstrations at scale, which are expensive to acquire. In this study, we propose
agentagentictool useself-evolvingbenchmark - arxiv:2610.15081 · cs.AIKnowledge transfer between simple discrete worldsSilvère Gangloff, Jan Hula
Adapting to a new world and reasoning correctly about which outcomes are possible under rules one has not met before is a basic capacity underlying general intelligence. We study it in small discrete worlds: sets of grid configurations consistent with a shared set of rules, concretely Sudoku-like pu
benchmark - arxiv:2610.15079 · cs.CVRegion Aware Squeeze-and-Excitation Networks for Image SegmentationJack Roberts, Praneeth Tummala, Kavi Sarna, Jeova Farias Sales Rocha Neto
Squeeze-and-Excitation (SE) modules improve convolutional networks by modeling channel-wise dependencies, but global average pooling collapses spatial structure into a single per-channel descriptor, limiting their effectiveness for dense prediction tasks such as image segmentation. We generalize SE
benchmark - arxiv:2610.15073 · cs.ROModeling and Control of a Rigid-soft Hybrid-link Humanoid RobotZewen He, Taiki Ishigaki, Ko Yamamoto
Physical flexibility plays a significant role not only in human motions but also in robot motions. For example, a humanoid robot with passive and flexible elements, which can be regarded as a rigid--soft hybrid-link system, has the advantages of storing energy and mitigating impact in a dynamic moti
humanoid - arxiv:2610.15058 · cs.LGThe Optimal Decoder is Tilted TemperatureNikolaos Papagiannis, Ananth Grama, Wojciech Szpankowski
Larger sample budgets benefit from higher sampling temperatures in LLMs, but what temperature tuning optimizes and why it works remain poorly understood. We seek to answer these questions: we pose decoding as pass@$k$ maximization subject to two constraints that encode critical decoding failure mode
benchmark - arxiv:2610.15055 · cs.AITrace Evolution Bench: Measuring Whether Agents Improve from Their Own Environment TracesFangfei Li, Chenyang Zhao, Long Wang, Feng Tian +1
Self-evolving agents are usually scored by endpoint Gain: held-out success after an update minus success before. That difference can credit a frozen rescore as evolution. We introduce Trace Evolution Bench (TEB), a protocol that separates that rescore from a context or weight update, together with t
self-evolving - arxiv:2610.15038 · cs.ROTraining-Free Multi-Agent Cooperative Embodied Search from Uncertain Vision-Language PriorsFan Pu, Keshu Wu, Jiwan Jiang, Shiyu Shen +4
Real-world Vision-Language Navigation (VLN) requires agents to act under partial observability, where map structure, semantic cues, and target locations may be incomplete or incorrectly inferred by foundation models. This paper presents a lightweight, training-free, and uncertainty-aware framework f
embodiedmulti-agent - arxiv:2610.15035 · cs.CVWhat Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training StrategiesDuo Li, Haojun Pan, Zuhao Yang, Xintong Wang +3
Contrastive learning has become the standard paradigm for multimodal representation learning, aligning semantically matched pairs while separating mismatched ones in a shared embedding space. Although various contrastive training strategies have been proposed across negative selection, batch constru
benchmark - arxiv:2610.15032 · cs.LGInCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly DetectionŁukasz Marcjan, Dorota Wilk-Kołodziejczyk, Roberto Corizzo
Industrial anomaly detection benchmarks have advanced rapidly, yet they remain largely static and therefore underrepresent deployment settings in which inspection systems must be extended from one industrial component to another over time. In our setting, continual industrial anomaly detection is fr
benchmark - arxiv:2610.15029 · cs.AIToward System One Clinical Intelligence: Parameter-Efficient Fine-Tuning and Counterfactual Explainability of Decision Foundation ModelsS. M. Mahbub Uddin Ahmed
Automated disease classification from patient-reported symptoms requires balancing the speed of lightweight discriminative models against the contextual understanding and latency costs of autoregressive large language models (LLMs). Decision foundation models offer an alternative by mapping represen
benchmark - arxiv:2610.15020 · cs.LGPATTERN Can Be Solved from Memory: Generator-Aware Baselines for Synthetic Node ClassificationAmin Najafgholizadeh, Mohsen Afsharchi
PATTERN, a synthetic node-classification benchmark still used to rank graph networks, plants a random subgraph in a random host graph and asks which nodes belong to it. Its generator draws 100 patterns once and plants the same ones in every split, so each test graph carries a pattern that also appea
benchmark - arxiv:2610.15012 · cs.CVCowCare: A Unified Visual Framework for Viewpoint-Robust Cattle Tracking and Spatiotemporal Heat-Stress AnalysisXin Liu, Weiran Li, Minghao Zhu, Mina Han +1
Reliable individual monitoring is essential for dairy cattle welfare and early heat stress detection in Precision Livestock Farming (PLF). However, translating raw multi-camera videos into automated diagnostics remains challenging due to disconnected tracking and recognition pipelines and complex ph
benchmark - arxiv:2610.15011 · cs.LGConformal Sequential Decision Making With Cost and Sublinear Policy Regret GuaranteesJayadev Naram, Giuseppe Durisi, Matteo Zecchin, Osvaldo Simeone
Automated decision makers, such as AI agents that control robots or network devices, must often satisfy hard cost constraints while maintaining competitive performance. In such settings, a cost measure, such as a safety violation rate, energy consumption, or a distortion level, must stay within a gi
agentai agent - arxiv:2610.15008 · cs.AICan You Trust Your Preference Models? Stackelberg Games for AlignmentRiccardo Formenti, Aditya Agrawal, Vikas Garg
Language models are frequently aligned to human preferences via imperfect evaluators whose errors can be exploited during optimization. We train policies to search for comparisons on which an evaluator may reverse the underlying preference, providing candidates for additional supervision. Since huma
evaluator - arxiv:2610.15005 · cs.CVRetraining-Free Depth Pruning of Vision Transformers: A Cross Domain Study of Layer RedundancyAlessandro Viespoli, Loris Nanni
Transformer architectures have achieved strong results in natural language processing and computer vision, but their deployment is limited by high computational, memory and storage costs, motivating research into model compression techniques, such as pruning, while preserving performance. Prior work
memory - arxiv:2610.15001 · cs.RODesign and Evaluation of a Collaborative Teleoperation Interface for Multi-Operator Multi-Robot Systems in Challenging EnvironmentsAlperen Kenan, Juan José García Cárdenas, Adriana Tapus, Paul Bremner +1
Multi-robot systems that combine the complementary capabilities of specialised platforms are critical for complex missions in challenging, inaccessible environments. Current deployments assign an operator or operator team to each robot, with collaboration relying on operator-to-operator communicatio
teleoperation - arxiv:2610.14997 · cs.ROBarrier-Constrained Guidance For Real-time Asynchronous VLA ExecutionWenbo Yu, Jifan Li, Wenke Xia, Yuchen Li +4
While asynchronous execution enables VLA models to operate in real-time environments by predicting the next chunk during execution, it inevitably introduces boundary misalignment with the committed action stream. Existing delay-aware solutions address this at the prohibitive cost of extensive retrai
vlavla modelaction head - arxiv:2610.14996 · cs.AICertify the Decision, Not Every Score: Exact Top-k Selection from Few EvaluationsYuhe Sui, Yingzhi Tang
Choosing a few robust algorithms or schedules can require evaluating every candidate under many adverse scenarios, even though an agent only needs to identify the winners. A predictor may suggest those winners, but cannot rule out a surprise in an unevaluated scenario. We introduce Certified Adaptiv
agent - arxiv:2610.14982 · cs.LGFitting Every Hint Is Not Enough: Premature Floods in Recurrent Algorithmic ReasonersAmin Najafgholizadeh, Mohsen Afsharchi
Recurrent algorithmic reasoners are trained with per-step hints on small graphs and run for many more steps on large ones. On breadth-first search in SALSA-CLRS, we find that a reasoner can reproduce every training hint and still fail on deep graphs: after some exact steps, it declares nodes reached
benchmark - arxiv:2610.14977 · cs.ROEnactOpt: Diagnosis-Guided Harness Evolution for Embodied AgentsZhiYuan Feng, Zhuoshi Pan, Zhaolu Kang, Zhongxia Chen +4
Embodied agents built on foundation models rely on an execution harness to organize observations, guide actions, and incorporate execution feedback into subsequent decisions. Failures can arise not only from limited model capabilities, but also from how the harness presents evidence and guides its u
embodiedagentembodied agent - arxiv:2610.14968 · cs.LGSEEK: Agentic Knowledge Graph Retrieval for Secure Code GenerationVarun Gadey, Markus Tohn, Lichao Wu, Ahmad-Reza Sadeghi +1
Large Language Models (LLMs) can generate functionally correct code while introducing exploitable security weaknesses, particularly when developers specify functionality without explicit security requirements. Recent state-of-the-art approaches mitigate this problem using external security knowledge
ragknowledge graphagentagentic - arxiv:2610.14966 · cs.AISkillHorizon: Constructing Long-Horizon Agent Tasks through Skill CompositionYuxuan Liu, Zhaochen Su, Hangyu Guo, Shijue Huang +10
Long-horizon agents must complete workflows in which later activities depend on earlier outputs and changes to shared state. A key challenge for evaluation is to expand task requirements and their dependencies while keeping tasks executable and outcomes verifiable. We introduce SkillHorizon, a bench
agentbenchmarkevaluator - arxiv:2610.14960 · cs.AIHumans vs. Machines: A Comparison With AI Agents in Software Reverse EngineeringSamuele Doria, Stefano De Rosa, Ruoyu Wang, Zion Leonahenahe Basque +2
Can an autonomous AI agent reverse engineer a binary as well as a skilled human analyst? Do they adopt the same strategy, and can agents scale to real-world programs? We compare human and LLM-based software reverse engineering (SRE): nine models tackle two compact static-analysis challenges previous
agentai agent - arxiv:2610.14958 · cs.LGWhat Survives When Activation Steering Is Used as MemoryVedant Jumle
Activation steering has recently been used as a memory for skills: past experiences of an agent are stored as steering vectors and retrieved later. Facts and preferences about the user are assumed to need text instead. We test that assumption with a training-free activation memory beside a frozen LL
memoryagent - arxiv:2610.14956 · cs.AIMCPDisCo: Seesaw-Mitigated Tuning for MCP-Native AgentsKounianhua Du, Jianxing Liu, Shuai Shao, Ke Li +2
The Model Context Protocol provides a shared interaction substrate for the agentic web, allowing heterogeneous services to expose capabilities through a standardized interface. Learning to use this ecosystem entails a hierarchy of capabilities: tool-format skills, server-level transfer, and cross-se
agenticbenchmark - arxiv:2610.14954 · physics.opticsPartial-coherence weak measurement of polarization qubits with a temporal pointer: theory and experimentShane P. Gervais, Lambert Giner
Weak measurement has proven useful for metrology, for probing quantum paradoxes, and for the direct measurement of quantum states. Weak measurements of polarization qubits with temporal pointers provide a direct route to weak-value characterization, including in regimes where the pointer is only par
benchmark - arxiv:2610.14941 · cs.LGCompositional Average Velocity Modeling for Efficient Diffusion Offline Reinforcement LearningThanh Nguyen, Abu Hanif Muhammad Syarubany, Hwanhee Kim, Chang D. Yoo
Diffusion-based policies have recently emerged as a powerful paradigm for offline reinforcement learning, but their reliance on multi-step denoising introduces substantial computational overhead and requires backpropagation through time during training, limiting scalability and practical deployment.
manipulationbenchmark - arxiv:2610.14940 · cs.CVEngram Neuron Inspired Adaptive Memorization for Lifelong LearningSuiyang Jiang, Bomin Wang, Hangqi Zhou, Jianfeng Feng +1
Lifelong learning enables artificial systems to adapt to changing environments. To achieve this, an effective artificial learner should dynamically reorganize past and new experiences, maintaining stability while enabling flexible knowledge integration. However, existing efforts primarily mitigate c
memorylifelong learningbenchmark - arxiv:2610.14939 · cs.AIVC-OPD: Visual Counterfactual On-Policy Distillation for Grounded Vision-Language ReasoningJiahe Feng
GRPO-style reinforcement learning has advanced vision-language reasoning, but its sequence-level rewards provide limited credit assignment for individual reasoning tokens. On-policy distillation (OPD) addresses this issue by providing dense supervision on student-generated trajectories, yet standard
benchmark - arxiv:2610.14936 · cs.ROCFG-Policy: Unified Diffusion Policy Learning Framework via Classifier-free GuidanceShutong Ding, Zhaoyang Liu, Yimiao Zhou, Jingya Wang +2
Diffusion policies are increasingly used in robotic control and manipulation, vision-language-action models, and world-model-based agents. Their ability to represent multimodal action distributions offers rich opportunities for exploration in reinforcement learning (RL). However, optimizing diffusio
vision-language-actionmanipulationdiffusion policyrobotwinpost-training - arxiv:2610.14935 · cs.LGBSIPipe: Bidirectional Sequence-Level Interleaved Pipeline Parallelism for Long-Sequence Large Language Model TrainingYuzhou Ni, Ling Chen
Pipeline parallelism has become a critical technique for training large language models. To address the throughput limitation of existing sequence-level methods that schedule subsequences unidirectionally, we propose bidirectional sequence-level interleaved pipeline parallelism (BSIPipe), which co-s
memory - arxiv:2610.14934 · cs.CLFidelityQ: Preserving Predictive Fidelity in Ultra-Low-Bit Language Model QuantizationYujie Chen, Weiyan Zheng, Yiwu Yao, Tianchi Hu +6
Ultra-low-bit scalar quantization offers a regular arithmetic structure for efficient deployment of large language models, but preserving predictive behavior at extreme precision remains challenging: with only a few reconstruction levels, accurate weight representation must translate into consistent
memory - arxiv:2610.14926 · cs.ROEdge-Native Temporal--Graph Dual-Expert Inference for Dexterous Grasp ClassificationJiajian Zheng, Bin Zhang, Wen Fan, Yanan Liu
Multi-finger hands contact objects at different times, so a global close command can overload early-contact fingers while others are still moving. Dense tactile arrays address this problem but add hardware, calibration, bandwidth, and computation. We present an edge-native pipeline that uses only bu
dexteroustactilegrasp - arxiv:2610.14923 · cs.LGTLA-Verus: Generating and Solving Temporal Proof Obligations with LLM AgentsKseniia Strelbytska, Balázs Szilágyi, Anna Mészáros, Ferenc Huszár
The TLA+ (temporal logic of actions) formal specification toolkit has seen widespread adoption in modeling modern distributed software systems including databases, storage systems, ledgers and coordination protocols. This popularity is despite its two significant limitations for software verificatio
llm agentagentic - arxiv:2610.14921 · cs.LGRepresentation Drift in Continual Learning: LoRA versus N-gram MemoryHao Wu
A language model that keeps learning online must not slowly erode its own base capabilities. We ask which part of a modern LLM should absorb continual updates, by measuring how each choice changes the model's internal representations. On a 48-layer mixture-of-experts model with a native pretrained N
memoryonline learning - arxiv:2610.14918 · cs.AISkillForgeBench: Evaluating Agent Reliability Grounded in Agent Skill SpecificationsZhengxi Li, Junyi Zhou, Qiyuan Zhang, Chuhan Wu +5
Evaluating an agent's ability to strictly satisfy diverse and procedurally complex requirements is essential to determining its practical reliability. Conducting such an evaluation demands agentic tasks that reconcile broad domain coverage with deep procedural complexity. Existing paradigms face a f
agentagenticbenchmark - arxiv:2610.14917 · cs.ROFunnel planning and control for manipulators using neural Signed Configuration Distance FunctionsBernhard Wullt, Christos K. Verginis, Per Mattsson, Mikael Norrlöf +1
Robust motion planning and control for manipulators is the problem of guaranteeing safe execution and collision-free motion under disturbances. The Kinodynamic Motion Planning via Funnel Control (KDF) is a recently proposed planner that comes with robustness guarantees and is particularly attractive
manipulator - arxiv:2610.14915 · cs.AIWorkBorne: Discovering Agentic Tasks from Real-World Artifact TransitionsJunyi Zhou, Yufei Wang, Qiyuan Zhang, Chuhan Wu +6
Developing terminal-based agents for long-horizon execution demands synthetic tasks that tightly couple complex instructions with authentic, interactive environments. However, current automated synthesis suffers from a fundamental asymmetry: instruction-oriented methods fabricate low-fidelity, synth
agentic - arxiv:2610.14913 · cs.LGBridging Risk Approximation Gaps in Model Predictive Task Sampling via In-Context ModelingJiarong Wen, Qi Tao, Kaiyu Zhang, Yun Qu +4
Reinforcement learning (RL) is essential for adaptive decision-making, e.g., robust Meta-RL and domain randomization (DR), and efficient post-training of large language models (LLMs), where tasks are commonly structured as Markov decision processes or prompts. A central bottleneck in all these setti
post-training - arxiv:2610.14911 · cs.ROScalable Multi-Human Cooperative Manipulation via Commander-Executor CoordinationBuzhen Huang, Xingchen Wu, Yuanwang Yang, Chenyi Xie +1
Multi-human cooperative object manipulation is a highly challenging topic in humanoid control. While existing methods can handle simple collaborative carrying tasks, they struggle with role-dependent, long-horizon, and sequential cooperative manipulation. To address this issue, we propose a hierarch
manipulationhumanoidgraspmulti-agent - arxiv:2610.14903 · cs.MAConflict-Aware Autonomous Resource Orchestration for 5G Edge Slicing Using Multi-Agent Systems and LLMsNathan R. Prediger, Rodrigo Moreira, Juliano Araujo Wickboldt, Tereza C. Carvalho +3
5G edge network slicing must reconcile two coupled requirements: understanding high-level operator intent and reallocating scarce computing and bandwidth resources under dynamic contention. Centralized optimization can face computational and responsiveness challenges under rapidly changing edge cont
multi-agentagent system - arxiv:2610.14896 · cs.ROTest-Time Compute for Humanoid Behavior Foundation ModelsWeiji Xie, Junsong Wu, Zibo Zhou, Yueru Chen +4
Humanoid Behavior Foundation Models (BFMs) can execute impressive motions, but downstream applications such as loco-manipulation demand high whole-body positional accuracy. Existing BFMs suffer from a systematic command-execution gap, where executed motions deviate from desired ones. We show that ad
manipulationhumanoidteleoperation - arxiv:2610.14894 · cs.LGPro-t-VAE: Closed-Form Student's t Mixtures for Robust Prototype DiscoveryZhikang Xu, Jiarui Xing, Jian Wang
We present a robust deep generative framework for unsupervised image prototype discovery, built on a heavy-tailed Student's t mixture prior, that bridges statistical robustness and mathematical tractability. Existing generative approaches typically rely on noise-prone Gaussian priors or computationa
benchmark - arxiv:2610.14893 · cs.CLDescription Complexity Bounds for Analytic Approximations of Lipschitz FunctionsArtem Vozmishchev
We study how many bits are needed to write down an elementary analytic expression that uniformly approximates a Lipschitz function. Expressions are finite trees over the basis $\{+,-,\times,÷,\exp\}$ with the variables and rational constants, and their size is $\dc(f)$, the length of the shortest de
evaluator - arxiv:2610.14889 · cs.ROLETS: Low-Overhead External-Torque Sensing for Force-Feedback Bilateral TeleoperationZonghao Guo, Dongchen Guan, Yubo Wang, Peng Yang
Bilateral teleoperation with force feedback is becoming increasingly important, and the quality of external-torque estimation is critical to this process. However, manufacturer-provided disturbance-torque estimates on commercial robotic arms are often inaccurate, while the use of end-effector force/
teleoperation - arxiv:2610.14886 · cs.CLAgent-Guided Evidence Acquisition for Emotion-Cause Pair ExtractionZhuangzhuang Pan, Yan Xia, Chee Seng Chan
Emotion-Cause Pair Extraction (ECPE) requires identifying the causes associated with emotions expressed in a conversation. Accurate extraction can depend on evidence that is not initially available, making it important to decide when further information is worth acquiring. We study agent-guided evid
agent - arxiv:2610.14885 · cs.ROVerify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action ModelsKaneyoshi Hiratsuka, Satoshi Yamamori, Jun Morimoto
Test-time compute (TTC), which scores multiple candidate actions at inference time, is a powerful way to improve the execution performance of vision-language-action (VLA) models, but existing methods keep the policy frozen, so the gain must be recomputed at every execution. We propose Verify, Act, D
vision-language-actionvlalibero - arxiv:2610.14883 · cs.LGMARL-BattleGrounds: A Novel JAX-native Benchmark for Heterogeneous and Competitive Multi-agent RLUlixes Tariq Hawili, Subramanian Ramamoorthy, Yoann Altmann, Ignacio Carlucho
We present MARL-BattleGrounds (MARL-BGs), an easy-to-use JAX-native benchmark for heterogeneous and competitive multi-agent reinforcement learning (MARL), designed to be sample-efficient, computationally efficient, and researcher-centric. MARL benchmarks are typically simple and efficient, like SMAX
multi-agentself-playcurriculum learningbenchmarkleaderboard - arxiv:2610.14879 · cs.CLSocial Bias in Chinese Large Language Models under Social Media ExpressionsMo Zhou
Existing Chinese social-bias benchmarks for large language models largely rely on normalized or template-based expressions, leaving the influence of internet slang, emoji, and sarcasm insufficiently examined. Safety alignment may further compress conventional plausibility ratings toward the lower en
benchmark - arxiv:2610.14878 · physics.opticsUltra-low-loss Ta2O5 waveguide for integrated nonlinear photonicsXiaotong Zhang, Shuxuan Guo, Ruixuan Yi, Zheze Shi +7
Integrated nonlinear photonics demands waveguides that simultaneously offer ultra-low-loss, relatively strong nonlinearity, and tight mode confinement. Tantalum pentoxide (Ta2O5) is attractive for this purpose because it combines a broad transparency window, relatively strong third-order nonlinearit
microring - arxiv:2610.14875 · cs.LGContrastive Linearized Laplace: Loss Geometry, Valid Repairs, and Reliable CoresFrancisco J. Sáez-Maldonado, Luis A. Ortega, Andrés R. Masegosa, Thomas D. Nielsen
While contrastive learning is widely used to learn deterministic representations, it does not inherently provide estimates of uncertainty. Post-hoc uncertainty estimation methods, such as the Linearized Laplace approximation (LLA), offer a way to quantify uncertainty without retraining the encoder.
benchmark - arxiv:2610.14873 · cs.LGDependency Reasoning for Aspect Sentiment Quadruple PredictionShima Esfandyar, Seywan Soleymanbaigi, Fardin Akhlaghian Tab, Fatemeh Daneshfar +1
Aspect Sentiment Quadruple Prediction (ASQP) is fundamentally a relational reasoning problem: identifying aspect-opinion-category-sentiment quadruples requires modeling structured dependencies, not merely sequential co-occurrence. Existing LLM-based approaches incorporate syntactic information indir
retrieval-augmented - arxiv:2610.14871 · cs.CVMORE: Learning Motion-Oriented Registers for Driving World Models and PlanningYingdong Gu, Haoyu Wang, Shaocheng Yan, Zhenjun Zhao +8
By predicting future states, latent world models provide rich self-supervision for end-to-end driving, but often inherit dense features from pretrained visual encoders as their world-state interface. This creates a mismatch: visual pretraining produces information-rich, spatially dense features, but
world modelv-jepamemory - arxiv:2610.14864 · cs.ROPredict in Chunks, Revise in Spans: History-conditioned Action Regeneration for Robotic ManipulationChenyang Wang, Yuntian Wang, Jiaying Tang, Chunzheng Li +2
Action chunking improves temporal coherence in visuomotor imitation learning by jointly predicting sequences of future actions. However, many existing policies regenerate entire action chunks even when errors are localized, potentially replacing otherwise valid predictions. We introduce Contextual M
manipulationaction chunking - arxiv:2610.14861 · cs.LGWorld2TS: Real-World Event Context for Time-Series ForecastingZhiding Liu, Jiqian Yang, Mingyue Cheng, Zhi Li +1
Time-series forecasting plays a vital role in a wide range of real-world applications. Recent approaches have achieved strong performance by modeling temporal correlations and cross-channel dependencies, yet most forecasting benchmarks still represent time series primarily as numerical histories. In
benchmark - arxiv:2610.14860 · cs.LGTerrarium: Growing Personal Assistants via Scalable Synthesis of App WorldsMengzhou Wu, Yuzhe Guo, Jiawen Tao, Xiaokun Yuan +8
Advances in large language models (LLMs) have brought the vision of a general-purpose personal assistant closer to reality. Such assistants could provide a unified natural-language interface to a broad ecosystem of applications. However, real-world applications do not readily support training at sca
agentevaluator - arxiv:2610.14856 · cs.CVCAST: Real-Time Motion Capture for Any Skeleton TopologyPu Pang, Zhongfei Qing, Ling-Hao Chen, Songnan Li +2
This paper studies the problem of recovering motion from monocular video, even when the target skeleton topology differs from the subject observed in the video. Although previous methods have made efforts to resolve this problem, they remain vulnerable to joint-rotation errors that are coupled acros
benchmarkevaluation protocol - arxiv:2610.14855 · cs.AIPEAM: On-device Communication Agent for Personalized, Efficient and Autonomous Mobile Network ConfigurationHaowen Wan, Shangzhuo Xie, Syue-Ju Syue, Xianyue Wu +2
The growing deployment of compact large language models on smartphones is enabling autonomous on-device decision-making, such as emerging mobile GUI agents. However, many mobile decisions, such as network configuration when users switch apps or connectivity conditions change, must satisfy stringent
agent - arxiv:2610.14854 · cs.RODEFT-VLA: Stage-Localized Force Specialization for Contact-Rich Robotic ManipulationZan Huang, Yibo Qiu, Kan Ding, Xuanping Zhang +3
Contact-rich manipulation poses a stage-dependent sensory conflict: vision and proprioception are essential for localization and approach, when force is largely uninformative and potentially distracting, whereas force becomes critical after contact, when irrelevant visual details and correlated robo
vision-language-actionmanipulationgrasppost-training - arxiv:2610.14852 · cs.CVGetting Space Right: Agentic Rewards for Image GenerationXiaofeng Tan, Jiaxin Lin, Haobo Yuan, Tiancheng Shen +4
Reinforcement learning (RL) is promising for spatial text-to-image generation, but existing work largely focuses on policy optimization, assuming accurate rewards and effective online rollouts. We observe in-domain gains with out-of-domain degradation, exposing two limitations: unreliable spatial re
agenticcurriculum learning - arxiv:2610.14848 · cs.ROLearning Dynamic GraspingNga Kiu Kelly Ho, Yilin Liu, Hao Xu, Niloy J. Mitra
Synthesizing believable 4D hand-object grasping sequences is important for character animation, extended reality, and embodied agents. Grasping requires a hand to approach, contact, and lift or manipulate an object coherently. Existing synthesis methods often model only the final static grasp or rel
embodiedgraspembodied agent - arxiv:2610.14846 · cs.CVTowards Privacy-Preserving Thermal Human Perception from Dataset to DeploymentKailai Sun, Yue Cui, Xinwei Wang, Zijie Gao +7
Indoor human perception is the foundation for smart homes and healthcare. However, RGB cameras raise privacy concerns in private indoor environments, while low-resolution thermal cameras provide limited spatial and semantic information. Large-scale indoor thermal datasets and real-world deployments
benchmark - arxiv:2610.14843 · cs.AIPre-Registered Genesis Accounting for Recursive Self-Improvement: First Audited Operator-Genesis Events in a Physics-Discovery RuntimeWenbin Mao, Pei Zou
Claims of recursive self-improvement are easy to make and hard to falsify: did improvement come from re-using designer-provided mechanisms, or from mechanisms the system authored itself? We introduce a pre-registered, two-tier accounting: M1 counts machine-generated content instances within a design
self-improvement - arxiv:2610.14840 · cs.ROPush the Limit: Learning High-Speed Sprinting and Aggressive Cornering on a Full-Scale HumanoidHuaxing Huang, Ruidong Li, Jinchao Han, Ziyang Dong +1
We present a learning-based framework that enables a full-scale bipedal humanoid to sprint at over 7 m/s with motion-prior-guided gait and perform aggressive cornering at high speed with a pronounced trunk lean of up to $25^\circ$. Achieving this combination is challenging: above approximately 4 m/s
humanoid - arxiv:2610.14835 · cs.LGSequence-Structured Decision Learning for Branch-and-Bound in Mixed-Integer Linear ProgrammingTengfei Liu, Jinrun Liu, Bai Yan, Yuhui Shi +1
Branch-and-bound (BnB) solves mixed-integer linear programs by constructing and exploring a search tree. Node selection and branching are two recurrent decisions that directly determine the tree. Learning-based approaches typically optimize either decision in isolation and condition on a mostly loca
benchmark - arxiv:2610.14833 · cs.ROScaling Latent Motor Adaptation to Coordinated Bimanual ManipulationSeung-Bin Joo, Alessandro Albini, Luca Wiehe, Perla Maiolino +2
Robust manipulation under uncertain object dynamics and geometry remains a core challenge in real-world robotics. Policies trained under fixed assumptions often degrade when object mass, friction, scale, or shape vary. Embedding-based adaptation methods such as Rapid Motor Adaptation (RMA) address t
manipulationsim-to-realfranka - arxiv:2610.14830 · physics.opticsStackelberg Game with Multi-Agent Transformer for ISRS-Aware Resource Re-optimization in Multi-Band Optical NetworksJiaxin Liu, Rentao Gu, Miao Zhu, Zeguang Zou +2
The rapid expansion of generative AI is driving computing infrastructures toward interconnected AI data centers. Multi-band optical networks have become the backbone of such connectivity. To accommodate increasingly dynamic traffic demands, these networks require continual resource re-optimization t
multi-agentbenchmark - arxiv:2610.14828 · cs.AIEvoCowork: A Benchmark for Real-World Office TasksZhou Liu, Xingyan Chen, ZhaoYang Han, Zewei Pan +5
Office workflows are a demanding testbed for self-improving agents because their deliverables must be correct, natively editable, and mutually consistent. A central question is whether gains on feedback tasks reflect reusable procedures rather than task-specific fixes. We introduce EvoCowork, a 240-
self-improvingself-improvementbenchmark - arxiv:2610.14826 · cs.AICross-Lingual TTS for Indian Languages: A Benchmark and Inference-Time Duration Estimation StrategiesAdarsh Arigala, Arjun Gangwar, Srinivasan Umesh
Recent alignment-free Text-to-Speech (TTS) models achieve high synthesis quality by removing explicit alignment modules and simplifying training. However, they often struggle in cross-lingual scenarios due to differences in character durations across languages, leading to duration mismatch and degra
benchmark - arxiv:2610.14818 · cs.AIA Reference Architecture for Integrating Agentic AI into Enterprise Software: Responsibilities, Interactions, and VariabilityMingqin Yu, Fethi Rabhi, Liming Zhu, Qinghua Lu
Agentic AI moves generative AI applications toward systems that coordinate reasoning, enterprise context, model inference, and external actions. Integrating these capabilities into enterprise software introduces diverse architectural concerns, from orchestration, knowledge and memory, and model acce
agentagentic - arxiv:2610.14813 · cs.LGMachine Learning-Based Prediction of Dengue Infection Using Clinical and Hematological Data from Bangladesh: A Comparative Analysis of Benchmark and Extended Prediction ModelsNeesha Naimeen, Victor Miranda Soberanis
Dengue is a viral infection caused by the dengue virus (DENV) and a major public health concern in Bangladesh, where symptom overlap with other febrile illnesses complicates early diagnosis. This study compares Logistic Regression, Random Forest and XGBoost to predict dengue from clinical and hemato
benchmark - arxiv:2610.14812 · cs.AIDepth-progressive 3D seismic velocity model building via 2D generative diffusion modelsShijun Cheng, Randy Harsuko, Mohammad H. Taufik, Tariq Alkhalifah
Three-dimensional (3D) velocity model building (VMB) remains computationally challenging with conventional approaches and native 3D neural networks, largely due to the prohibitive memory and training cost of full 3D formulations. Therefore, we introduce a depth-progressive framework that achieves 3D
memorybenchmark - arxiv:2610.14810 · cs.LGKedgeFlow: A Programmable Control Plane for Enterprise AI AgentsJun He, Deying Yu
Enterprise AI agents can execute altered versions of approved actions, concurrently reuse the same permission, or retry operations whose responses were lost. These failures arise when approval and resource execution lack a shared contract. KedgeFlow specifies language-independent contracts for trust
agentai agentagent system - arxiv:2610.14807 · cs.AIASIWorld: Benchmarking Autonomous Agents in a Persistent Survival SandboxJianfeng Lin, Runze Wang, Haorong Hong, Yuxuan Xie +2
Evaluating autonomous agents requires examining whether they can translate goals into coherent action across extended interactions. In persistent environments, this requires coordinating interdependent actions under incomplete observations, accounting for delayed effects, and adapting as earlier dec
agentautonomous agentbenchmark - arxiv:2610.14800 · cs.MAEAIR: Interaction Risks in Enterprise LLM Multi-Agent SystemsYuchen Ma, Swadheen Shukla, Paulo Salem, Yue Huang +4
LLM agents increasingly divide enterprise work across roles with different goals, authority, access, and information. Because each agent often sees only a local snapshot of the enterprise workflow, locally reasonable actions can still violate organization-wide constraints (e.g., several service agen
agentllm agentmulti-agentagent system - arxiv:2610.14799 · cs.LGCESNET-TimeSeries26: A Multi-Year Network Traffic Time Series Dataset for Annual Seasonality, Concept Drift, and Long-Term Anomaly DetectionJosef Koumar, Karel Hynek, Kamil Jerabek, Pavel Siska
Forecast-based models are fundamental to network traffic management and anomaly detection. However, it is essential to evaluate these forecasting models over extended periods of time in order to assess their adaptability to unexpected phenomena and shifts in underlying patterns. This kind of long-te
benchmark - arxiv:2610.14790 · cs.AICounterfactual Regret for Reflective Agents with Lean-Verified GuaranteesZhongnian Li, Shi Xu, Mengting Xu, Peng Ying +1
Reflective language agents reuse feedback and memory to adapt across tasks without updating model weights. However, existing guarantees for policy selection and action generation do not by themselves establish when local reflection yields long-term performance guarantees for the complete agent under
memoryagent - arxiv:2610.14789 · cs.CLTraining-Free BioNEL-RAG: Advancing Biomedical Entity Linking through Retrieval-Augmented Generation and Large Language ModelZiyi Chen, Hui Liu, Ziyang Ling, Ronald X. Xu +1
Biomedical Named Entity Linking (BioNEL) is the task of mapping mentions extracted from biomedical texts to canonical entities or concepts in the standardized knowledge base. Although existing methods have achieved remarkable performance in mapping accuracy, most rely on training with domain-specifi
retrieval-augmentedrag - arxiv:2610.14777 · cs.AILYNX: Analytical Program Synthesis via Grounding and Closed-Loop RefinementRuichen Xia, Xiao Han, Dingyu Yang, Luwei Yang +4
LLM-powered data analytics aims to translate natural language instructions into executable analytical programs over underlying data. However, existing approaches, whether treating LLMs as intent-to-code translators or as standalone analytical engines, often fail to reliably align domain semantics wi
multi-agentagent frameworkbenchmark - arxiv:2610.14774 · cs.AIProgAgent: Learning Tool Orchestration through Programmatic TrainingHao Li, Keqing He, Longtao Zheng, Junlong Li +3
Large language models (LLMs) are increasingly expected to orchestrate diverse tools over long-horizon tasks. Existing agents typically use Direct Tool Calling (DTC), where the model invokes tools sequentially, and tool responses accumulate in the model context, which can lead to context rot and orch
agentictool callingbenchmark - arxiv:2610.14772 · cs.AIHow Should a Camera-First AI Agent Work?Xiaofan Wu, Xi Zeng, Miaoxia Chen, Peishan Chen +5
Camera-first agents must act on the visual world as it unfolds. We argue that the central obstacle for future camera-first agents is not recognition ability - multimodal models (vision-language and live/streaming alike) generalize well through language and remain the right tool for semantic reasonin
memoryagentai agentagentic - arxiv:2610.14769 · cs.CLWho Receives Respect? A Counterfactual Benchmark for Honorific Allocation Bias in BengaliAjwad Abrar, Md Kamrul Hasan, Hasan Mahmud
Bengali has no grammatical gender, but honorificity is obligatory: every third-person reference must choose ordinary or respectful register. English has no such signal, so the system must choose a level of deference that the source never specified. We use this property to measure how large language
benchmark - arxiv:2610.14766 · cs.AILast Secure Code Benchmark: Is the Code You Vibe Secure?Tianyi Wu, Chao Ge, Peng Chen, Fengji Zhang +11
AI coding agents now write much of the code that ships, yet whether that code is secure remains unclear. We introduce the Last Secure Code Benchmark, a comprehensive evaluation of secure code generation by coding agents. Built from 150 real vulnerabilities, its 450 tasks span seven widely used langu
benchmark - arxiv:2610.14762 · cs.LGCrossMean: Model Merging by Pairing Merged and Expert ActivationsMoshe Eliasof
Model merging combines fine-tuned experts into one network without retraining, and closed-form methods compute each merged layer from statistics of its input activations. However, these statistics are measured on a single expert, while in the merged model, each layer receives activations from the me
benchmark - arxiv:2610.14760 · cs.CLThe Constitution Speaks Nheengatu: An Open MT System and Reproducible Corpus Pipeline for the Amazonian Língua GeralFabrício Carraro, Rodolfo Joel Zevallos, Rodrigo Gonçalves de Souza, Caio Henrique Faustino da Silva +1
Nheengatu (yrl), the Amazonian Língua Geral, has 6--10 thousand speakers and is co-official in São Gabriel da Cachoeira (Brazil), yet remains poorly served by evaluated Portuguese--Nheengatu MT resources. We present the largest parallel corpus assembled for the language (14.5k Portuguese--Nheengatu
benchmark - arxiv:2610.14757 · cs.LGRethinking Opinion Measurement Interfaces for Human-in-the-Loop OptimizationChangkun Ou, Sven Mayer, Daniel Buschek, Andreas Butz
Human-in-the-loop optimization refers to systems that present choices to a human decision-maker and query their judgment to optimize system parameters iteratively. The user interface typically provides either an absolute rating scale or asks for a preferred option. However, these approaches can pote
human-in-the-loop - arxiv:2610.14754 · cs.AIGateSight: Recurrent Memory Anticipates Attention Memory for Query-Agnostic KV Eviction in Hybrid LLMsJunjie Li, Jiong Lou, Jie Li
Hybrid language models interleave Gated DeltaNet (GDN) layers, whose recurrent memory has a fixed size, with full-attention layers, whose key-value (KV) cache grows with the context and dominates serving memory. In this paper, we ask whether the recurrent memory already knows what the attention memo
memorybenchmark - arxiv:2610.14753 · cs.ROBeyond Fixed Visual History: Learning Motion-Adaptive Temporal Receptive Fields for Complex-Terrain Quadruped LocomotionHao Zhang, Yuping Yan, Yitian Hong, Luying Feng +4
Versatile quadruped locomotion depends on a tight coupling between what the robot perceives and what its body can execute. Yet existing visual controllers typically organize depth history using fixed temporal patterns, overlooking that the same temporal span corresponds to different spatial ranges a
quadrupedmemory - arxiv:2610.14750 · cs.AIEqual-information evaluation of credit assignment in modular financial reinforcement learningGennadiy Chistiakov, Yury Yanovich
Credit assignment in modular financial reinforcement learning is difficult to evaluate because an attributed signal may improve a component loss without adding decision-relevant information or improving economic outcomes. We propose an evaluation framework combining equal-information reconstruction
evaluation framework - arxiv:2610.14746 · cs.ROBridgeWAM: Bridging Video and Action Experts via Latent Bridge QueriesJiaxin Fang, Ruitong Liu, Zeyu Lai, Xingyu Wang +5
World Action Models (WAMs) commonly couple video and action experts through dense, layer-wise Mixture-of-Transformers (MoT) attention, exposing action prediction to irrelevant visual features while enforcing matched expert depths. This coupling hinders selective visual-feature use and introduces red
action headlibero - arxiv:2610.14745 · cs.AISAGE: Pushing 2-Bit Vector-Quantized LLMs toward Real-World UsabilityXianglong Yan, Zenan Wu, Kewei Zhang, Kaicheng Yang +1
Extreme low-bit weight-only quantization is an effective approach to reducing the deployment costs of large language models. Below 3 bits, scalar quantization often suffers severe accuracy loss, while vector quantization (VQ) better preserves accuracy through more expressive joint encoding. However,
tool usebenchmark - arxiv:2610.14738 · cs.MAsGLP: Specifying Stochastic Agents that Simulate PeopleEhud Shapiro
Grassroots platforms are designed to be owned, operated and governed by their participants, and thus are constituted from person-to-person smartphone apps that run on participants' smartphones. A grassroots app is specified using a grassroots social contract, which is compiled through an abstraction
agent - arxiv:2610.14737 · cs.LGPTGPO: Prefix Tree Group Policy OptimizationRuolin Chen, Yang Li, Jihang Wang, Zhuoya Zhao +3
Training language agents on long-horizon interactive tasks is challenging because sparse terminal rewards provide weak credit signals for intermediate actions. Existing group-based methods either assign a shared trajectory-level advantage or estimate step-level advantages from state groups, which ma
benchmark - arxiv:2610.14736 · cs.AIThink in Strategies, Act with Tools: Traceable Agentic Optimization for Combinatorial ProblemsZhongnian Li, Shi Xu, Peng Ying, Weikang Gong +2
Optimization by PROmpting (OPRO) directly leverages Large Language Models (LLMs) to generate candidate solutions through natural-language prompting. However, OPRO's optimization process is difficult to track, which prevents LLMs from capturing useful information accumulated over previous iterations,
agentagentic - arxiv:2610.14733 · cs.ROB-Human '26 - Scalable Team Play up to 11 vs. 11Arne Hasselbring, Tim Laue, Thomas Röfer, Philip Reichenberg
B-Human won multiple competitions in the Middle Division of the new Humanoid Soccer League in 2026. In addition to regular matches with three or five players per team, we ventured into uncharted territory by playing demo matches with eleven humanoid robots, highlighting the tactical flexibility of o
humanoid - arxiv:2610.14731 · cs.LGVitaBench: Memory Lifecycles Beyond RetrievalMuquan Yu, Dihan Hu
Long-term memory benchmarks for language agents primarily evaluate whether information from earlier interactions can be retrieved and used when relevant. Persistent memory systems face an additional problem before future use is known: deciding whether a memory should remain exact, be compressed, con
memorypersistent memorybenchmark - arxiv:2610.14730 · cs.LGSelf-improving data synthesis can scale embedding models across domains and tasksZucan Lyu, Rui Tang, Jianlyu Chen, Peisheng Jin +2
Adapting embedding models to new tasks requires training data that reflect task-specific information needs and relevance criteria. However, manual data design is costly and difficult to scale. Here we introduce SYNTRA, a novel retrieval data synthesis method whose self-improvement loop iteratively r
self-improvingself-improvement - arxiv:2610.14727 · cs.AIMHWA: Multi-timescale Hierarchical World-Action ModelPengcheng Pan, Guoqing Ma, Yuhan Zhang, Yang Chen +3
Autoregressive world models have emerged as a powerful paradigm for sample efficient decision making, yet their open loop rollouts often exhibit increasing drift that leads to error accumulation even within a relatively small planning horizon. Hierarchical temporal abstraction is a natural candidate
world model - arxiv:2610.14720 · cs.ROCAG-WAM: Causal Action-Gain World Action Models for Robot ManipulationPeizheng Guo, Cheng Zhang, Xingyu Zhang, Changwen Zheng +1
World Action Models (WAMs) combine action generation with future prediction, allowing predicted futures to support action evaluation and robot decision making. However, the future associated with a candidate action may contain action-induced and environment-driven changes. Directly evaluating the pr
manipulationrobot policyaction-conditionedbenchmark - arxiv:2610.14713 · cs.CVPixARG: Pixel-Asynchronous Retrieval-Augmented Diffusion for Fast and Realistic Image GenerationZewei Xin, Qinya Li, Chaoyue Niu, Fan Wu
While diffusion models achieve impressive image generation capabilities, they suffer from high computational costs, implausible artifacts, and limited adaptability to dynamic or private data due to static, centralized training. To address these, we propose PixARG, a novel retrieval-augmented generat
retrieval-augmented - arxiv:2610.14710 · cs.LGDeKoR: Decomposed Knowledge Routing for Efficient GNN--LLM FusionPeiyuan Li, Yongqi Huang, Jitao Zhao, Di Jin +1
Large Language Models (LLMs) have shown strong potential for text-attributed graph learning by leveraging rich pre-trained knowledge and language understanding capabilities. Combining LLMs with Graph Neural Networks (GNNs) offers a promising paradigm for expanding the modeling capabilities of graph
benchmark - arxiv:2610.14709 · cs.AIVela: User Behavior as Supervision Sparks LLM IntelligenceYang Xiao, Yusong Sun, Haoyi Wu, Wenyang Hui +16
Large language models are increasingly deployed in applications that act on users' long-term behavior histories, from personal assistants to recommendation and advertising systems. These applications demand the user dimension of LLM intelligence: inferring a person's preferences, intent, and likely
benchmark - arxiv:2610.14707 · cs.CVHow to Build a Benchmark That Actually Measures TasteXiaofan Wu, Xi Zeng, Miaoxia Chen, Peishan Chen +5
Expert-aligned "taste" benchmarks should reward craft judgment, not the ability to game the test. We introduce TasteBench, a pairwise good-vs-bad benchmark across visual, brand, book, and tech items under a frozen rubric with expert-only gold (panel n>=3; percent agreement >=0.60 primary gate; weak
benchmarkeval - arxiv:2610.14705 · cs.AILingdan-V2: Large language models for clinically grounded reasoning in traditional Chinese medicineRui Hua, Yu Wei, Ziri Jin, Zhuo Sun +14
Large language models have advanced medical reasoning, but clinically grounded reasoning in traditional Chinese medicine (TCM) remains challenging because prescription decisions require integrating domain knowledge, syndrome differentiation, treatment principles, herbal prescription composition, and
post-trainingbenchmark - arxiv:2610.14704 · cs.ROPRISM: Precision Forcing for Generalist Vision-Language-Action PoliciesChenghao Yue, Peiyan Li, Hao Wang, Lei Lv +4
Coupling a pretrained vision-language model (VLM) with a diffusion or flow-matching action expert has become the mainstream vision-language-action (VLA) paradigm for embodied policies, yet they remain unreliable once a task demands precise manipulation, e.g., submillimeter cable insertion. We attrib
vision-language-actionembodiedmanipulation - arxiv:2610.14702 · cs.ROBGC-Nav: Cross-Altitude Multi-UAV Collaborative Navigation via Belief-Guided CouplingJiuwu Hao, Yueyang Wu, Liguo Sun, Haolin Song +2
Despite recent advances in Aerial Vision-and-Language Navigation (VLN), single Unmanned Aerial Vehicle (UAV) search-and-rescue in outdoor scenarios remains challenging due to partial observability, making cross-altitude collaboration a promising direction. However, existing collaborative navigation
memoryagent - arxiv:2610.14700 · cs.AIDSABench: Benchmarking Source Discovery, Coordination, and Execution for Multi-Database AgentsJinlong Luo, Haoran Wu, Bo Xu
Most text-to-SQL benchmarks provide the target database. In a large data catalog, however, choosing the sources is part of the problem: an agent must find every database needed for a request, determine how their results fit together, and execute the resulting workflow. We present DSABench to evaluat
agentbenchmark - arxiv:2610.14698 · cs.CVPURE: Refining Feed-forward 3D Reconstruction with Unlabeled Post-trainingYaolong Zhu, Peng Li, Jiahao Lu, Chengfeng Zhao +7
Feed-forward 3D reconstruction models have achieved remarkable progress through large-scale training and strong learned priors. However, they remain fragile when applied to dynamic in-the-wild videos, where non-rigid motion, occlusions, and illumination changes are prevalent. A key limitation lies i
post-trainingbenchmark - arxiv:2610.14697 · cs.AIAlignAudio: A Unified Dual-Alignment Framework for Noise-Robust Speech-Conditioned GenerationYixuan Li, Xuenan Xu, Zeyu Xie, Mengyue Wu
Speech-conditioned generation maps speech signals to target audio outputs. Prior end-to-end speech-conditioned generative models achieve high fidelity in clean acoustic conditions, but their performance degrades substantially under environmental noise. While prior noise-robust approaches have achiev
post-training - arxiv:2610.14696 · cs.LGUnderstanding Hidden Spatial Preferences in Bayesian OptimizationMaria Bånkestad, Jens Sjölund
Bayesian optimization uses posterior uncertainty to decide where to explore, but this uncertainty is not spatially neutral. On bounded domains, Gaussian-process posterior variance is typically larger near faces, edges, and corners, giving variance-based acquisition functions a systematic boundary bi
benchmark - arxiv:2610.14691 · cs.RODexCast: Learning Real-World Dexterous Control from Human VideosZihan Gao, Yuhong Zhang, Mengzhe Wang, Kaisheng Liu +6
Recent advances in 3D hand-object interaction (HOI) perception have opened new opportunities for learning dexterous robotic skills from human videos. However, existing dexterous imitation learning methods often rely on costly and manually teleoperated robot demonstrations, requiring substantial huma
manipulationdexterousteleoperationdiffusion policy - arxiv:2610.14690 · physics.opticsSelective spatiotemporal mode-locking in multimode Mamyshev oscillatorsChangfeng Wang, Jia Liu, Yutai Zhao, Ran Xia +5
Spatiotemporal mode-locking (STML) in multimode fibers introduces spatial degrees of freedom for ultrashort pulse generation, yet flexible manipulation of transverse mode compositions within multimode solitons remains challenging. Here, we propose a spectrum-to-mode mapping strategy in multimode Mam
manipulation - arxiv:2610.14688 · cs.LGAccelerating Cross-Model LLM Inference via Attention Map ReuseZhan Jin, Yuze Luo, Bin Cui, Xupeng Miao
LLM-augmented data workflows often involve multiple heterogeneous models operating over substantial shared context, such as schemas, tool specifications, interaction history, and intermediate results. Although prefix caching can reduce prefilling overhead, existing approaches are largely limited to
long context - arxiv:2610.14685 · cs.CVDemiurge: Forging Navigable 4D Worlds via Autoregressive Video RerenderingHanxin Zhu, Jianqi Ma, Cong Wang, Zhihao Hu +9
Recent advances in video diffusion enable novel-view video synthesis from a single input video, supporting world modeling, embodied intelligence, and robotic simulation. However, existing methods process camera trajectories independently, lacking the cross-trajectory consistency needed for coherent,
embodiedworld modelmemory - arxiv:2610.14682 · cs.ROSuccessful but Unsafe: An Analytical Benchmark for Revealing Collisions in Vision-and-Language NavigationHieu D. Pham, Quoc-An Luong, Dung D. Le
Vision-and-language navigation (VLN) agents are typically evaluated by whether they reach a target and how efficiently they follow an instructed route, yet these metrics provide little information about collision behavior. Moreover, continuous VLN evaluation commonly uses a sliding regime, where an
vision-language-actionagentbenchmark - arxiv:2610.14680 · cs.CVA Hybrid Kinematic GCN and Elastic Transformer for Efficient Indian Sign Language RecognitionPushp Garg, Himangshu Sarma, Mrinmoy Ghorai
Sign Language Recognition (SLR) presents complex challenges due to variable signing speeds and spatial intricacies. Specifically, Indian Sign Language (ISL) requires precise modeling of its inherently bimanual nature and is bottlenecked by the scarcity of large-vocabulary benchmarks and the inabilit
benchmark - arxiv:2610.14679 · cs.CVHumanPhys: A Human-Centric Physical-World Perception Benchmark for Real-World Visual ReasoningYiming Zhang, Enzhi Zhou, Junan Lin, Dulan Zhou +3
We introduce HumanPhys, a human-centric benchmark for real-world physical perception, and a training corpus emitted by the same pipeline. Each question targets a physically grounded quantity related to people, including depth, inter-person distance, gaze angle, displacement, velocity, heading-step t
benchmark - arxiv:2610.14663 · cs.CVTETRIS-4DGS: Efficient Post-Training Compression of 4D Gaussian SplattingChengzhen Wu, Kelei Liu, Sicheng Li, Lu Yu +1
4D Gaussian Splatting (4DGS) has emerged as a powerful dynamic scene representation, supporting real-time high quality rendering but suffers from prohibitive storage overhead. Existing methods typically attempt to solve this by treating compression as a per-scene optimization problem, limiting gener
post-training - arxiv:2610.14662 · cs.MADoes Failure Feedback Direct Agent Self-Modification?Rong He
Self-improving agents use failure feedback to revise their code or instructions, but gains after selection do not establish target-specific repair or generalization. We measure directedness by comparing repair of a randomly selected target failure with repair of the parent's other failures. Across 1
agentself-improving - arxiv:2610.14659 · cs.AIRASSG: Repair-Aware Semantic Survival Gating for Streaming Tool ExecutionDaijin Hu, Xi Wang
Streaming speech assistants can launch tools before users finish speaking, overlapping execution with incoming speech to reduce user waiting time. However, subsequent self-corrections can invalidate apparently stable arguments, causing stale execution whose irreversible effects persist even after th
tool useself-correction - arxiv:2610.14647 · cs.ROCan Synthetic UAV Benchmarks Be Trusted? Real-Synthetic Validation for UAV Model SelectionYuyi Zhi, Min Du, Bill Shi, Guodong Yuan +3
Synthetic UAV imagery can expand pre-deployment test evidence when task-specific Real data are scarce, but visual plausibility or semantic label alignment alone does not establish validity for model selection. We study when Synthetic evaluation preserves the model-selection conclusions obtained from
benchmarkevaluation protocol - arxiv:2610.14636 · cs.AIPEARL: Process-Experience Aware Reinforcement Learning for Multi-Turn AgentsJie Yang, Mianqiu Huang, Yiwei Zhao, Bo Wang +2
Terminal rewards make multi-turn agents trainable without step-level annotations, but group-relative advantages provide little guidance about which intermediate decisions to adjust. Failed trajectories also contain reasonable actions, while unfavorable decision patterns can recur as training proceed
memory - arxiv:2610.14635 · cs.CVPVFH-Bench: Benchmarking Premise-Induced Visual Fixation Hallucinations in Vision-Language ModelsXinhao Wang, Junbo Cui, Zhenghao Liu, Yuan Yao +3
Large Vision-Language Models (LVLMs) have achieved remarkable progress across a wide range of multimodal tasks. However, object hallucination remains a persistent challenge, undermining model reliability in applications such as visual question answering, embodied agents, and human-AI decision suppor
embodiedembodied agentbenchmark - arxiv:2610.14630 · cs.CVDuSRFlow: Dual-Lens Super-Resolution via Parallax-Aware Flow Estimation and Complementary Reference WarpingFangpu Zhang, Huanjing Yue, Zifan Cui, Shaoxiong Tu +2
Dual-lens super-resolution (DuSR) aims to enhance a wide-angle low-resolution (LR) image by exploiting a telephoto reference (Ref) image. A central challenge in DuSR is reliable reference utilization under large parallax, occlusion, and cross-camera imaging discrepancies. Existing methods often empl
benchmark - arxiv:2610.14625 · cs.AIFrom Existing Methods to New Research Ideas: Two Perspectives on Recursive Self-ImprovementLeijun Zhou, Yanke Yu, Zhihao Liu, Qu Xiang +9
Recursive self-improvement (RSI) offers a route to AI systems that participate in their own continued development. Recent methods allow language agents to retain changes to their parameters, memory, skills, and code for subsequent execution and improvement. We examine how these methods can inform ne
self-improvementbenchmark - arxiv:2610.14622 · cs.LGFrom Rainfall to Road-Network Functionality: Transferable Graph Forecasting for Flood-Disrupted Transportation NetworksJunyan Wang, Naiyu Wang
Extreme rainfall can rapidly degrade road-network functionality (RNF), disrupting emergency mobility, accessibility, evacuation, logistics, and infrastructure restoration. Existing flood-transportation impact models are often difficult to deploy across regions because they rely on local calibration,
benchmark - arxiv:2610.14615 · cs.AILimitations on Autoregressive Token Generation via Sequential SamplingGal Arnon, Andrej Bogdanov, Paul Lou, Giulio Malavolta +2
We study autoregressive token generation through the notion of sequential samplers: A circuit that samples the next symbol only given the previous ones and possibly a small amount of state. We show that (1) a sequential sampler with $k$ bits of state can be simulated by a stateless one at a roughly
llm agent - arxiv:2610.14609 · cs.CVEviScene: Beyond "Is It AI?" for Agentic Image Factuality AssessmentZiAn Wang, Ruiqi Liu, Zhengbo Zhang, Yi Han +9
Determining whether an image is AI-generated does not establish whether the situation it depicts is factually supported. We formulate Image Factuality Assessment: given an image and a generic verification instruction, a system returns a supported or refuted verdict with an evidence-grounded explanat
agentagenticbenchmark - arxiv:2610.14603 · cs.ROAnyRank: Reliable Anytime Policy Ranking via Sequential TestingApurva Badithela, David Snyder, Tenny Yin, Jade Sceats +5
Due to the high cost of hardware experiments, robot policies are predominantly evaluated by comparing their empirical performance on a limited number of hardware trials. However, finite-sample statistical uncertainty can lead to erroneous conclusions, especially when ranking multiple policies using
world modelaction-conditionedbenchmarkpolicy evaluation - arxiv:2610.14601 · cs.ROStateSkill-Agent: An Effective and Efficient Embodied Agent via State-conditioned SkillChangxu Wei, Jingyi Zhang, Zhiyu Zheng, Tianyi Lin +5
In this work, we aim to develop an embodied MLLM agent that understands and solves embodied tasks by learning, abstracting, and reusing transferrable and generalizable behavioral patterns. To this end, we propose StateSkill, a new skill representation for embodied agents, which introduces the concep
embodiedagentllm agentembodied agent - arxiv:2610.14596 · cs.ROLoad-Aware Quadruped Navigation with Online Contact-based Geometry EstimationQi Yin, XiaoLi Zhang, Yangming Shi
Navigating cluttered environments while carrying objects remains a challenging problem for quadruped robots. Compared with works that focused on load-aware locomotion adaptation to maintain gait stability, existing autonomous navigation systems generally ignore the influence of carried objects' geom
quadruped - arxiv:2610.14589 · cs.CVAttribute-Centric Vision-Language Alignment for Fine-Grained Gait UnderstandingSaihui Hou, Chenye Wang, Wenpeng Lang, Qiannan Guo +4
Walking sequences reveal human-interpretable attributes related to body structure, posture, and motion. Gait attribute recognition provides a concrete route to fine-grained gait understanding by predicting these attributes. However, existing benchmarks provide limited fine-grained annotations, while
benchmark - arxiv:2610.14588 · cs.ROFlow-Support-Weighted Proposals for Test-Time Verified VLA PoliciesChen Zhao, Zipeng Ye, Haitao Shen, Guanlin Li +2
Test-time sampling promises to turn one action prediction into a set of distinct executable candidates that a verifier can choose among. Downstream fine-tuning works against that promise. A vision-language-action model trained to solve one task distribution concentrates its flow-matching decoder aro
vision-language-actionvlamanipulation - arxiv:2610.14583 · cs.RORemembering What Matters for Long-Horizon Robotic World ModelsHao Wu, Tianyu Shao, Shengju Qian, Yizhou Zhao +9
Autoregressive robot video world models based on discrete visual tokens remain unreliable over long rollout horizons. Retaining the entire generated history does not necessarily improve prediction quality. We investigate a possible memory-role conflict in Full Context decoding: persistent scene info
liberoworld modelaction-conditionedbenchmark - arxiv:2610.14579 · cs.LGFine-Tuning Strategy, Not Just Data: A Statistically Rigorous Audit of Gender and Stereotypical Bias in Embedding Models for HR Recommendation SystemsEkta Yadav
Embedding models underpin modern job-candidate matching: resumes, job descriptions, and queries are encoded as dense vectors, and recommendations come from nearest-neighbor retrieval over that space. Any social bias in the embeddings therefore feeds directly into HR recommendations. We audit six mul
benchmark - arxiv:2610.14577 · cs.ROIn View or Out of View? A Vision-Aware Filter that Infers and Reacts to the Human's Field of ViewMengxue Fu, Ya-Chuan Hsu, Yinlong Dai, Erdem Bıyık +1
During human-robot interaction, sight plays an important role. When the human sees a robot, they can observe the robot's actions and better predict the robot's future behaviors. Conversely, when the human cannot see the robot, they become uncertain about the robot's intent and plans. In this paper,
robot policy - arxiv:2610.14576 · cs.AIIterate on an LLM, Serve on an SLMKeegan Wang
As frontier LLMs gain more capabilities, the cost to run them also goes up. At the same time, small language models (SLMs) also gain more capabilities but have largely been able to keep their cost stable, offering 5-40 times cheaper inference for the narrow, repetitive subtasks that agents decompose
agent - arxiv:2610.14574 · cs.LGCB-Agent: Concept-Based In-Context Learning for Sequential Action SelectionQingyue Zhang, Huaze Tang, Minyu Zhang, Yuxiang Yan +2
In-context learning (ICL) allows large language models to use demonstrations without updating their parameters. For interactive agents, demonstrations are trajectories of reasoning, actions, and observations that illustrate how to complete a task. Reusing demonstrations in new situations is challeng
agentbenchmark - arxiv:2610.14572 · cs.LGReward-Aligned DistillationYilan Chen, Zian Su, Manish Prajapat, Dingmin Wang +3
Reinforcement learning with verifiable rewards improves the reasoning capabilities of large language models, but remains computationally expensive and sample inefficient. On-policy distillation (OPD) provides dense token-level supervision from a stronger teacher, yet standard OPD distills every toke
benchmark - arxiv:2610.14570 · cs.AIPsyMemBehav: Psychology-Informed and Memory-Grounded Two-Level Identity-to-Action Reasoning for Social-Media Behavior SimulationWenji Zhang, Xilong Cheng, Yunxiao Qin, Mai Tu +7
Large language models (LLMs) support social simulation, but faithfully reproducing real users requires grounding both structured platform decisions and language in sparse, noisy behavioral evidence. We introduce PsyMemBehav, which maps psychology-informed profiles and retrieved memories to target-fi
benchmark - arxiv:2610.14568 · cs.CVFrameFold: Response-Guided KV Cache Consolidation for Autoregressive Long Video GenerationMuyu Liu, Mingming Liu, Hong Gu, Yuyao Zhang +1
In autoregressive video generation, retaining all historical key-value (KV) states increases memory usage and per-step attention costs as videos grow longer. A fixed cache budget bounds these costs but raises the challenge of preserving historical content that supports consistent, high-quality gener
memory - arxiv:2610.14564 · cs.ROChunk-wise Model-based Policy OptimizationWeixi Song, Donglin Wang, Yong-Lu Li
Model-based reinforcement learning has shown superior sample efficiency by leveraging the learned latent dynamics either in the form of derivative-free planning/imagined rollout training or dynamics gradient-based policy updates. While these approaches achieve comparatively fast learning, they all s
humanoidquadrupedworld modellatent dynamicsevaluator - arxiv:2610.14561 · cs.ROFrom Video to Vision: Learning Generalizable Humanoid Carrying from Human MotionZiluo Ding
Most existing approaches rely on video or motion-capture data for reinforcement learning (RL) training; however, all of these methods require an upper-level module to generate reference trajectories in real time. In object carrying tasks, the humanoid can only lift objects from designated locations
humanoid - arxiv:2610.14558 · cs.LGPGSB: Pretrained-Guided Shared Basis for LoRA Model MergingMuqing Liu, Chongjie Si, Zhuoya Liu, Yuheng Jia
Large pretrained models have achieved remarkable success across diverse domains, yet adapting them to multiple downstream tasks remains challenging: full fine-tuning is costly, while maintaining separate task-specific models is storage-inefficient and fails to produce a unified multi-task model. Low
benchmark - arxiv:2610.14551 · cs.ROBefore We Act: Learning Action-Conditioned Team-Response Beliefs for Decentralized Multi-Robot ManipulationZihao Li, Yanan Zhou, Ranpeng Qiu, Guoqiang Ren +1
In multi-robot manipulation, an action's value depends on how teammates respond. We present \method{} (Cooperative Action--Response Belief Estimation), a framework for learning \emph{team-response beliefs} from local history. A predictive belief encoder represents the interaction and conditions a no
manipulationaction-conditioned - arxiv:2610.14550 · cs.LGLLM-based User-Product Relevance Evaluation for Dynamic Product AdvertisingWen Zhang, Xiao Bai, Hsiang-wei Chao, Mengfan Zhu +8
Dynamic Product Advertising (DPA) retrieval systems are a core component of modern e-commerce advertising; however, their evaluation remains challenging. Conventional offline engagement metrics are limited by sparse, biased feedback, while online evaluation is costly and time-consuming. In this work
agenticscalable evaluationscalable eval - arxiv:2610.14544 · cs.CVInfraWorld: Motion-Aware World Modeling for Roadside 4D Occupancy ForecastingSongkai Wang, Xiaokai Bai, Lei Yang, Hui-liang Shen
Roadside 4D occupancy forecasting predicts future semantic occupancy in a fixed infrastructure coordinate system. Roadside scenes are globally structured yet locally evolving, requiring comprehensive scene-level context while precisely modeling sparse localized motion. However, existing forecasting
world model4d occupancy - arxiv:2610.14542 · cs.AIAre Multimodal Benchmarks Really Useful? Item-Level Multimodal Benchmark Diagnosis via Structure-Response Co-CalibrationShiqi Zhang, Weixin Zeng, Ziheng Zhang, Jiuyang Tang +1
Multimodal benchmarks are central to evaluating whether models can reason over information from multiple modalities. However, existing literature has revealed that some benchmarks may contain shortcut items that can be answered from a single modality or from answer-option regularities alone, thus fa
benchmark - arxiv:2610.14541 · cs.ROGG-Nav: Goal-oriented Grounding Chain-of-Thought Elicits Reasoning in Navigation Foundation ModelsShaohao Zhu, Wensi Huang, Siqi Zhang, Yuqiang Yang +5
Emerging strong Vision-Language Models (VLMs) have advanced navigation foundation models. However, these models still have unsatisfactory performance on goal-oriented navigation settings. In this paper, we first conduct a motivating study revealing that the key issue lies in the insufficient utiliza
benchmark - arxiv:2610.14539 · cs.AIFDE-Bench: Evaluating End-to-End Delivery from Underspecified Real-World Business RequestsHuaiming Li, Can Huang, Chufan Wu, Danyu Liu +17
Real-world business requests often leave key requirements unstated because frontline practitioners struggle to articulate rules they know from experience. Forward-deployed engineers (FDEs) must elicit these requirements through dialogue and iteratively optimize solutions until they meet customer acc
agentbenchmarkevaluator - arxiv:2610.14537 · cs.LGLearn the Dynamics, Enforce the Physics: Generative Latent Operators for High-Dimensional Beam EvolutionHaohan Zou, Alexander Scheinker, Yuanyuan Shi
Particle-in-cell (PIC) simulation is indispensable for modeling beam dynamics in particle accelerators, but its computational cost limits many-query workflows, such as configuration exploration, virtual diagnostics, and controller design. We introduce GLOBE, a physics-constrained Generative Latent O
benchmark - arxiv:2610.14531 · cs.CLLongHisDoc: A Comprehensive Benchmark for Chinese Long Historical Document UnderstandingZeyu Shan, Yongxin Shi, Shihang Wu, Peirong Zhang +3
Chinese historical documents are invaluable cultural treasures, but their complexity poses significant comprehension challenges for non-specialists. Despite the success of Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) in modern document tasks, their capabilities in Long (full
retrieval-augmentedragbenchmark - arxiv:2610.14530 · cs.ROWho to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-ValidationRuiyang Wang, Hao-Lun Hsu, Haocheng Meng, Songqiao Xie +1
Multi-robot systems (MRS) offer significant advantages over single-robot platforms by enabling information sharing and coordination for different tasks in complex environments. However, reliable cooperation critically depends on inferring trust over both information sources and their reported hypoth
benchmark - arxiv:2610.14523 · cs.CVScienceBuddy-PV: A Verifier-Governed Recursive Improvement Protocol for Panvascular Intervention ResearchLingsen You, Yujun Guo, Xinyu Zhong, Zijie Feng +3
ScienceBuddy-PV adapts the ScienceBuddy recursive-in-recursive design to panvascular intervention research through a structured evidence ledger, device-vessel ontology, abstention contract and retained failure bank. Recursive harness optimization and outer-loop model learning, including reinforcemen
agentagent system - arxiv:2610.14519 · cs.LGQuery-Adaptive Neuron Steering for LLMsYanxi Pan, Jingwen Wu, Yan Ma, Xiang Chen +7
Effectively eliciting the capabilities of large language models (LLMs) at inference time remains an important challenge. Existing studies address this challenge through context engineering, which shapes user queries, or fixed steering, which directly intervenes in model computation. However, predefi
benchmark - arxiv:2610.14517 · cs.CVWe Have Been Interpreting Diffusion Transformer Backbone Behaviors WrongBenyuan Meng, Longtao Huang, Yang Liu, Liang Lin
During a network forward call, diffusion backbones naturally produce many intermediate activations. If we evaluate each activation against an identical downstream benchmarking protocol, we can plot an inverted-U-shaped curve as the depth profile: the activation's performance goes up first and then d
benchmark - arxiv:2610.14515 · cs.MASEPM: Self-Evolving Procedural Memory Grounded in Task Graphs for Multi-Agent CollaborationYi Liu, Junwei Zhao, Li Sun, Jiansong Zhang +1
Long-horizon multi-agent collaboration requires memory that preserves both successful actions and the coordination structure underlying them. Existing memory systems support information sharing or procedural reuse, but often leave the evolution of shared procedures disconnected from the task depende
memorymulti-agentself-evolvingbenchmark - arxiv:2610.14513 · cs.ROTartanMatch: Towards Universal Dense Correspondence Across ModalitiesHyeokjoon Kwon, Jiting Cai, Ruogu Li, Kritan Bhandari +6
Robotic platforms combine multi-modal sensors to operate across diverse conditions, but most dense correspondence methods are limited to RGB-RGB or a small set of modality pairs. Training a model for each modality pair scales poorly with the number of sensors and forgoes potential positive transfer
benchmark - arxiv:2610.14507 · cs.AIMontageFlow: Physically Valid High Fidelity EEG GenerationSudarshan Regmi, Nabaraj Subedi, Arvind Pillai, Yu Yvonne Wu +5
Generative modeling offers a promising solution to the scarcity of labeled clinical electroencephalography (EEG) data under strict privacy constraints. Although recent generative models report substantial improvements in generation fidelity, we show that these gains conceal a categorical failure. EE
benchmark - arxiv:2610.14495 · cs.AIReels-RSI: Program-Level Self-Improvement for Code-Rendered Explainer VideosWei Lin
Explainer videos -- a worked geometry problem, why the sky is blue, how a dynasty fell -- are usually one more target for pixel video generation. We argue they are a different kind of artifact: a program (script, storyboard and per-frame animation code) that a deterministic renderer turns into frame
memoryagentself-improvementbenchmark - arxiv:2610.14493 · cs.LGA Gated Catchment Hydrology Model (GCHM) for Physically-Interpretable Rainfall-Evapotranspiration-Runoff Simulation Across Diverse Hydroclimatic RegimesM Jawad, HV Gupta, MA Farmani, A Behrangi +1
Recent advances in machine learning have improved rainfall-runoff prediction, yet most data-driven models remain difficult to interpret and do not explicitly enforce physical conservation laws. We present the Gated Catchment Hydrology Model (GCHM), a mass-conserving architecture based on the Mass-Co
benchmark - arxiv:2610.14492 · cs.ROMulti-modal Representation Fusion for Long-term Microsurgical Anastomosis Trajectory AnticipationYuxuan Liu, Yating Luo, Yunfei Luan, Xinyao Zhou +3
Microsurgical anastomosis is one of the most challenging and sophisticated procedures for surgical robotics, requiring accurate manipulation of surgical instruments within a limited space to achieve precise operations. Accurate surgical trajectory anticipation is essential for surgical scene underst
manipulationmemory - arxiv:2610.14487 · cs.AIAlignPoison: Conditional Targeted Poisoning of Generative Recommenders via Decoding-Aligned PrefixesZhuoxuan Zhang, Jinhao Duan, Xinyu Zhao, Jinduo Guo +6
Generative recommendation models based on autoregressive decoding have been adopted as alternatives to classical ranking-based systems. While their modeling flexibility has been widely studied, their resilience to data poisoning remains largely unexplored. In this work, we ask: how susceptible are g
benchmark - arxiv:2610.14481 · cs.MAOptimize Once, Execute Fast: Latency-Aware Multi-Agent Workflow Learning for Recurrent QueriesEnpei Zhang, Feiyu Qu, Zheng Huang, Dawei Zhou +2
Large Language Model-based Multi-Agent Systems (LLM-based MAS) have shown remarkable success in solving complex tasks by coordinating specialized agents through multi-step workflows. To overcome the high cost and specificity of manually designed workflows, recent research has shifted toward learning
agentmulti-agentagent systembenchmark - arxiv:2610.14478 · cs.ROShared-Control Teleoperation of Omnidirectional Aerial Robots for Continuous Valve RotationJinjie Li, Yicheng Chen, Haokun Liu, Moju Zhao
Omnidirectional aerial robots offer full 6-DoF independent control over position and orientation, making them popular for aerial manipulation. Although there have been advancements in robotic autonomy, human operation remains essential in complex aerial environments. Previously, we proposed an aeria
manipulationteleoperation - arxiv:2610.14477 · cs.AIWhat Counts as the Same Task? Executable Contracts for Robustness Evaluation of Tool-Using AgentsShaikh Akib Shahriyar, MD. Akil Raihan Iftee, Armon Barton, Matthew Wright
Robustness tests for tool-using agents often change the agent's environment, for example by hiding a hostile note in an invoice, and then judge the agent by what it does. The result tells us about the original task only if the changed test case is still a valid test of that task. Many evaluations de
agentbenchmark - arxiv:2610.14475 · cs.CLEdgeLLM-Nano: Auditing Polar-Guided Key-Cache Compression and Consistency-Anchored Pruning on a 33.6 MB On-Device Language ModelS M Mahbub Murshed, Hayden Bright, Rezaul Karim, Ali Shiri Sichani
Deploying language models (LMs) on the edge presents unique memory, latency, and power challenges absent in high-performance computing. KV-cache compression emerges as a promising approach to address these challenges. However, the effectiveness of current KV-cache compression methods has not been th
long-context - arxiv:2610.14474 · cs.LGRES-MR: Risk-Aware Reasoning for Explainable and Safe Medication RecommendationCong Wang, Jin Li, Shoujin Wang, Yishuo Li +2
Medication recommendations seek to deliver personalized drug combinations based on patients' clinical information, where ensuring transparent decision-making and personalized safety are of paramount importance. However, existing approaches predominantly operate as ``black boxes'', offering limited i
benchmark - arxiv:2610.14465 · cs.AICo-adaptive Rule Routing and Constraining for Structured Knowledge RetrievalZhisheng Qi, Haoyu Han, Jiaxi Li, Bing He +8
Structured knowledge retrieval acquires task-relevant evidence from structured knowledge sources for downstream tasks. Existing methods rely on unconstrained retrieval or predefined query-rule supervision, such as author-paper rules for scientific discovery and anatomy-drug rules for precision medic
agent - arxiv:2610.14464 · cs.CLEvaluating Bias in Chinese Large Language Models across Tasks and Registers: Task Operationalization Effects and Irony Classification RisksMo Zhou
In Chinese content governance, large language models are increasingly used to identify toxic and discriminatory expressions, yet both task operationalization and ironic expression can alter bias measurement readings. We conducted a preregistered multi-task behavioral audit of multiple Chinese large
manipulation - arxiv:2610.14460 · cs.CLRe-executing Coding-Agent Evaluations: Unchanged Decisions and Unresolved Policy RiskTianyu Ding, Yaocheng Tong, Muzhi Yang, Fei Jiang +2
Executable tests grade large language model (LLM) coding-agent patches for leaderboards, rewards and self-improvement. Two preregistered re-execution studies returned negative results. SWE-bench Pro v1's offline option rejected none of 59 accepted Go patches whose grading configuration changed. Thei
self-improvementleaderboard - arxiv:2610.14449 · cs.LGJevolution: A Framework for Studying Communication and Collective Dynamics in AI AgentsXirui Huang, Nikhil Hooda, Utkarsh Gupta
Experiments with remote decision-making agents depend on more than the policy: the world may change during inference, the simulator determines which actions are available, and births and immigration alter who can act. We present Jevolution, an instrumented spatial predator-prey framework that makes
ai agent - arxiv:2610.14448 · cs.LGKV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache QuantizationSihwan Park, Sung-Yub Kim, Doohyuk Jang, Eunho Yang
Asynchronous rollout improves the efficiency of reinforcement learning with verifiable rewards (RLVR) by reducing under-utilization during long-tail generation. Sustained concurrency also concentrates context processing into fewer decode steps, increasing the KV workload per step. We characterize th
benchmarkevaluator - arxiv:2610.14443 · cs.AISpecification-Based Testing of Large Language ModelsKausar Moshood, Jace Bolante, Roshan Aditya Murali, Manish Motwani
Model cards, system cards, and safety reports increasingly document LLM safety behaviors, yet safety evaluations typically start from predefined benchmarks rather than asking which documented claims those benchmarks actually test. We introduce LEST (LLM Evaluation through Specification-driven Testin
benchmark - arxiv:2610.14440 · cs.LGTarget-ESS Tempered Policy Optimization (TETPO): Smooth Variance Control via Bridge Importance SamplingShayan Mohajer Hamidi, Kayhan Behdin, Wenhui Zhu, Aida Rahmattalabi +3
Asynchronous reinforcement-learning post-training of large language models is computationally efficient but inherently off-policy: the learner updates on rollouts generated by a stale policy, while correcting this mismatch requires importance weights that can be extremely heavy-tailed. A small numbe
post-trainingbenchmark - arxiv:2610.14439 · cs.CVSceneForesee: Learning Scene Evolution for Spatiotemporal ReasoningChaoyue Li, Shoudong Han, Haoyang Wu, Boxue Yang +3
Large multimodal models (LMMs) need to understand how entities move and interact as a scene develops. This requires following the same entities over time and relating their observed motion to subsequent changes. We formulate Scene Evolution Understanding (SEU), a task for predicting and interpreting
benchmark - arxiv:2610.14438 · cs.CLNeither True nor False: Do LLMs Represent Truth-Value Gaps in Plural Predication?Duaa Nakshbandi, Zhiyu An, Yiping Li, Sachin Giri +1
Do Large Language Models (LLMs) genuinely grasp the truth conditions of plural descriptions, or do they force every sentence into true or false? We investigate the Homogeneity Gap, a semantic phenomenon where a plural definite (e.g., the kids laughed) and its negation (the kids didn't laugh) are bot
grasp - arxiv:2610.14437 · cs.LGFrom Transfer Probes to Expert Grouping: Evidence from LoRA Post-TrainingXueyang Wu
A platform that post-trains one base model for many business lines often has more lines than experts it can maintain, so lines must share experts, yet lines seldom come with a taxonomy saying which should share. We ask whether transfer probes (train on one source, measure paired changes on all sourc
post-trainingevaluator - arxiv:2610.14431 · cs.CLMemForm: Benchmarking Memory Formation Poisoning in LLM AgentsHaowen Pan, Pengyang Shao, Yunshan Ma, Yanzheng Jin +3
Long-term memory enables LLM agents to learn from past interactions but exposes memory formation to adversarial influence. Existing studies primarily organize memory poisoning by attack channels, corruption patterns, or lifecycle stages, leaving formation interfaces insufficiently characterized. We
manipulationmemorymemory architecturepersistent memoryagent memoryagent - arxiv:2610.14428 · cs.AIFrustratingly Easy Agentic RAG via Portable Skills: A Case Study in Customer SupportIstok Bako, Bonan Min, Tiffany Shao, Claire Cheng +1
Enterprise workflow automation typically requires retrieving a complete and accurate set of relevant information from large operational document corpora (SOPs, policy documents) given a specific scenario or request. In customer support (CS), for example, human specialists must determine the correct
ragagentic - arxiv:2610.14415 · cs.RORed-Teaming of Vision-Language-Action Models via Quality Diversity Environment GenerationShihan Zhao, Henry Chen, Ya-Chuan Hsu, Minjune Hwang +6
A long-standing goal in robotics is to develop generalist robots that can perform tasks across a wide range of settings, flexibly specified with images and text. Vision-Language-Action (VLA) models offer a promising path toward this goal by training on large collections of robot demonstrations. Howe
vision-language-actionvlavla modelsim-to-real - arxiv:2610.14413 · cs.LGCoarse-Graining Recursive Expert Computation: Continuation-Relative Interfaces and Failure ModesShihua Yu, Yanzhi Wang, Tianchi Chen
Computational coarse-graining aims to replace a sequence of updates with a cheaper module while preserving task performance. We present a continuation-relative evaluation framework that assesses replacements through the computation that follows them, using downstream task quality and total execution
evaluation framework - arxiv:2610.14406 · cs.ROMemory Without Curation: Learning Multi-Stream Latent Memory for Robot PoliciesYilin Wu, Rutav Shah, Shuaiyi Huang, Sumedh A. Sontakke +4
It is widely appreciated that robotic policies require long-term memory to perform complex tasks. However, directly conditioning policies on all past observations leaves policy learning susceptible to learning spurious shortcuts, and introduces large latency at policy deployment. As such, today's me
vision-language-actionvlamemory - arxiv:2610.14405 · cs.LGSPROUT: Scaling Agentic Reinforcement Learning for Long-Horizon Software EngineeringMengyang Liu, Weiming Wang, Haixin Nan, Yu Wu +5
Reinforcement learning (RL) has driven substantial progress in coding agents by relying on verified successful trajectories as learning signals. However, such trajectories are scarce in long-horizon software engineering tasks, where a minor error can invalidate the entire rollout. Repeatedly restart
agentagenticbenchmark - arxiv:2610.14403 · cs.CVVisualCoding: Learning Executable Skills for Reliable Visual ReasoningLiu Jiaqi, Jin Ruinan, Li Xiaoxiao
Learning reusable skills can improve visual reasoning agents, yet applying a skill still requires the agent to implement or invoke a procedure and produce the final answer. Errors and hallucinations at these steps can prevent acquired experience from translating into reliable single-run performance.
agentbenchmark - arxiv:2610.14400 · cs.LGSparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task RepresentationsDuaa Nakshbandi, Zhiyu An, Wan Du
Successor-feature-based zero-shot reinforcement learning methods pre-train a task encoder and successor-feature predictor on offline, reward-free data, then recover a policy at test time by solving a closed-form linear regression over a small labeled dataset. We identify an underexplored limitation
manipulation - arxiv:2610.14397 · cs.CVRealityBench: Simulated Environments to Evaluate a Vision Language Model's Ability to Document Human Physical WorkAndrew Perrault, Anmol Takiar, Dominik Wienecki, Arnab Nandi
Augmenting worker capabilities in the physical world through guidance has been a decades-long goal for AI agents. One key supporting capability is to understand what the worker is doing, seeing through the human's camera or smart glasses, and assist them in documentation. Despite advances in vision
agentai agentbenchmark - arxiv:2610.14395 · cs.LGClinical value claims in multimodal AI outpace supporting evidenceJiayu Feng
Multimodal prediction models add a data source, such as imaging or an assay, to clinical information, and health systems must decide whether that added test is worth acquiring. We audited 177 studies comparing multimodal with unimodal models: 81 from a systematic review and 96 from a probability sam
benchmark - arxiv:2610.14393 · cs.CLSciCurateBench: Curating Experimental Data from Heterogeneous Scientific Literature with Verbal-Action AgentsMinuk Ma, Kaiyun Guo, Jiarui Ding
Scientific research increasingly depends on curating experimental data from literature. Scientific data curation requires verifying exact experimental conditions (e.g., specific assays, clinical outcomes, and data availability) across long, heterogeneous files, which is a highly time-consuming proce
agentbenchmark - arxiv:2610.14389 · cs.AIMeasuring the Inference Energy Efficiency of Open-Weight LLMs across Benchmarks, Reasoning Modes, and HardwareZeyu Yang, Wes Armour
Benchmark scores tell us what an LLM can do, but not how much energy is required to deliver that capability. We directly measure this trade-off for 86 open-weight model checkpoints, evaluated on seven benchmarks and six hardware platforms ranging from a consumer GPU to an eight-GPU server. We find l
benchmark - arxiv:2610.14382 · cs.AIANASSA: From High-Level Framework to Low-Level Architecture for Spatial Agentic AIConstantinos Papantoniou, Brian Hilton
Agentic AI systems operating over geospatial data require more than autonomous reasoning and tool use; they must also support explicit orchestration, spatial validation, governance, provenance, failure handling, and human decision authority. Prior work introduced ANASSA, an agentic AI orchestration
agentai agentagentictool use - arxiv:2610.14381 · cs.AIACED-Bench: Evaluating How LLM Agents Acquire and Act on Causal EvidenceYiwei Chen, Amila Weerasinghe, Siddharth Biswal, Utkrisht Rajkumar +3
Large language models are increasingly used as scientific assistants, yet we still lack reliable ways to evaluate whether they can investigate a system rather than merely answer a prompt. A useful evaluation must let agents choose experiments, interpret finite evidence, and make causal decisions, wh
agentllm agentbenchmarkleaderboard - arxiv:2610.14379 · cs.ROOATH: Object-Agnostic Tracking for Dexterous Hands, Enabling Rapid Physics-based RetargetingSolvin Sigurdson, Prasanna Sriganesh, Juan Borbon, Joel Burdick +2
Retargeting human demonstrations into dynamically feasible dexterous hand trajectories requires resolving complex contact physics, yet current methods either omit physical constraints or rely on computationally expensive sampling-based optimization in simulation. Reinforcement learning has previousl
dexterousgrasp - arxiv:2610.14372 · cs.RODexFLEX: Contact-Aware Foundation Controller for Command-Guided DexterityWeikang Wan, Zeyu Lai, Jianyu Hou, Yuchen Zhou +3
Dexterous robot hands can receive useful motion intent from teleoperation or learned policies, but successful execution also depends on contact decisions that these commands rarely specify. We introduce DexFLEX, a contact-aware foundation controller that turns upstream fingertip-motion drafts into c
dexterousteleoperationtactile - arxiv:2610.14369 · cs.LGWhen More Memory Fails: Hidden Feasibility Holes in Dynamic Tensor RematerializationMahesh Reddy Pagadala
Dynamic Tensor Rematerialization (DTR) trains neural networks under a memory budget by evicting tensors and recomputing them on demand. A natural expectation is that a budget large enough to complete training stays sufficient when it is increased. Using DTR's reference simulator, simrd, to replay th
memory - arxiv:2610.14360 · cs.AIWhen Pleasing the User Compromises Care: AI Sycophancy in Caregiver-Mediated HealthcarePriyadarsi Mishra, Arjun Beniwal, Ziyu Yi, Yu Ding +1
Large language model (LLM) alignment often assumes that the user interacting with a model is also the primary beneficiary of its behavior. Yet in many consequential settings, one person directs the system while another bears the consequences. We study this problem in caregiver-mediated healthcare, w
benchmark - arxiv:2610.14352 · cs.LGClass-Conditional Conformal Prediction for Signal Temporal Logic Inference with Many ClassesYixuan Wang, Danyang Li, Matthew Cleaveland, Roberto Tron +1
Signal Temporal Logic (STL) inference provides interpretable formulas for time series classification. For multiclass classification, however, accurate predictions do not necessarily ensure that the inferred formulas are satisfied consistently with the true class. We introduce the positive satisfacti
manipulation - arxiv:2610.14351 · cs.ROLearning Transferable World-Action Models from Task-Paired Human VideosJeongeun Park, Taekyung Kim, Dongyoon Han, Sangdoo Yun
World-action models inherit broad visual and physical priors from video pre-training. However, standard within-video next-frame prediction allows them to rely on appearance shared between the context and future, rather than learning interactions that transfer across scenes and embodiments. We introd
post-training - arxiv:2610.14350 · cs.LGNatural Code Transformation Confuses Agents in Vulnerability AnalysisJie Zhu, Ronik Bhaskar, Heather Zheng, Ben Y. Zhao +1
Having demonstrated a strong potential for automating vulnerability discovery, AI agents can also be used to attack software systems at scale. Today's software systems need mechanisms to level the playing field by slowing down potential attacks, giving them time to patch vulnerabilities and strength
agentai agentagentic - arxiv:2610.14346 · cs.CLEvaluating LLMs as Agentic Workflow Executors in Real-World Customer Support SkillsMd Tahmid Rahman Laskar, Gundeep Singh, Xue-Yong Fu, Karol Chang +10
Enterprise customer-support agents increasingly rely on Large Language Models (LLMs) to execute bounded, tool-using workflows. We study this problem through the lens of Skill LLMs: specialized executors that operate within a selected skill context and must complete a narrow, stateful workflow under
agentictool usebenchmarkevaluator - arxiv:2610.14335 · cs.ROPILOT: Guiding Long-Horizon World-Model Planning for Robot Manipulation with Visual Subgoals and Policy PriorsYang Liu, Hanchen Cui, Karthik Desingh
Latent world-model planning has shown promising results in goal-conditioned robotic manipulation, but remains challenging for long-horizon planning due to limited temporal guidance and inefficient action search. A distant visual goal provides little guidance for preparatory actions, while uninformed
manipulationliberoworld modelaction-conditioned - arxiv:2610.14334 · cs.AIWhich Metadata Matters for LLM Tool Selection in MCP? Evidence from Controlled Factor-Level InterventionsXiaofan Li, Fengchun Qiao, Xing Gao
In Model Context Protocol (MCP) systems, LLMs select tools from metadata rather than from their implementations. Names and descriptions summarize what a tool does, while the parameter schema specifies the fields a tool accepts and how they are organized. Our analysis of metadata for 367,174 MCP tool
evaluator - arxiv:2610.14332 · cs.CVTraceSampler: Making Sampling Errors Predictable and Correctable for Accelerated Generative SamplingGeorge Matthew, Guoqiang Zhang, Kenta Niwa
Accurate sampling from continuous-time generative models can require many model evaluations. Low-step sampling reduces cost, but coarse updates introduce endpoint errors that can accumulate and degrade generation quality. We introduce TraceSampler, a post-training method that corrects these errors w
post-training - arxiv:2610.14326 · cs.CVReal-Time Traffic Density Estimation: A Hybrid Computer Vision and Predictive Machine Learning PipelinePrince Kwabena Appiah Boadu
Traffic density estimation from camera feeds requires connecting visual perception with short-horizon forecasting. We present a two-stage pipeline that combines vehicle detection and tracking with a spatial logic layer, followed by a temporal estimator. The spatial layer filters bounding boxes by th
benchmark - arxiv:2610.14324 · cs.LGEV-JEPA: Elastic V-JEPA for Any Device and Cross-Device InferenceAnxhelo Diko, Jieyu Lin, Stefano Ambrogio, Ziyun Li +4
Foundation models like V-JEPA are increasingly deployed across a heterogeneous edge-cloud continuum spanning smart glasses, phones, robots, and servers. Yet, they are typically trained as rigid, fixed-capacity networks bound to one device, supporting neither reconfigurable compute budgets nor collab
v-jepabenchmark
02 US SEMI · SEC 8-K FILINGS
0 itemsscanned: NVDA / AVGO / MRVL / COHR / LITE / AMD / TSM / SMCI / ANET / CRDO / POWL / VECO
03 HUMANOID · COMPANY NEWS
62 itemsscanned: figure-ai / 1x / boston-dynamics / unitree / apptronik / sanctuary-ai / neura-robotics / agility-robotics / physical-intelligence / agibot
Figure AI (10)
- Figure AISeptember 30, 2026F.02 Decommission
- Figure AISeptember 17, 2026Helix 2.5: Zero-Shot 30-Home Generalization
- Figure AIOctober 09, 2025Introducing Figure 03
- Figure AISeptember 03, 2026Figure and Nscale Sign Strategic Partnership For Up to 100,000 GPUs on the NVIDIA Vera Rubin Platform
- Figure AIAugust 25, 2026Introducing Index: Building The World’s Largest and Most Diverse Physical Dataset
Boston Dynamics (10)
Unitree 宇树 (10)
- Unitree 宇树Components
- Unitree 宇树Kung Fu Meets Spring, Unitree SFG Robots Present "Cyber Real Kung Fu" in the Year of the Horse2026-05-31Media Coverage
- Unitree 宇树Welcoming Myanmar President Min Aung Hlaing to Unitree2026-08-05Media Coverage
- Unitree 宇树Unitree founder Wang Xingxing graces the cover of Time magazine2026-08-05Media Coverage
- Unitree 宇树Unitree Announces H2 Plus, an NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research2026-06-01Media Coverage
Apptronik (1)
Sanctuary AI (6)
- Sanctuary AIBlog PostTIGER: Task-Space Imitation Guidance for Efficient Reinforcement LearningDemonstrations as a compass, not a scriptRead More
- Sanctuary AIPress ReleaseProduct UpdatesSanctuary AI Expands Physical AI Strategy to Industrial Robotics, Demonstrating Production-Ready AI PerformanceRead More
- Sanctuary AICorporate NewsDaniel Friedmann Appointed CEO of Sanctuary AIRead More
- Sanctuary AIPress ReleaseZeon Invests in Sanctuary AI and Partners to Advance Specialized Materials for Dexterous RoboticsRead More
- Sanctuary AIThought LeadershipWeb Summit Reflections: Canada’s Physical AI Moment Can’t WaitRead More
Agility Robotics (10)
- Agility RoboticsAgility’s Humanoid Deployment ProcessPras VelagapudiAugust 04, 2026
- Agility RoboticsThe Realistic Pathway to HomeInsightMay 26, 2026
- Agility RoboticsAgility and AIInsightMarch 16, 2026
- Agility RoboticsAgility Gets a New BrandInsightMarch 5, 2026
- Agility Robotics2026: The Automation EvolutionInsightJanuary 16, 2026
Physical Intelligence (8)
- Physical IntelligenceNewFrontier Models for Every RobotOctober 13, 2026Our latest robotics models are now available through the Pi API beta. Upload data, fine-tune models, and deploy cloud-hosted inference on your own robots.
- Physical Intelligenceπ0.7: a Steerable Model with Emergent CapabilitiesApril 16, 2026A steerable robotic foundation model that exhibits a step-change in generalization.
- Physical IntelligenceThe Physical Intelligence LayerFebruary 24, 2026General-purpose physical intelligence models will enable a Cambrian explosion of robotics applications. See how our partners are already solving real-world problems.
- Physical IntelligenceMoravec's Paradox and the Robot OlympicsDecember 22, 2025By fine-tuning our latest model, we were able to solve a series of very difficult manipulation challenge tasks.
- Physical Intelligenceπ*0.6: a VLA that Learns from ExperienceNovember 17, 2025A method for training our generalist policies with RL to improve success rate and throughput on real-world tasks.
智元 AgiBot (7)
- 智元 AgiBotAGIBOT Opens Hong Kong Office in Causeway Bay2026-10-09
- 智元 AgiBotAGIBOT Partners with Vietnam National Innovation Center(NIC) to Advance Embodied AI Industry and Talent Ecosystem2026-10-02
- 智元 AgiBotIDC Data Shows AGIBOT Led Global Humanoid Robot Shipments in H1 20262026-09-30
- 智元 AgiBotAGIBOT and ASD Deploy 100 AGIBOT X2 Humanoid Robots Across 100 Retail Stores2026-09-29
- 智元 AgiBotAGIBOT and Chimelong Launch Large-Scale Embodied AI Theme Park with More Than 300 Robots2026-09-23