Physical AI Brief

Action-conditioned world models are increasingly used as learned simulators for policy evaluation and improvement, yet their effectiveness rests on an unverified assumption: generated futures faithfully reflect arbitrary valid actions. Existing benchmarks are typically confined to expert demonstrations, leaving off-expert action following inadequately evaluated. To address this gap, we introduce WorldEcho, which probes action following over a broader action distribution using visual integrity and SE(3) trajectory alignment. Our diagnosis shows that current world models reasonably execute expert actions but struggle with diverse off-expert trajectories, either ignoring the commanded actions or producing visually invalid rollouts. We further propose WorldSync, which strengthens action following along three complementary axes: distributional coverage, representational grounding, and intervention-effect alignment. It broadens the training distribution over action consequences, grounds intermediate video representations in action-induced robot dynamics through an Action-Forcing Expert, and aligns predicted changes under action interventions with the corresponding changes in ground-truth futures. Experiments on RoboTwin benchmarks and real-robot tasks show that WorldSync improves WorldEcho metrics and serves as a more reliable simulator for iterative policy improvement, enabling policies to achieve higher success rates.

World action models (WAMs) improve robot control by modeling how observations evolve, but generating future observations at test time incurs substantial latency. Fast-WAM removes this process for efficiency; however, our matched implementations show lower generalization for Fast-WAM than for future-aware alternatives, especially with scarce robot demonstrations and in out-of-distribution scenarios. To bridge this gap, we introduce **LAWA**, a WAM architecture that uses compact latent actions as an operational representation of future intentions, enabling efficient test-time future imagination without generating future observations. Specifically, a discrete tokenizer enhanced by action-free pre-training produces manipulation-centric codebook targets. LAWA jointly denoises a continuous latent state anchored to these targets with executable action chunks while omitting the future-video branch at inference. On RoboCasa, LAWA achieves state-of-the-art average success rates of 65.6% and 80.8% in the few-shot and full data settings, improving over the matched Fast-WAM baseline by 9.6 and 4.5 points, respectively. It also preserves the performance level of the matched Joint-WAM variant while requiring 42.9% lower inference latency. LAWA also demonstrates competitive zero-shot robustness on LIBERO-Plus and superior performance on real-world tasks. These results show that future imagination need not be discarded: retaining it with compact latent actions yields an effective trade-off among performance, generalization, and latency. Code and models will be released.

Smart glasses are evolving from capture and display accessories into first-person intelligence platforms that connect human perception, persistent context, and digital or physical action. Their on-body viewpoint aligns with the wearer's vision, audition, motion, and hand-object interaction, but must operate under tight energy, thermal, privacy, and feedback constraints. Despite rapid progress in augmented reality, egocentric vision, multimodal models, human-computer interaction, and embodied intelligence, the literature remains fragmented across devices, tasks, and benchmarks. \textit{The key challenge is not whether a model can recognize, answer, remember, or act in isolation, but whether a complete system can sustain a reliable, temporally valid, correctable, and governable perception-state-interaction-action loop.} This survey is \textit{the \textbf{first} to systematically study smart glasses through such a unified framework}. We formalize first-person data flow and constrained task utility, characterize devices along eight verifiable hardware capability axes, organize the literature around seven interdependent foundational capabilities, and introduce an L0-L5 framework spanning capture, reactive perception, contextual assistance, persistent state, governed action, and embodied coupling. Across nine application scenes, we connect tasks with datasets, systems, products, stakeholders, failure consequences, and evidence gaps. We further present a nine-dimensional deployment framework, a claim-conditioned evaluation protocol, and an evidence ladder from controlled measurement to longitudinal field validation and audit. Together, these elements make smart glasses more comparable, deployable, and reproducibly evaluated, while outlining a roadmap toward trustworthy first-person intelligence.

Recursive self-improvement (RSI) remains hard in long-horizon tasks, where growing histories obscure the task state and misalign skill invocation. We introduce Recuris, a recursive Experiential-Working Memory architecture for long-horizon agent harnesses, in which Working Memory tracks task progress and guides skill selection from Experiential Memory, grounding skill use in current needs rather than the full history. This coupling also turns execution into structured evidence that localizes failures to specific memory components. Across tasks, a fixed Meta-Agent turns that evidence into localized, validation-gated updates to Skill Memory that reshape execution and yield new evidence, forming a bounded recursive memory-evolution loop. Across four long-horizon benchmarks and ten models, Recuris improves task success in 35 of the 37 completed model-benchmark pairs, carrying frontier models to SOTA-level task success: on tau-bench it adds +17.8 points to GPT-5.6 Sol and +15.6 to Claude Opus 5, taking Opus 5 to 87.9%, and +16.6/+13.5 points on Qwen3.6-27B/35B on SkillFlow. The advantage widens as the interaction horizon grows, to +32.2 points on the longest tasks, and common long-horizon failures fall by up to 80%. These results position recursively evolving memory as a scalable foundation for RSI, enabling agents to continuously transform accumulated experience into increasingly effective long-horizon behavior. Code: https://github.com/Gen-Verse/Recuris

Group-relative reinforcement learning waits for sibling rollouts of the same prompt, which is costly for long and variable tool-use trajectories. Single-stream Policy Optimization (SPO) removes this dependency with a persistent prompt-level value estimate, but its recipe whitens one advantage per trajectory before optimizing a token-mean actor loss. We show that trajectory centering generally does not center the token-weighted quantity consumed by the actor, and fix the mismatch by standardizing terminal-outcome advantages under the action-token measure. We additionally organize prompt evidence by the policy event that generated it rather than learner receipt order. Across matched runs on ALFWorld at two model scales and on Math-TIR, SPO++ improves online learning efficiency over SPO. A paired ablation identifies action-token-measure normalization as the strongest tested component.

Latent world models are inherently strong encoders that transform image pixel to latent embedding, yet existing world models still rely on online trajectory optimization for action planning: for every state-goal pair, an iterative optimizer is run from scratch to search for optimal action sequences, treating the world model as a black-box simulator. This approach pays the full iterative optimization cost anew at every replanning step and reuses no planning experience across queries. In this work, we ask whether planning itself can be amortized once a latent world model has been learned. We present LeFlow, which learns a reusable latent trajectory prior operating directly in the latent dynamics space from the world model. LeFlow recasts planning as conditional latent trajectory generation: a rectified-flow model imagines a future latent path between the current and goal embeddings, an inverse dynamics decoder turns latent transitions into action chunks, and the frozen world model verifies each candidate by autoregressive rollout. Across four major goal-conditioned pixel-control benchmarks, LeFlow replaces iterative action-space optimization with amortized latent planning and fixed-budget rollout selection, achieving consistent success-rate gains with an order-of-magnitude reduction in planning time. Our results argue that latent world models should support not only prediction but reusable planning priors. Our code is available at https://github.com/hsiangwei0903/LeFlow.

Web agents that act from rendered pixels avoid the fragility and heavy token cost of reading a page's HTML or accessibility tree, but training them depends on large amounts of high-quality interaction trajectories, and how to produce such data at scale remains an open problem. Public datasets typically contain only a few thousand trajectories drawn from a fixed and narrow set of websites, and even recent automated synthesis pipelines stay bound to predefined site lists or tutorial sources, so the number of distinct websites the agent ever sees barely grows. We present BrowserForge, a framework that generates web interaction data at scale by driving many browser sandboxes in parallel over the open web. BrowserForge couples three components: an open-web sourcing stage that exposes the agent to hundreds of thousands of real, openly reachable websites; a sandbox cluster manager that schedules hundreds of concurrent browsers with high utilization; and a Proposer-Solver dual-agent loop that turns a raw page into an executable task and then collects a verified trajectory for it. A rule-plus-model cleaning pipeline removes failed runs and rewrites the surviving reasoning into a single unified chain-of-thought style. Page structure such as the accessibility tree is used only as a synthesis-time signal; the agent we train and release acts purely from the screenshot. The resulting corpus contains 203,238 trajectories, each collected from a distinct website, larger and more diverse than prior trajectory datasets. Fine-tuning a compact multimodal model on this corpus raises its success rate on the live Online-Mind2Web from 25.66% to 33.33% and consistently improves step accuracy on the static Multimodal-Mind2Web, with the gain growing as the corpus scales. Controlled analyses further confirm that open-web sourcing and broad website coverage are key contributors to the observed improvement.

Real-world data for knowledge graph question answering is often distributed across different organizations due to governance and data sovereignty constraints. While centralized systems exist, they cannot answer multi-hop questions when the required facts are split across vertically partitioned silos. In this paper, we propose FedV-KGQA, a framework for multi-hop reasoning over knowledge graphs in which organizations share entities but own disjoint sets of relations. Our approach combines local graph enrichment and knowledge graph embeddings to ensure raw triples and relation parameters never leave each silo, establishing a structural data boundary without requiring centralized graph access. We further introduce a topic entity anchoring mechanism that grounds questions in the correct graph neighborhood without any runtime inter-silo communication. We evaluate 12 model configurations across three benchmarks and show that FedV-KGQA performs strongly, remains close to centralized performance, generalizes to 3-hop reasoning, and is robust to embedding perturbations.

We present LAION-BVD, a large-scale open video dataset for multimodal learning, which contains 1.3B platform-specific video URLs collected from CommonCrawl. From these, we download 80M videos with a total duration of 10 million hours. The dataset is designed for multimodal pre-training across the video, audio, and image modalities. Using content-aware scene detection, we extract clips for which we synthetically generate video and audio captions. Models trained on these data achieve competitive performance on standard video-text and audio-text benchmarks, with consistent improvements as training or model scale increases. Additionally, we explore video frames as an alternative source of image-text data by extracting scene-changing frames. These frames exhibit a visual distribution distinct from standard web image corpora, and models trained on this dataset achieve strong image-text retrieval performance. We release LAION-BVD to the research community. It significantly expands open access to multimodal videos at an unprecedented scale.

Large language models (LLMs) are increasingly deployed as AI analysts to process financial disclosures and support AI-assisted investment decisions. Yet such systems are usually evaluated by what they can retrieve, not whether retrieved information affects their judgments. We identify a retrieval-integration gap in long-context financial analysis. Holding focal-firm information fixed and varying only unrelated context from 2,000 to 128,000 tokens, we find that a risk disclosure's influence on investment judgments falls to the experimental noise floor even as direct retrieval remains accurate. The pattern replicates across model families and judgment tasks and in experiments removing real disclosures from actual 10-K filings. More capable models postpone but do not eliminate the gap. Causal memory interventions show that compressed summaries and source-text lookup jointly transmit disclosures into judgments. Workflow architecture determines whether this transmission succeeds: chunk-and-summarize pipelines evict relevant information, whereas a targeted, structured restatement adjacent to the decision restores its influence. AI analyst performance is therefore jointly determined by model capability and workflow architecture. Retrieval-based evaluations can certify systems whose investment judgments ignore information they demonstrably retrieved.

Large language models are increasingly used for knowledge graph question answering (KGQA), but can fail to correctly ground answers in the underlying graph. Current approaches to LLM-based KGQA either rely on full semantic parsing into executable queries such as SPARQL, which is brittle in practice due to complex schemas or incompleteness of real-world KGs, or on LLM-reasoning and answer generation over KGs, which can be more robust but lacks formal guarantees. In this work, we study a complementary setting in which \emph{candidate} answers are generated by an LLM-based system and subsequently verified using lightweight symbolic constraints derived from the question. We introduce \emph{Constrained Entity Selection under Partial Knowledge (CES-PK)}, a problem formulation that focuses on eliminating invalid answers and providing symbolic support for valid ones without requiring construction of executable logical forms. To account for incomplete KGs, we employ a three-valued constraint semantics (\emph{satisfied, violated, unknown}) that avoids incorrect rejections under open-world assumptions. To demonstrate the effects of our method, we instantiate this framework over the Hetionet biomedical knowledge graph and evaluate the impact of type, relation, and exclusion constraints. Experiments show that precision improves by filtering invalid candidates, while recall is preserved due to retaining candidates whose constraints are not explicitly violated. Satisfied constraints provide additional positive symbolic evidence to rank remaining candidates.

Neighborhood-based fairness audits evaluate individual fairness by comparing predictions among similar individuals in feature space. Despite their widespread use, little is known about the robustness of the auditing procedure itself. Because these audits rely on nearest neighbor relationships, small perturbations in feature space can alter local neighborhoods and produce different fairness assessments even when model predictions remain unchanged. We develop a geometric framework for analyzing the robustness of neighborhood-based fairness audits under bounded perturbations. Our analysis establishes sufficient conditions for neighborhood invariance, quantifies how neighborhood replacement propagates to audit instability, and introduces audit volatility, a measure of the expected sensitivity of fairness audits under repeated perturbations. Experiments on benchmark datasets support the theoretical analysis and show that the proposed framework explains the observed stability of neighborhood-based fairness audits.

Recent work has applied Mamba style state space models (SSMs) to video anomaly detection, yet existing approaches still rely on buffering clips or windows internally, lack a theoretical account of how temporal memory relates to detection latency, and benchmark efficiency only through GPU throughput rather than the edge hardware these methods are intended to target. We introduce a strictly causal streaming anomaly detector whose fixed size state is updated in O(1) time and memory per incoming frame, with no lookahead and no clip buffering. Its temporal core is a diagonal linear state space recurrence with an input and state dependent decay gate, trained self supervised through causal next embedding prediction on a frozen visual backbone. We derive a closed form relationship between the recurrence decay spectrum and both detection delay and the shortest anomaly it can reliably capture, then validate empirically on UCSD Ped2 and CUHK Avenue. The settling delay bound predicted from the learned base decay (57 to 59 frames) sits far above the measured detection delay (1.6 and 18.4 frames), showing that the event boundary gate, not the base decay, governs responsiveness. We further report end to end latency and throughput measured directly on Apple M3 Pro hardware, 0.74 ms and 0.77 ms per frame (over 1300 FPS), rather than simulated GPU numbers. With an untuned initial configuration the method reaches 67.9 percent and 70.2 percent frame level AUC on Ped2 and Avenue, trailing prior non causal SSM baselines in accuracy. Ablations over decay rate, state size, and gating reveal that the gate contribution is dataset size dependent, hurting accuracy on the smaller Ped2 training set but helping on the larger Avenue one. Closing this accuracy gap and extending evaluation to a third, larger benchmark are immediate next steps.

We present Crase, a bounded and inspectable alternative to deep research agents for scholarly search. Instead of an open-ended search loop, Crase queries a search engine once for seed papers, expands them along their 1.5-hop citation neighborhood, prunes citation edges whose claims lack entailment support, and ranks the remaining papers with a recency-aware random walk. This makes the candidate set, the reason each paper is kept, and the stopping condition explicit and fixed before inference. On LitSearch and one further benchmarks over a 500K-paper arXiv corpus, Crase outperforms deep research agents built on proprietary models by up to 3$\times$ recall@50 at roughly a third of the cost.

We present StarHarness, a framework for evolving environment-specific agent harnesses while keeping model weights fixed. The evolved harness can include prompt and task framing, tool interfaces, skills, MCP-backed providers, subagent structure, and agent-loop configuration. StarHarness constructs a compact evolution pool by stratifying tasks according to baseline failure behavior, separates proposer-visible search tasks from proposer-hidden selection tasks, and reserves held-out tasks for evaluating generalization. Across ITBench SRE, EnterpriseOps-Gym ITSM, and AutomationBench Finance, harness evolution improves full-benchmark performance by 20-35 percentage points over the default harness after 4-12 accepted changes per environment. These gains persist on tasks excluded from evolution and transfer without re-evolution across GPT and Qwen model families. Trace analysis links the improvements to interface repairs, environment conventions, and operational knowledge that compresses search, with fewer false-positive diagnoses and shorter trajectories in several settings. StarHarness therefore offers a practical way to reduce persistent model-environment mismatch in tool-rich enterprise tasks.

Outcome-supervised search agents learn when and how to retrieve evidence, but terminal rewards neither localize intermediate errors nor redirect an ongoing trajectory before those errors compound. Treating corrective feedback as a learned in-trajectory intervention couples the two roles: the agent must decide when to request and use feedback, while the critic must infer useful corrections from outcome-confounded rollouts whose failure patterns shift as the agent improves. We introduce CAFE (Coupled Agent--Feedback Evolution), a framework in which a shared-parameter model alternates between search-agent and critic roles. CAFE initializes feedback-conditioned recovery from trajectories built around the base agent's own failures, then couples online and offline optimization. During online RL, a comparative feedback estimate uses a prompt-level call--skip success gap to shape request returns, while feedback-aware advantage shaping reweights token advantages before and after feedback. Offline, rollout-derived preference optimization learns feedback from matched successful and unsuccessful trajectories. On seven agentic search benchmarks, CAFE outperforms the evaluated RL-based search agents on average, retains its gains across all six out-of-domain benchmarks, and reduces answer-level hallucinations. One-sided ablations show that improving only the agent or only the critic eventually plateaus, whereas alternating the two updates continues to improve performance. These findings suggest that a self-improving search agent needs feedback that co-evolves with the policy it guides.

Clinicians read chain-of-thought (CoT) rationales as evidence of medical reasoning, but whether the visible chain plays that role is rarely tested. General-domain CoT-faithfulness probes ignore clinical cost, and medical LLM evaluations treat the chain as a black box. We close this gap with a medical perturbation audit: a 30-operator battery edits both the chain and the question with clinically motivated operators (severity reversal, negation flip, demographic swap, evidence ablation), paired with a chain-update times answer-flip joint analysis that classifies each model by its failure mode. Applied to 14 LLMs on four medical QA benchmarks, three independent tests converge: the Chain-Decoupling Rate (CDR; chain does not register the edit and the answer does not flip) is 72.9% panel-wide on clinically meaningful destructive edits, chain corruption leaves accuracy unchanged, and removing CoT prompting does not reduce accuracy. Two board-certified clinicians re-annotate N=197 perturbed questions; 98.5% leave the gold defensible. The pattern holds across medical and reasoning fine-tuning and scale; on the closed-source tier, where the chain text is unavailable, the answer-side signals are consistent with the same decoupling. Our framework and CDR provide a reusable yardstick for auditing whether medical CoT is faithful or merely documentation.

The increasing heterogeneity of multi-agent systems poses significant challenges for jointly training a global model across agents. At the same time, cooperative inference between agents has long been recognized as a powerful mechanism for distributed decision making over networks. Motivated by these observations, we propose a collaboration framework for distributed binary classification over multi-agent networks, where a set of independently trained agents, potentially differing in architecture, feature space, or modality, coordinate their actions during test time to form collective predictions. This coordination is achieved by exchanging local decision statistics through a distributed learning protocol. We develop a theoretical and experimental study of this independent training and cooperative inference paradigm, and examine its performance under different communication budgets and distributed learning rules. We establish classification error guarantees under sufficient, finite-round, and finite-precision communication, together with PAC-style generalization bounds. These results capture the influence of model heterogeneity, network topology, combination policy, and communication constraints on prediction accuracy. Taken together with the experimental results, they reveal both the price of independent training and the benefit of collective prediction for the proposed distributed decision making framework with models learned from data.

Distinguishing machine-generated text (MGT) from human-written text (HWT) becomes increasingly important due to potential misuse. However, most supervised detectors often degrade out-of-domain (OOD) and require large, diverse training sets. In this work, we analyze the linearity and quality of MGT representations and show that simple linear probes outperform a wide range of detectors while being substantially more sample-efficient. We first show that MGT and HWT latent representations are linearly separable in low-dimensional space, and provide a plausible explanation for this separability through systematic differences in their representation quality. Motivated by these insights, we train two variants of simple linear probes and evaluate them across 4 benchmarks against 16 baselines. Probes consistently improve OOD detection (+11 AUC), requiring solely ${<}100$ samples to reach near-peak performance. We show that this transferability arises because probes recover a shared latent MGT direction that generalizes across diverse settings. Finally, we demonstrate that probing vectors capture a continuous spectrum of ``machineness'', highlighting their potential for fine-grained estimation of AI-edited text. Overall, our work provides insights into latent-space differences between MGT and HWT and demonstrates the potential of linear probes as as robust and sample-efficient MGT detectors. We release our code on~\href{https://github.com/gerritq/mgt_probes}{github}.

LLM-based agents can interact with external environments through tool invocation, but this capability also introduces security risks such as file modification, information leakage, and unauthorized actions. Existing guardrails often evaluate completed trajectories, leaving pre-execution monitoring of step-level actions underexplored. We propose StepGuard, a step-level guard model that can audit completed agent trajectories and check tool actions before they are executed. To train StepGuard, we introduce StepGen, an automatic data engine that generates safe and unsafe trajectories with the same context but different actions at the risky step. To further reduce over-defense and under-defense, we propose Balance-GRPO, which dynamically balances learning between safe and unsafe actions based on their observed accuracy. Experiments show that StepGuard achieves the highest average accuracy among open-weight guard models, with performance comparable to GPT-5.4. When used to guard agents on AgentDojo and AgentDyn, StepGuard reduces mean attack success rate by 77.3% relative to the no-guard setting, while mean utility drops by only 2.8 percentage points.

Large language model agents are moving beyond conventional retrieval-augmented generation toward direct interaction with external corpora. Direct Corpus Interaction (DCI) keeps the full corpus accessible, yet reachable evidence can remain unusable under finite interaction budgets. Required evidence may fail to surface, a surfaced supporting document may remain unopened, or an opened document may fail to expose its decisive fragment. We call this progressive silent loss Evidence Blindness and quantify it through stage-wise evidence realization. Within the DCI paradigm, raw interaction adds little reusable corpus organization, while dynamic-workspace methods reconstruct a query-conditioned interaction space from each query and trajectory. In both cases, useful structure is recovered largely online. We instead formulate large-scale agentic search as finite-budget navigation over reusable corpus structure. We introduce AtlasNav, a persistent multi-view corpus-navigation framework that retains direct corpus interaction but organizes the corpus once into a Corpus Atlas, allowing each query to navigate adaptively rather than reconstruct shared structure. On BrowseComp-Plus, AtlasNav achieves 92.05% strict accuracy while reducing recorded online inference cost by 30.21% relative to the prior dynamic-workspace state of the art. Under matched budgets, it realizes the complete required evidence earlier and approaches the same model's evidence-supplied empirical reference more rapidly. The same representation principle remains effective under PhantomWiki's distinct corpus organization and controlled 10K-1M scaling, and transfers competitively to heterogeneous enterprise knowledge. These results show that agentic search depends not only on accessible evidence, but also on how the corpus is represented so that limited interaction becomes effective navigation.

Procedural video-language models must solve heterogeneous tasks from the same visual evidence, including action recognition, forecasting, and procedure prediction. Dense transformer decoders share the same feed-forward networks across tasks, which can entangle task behavior and make controlled capability expansion difficult. Sparse Mixture-of-Experts (MoE) decoders provide conditional computation, but token-level learned routing is not naturally aligned with task-level procedural objectives. We propose MoTE (Mixture of Task Experts), a decoder architecture that converts large language model feed-forward networks into task-specific experts while keeping the multimodal backbone shared. Each example follows one sample-level task route, so active task-expert computation remains independent of the number of stored task experts. We instantiate this design as VideoLLM-MoTE and evaluate it on five COIN benchmarks using explicit task routes. The five-expert model activates ~2B LLM parameters per sample and achieves higher average top-1 accuracy than recent VideoLLM baselines. Under the same expert topology, it improves over dense all-expert activation and learned sparse-routing controls. These results show that task-structured routing provides an interpretable and compute-efficient decoder alternative for multi-task video-language learning.

Text-to-CAD aims to generate executable CAD programs from natural-language descriptions. However, real-world descriptions are often underspecified and omit critical spatial constraints required for valid CAD construction, a challenge that has been largely overlooked by existing methods. In this paper, we argue that missing spatial constraints should be inferred with respect to the underlying construction structure and informed by reusable design experience. Based on this insight, we propose ExpConCAD, an experience-enhanced framework for implicit spatial constraint completion. ExpConCAD first recovers the intended construction structure and constraint scopes, then retrieves relevant constraint-completion experience for similar scopes to complete the missing spatial constraints, and finally generates executable CadQuery programs. Extensive experiments demonstrate the effectiveness of ExpConCAD and provide insights into the role of construction structure understanding and experience memory in spatial constraint completion. Our code is available at: https://github.com/Hotjiashell/ExpConCAD.

Evaluating Retrieval-Augmented Generation (RAG) systems requires assessing not only end-to-end correctness but also how individual components interact and how errors propagate through the pipeline. We introduce a Bayesian evaluation framework that jointly models retrieval success, abstention behavior, and answer correctness, factorized according to the pipeline's information flow. The model distinguishes task success. Whether the user received a correct answer (from generator success) and whether the generator behaved appropriately given the retrieval outcome. We apply the framework to 27 RAG configurations across three datasets, three retrievers, and three generators, and show that the conditional decomposition reveals substantial behavioral differences between systems that appear equivalent under marginal metrics. We further analyze the annotation allocation problem, demonstrating that retrieval-success annotations are more informative than task-success annotations for estimating policy adherence, and provide an information-theoretic explanation for this asymmetry. Finally, we extend the model to incorporate LLM-as-a-judge annotations as calibrated noisy observations, enabling practitioners to combine limited human judgments with cheaper automated assessments within a unified probabilistic model.

Large language model (LLM) agents are trained with reinforcement learning (RL) for complex decision-making tasks. However, most RL-trained agents remain episodic and cannot accumulate reusable knowledge across episodes. Recent skill-based approaches, such as SkillRL, attempt to address this issue by extracting skills from raw trajectories, but treat the skill bank as an append-only repository without verifying whether stored skills remain effective. In this paper, we propose SkillForge, a framework for continuous skill evolution that enables skills to be verified and refined through environment interaction. By making skill usage explicit during agent interaction, RL can directly optimize both environment actions and skill invocation decisions. SkillForge further introduces evidence-based skill verification and multi-pathway skill induction, allowing the skill bank to continuously grow while maintaining its quality. Extensive experiments on ALFWorld, WebShop, and AppWorld show that SkillForge consistently outperforms SkillRL, demonstrating the effectiveness of continuously verified skills in training stronger LLM agents.

Physical implementations of Turing Machines remain rare, and existing electromechanical demonstrators and mechanical logic games typically require manual operator intervention, either to trigger each computational step or to reconfigure the state table, or both. This restricts prior physical models to short, operator-paced demonstrations and prevents autonomous execution of extended computations. This paper addresses that gap with a hardware Turing Machine that enables autonomous multi-step execution and reprogrammable optical input without manual intervention between programs. The system integrates an Arduino Mega for state-transition logic, dual NEMA 17 stepper motors for bidirectional tape actuation, infrared reflectance sensors for symbol detection, and an ESP32-CAM-based optical punched-card reader for automated state-table loading. Hole detection under non-uniform illumination used a Breadth-First Search flood-fill algorithm with local adaptive thresholding rather than fixed global thresholding, driven by the memory and library constraints of the ESP32-CAM's microcontroller environment; this improved card-decoding accuracy from 75% to 90% (100% with mechanical card flattening) on a 20-card test set. Mechanical evaluation showed fabrication accuracy of +/-0.15 mm, rack-and-pinion positional error below 0.3 mm across 50 trials, and voltage supply stability within +/-0.2 V under full system load. End-to-end computation was validated against a parallel software simulator (tlang), with all hardware outputs matching the simulated reference exactly across multiple test programs. The system advances prior physical Turing Machine demonstrations through autonomous execution, reprogrammable optical input, and quantitative evaluation of its mechanical, optical, and computational performance.

Learning from Demonstration (LfD) allows robots to learn manipulation tasks directly from humans, thereby supporting the versatile application of robots. Most LfD methods do not explicitly model the physical interactions between a robot and its environment, such as the making and breaking of contact, while these are crucial during manipulation tasks. Because the same basic physical interactions recur often, they can be a basis for robust, generalizable, and adaptive task reproduction. We propose an LfD method that explicitly uses what physical interactions take place where and when. Using that information, a hybrid position-force controller tracks demonstrated trajectories until contact-based transition conditions from the demonstrations are met. We evaluate our method in real robot experiments consisting of opening doors and locks, bolt picking and screwing, dislodging, and surface contouring. We show that explicitly modeling physical interactions benefits LfD in four ways. First, by allowing reproduction of complex, sequential, and contact-rich manipulation tasks using only a single demonstration and no prior knowledge of the task. Second, by facilitating robustness to unknown geometric variations in the environment. Third, by facilitating generalization when geometric variations are known. Fourth, by facilitating online adaptation using geometric information explored during task reproduction. We discuss how robustness, generalization, and adaptivity can be explicitly implemented, which is generally lacking in the LfD literature. Thereby, our work aims to close a gap in interpretable few-shot LfD of robotic manipulation.

Self-improving LLM agents refine answers, not the process that produces those answers. Systems that add a meta-level hold that level fixed, and those that edit themselves must leave part of their own editing machinery untouched to stay stable, capping the meta-depth they realize at roughly two. We present Meta$^n$, which keeps the meta-operation fixed and recurses on its input instead. That operation, $Ω$, is applied repeatedly to its own products, reading the traces of the solver stack below together with the code that produced them, then writing the next layer as a strategic pre-process and a library of callable helpers. Because $Ω$ never changes, it cannot destabilize the system, and because its input strictly grows, each layer reasons from a higher vantage than the last. Depth is set by convergence rather than fixed in advance, and an evolutionary archive searches over layer chains. Across two backbones, Meta$^n$ outperforms prior self-improving agents on all eight benchmark families. The sharpest case is ARC-AGI-2, built to resist skill memorization, where it alone scores above zero. Ablations indicate that most of the gain from recursion comes from the conditioning each layer passes to the next, and distinct layer roles emerge with depth although no prompt prescribes them. Code available at https://github.com/minnesotanlp/meta-n

We settle the minimax-optimal alternating regret, a regret notion motivated by alternating learning dynamics in games, for both online linear optimization (OLO) and online convex optimization (OCO). For OLO over the probability simplex $Δ_d$, we give an algorithm with $O(\log d)$ alternating regret that remains a constant for any time horizon $T$, and a matching lower bound. Our constant regret bound significantly improves previous results with $O(\log ^{2/3}d \cdot T^{1/3})$ regret [Cevher, Cutkosky, Kavis, Piliouras, Skoulakis, Viano, NeurIPS 2023, Hait, Li, Luo, Zhang, COLT 2025]. As a result, we obtain alternating learning dynamics with $O(\log d /T)$ convergence to Nash equilibria in two-player zero-sum games and $O(\log d /T)$ convergence to coarse correlated equilibria in two-player general-sum games. This is the first uncoupled learning dynamics with $O(1/T)$ convergence to CCE in two-player general-sum games, while all prior works suffer additional $\log T$ factors. For general OCO over a $d$-dimensional compact convex set, we give an algorithm with $O(d\log (1+T/d))$ alternating regret, improving the previous best of $\widetilde{O}(d^{2/3}T^{1/3})$. We also prove a matching lower bound of $Ω(d\log (1+T/d))$, showing that the $Ω(\log T)$ factor is unavoidable.

Hyperparameter selection remains a key challenge in Bayesian optimization (BO) and Bayesian active learning (AL), as model misspecification can lead to suboptimal performance, while more accurate fully Bayesian treatments typically rely on computationally expensive MCMC sampling. This paper proposes a unified framework, KENDO (Kernel ENsemble Disagreement-aware Operator), that integrates Ensemble Gaussian Processes (EGP) with disagreement-aware acquisition strategies. The central idea is to replace hyperparameter sampling with a kernel ensemble and adaptive Bayesian weighting, combined with disagreement-aware acquisition strategies. Within this unified framework, we instantiate KENDO-BO for BO and KENDO-AL for Bayesian AL, demonstrating that both arise from a common self-correcting mechanism with task-specific acquisition objectives. We further extend the approach to multi-objective optimization via random scalarization that preserves the single-optimizer conditioning structure. Thorough numerical tests on synthetic and real-world benchmarks across single-objective optimization, multi-objective optimization, and active learning demonstrate that (i) KENDO-BO achieves competitive or superior optimization performance compared to state-of-the-art methods while reducing computational overhead by up to $5\times$ and (ii) KENDO-AL achieves superior predictive calibration over MCMC-based active learning baselines with up to $27\times$ speedup.

World-Action Models (WAMs) jointly learn future visual prediction and action generation, using video dynamics as a representation-learning signal for robotic manipulation. However, their video latents are primarily optimized for visual prediction and are not explicitly encouraged to preserve cross-view geometric structure or spatially localized, object-relevant semantics. We propose \textbf{GaussianWAM}, a training-time representation-enhancement framework that organizes geometric and semantic supervision through a 3D Gaussian field. Given synchronized multi-view observations, frozen geometry and vision foundation models provide depth, camera parameters, and dense semantic features. GaussianWAM binds these heterogeneous signals to shared Gaussian primitives and renders spatially aligned semantic, depth, and coverage targets, which are distilled into the current-observation representations of the WAM. All teacher models, Gaussian components, and auxiliary prediction heads are removed after training, leaving the original WAM inference path without additional modules or forward computation. On LIBERO-Plus, GaussianWAM improves FastWAM from 52.05\% to 71.29\% and Cosmos Policy from 71.52\% to 77.30\%. Direct CLIP and VGGT distillation already establishes a strong FastWAM baseline of 69.37\%, while Gaussian-field unification further improves it to 71.29\%, supporting the benefit of spatially organizing heterogeneous teacher signals. GaussianWAM also improves performance on standard LIBERO and shows positive transfer trends on RoboTwin and real-world manipulation. These results suggest that training-time Gaussian distillation provides a practical way to inject geometry- and semantics-related supervision into WAM representations without changing their deployment architecture.

While text-based hallucination detection has been extensively studied, spoken hallucination detection remains largely unexplored, particularly for low-resource languages. We present the first multilingual spoken hallucination benchmark comprising 12,013 news samples across English, Russian, and Kazakh with controlled hallucinations of three types and three severity levels. Samples comprise original articles and aligned hallucinated counterparts in text and audio. We complement the synthetic corpus with 290 fact-checked fake news items collected natively in Russian (225) and Kazakh (65), translated into the other language and rendered through the same TTS-ASR pipeline. We assess fine-tuned multilingual encoders and, in zero-shot in-context settings, multimodal decoder models on transcript-based versus direct audio processing. Transcript-based detection generally outperforms direct audio processing, with binary-task degradation for strong encoders tracking per-language ASR error. On real-world fakes, synthetic-trained detectors transfer strongly (macro-F1 0.82-0.88 on original text), while Russian provenance analysis reveals both veracity-related and model-dependent machine-style signals, quantifying a key confound in synthetic hallucination benchmarks.

Predictive Coding (PC) is a neural learning paradigm that enables parallelizable neural network layer updates. However, the main bottleneck of PC Networks (PCN) is the sequential backwards error propagation. To tackle this, we introduce a training technique that pairs a Generative PCN with a support Encoding PCN. The two PCNs are trained in parallel to match their neural activations, without sequential propagation. We apply this to time series anomaly detection and show that our approach results in more stable, continuous, online learning.

Reinforcement Learning with Verifiable Rewards (RLVR) and on-policy distillation (OPD) have become two widely adopted paradigms for post-training large language models. However, RLVR suffers from sparse task-level feedback, while OPD provides dense token-level guidance but ignores trajectory correctness, limiting its performance to that of the teacher. Combining them is a promising direction: OPD supplies dense supervisory signals, while RLVR provides task-level correctness. Nevertheless, existing integrations often rely on weighted combination or heuristic switching, introducing extra hyperparameters and trade-offs. We propose On-policy Distillation with Verifiable Reward (OPDVR), a simple yet effective method that seamlessly combines OPD and RLVR without adding any hyperparameters. We first reformulate the implicit reward of sampled-token OPD based on trajectory correctness, then apply a ReLU gating mechanism to ensure that correct trajectories receive non-negative rewards and incorrect ones receive non-positive rewards---thereby aligning the distillation signal with task success while preserving the teacher's distributional guidance. Furthermore, our modification transforms sampled-token OPD into a proper RLVR method, making it readily combinable with any policy gradient algorithm, such as GRPO. Experiments on six reasoning benchmarks show that OPDVR consistently outperforms standard OPD. Our code is available at https://github.com/LeapLabTHU/OPDVR.

Agentic systems increasingly gate actions on a model's own stated confidence, which assumes confidence tracks correctness at the moment of acting. We test this in a hidden-information chess variant where royal status can be secretly, repeatedly relocated between pieces, and where an agent's stated probability distribution over the opponent's hidden royal piece -- elicited every turn, separately from the move it chooses -- is scored against ground truth recoverable after the game. Across two independent batches, captures made at high stated confidence ($\geq 0.5$) about the hidden piece's location were correct in 1 of 62 cases. The calibration deficit is concentrated almost entirely in these events: 99.3% of it in the original batch, 98.7% in the replication. The same pattern, in weaker form, orders consistently (point estimates only; most pairwise gaps are not statistically distinguishable at this sample size) across four further model configurations spanning a second provider -- reported as scope for the finding, not as evidence that capability predicts calibration: a same-model comparison at a fixed external leaderboard score shows a deliberation-budget change alone moves the metric by nearly as much as a large cross-model gap. In a separate seat, conventional evaluation axes -- legality, cost, latency, completion rate -- can dissociate entirely from belief quality, with the configuration winning on every conventional axis producing the worst belief quality tested. A model exhibiting this pattern can still win the game its belief was about, which is why outcome-only evaluation would not detect it.

Precision oncology necessitates a longitudinal model of patient state that captures cancer evolution and treatment over time, integrating multimodal observations. We introduce the oFM, a foundation model developed on a real-world oncology cohort of 1.67 million cancer patients that integrates clinical trajectories with DNA, RNA, and H&E pathology. Patient-level partitions were reserved for training, validation, and testing, with over one million patients used for training. The oFM encodes daily clinical and molecular episodes and, along with pathology images, integrates them over time to produce a patient state embedding. We evaluate frozen oFM embeddings against expert-curated clinical and molecular baseline features. In prognostic benchmarks, the oFM improved AUC for treatment response, progression-free survival, and overall survival (0.774 vs. 0.563 for overall survival). Across 11 comparative-treatment cohorts, the oFM embeddings achieved a three-fold higher pooled and scale-normalized treatment-benefit AUTOC than baseline features with improved benefit ranking in 9 of 11 cohorts, and provided stronger prognostic discrimination within both treatment arms. We also evaluated a mechanism discovery framework that interprets downstream models built on oFM embeddings by linking their predicted outcomes to clinically and biologically grounded mechanisms through an evidence-grounded temporal graph, enabling evaluation in clinical and drug-development applications.

Video games provide a scalable source of training data for video world models, offering diverse environments, complex interactions, and abundant in-the-wild gameplay videos. However, raw gameplay footage entangles the game world with screen-space interfaces, introducing game-specific biases and irrelevant dynamics that hinder world-model training. To address this problem, we introduce GameUI-Taxonomy and G2WEngine, a full-stack framework that formalizes gameplay UI grounding and removal. G2WEngine automatically extracts reusable UI assets from real gameplay videos and synthesizes temporally coherent UI overlays on clean footage. Using this engine, we construct Game2World, comprising 96K synthetic paired videos with precise reconstruction targets and 1,079 in-the-wild clips from 303 games for realistic evaluation. Its asset library contains 5,132 verified UI elements across 21 taxonomy categories, collected from 1,010 representative gameplay frames. Based on Game2World, we propose GameCleaner, a mask-free gameplay UI removal model that combines multimodal semantic understanding with video editing capabilities. Unlike mask-based methods, GameCleaner directly identifies and removes diverse HUD elements while preserving the underlying scene content and temporal dynamics. In a controlled pilot, world models trained on UI-free gameplay improve overall VideoReward by 6.83% over those trained on UI-overlaid data. On UI-removal evaluation, GameCleaner achieves an average AAR of 95.36 on synthetic videos, outperforming the strongest temporal mask baseline by 57.3%, and obtains the best in-the-wild AAR of 80.05 with 99.8 background preservation. These results demonstrate the scalable potential of transforming Internet gameplay videos into high-quality world-model training data. Code, dataset, and model will be available at https://github.com/Dongping-Chen/Game2World.

Power outage prediction models are increasingly used in assessments of climate-driven infrastructure risk, yet current evaluation practices obscure whether these models generalize to the novel conditions such applications require. We identify three common methodological choices in power outage prediction models that influence their ability to generalize across spatial, temporal, and event-based settings. We compare the predictive performance impacts of different methodological decisions using publicly available data for the U.S. East Coast from 2018 to 2023 and feature sets derived from weather reanalysis and land-cover data, and embeddings from a GeoAI foundation model (Prithvi WxC). Specifically, we assess model performance under multiple test selection strategies, including unfiltered random splits, leave-one-state-out, and leave-one-event-out designs, which increasingly approximate real-world deployment conditions. While random train-test splits yield strong performance, we show that these results are inflated by spatial and temporal autocorrelation. Under spatial and temporal holdout experiments, predictive accuracy degrades substantially, with models often failing to outperform a simple null baseline. Incorporating GeoAI foundation model embeddings yields limited and inconsistent improvements, primarily for spatial generalization, and does not resolve poor event-level transferability. These findings suggest that, given current data availability and evaluation practices, publicly trained outage prediction models offer limited and uncertain operational value. Progress will likely require improved data coverage, more realistic evaluation protocols, and a shift in focus from marginal modeling advances toward addressing structural data constraints.

Scaling test-time reasoning has substantially improved the problem-solving ability of large language models (LLMs), but standard autoregressive decoding still executes long reasoning traces sequentially, creating severe latency for difficult tasks (up to days and weeks). Parallel reasoning offers a natural remedy. However, prior systems primarily focus on Subtask Parallelism, where the model learns to decompose a high-level task into smaller chunks that can be solved independently. This approach overlooks another pervasive form of parallelism: Trial Parallelism, where multiple speculative attempts explore, verify, and aggregate competing hypotheses in parallel. In this paper, we introduce Parason, which reveals and learns both forms of parallelism in LLM reasoning. Our analysis identifies Trial Parallelism as the majority of parallelizable reasoning computation (65.5% in DeepSeek-V4's reasoning steps in HLE), and it becomes increasingly dominant on hard problems. Guided by this taxonomy, Parason converts sequential reasoning traces into structured parallel trajectories with a context-free grammar, then trains models with Parallelism-Aware Group Relative Policy Optimization (PA-GRPO), whose reward jointly balances accuracy, latency, and the two parallelism ratios. At inference time, Parason executes the learned parallel structure through tool calls, translating theoretical savings to real-world wall-clock acceleration. Experiments on mathematical reasoning benchmarks including AIME24 and AIME25 show that Parason achieves an average acceleration about 1.7$\times$ while maintaining competitive accuracy.

System-level simulation is an essential tool for exploring the rapidly expanding design space of LLM serving systems, where real deployments remain costly and often infeasible. However, modern LLM serving now evolves faster than human-driven simulator development can track, and emerging workloads and mechanisms, from agentic workflows to disaggregated serving, no longer fit the monolithic simulation pipeline that existing simulators assume. Each new mechanism therefore demands an invasive rewrite, leaving a widening development gap between deployed serving systems and the simulators that model them. To close this gap, we present Borg, a framework that realizes agent-driven simulator development. Borg introduces a composable simulator infrastructure that uniformly expresses the complete serving workflow, including the control decisions that coordinate it, and realizes it as a unified dynamic graph in Borg simulator. Synthesizer agent, a harnessed coding agent, then lowers natural-language feature requests onto this abstraction under simulator-specific guardrails and fidelity validation, evolving one shared simulator instead of building a new one for every feature. Under the same coding agent and harnesses, extensions built on Borg follow a vLLM-based real system with 2.51% average throughput error, versus 6.03% for extensions built on existing simulators. On identical workloads, Borg also simulates up to 284.96x and 23.19x faster than two state-of-the-art simulators, LLMServingSim2.0 and Vidur, respectively.

Reinforcement learning can align diffusion models with human preferences and task-specific objectives, but endpoint rewards do not specify how an intermediate denoising prediction should change. We introduce DiffusionOPSD as an on-policy self-distillation framework that converts image-level reward guidance into explicit targets for clean-output predictions at sampled queries. At each outer iteration, a frozen behavior policy generates trajectories and supplies query states and anchors. Reward gradients construct bounded positive and negative targets around each anchor. The trainable policy fits these targets as detached supervision through finite fitting before an exponential moving average update refreshes the behavior policy. This setup lets us measure target construction and finite realization separately. Controlled same-query experiments show that larger target-construction gains do not necessarily translate into larger realized gains after a single fitting update. Across SD 3.5-M and the step-distilled Z-Image-Turbo, our approach achieves the best final held-out scores in 19 of 20 reward-matched settings across two backbones and ten evaluators. It outperforms the strongest competing method by up to 44.0% and reduces training GPU-hours relative to DiffusionNFT by 40% on SD 3.5-M and 63% on Z-Image-Turbo. These results support on-policy self-distillation as an efficient and analyzable approach to diffusion post-training by converting image-level reward guidance into explicit and continually refreshed intermediate supervision, thereby opening a path toward more efficient and diagnosable alignment.

Similarity in many decision systems is governed not by distance alone but by interactions among variables. In fraud and anomaly detection, small local perturbations can cross interaction-sensitive decision boundaries while leaving ambient distance almost unchanged. Motivated by this setting, we introduce a thin-slab interaction model and an interaction-driven quantum kernel constructed from entangled Pauli-string feature maps. The feature map explicitly encodes sparse high-order block interactions. We show that the resulting fidelity kernel is positive semidefinite, admits an exact block-factorized formulation, and induces a geometry sensitive to changes in interaction regime. Across balanced and imbalanced synthetic experiments spanning third-, fourth-, sixth-, and eighth-order interactions, the proposed kernel consistently outperforms linear, radial basis function, Laplacian, and polynomial kernels, as well as an engineered-interaction linear baseline supplied with the planted block products. On real fraud-detection benchmarks, it achieves the highest mean accuracy and F1 on Credit Card Fraud Detection and ranks second on IEEE-CIS Fraud Detection. These findings show that quantum-kernel performance depends on alignment between feature-map geometry and the underlying predictive structure, rather than on Hilbert-space dimension alone. Because the prescribed block-factorized kernel can also be evaluated exactly on a classical computer, the results establish predictive and representational value rather than computational quantum speedup.

Liver fibrosis, the principal predictor of long-term outcome in chronic liver disease, is staged from histological estimates of collagen content. Sirius Red (SR) provides the standard quantitative readout (collagen proportionate area, CPA) but is not acquired at every clinical centre and consumes tissue, time, and reagent cost beyond the routine Hematoxylin and eosin (H&E) stain. AI-based virtual staining can generate SR directly from H&E, yet systematic benchmarks of unsupervised models are scarce and their predictive uncertainty has not been quantified, even though visually plausible outputs may not faithfully reproduce the underlying tissue structure. We therefore benchmark six unsupervised image-to-image architectures (GAN-based and diffusion-based) across 54 scaling configurations on a newly released paired H&E to SR mouse liver dataset, the first open resource for this translation task. Each configuration is evaluated jointly on perceptual, distributional, and task-specific axes plus a blinded expert reader study; the best per family is then retrained as a deep ensemble, the first systematic comparison of epistemic uncertainty across unsupervised stain-to-stain architectures. Across families, perceptual quality, task-specific error, and ensemble agreement measure largely independent axes of model fitness: GAN-based methods cluster tightly on perceptual metrics yet differ substantially on task error and ensemble agreement, while the diffusion-based method (CycleDiffusion) is qualitatively different on all three. No single metric captures these differences, so reliable virtual staining requires reporting and selecting on all three jointly. The dataset, tiling pipeline, models, and evaluation code are released publicly.

Can language models be trusted in safety- critical operations? In such settings, strong per- formance on semantic metrics does not guaran- tee operational reliability: a misread altitude, a dropped execution condition, or a confused call- sign may score well under standard F1 yet carry sharply asymmetric operational consequences. We study this problem in air traffic control (ATC), where controller-pilot communication demands near-zero error tolerance, and use consequence-aware evaluation to test whether semantic scores misstate operational reliabil- ity. The framework is instantiated in a con- trolled diagnostic ATC benchmark grounded in aviation standards and feedback from 40 air traffic controllers across three countries. Evaluating 8 models, we uncover a system- atic semantic-safety gap: conventional scores give substantially higher performance estimates than consequence-aware evaluation, even for models that appear reliable under standard met- rics. Risk-aware fine-tuning narrows but does not close this gap, showing that consequence- aware evaluation is a necessary complement to standard NLP metrics before any real safety- critical deployment claim

Post-training quantization lowers the memory footprint of Large Language Models (LLMs) and speeds up inference, which is why it is now common for on-device deployment. Most of what we know about its effects, however, comes from English benchmarks. It is not clear whether the same holds for morphologically complex, low-resource languages such as Bangla, and this gap is what we address here. We evaluate three model families---Qwen-2.5-7B, LLaMA-3.1-8B, and GPT-OSS-20B---in full precision and in three quantized formats (GPTQ-Int8, GPTQ-Q8, GGUF-W8A16) across five Bangla natural language understanding benchmarks (Bangla MMLU, CommonsenseQA-BN, OpenBookQA-BN, PIQA-BN, and BoolQ-BN), using zero-shot evaluation through lm-evaluation-harness. To our knowledge this is the first controlled comparison of quantization formats on Bangla NLU. The three families do not respond the same way: GPT-OSS loses up to 57.35% accuracy on reasoning-heavy tasks under GGUF-W8A16, while Qwen and LLaMA hold steady under GPTQ, and in a few cases the quantized version edges out the full-precision one. BoolQ-BN, a comprehension task, stays stable across all three families regardless of format. Taken together, these results suggest quantization can work well for Bangla deployment, but the choice of architecture and quantization method matters more than the bit width alone. We discuss what this means for practitioners choosing a model to run on constrained hardware.

Vision language action models (VLAs) have advanced general purpose robotic grasping and manipulation by enabling robots to interpret visual observations and natural language instructions to generate executable action sequences. However, existing VLAs often implicitly assume gripper invariance, despite grasping strategies being inherently embodiment-dependent. Different gripper types, such as parallel-jaw and suction, usually require distinct interaction strategies to achieve the same grasping objective. Moreover, current datasets for VLAs predominantly rely on parallel-jaw grippers, limiting gripper-aware learning. To address this gap, we introduce MiGA, a multi-gripper-aware dataset spanning five distinct gripper types across multiple robots with 103,000 demonstrations, explicitly capturing strategy divergence under shared task objectives. We further propose GVLA, which combines a new multi-gripper tokenizer with adapter-based policy routing. Our new gripper encoding induces structured embedding information that balances parameter sharing and strategy differentiation, while layer-wise probing confirms meaningful gripper-conditioned representations for VLAs. Intensive experiments in both simulation and real-world robots show that our GVLA outperforms the current baselines across evaluated settings. Our method also improves zero-shot generalization or few-shot adaptation to new objects or unseen tasks, and enable more efficient gripper adaptation.

Electroencephalography (EEG) is a widely used window into human brain function, but most EEG models remain tied to a one-dataset-one-model supervised paradigm. Recent EEG foundation models offer a route toward reusable representations, but most remain reconstruction-centered, assuming that EEG content predictable from local context is necessarily transferable neural information. Here we present INCEPT, an invariance-oriented EEG foundation model trained on over 11,000 hours of unlabelled clinical EEG. Rather than prioritizing signal recovery alone, INCEPT learns representation-level stability across correlated EEG observations, separating stable neural structure and essential subject-sensitive information from the nuisance variability that dominates scalp recordings while preserving subject-, state- and condition-discriminative information. We evaluate INCEPT on a broad-spectrum benchmark of ten datasets spanning three levels of post-acquisition EEG analysis: signal-level assessment, brain-state decoding, and brain-health evaluation. INCEPT ranks first among recent EEG foundation models on 26 of 30 linear-probing metrics and 24 of 30 fine-tuning metrics, and also surpasses strong task-specific specialist encoders across diverse downstream settings. Objective ablations and representation analyses further show that invariance-oriented pre-training improves transfer and organizes subject-sensitive neural representations beyond reconstruction alone. These results establish invariance learning as a promising principle for building reusable EEG foundation models.

Submerged kelp forests are vital coastal ecosystems that support marine biodiversity and ecosystem dynamics, yet accurate underwater kelp segmentation remains challenging due to optical degradation, illumination variability, turbidity, overlapping vegetation, and complex benthic backgrounds. We systematically evaluated three deep learning semantic segmentation frameworks, ResNet34-U-Net, ResNet50-DeepLabV3, and a hybrid ResNet50-ASPP-Transformer architecture, for kelp detection using high-resolution underwater RGB imagery collected from northeastern U.S. coastal waters. A dataset of 3,395 SSeg assisted annotated image-mask pairs was developed for model training and validation, while geographically independent sites were used for quantitative and qualitative evaluation. All models used consistent preprocessing, augmentation, and evaluation protocols. On independent test data, ResNet50-DeepLabV3 achieved the highest Dice (0.7120) and Intersection over Union (IoU; 0.6267), followed by ResNet34 U Net (Dice 0.6868; IoU 0.5978). The hybrid ASPP Transformer achieved the highest pixel accuracy (0.8528) but lower Dice (0.6437) and IoU (0.5746). External qualitative evaluation further showed that DeepLabV3 produced more consistent segmentation across varying environmental conditions, image qualities, and benthic habitats. Overall, ResNet50-DeepLabV3, termed Kelp-O-Tron, provided the best balance of segmentation accuracy, robustness, and generalization. The dataset, annotation workflow, and comparative evaluation provide resources for advancing automated underwater habitat mapping and ecological monitoring.

Adaptive optimizers retain gradient history in moment variables, allowing a local change in loss weighting to alter later updates. We examine whether this delayed transport is large enough to change prospective short-horizon decisions. On committed future-minibatch sequences, we differentiate eight-step AdamW trajectories through the complete model--optimizer state and select exposure-matched Math--Code loss schedules before independent evaluation. Across 12 unused 0.3M Transformer histories, full transport lowers token-disjoint loss relative to an optimizer-aware immediate derivative in 10/12 histories (mean benefit $4.71\times10^{-4}$; exact one-sided sign test, $p=0.0193$). The two controllers act equally often but select different schedules in 60/96 windows. Crossed checkpoint--future-path tests attribute this reordering to the interaction between optimizer state and near-future data, while an independent Ising--CNN experiment shows that deleting moment-state transport destroys accurate response prediction. Full-transport scores also concentrate exact-rollout winners in larger candidate libraries, focusing finite-amplitude evaluation on a shortlist. On these committed short paths, optimizer memory and near-future data order are therefore actionable components of the training state, providing a mechanism-based criterion for when finite-horizon rather than one-step intervention is required.

Self-Consistency (SC) is a decoding strategy that samples diverse reasoning paths and selects the most consistent answer, demonstrating strong performance on complex reasoning problems. However, the excessive token consumption incurred by generating multiple reasoning paths has been identified as a major limitation of SC. To improve computational efficiency, several studies have proposed strategies that adjust the number of reasoning paths or allocate resources differentially according to problem difficulty. Nevertheless, most existing methods categorize difficulty into a few fixed levels, failing to fully capture the continuously varying nature of reasoning complexity. In this work, we propose Flexible Self-Consistency (FSC), which estimates problem difficulty as a continuous signal and dynamically adjusts the number of generated reasoning paths accordingly. FSC predicts the output entropy of an input question using a pre-trained probe and leverages it as an indicator of model uncertainty to flexibly control the sampling budget. Experimental results show that, across various models and benchmarks, FSC maintains accuracy comparable to SC while achieving token savings of up to 76%.

Large Language Model (LLM) agents increasingly solve long-horizon tasks through multi-turn interactions with users and external tools. In these settings, relevant task information often unfolds over time rather than being fully specified at the initial prompt. Service agents make this challenge especially concrete: users may clarify or revise their goals, while tool responses provide information needed for subsequent decisions. Thus, a final reward alone cannot indicate which actions contributed to resolving the task. Recent methods rely on comparative evidence from other trajectories or resampled continuations, or on separately constructed step-level learning signals, to refine credit. However, a completed rollout already records how information and errors flow between agent actions. We introduce Influence-Aware Policy Optimization (IAPO), which represents each rollout as a typed influence-dependency graph over trainable agent actions, with user and tool observations serving as evidence. IAPO converts support-use and failed-use structure into routing weights that redistribute the same trajectory-level advantage. Experiments with Qwen3-4B and Qwen3-8B demonstrate superior performance over multi-turn reinforcement learning (RL) baselines across three service-agent benchmarks: {τ^2}-Bench, UserBench, and AgentChangeBench. BFCL-v4 Multi-Turn further shows that these gains do not compromise multi-turn function-calling performance. This work advances the understanding of credit assignment in multi-turn user interactions and provides a principled approach to training service agents from sparse outcome feedback.

Automated high-density storage systems (warehouses, robotic parking, plant logistics, etc.) require fleets of agents to move through scarce task-critical resources and then park without obstructing future operations. We introduce Pivot-and-Station Multi-Agent Path Finding (PS-MAPF), a MAPF variant in which a subset of tasked agents must each visit one of a set of interchangeable pivots (e.g., workstations) before the entire fleet terminates at anonymous stations, one agent per station. We characterize solvability completely: every instance on a 2-edge-connected graph is solvable, and, on arbitrary connected graphs, a structural effective-distance measure relative to the number of unoccupied vertices gives a necessary and sufficient condition. We prove that minimizing station-makespan or station-flowtime is NP-hard already with a single pivot. We present three algorithms, a complete baseline, a SAT-based optimal solver, and Pivot-Prioritized Planning (PPP), the last solving 74-89% of benchmark instances with makespan and flowtime orders of magnitude below the baseline.

Video multimodal large language models support language guided video segmentation, but they often show spatio temporal inconsistencies, e.g., jitter, drift, and identity switches. These failures are more common when targets are partly hidden or when similar objects appear nearby.One likely reason is that current training lacks explicit spatial priors, which makes it difficult to maintain stable spatial identity and shape over time. We present PhysMLLMs, a training-stage prior injection architecture that injects physics-inspired spatial continuity priors into Video MLLMs. PhysMLLMs is designed to encourage more stable object-centered representations by aligning the student global visual representation with a frozen teacher model during training. Our core mechanism, Global Representation Prior Alignment (REPA-Global), distills global visual representations from a frozen DINOv2 teacher using an offline embedding cache and a scheduled distillation plan. This design keeps inference unchanged and does not add inference time cost. Across multiple video benchmarks, PhysMLLMs improves video segmentation mask quality and cross-frame consistency, with larger gains on challenging cases involving small targets, fast motion, occlusion, distractors, and reasoning queries. On single-frame referring image segmentation and representative general VLM benchmarks, PhysMLLMs maintains comparable performance, demonstrating that the injected spatial prior improves video consistency without compromising image-level grounding or general multimodal capability. These results suggest that physics-inspired spatial prior injection can improve temporal stability while preserving general capability. The code is available at https://github.com/tusu-code/20260121-icml2026-2.git.

Capturing hand pose during dexterous manipulation remains difficult: vision-based methods degrade under occlusion and challenging lighting, while sensorized gloves, though occlusion-free, are prone to drift and magnetic interference and rarely match motion-capture accuracy. We introduce a fiber optic sensing glove for full hand pose tracking that targets these failure modes, using multi-core shape-sensing fibers that capture each fiber's full 3D shape rather than curvature alone. A novel pipeline registers each reconstructed fiber shape to a common hand reference frame, and a new inverse-kinematics solver reconstructs full hand pose at 60 Hz using curve constraints. Benchmarked on a 2-hour dataset of dexterous object manipulation tasks across 5 subjects, the glove achieves 7.2 mm mean fingertip position error against motion capture ground truth, reduced to 4.9 mm by a one-time factory calibration of the fiber routing hub that transfers across users and sessions. These capabilities enable high-fidelity data capture and bimanual virtual teleoperation - both essential to advancing the robotics field.

Tool-augmented language models are bounded by the APIs humans bothered to write; existing tool-creation systems patch this by prompting a frozen LLM at inference time, leaving the model that writes a tool decoupled from the one that uses it, with no signal that the schemas it produces are schemas it can invoke. We propose SMITH (Schema-grounded Multi-task Iterative Tool Honing), a reinforcement learning framework that jointly trains tool creation and tool use inside a single policy. Each rollout is either a build task (write a tool from a few examples) or a use task (invoke a pooled tool on a held-out question). Three separate reward axes catch schema, code, and outcome failures independently, so each failure mode contributes its own gradient. A 4B Qwen3 trained with SMITH on 13 procedural reasoning tasks with exact verifiers reaches 79.8 macro-average accuracy on held-out tasks, the best across all evaluated methods and ahead of an untrained 30B-A3B tool-writer. It also reaches 40.4 on TabMWP-Hard and 42.6 on out-of-domain GQA (+7.6 over the best same-backbone inference-time baseline), without any visual or tabular training data. Tools written by our 4B models also lifted the performance of LFM-2.5-350M and Qwen3-30B-A3B under same reasoning tasks.

Clinical diagnosis is an active evidence-seeking process in which clinicians acquire evidence, update competing hypotheses, and decide when the available evidence is sufficient for diagnosis. Yet many medical diagnosis systems built around large language models (LLMs) still formulate diagnosis as static case-to-answer prediction, with limited support for evidence acquisition. Agentic LLMs offer a dynamic alternative through tool use and intermediate diagnostic trajectories, but existing systems often under-specify how patient evidence should be exposed, scaffolded, and controlled at runtime. We introduce EviDx, an evidence-aware active diagnosis framework that pairs patient-specific diagnostic environments with a clinical diagnostic scaffold and an observer-guided runtime harness. In EviDx, $\mathcal{E}$-Synthesis constructs interactive environments from raw clinical cases; the scaffold organizes role-specialized agents, evidence tools, and evolving evidence states; and the harness regulates diagnostic termination by tracking uncertainty and evidence coverage. A 3-level evaluation pyramid assesses execution robustness, reasoning dynamics, and diagnostic outcomes. Experiments show that EviDx improves diagnostic performance and process stability while revealing model-dependent capability boundaries.

Large language model (LLM) agents coordinate complex tasks through multi-role and multi-stage workflows. Upstream state is repeatedly transformed into intermediate language artifacts, such as summaries, plans, tickets, memories, and handoff notes, from which downstream components act. For action-constraining state, topical retention is insufficient: an artifact may mention an unresolved condition while changing it from a requirement that must be resolved before execution into information that may merely inform the next action. We study this action-binding role as operational state preservation. Safety blockers provide a controlled instance because each source state has an explicit prerequisite, authority, fallback, and execution consequence. We condition on correct upstream identification, vary the handoff transformation, and evaluate an executor restricted to the resulting artifact. Across 1,296 controlled synthetic episodes, direct-handoff controls preserve every blocker, whereas compression, plan assimilation, convergence, ownership deferral, and precedent substitution repeatedly turn binding state into caveats or non-binding considerations. Normal handoff compression produces 100.0% deactivation and 54.2% forbidden action. Restoring all four state fields raises preservation to 100.0% and reduces forbidden action to 0.0%. Fixed-artifact interventions further separate preservation from containment: downstream verification eliminates forbidden action while artifact deactivation remains 95.3%. These results identify a state-transmission failure between information extraction and action. Handoff transformations can retain state content while weakening its constraints on downstream action. Semantic availability does not guarantee operational preservation.

Rapid earthquake magnitude estimation is central to earthquake early warning, yet many operational systems depend on dense regional seismic networks and region-specific calibration. This creates a spatial coverage barrier for high-risk areas with sparse sensing infrastructure. Single-station learning offers a lower-cost alternative, but existing models often face an accuracy--latency trade-off and may degrade under regional distribution shift. We present SeisMamba, a lightweight Mamba-based architecture for low-latency magnitude estimation from minimally processed three-component seismic waveforms recorded at a single station. SeisMamba combines hierarchical convolutional encoding, sparse selective state-space modelling, multi-scale feature fusion, and an auxiliary temporal prediction head to support efficient long-sequence waveform analysis. On the STEAD benchmark, SeisMamba achieves the best MSE, RMSE, and $R^2$ among tested baselines while requiring only 0.55 ms for a batch of 32 waveforms on an NVIDIA T4 GPU, making it about three times faster than transformer-based baselines. We further conduct a Chile--Taiwan regional hold-out experiment as a diagnostic test of cross-region deployment, where SeisMamba retains useful performance on geographically unseen seismic regions. These results suggest that selective state-space waveform modelling provides a promising accuracy--latency backbone for spatially distributed, low-cost earthquake early warning.

Despite the critical role of grey literature in scholarly communication, artefacts such as Calls for Papers (CfPs) remain largely isolated from modern Scholarly Knowledge Graphs. The unstructured and highly heterogeneous nature of these documents has traditionally hindered their large-scale processing. In this demo paper, we present the Conference Organisers and Content Identifier (COCI), an AI-based framework designed to extract fine-grained, structured metadata from raw CfP texts. COCI employs a multi-stage pipeline that combines Large Language Models (LLMs) with semantic mapping techniques to integrate extracted entities with established knowledge bases, including OpenAlex, DBLP, TIB ConfIDent, and the AIDA Dashboard. By disambiguating authors and semantically aligning topics and conference series, COCI bridges the gap between informal scholarly dissemination and structured Semantic Web resources, laying the foundation for systematic analysis of non-publisher-based academic events.

Prehospital stroke assessment aims to accurately identify stroke symptoms and make rapid decisions through standardized procedures within an extremely narrow time window, thereby saving valuable time for subsequent treatment. In clinical practice, FAST-based scales are widely used for prehospital stroke assessment by issuing instructions that guide subjects to perform specific actions to screen facial, arm, and speech functions. However, in home and community settings, non-clinical users often encounter challenges such as inaccurate descriptions, incomplete symptom observation, and difficult operational procedures, which may lead to inaccurate or biased assessment results. To address these challenges, this paper presents StrokeGuard: a multi-agent guided system designed for prehospital stroke assessment that makes mobile FAST screening more standardized and executable. Specifically, to overcome the limitations of traditional single-agent systems in terms of procedural fault tolerance and user guidance capability, StrokeGuard adopts a dual-channel agent mechanism that separates formal assessment (i.e., facial palsy, arm weakness, speech impairment) from procedural support (e.g., step prompts, error correction, and real-time feedback). It guides the assessment process through multi-agent collaboration, dual-channel interaction, state-machine control, and stage-local fallback recovery mechanisms. Stage-specific scoring is delegated to constrained pretrained video assessment modules, while evidence source records are integrated with structured report generation. The user evaluation uses MATES-9, an exploratory scale for measuring user experience in multistep AI-guided tasks. In a simulated prehospital scenario, StrokeGuard improves the MATES-9 total score over a paper FAST-style form by 10.83 points, corresponding to a 23.8% relative increase.

Parliamentary systems pass more bills than presidential systems at baseline, but collapse to near-zero passage under party-system fragmentation. The literature offers three competing micro-explanations: coalition-formation failure, party discipline, and committee gatekeeping. These operate simultaneously in any real legislature, so observational studies struggle to separate their contributions. We present an agent-based model that compares four democratic institutions: pure parliamentary, pure republican/presidential, premier-presidential (France), and president-parliamentary (Russia). Across four scenarios and N=200 seeds per cell we report bootstrap confidence intervals, Morris screening, Sobol variance decomposition, mechanism ablations, and a hung-parliament variant comparison. Three findings emerge. First, government formation failure alone does not halt legislation: when a fragmented parliament reverts to personal voting, parliamentary passage (46.4%) is statistically indistinguishable from the presidential benchmark (44.8%); collapse requires cohesive opposition obstruction, which drives passage to 0.05%. Second, disabling discipline restores fragmented passage to 46.7%, and the rescue magnitude is monotone across the four institutions in a pattern that survives varying the common discipline level. Third, the passage-representation tradeoff is a single spectrum: parliamentary maximises throughput at the cost of representational fidelity; republican maximises fidelity via the presidential veto; semi-presidential variants split the difference.

Machine learning models are widely used in financial fraud and credit-risk detection, yet their adversarial robustness remains difficult to evaluate because financial tabular data involve domain-specific constraints, severe class imbalance, and asymmetric attacker capability. We argue that, in this setting, robustness is not only an attribute of the model, but also an attribute of the evaluation protocol. Different ways of enforcing constraints and capability can lead to substantially different robustness conclusions. This paper presents FraudBench, a protocol-sensitive benchmark for adversarial robustness evaluation in financial fraud and credit-risk detection. Rather than treating domain constraints as post-hoc validity checks, FraudBench evaluates the same dataset--model--attack--defence setting under three matched protocols: unconstrained attacks, post-hoc feasibility filtering, and deployment-aware constraint-integrated attacks. FraudBench covers four public financial datasets, and evaluates neural, tree-based, and ensemble models using three attack settings. Our results show that robustness conclusions are highly protocol-sensitive. On Lending Club Loan Data under the white-box setting, post-hoc filtering leaves only 3.7 feasible-flipped examples on average, whereas in-attack projection with attacker mutability masking produces 2,832.3 feasible-flipped examples under the same perturbation budget. The results on IEEE-CIS further show that feasibility and attacker capability are separate axes, while black-box evaluation shows that protocol choice can alter model-family rankings. These findings suggest that fraud robustness evaluation should report predictive degradation and attack feasibility jointly, and should incorporate domain constraints into attack generation rather than treating them as post-processing checks.

Simulation is central to modern engineering and science, but the cost of numerical solvers for partial differential equations (PDEs) remains a bottleneck whenever fast or many-query evaluations are required. Neural emulators trained on solver-generated data promise significant speedups, yet they are usually framed as opaque alternatives to the very methods that produce their training signal. This thesis argues the two paradigms are more alike than different: neural architectures mirror classical discretizations, their errors are amenable to the same spectral analysis, and insight flows profitably in both directions. We approach the relationship by disentangling the multiple roles a solver plays in the emulator learning pipeline. Mode-wise Fourier analysis then provides a common language in which solver errors, architectural inductive biases, and training objectives can all be read off simultaneously. Taken together, this allows synthesizing three contributions. (1) APEBench, a comprehensive benchmarking suite for autoregressive neural emulators of PDEs that uses fast differentiable pseudo-spectral solvers in JAX. (2) Progressively Refined Differentiable Physics, an investigation of the effect of unconverged solvers on surrogate training. (3) Neural Emulator Superiority, an analysis of the influence of numerical errors and architectural inductive biases.

Human collective intelligence depends on transmission processes: who shares what with whom, how, and when. While these processes emerge from individual cognition, they can also be directed by deliberate top-down protocols. Prior work has studied how transmission shapes collective outcomes primarily through the lens of network structure, varying who shares with whom and when. But networks are state-agnostic: they cannot condition transmission on what agents know or on the state of the collective. Here, we formalize transmission protocols as state-aware programs that route information and resources based on agent and collective states, and we use LLM-guided evolutionary search to design effective protocols in a collective discovery task. Evolved protocols increase collective performance over standard baselines from the literature by up to 37%. Ablations confirm that state-awareness drives this advantage: removing content-dependence while preserving network topology and timing eliminates performance gains. We find that evolved protocols also transfer across domain variations and agent populations. These results demonstrate that effective and generalizable transmission protocols can be discovered in silico, suggesting a path toward AI-assisted design of coordination infrastructure that enhances human collective intelligence.

Many feature-based visual-inertial odometry (VIO) systems rely on sparse feature tracking, whose accuracy and robustness directly affect state estimation. Classical KLT trackers rely primarily on local image patches and can become unreliable under rapid motion or in low-texture environments. We propose KLTNet, a lightweight learning-based, plug-and-play sparse feature tracker designed to replace classical KLT trackers in KLT-based VIO front ends. KLTNet follows a coarse-to-fine, dense-to-sparse architecture that combines low-resolution dense optical flow for robust global motion initialization with triplet-patch refinement for accurate and temporally consistent tracking. A fixed reference patch provides a stable anchor throughout each feature track and helps reduce accumulated tracking drift. In addition, KLTNet predicts anisotropic confidence weights supervised through differentiable multi-view triangulation, which can be used as observation weights in compatible VIO estimators. Experiments with VINS-Mono and OpenVINS on public benchmarks and a self-collected low-texture dataset demonstrate improved tracking and odometry accuracy over classical KLT, while maintaining real-time performance on an embedded platform.

Surgical instrument segmentation (SIS) is fundamental for computer-assisted surgery, where reliable instrument masks enable precise scene understanding and clinical assistance. Recently, adapting foundation models like the Segment Anything Model (SAM) to the surgical domain via prompt-learning has shown encouraging results. However, the performance of these adapted models under challenging surgical conditions is constrained by suboptimal adaptation mechanisms. Specifically, optimizing prompts or prototypes purely via downstream segmentation loss tends to cause them to degenerate into task-specific parameters rather than serving as persistent, stable category memory, thereby degrading their robustness against complex intraoperative variations. Moreover, routing multi-scale visual cues through a single prompt pathway creates a bottleneck that hinders effective scale-matched coupling. To address these limitations, we propose HPMA, a Hierarchical Prototype-Memory Adaptation framework for SAM. Specifically, HPMA constructs a frozen, multi-scale visual prototype memory bank from annotated surgical scenes and integrates it into SAM's feature space using lightweight adapters to preserve stable category evidence. To maximize the utility of multi-scale cues, we introduce a scale-matched coupling mechanism where global prototypes calibrate class-level prompt features, structural prototypes guide decoder object queries, and local prototypes align high-resolution feature maps through a local alignment objective. Extensive experiments on the public EndoVis2017 and EndoVis2018 datasets demonstrate that our approach achieves state-of-the-art performance, outperforming existing foundation model adaptation methods.

Data visualizations are widely used for communicating information, but they are also vulnerable to intentional manipulations that induce misleading interpretations. Existing methods focus on locating tampered regions or recovering hidden information, without explaining how the visualization has been manipulated or why the resulting changes may mislead viewers. We propose \textbf{VizAnchor}, a framework for visualization manipulation understanding through dual-anchor evidence construction and VLM-based reasoning. In the first stage, VizAnchor constructs a semantic anchor to recover authentic chart information and a spatial anchor to localize tampered regions. In the second stage, three specialized agents decode the manipulation. The misleader grounding agent analyzes a four-panel visual prompt to predict the misleader information. The chart narrative reconstruction agent takes the original and tampered charts as inputs and reconstructs their respective visual narratives. Finally, the intent inferring agent integrates the visual evidence and misleader information to infer the misleading intent. We further construct a dataset for tampering localization and a dataset for misleading intent inferring. Evaluation shows that VizAnchor accurately localizes manipulations and produces faithful explanations of their manipulation, misleaders, and misleading intents.

Clinical LLMs can generate recommendations that are factually plausible yet physiologically unsafe. We investigate whether safety alignment can be improved by grounding preference optimization in structured physiological knowledge rather than text-only supervision. Methods: We propose Neurosymbolic Alignment, a training-time framework that couples a 7B clinical LLM with an HGNN-based Physiological World Model over an 847K-node biomedical knowledge graph. Candidate responses are scored using homeostatic constraints, multi-hop path plausibility, and drug-interaction penalties, and the resulting rankings drive iterative on-policy ORPO updates. Evaluation is performed on the Clinical Safety Benchmark (CSB), a 2,500-scenario benchmark for physiological constraint violations in generative clinical reasoning. Results: Relative to ORPO, the proposed method improves CSS from 69.5% to 90.8% (+21.3 pp), reduces physician-evaluated HR from 14.1% to 5.1% on the blinded subset, and improves DID from 72.8% to 91.6%. These gains are corroborated by an HGNN-independent Rule-Engine Safety Score (RSS: 86.4%, +21.2 pp over ORPO; r=0.97 concordance with CSS). The method also exceeds GPT-4 (5-shot) on all safety metrics despite a 10x parameter disadvantage, and outperforms an inference-time self-correction pipeline (SFT+SelfCorrect) by 11.4 pp CSS. Under synthetic EHR-style noise, 84.2% CSS is retained. Ablation analysis shows that HGNN scoring (-16.2 pp) and iterative training (-11.5 pp) are the dominant contributors. PhysioScore calibration against 200 clinician labels yielded ECE = 0.038 and kappa = 0.91. Conclusion: Training-time physiological grounding produces measurable and independently verifiable safety improvements in open-weight clinical LLMs under controlled evaluation. External validation on real clinical data is required to determine whether these gains transfer to deployment settings

Probabilistic full-field reconstruction provides uncertainty-aware response evidence for structural reliability assessment, yet inference from sparse and noisy measurements remains underdetermined. Most existing methods overlook shifts between offline training and operational distributions. Under such shifts, posterior intervals may become miscalibrated, causing the reported uncertainty to lose its probabilistic meaning. This study proposes Modal Residual Flow Matching with Context-Conditioned Affine Spread Transport (MoRF-AST) for calibrated structural virtual sensing under changing operating conditions. MoRF constructs an analytic Gaussian reference posterior in normalized modal coordinates and trains a conditional flow only on posterior-whitened residuals. At deployment, AST estimates response scale from historical measurements at installed sensors and uses gated, mean-preserving Bures-Wasserstein transport to adjust posterior spread. On a bridge-deck benchmark, MoRF achieves a posterior-mean normalized root-mean-square error (NRMSE) of 7.20%, compared with 16.1% and 17.9% for two direct conditional flows. Across eight shifted traffic domains, AST reduces MoRF's cross-domain average coverage error from 0.0535 to 0.0236, a 55.9% reduction, while preserving posterior-mean accuracy. The same transport does not improve the tested alternatives in aggregate, showing that calibration gains require its direction to match the base posterior's dispersion bias. MoRF-AST provides a data-efficient framework for probabilistic full-field reconstruction whose uncertainty remains interpretable under scale-dominated operational distribution shifts. More broadly, this work highlights the need to calibrate uncertainty under changing operational distributions, thereby supporting trustworthy probabilistic modeling and reliability-informed decision-making in civil and infrastructure engineering.

Recent end-to-end driving systems demonstrate strong performance on closed-loop benchmarks, yet are still predominantly trained on fixed expert-collected data using open-loop imitation learning. This training-inference mismatch leaves the policy vulnerable in policy-induced states, where accumulated errors can lead to safety-critical failures. A promising post-training approach to overcome this issue is Dataset Aggregation (DAgger), which gathers expert demonstrations in policy-induced states and subsequently fine-tunes the policy on the resulting aggregated dataset. Existing driving DAgger pipelines, however, face three challenges: i) the expert is restricted to a limited trajectory-and-speed solution space, ii) takeover may occur too early or too late relative to impending failures, and iii) privileged expert decisions may rely on information unavailable to the student. To address this, we introduce RoG-DAgger, a post-training framework that uses short-horizon kinematic rollouts to construct high-quality expert demonstrations in safety-critical states. Specifically, RoG-DAgger expands the expert's trajectory-and-speed solution space and evaluates candidate plans through rollout to construct preventive supervision. Moreover, it uses rollout solvability to time the takeover near the estimated point of no return. Lastly, it aligns the expert's field of view with that of the student to provide student-compatible supervision. Across in-distribution (including long-horizon) and out-of-distribution evaluations, RoG-DAgger improves the end-to-end model SimLingo by 5.3 driving-score points and 6.2 percentage points in success rate on Bench2Drive, doubles its driving score from 22 to 44 on Longest6 v2, and improves out-of-distribution success rate from 55\% to 66\% on Fail2Drive.

Proactive medical dialogue requires an agent to decide what to ask from incomplete patient information. Existing information-seeking approaches commonly prioritize questions that most reduce diagnostic uncertainty. While effective for acquiring informative evidence, this criterion overlooks an important property of medical diagnosis: different diagnostic errors can carry substantially different consequences. Missing a severe condition may matter more than reducing uncertainty among less consequential alternatives. Question acquisition should therefore consider not only how informative new evidence is, but also how it is expected to affect the downstream diagnostic decision. To this end, we propose Expected-Severity-Risk (ESR), a consequence-aware question-supervision objective that values each candidate by its expected reduction in severity-aware terminal risk. Because questions must be selected before their answers are observed, ESR marginalizes over possible answers using train-only population statistics. Its rankings are then distilled into a prefix-only language policy, so next-question selection requires no teacher-side computation at deployment. Across three Qwen3-4B training seeds on DDxPlus, matched ESR supervision reduces mean high-severity diagnostic miss from .0645 to .0455 (-29.5%) and improves mean diagnostic accuracy from .9123 to .9320 while requiring only 0.14 additional questions per dialogue. Fixed-budget analyses show that the two objectives remain behaviorally distinct when question count is controlled, while a matched expected-0/1-risk control shows that severity-aware weighting improves the high-severity error profile beyond generic decision-aware supervision. These results support moving proactive medical dialogue beyond uncertainty reduction toward consequence-aware evidence acquisition.

LLM agents increasingly solve tasks by invoking multiple tools, where parallel execution is essential for low latency but difficult to manage safely. Existing agent benchmarks primarily evaluate tool selection, argument generation, and end-to-end success under mostly serial execution, largely overlooking valid parallelization and resource-constrained scheduling. This missing scheduling dimension creates a practical failure mode: serial execution is safe but slow, while resource-agnostic parallel execution is fast but prone to avoidable resource overflows. To address this gap, we introduce PeakBench, a benchmark of executable multi-tool workflows with execution-grounded dependency annotations and measured resource profiles. A central challenge in evaluating such workflows is attribution: failures and inefficiencies may arise from incorrect dependency planning, poor resource-constrained scheduling, or both. PeakBench addresses this challenge with a two-part evaluation framework that disentangles logical planning from physical scheduling, with dedicated metrics for each dimension. Using this framework, we show that strong logical planning does not reliably translate into safe or efficient execution under resource constraints. We further show that exposing resource information can reduce avoidable overflows and improve resource utilization, making PeakBench a useful testbed for diagnosing resource-aware agent behavior. Code is available at https://github.com/Czzzk/Staggering-the-Peaks.

The STAR (Student-Teacher Achievement Ratio) experiment (1985, Tennessee, USA) is a landmark hierarchical dataset designed to assess the impact of class size on student outcomes, with observations nested within classes. To encode class-level interventions in such hierarchical settings, we develop a complete, scalable, open-source pipeline for Hierarchical Structural Causal Models (HSCM) that bridges symbolic identification and practical estimation. Our approach integrates graph transformations, pyAgrum's do-calculus for automatic identification of causal effects, adaptation of symbolic expression into closed-form HSCM formulas, and numerical estimation from fitted local probability models. A key innovation is our adapted Abstract Syntax Tree (AST), which decomposes pyAgrum's identified formulas into independent density, expectation, and marginalization tasks, enabling parallel and scalable computation. We validate the pipeline on canonical HSCM motifs and benchmark scenarios with known ground truth, then apply it to STAR kindergarten mathematics outcomes. The results show that flat baselines (ignoring hierarchy) recover associations but fail to encode class-level interventions, and that symbolic identification alone is not enough for practical Hierarchical Structural Causal inference; scalable estimation and numerical stability checks are central parts of the scientific object.

Purpose: To quantify how evaluation annotations influence measured pulmonary embolism (PE) segmentation performance relative to model training changes, and to establish a human-referenced framework. Materials and Methods: This retrospective study screened 166 voxel-annotated CT pulmonary angiography cases from CADPE (n=91), FUMPE (n=35), and READ (n=40); 149 were included. A primary rater annotated PE by protocol, and a senior thoracic radiologist reviewed and revised all segmentations. Three additional raters at three centers annotated a 15-case subset. The label effect was measured by evaluating two pretrained nnU-Net models (nnU-Net-A, nnU-Net-B) against original and refined annotations. The model effect was measured by comparing the same architecture trained on different dataset combinations with annotations fixed. The benchmark model (nnPE) was trained with leave-one-dataset-out and pooled five-fold cross-validation. Four metric categories were analyzed with case-paired Wilcoxon signed-rank tests, Benjamini-Hochberg correction, and bootstrap 95% CIs. Results: Changing only the annotation increased mean DSC by 0.143 (0.122-0.166) for nnU-Net-A and 0.188 (0.163-0.213) for nnU-Net-B (both P < .001), whereas changing training-dataset composition changed DSC by 0.028. The label effect exceeded the model effect on CADPE and FUMPE and was 0.045 on READ. Within-mask attenuation SD fell in all three datasets after re-annotation (all P < .001). nnPE reached DSC 0.72 +/- 0.22 on pooled cross-validation but scored below all four annotators across 52 paired comparisons (all corrected P < .05). Conclusion: Evaluation annotations affected measured PE segmentation performance at least as much as model training choices. A human-referenced evaluation framework is publicly available for future study.

Automated parking commonly assumes marked slots and short approach maneuvers. Delivery and service vehicles, however, may need to reach an operator-specified pose in an irregular bounded environment from a distant start. Existing learning-based parking planners often rely on local observations, which can restrict long-range route reasoning. To address this problem, we present NeuralParker, a reinforcement learning-based hybrid planner for arbitrary-pose parking. NeuralParker encodes full-environment obstacle and boundary geometry in a target-relative vertex representation, allowing the policy to retain route-defining context throughout the approach. It further couples a learned curvature--length arc policy with an in-loop terminal ensemble that selects from diverse cubic Hermite connections using a curvature-regularized cost. We also establish factorial and long-range route-choice benchmarks to evaluate planning success and trajectory quality. Experiments on these benchmarks show that NeuralParker achieves higher planning success and better overall trajectory quality than the evaluated baselines, while ablation studies support the benefits of the target-relative global representation and terminal ensemble. Finally, a real-vehicle evaluation confirms that the planner transfers effectively to real delivery-vehicle perception at a working parking site, planning successfully at low computational cost.

Mechanistic Localization bridges mechanistic interpretability and post-training optimization by isolating critical parameters via interpretative approaches and then guiding parameter-efficient Supervised Fine-Tuning (SFT) in a ``locating-then-tuning'' paradigm. However, due to the retrospective nature of mechanistic interpretability, directly interpreting pre-SFT models introduces misleading conclusions. Specifically for novel tasks, initially identified neurons differ drastically from those governing the final model, introducing biases that actively disrupt SFT. To address this, we propose a forward-looking localization framework that accurately estimates the post-SFT interpretability state using only pre-SFT parameters and the target dataset. Theoretically, we model SFT as a continuous parameter evolution, leveraging Taylor expansion to rigorously bridge the post-tuning mechanistic objective with the pre-SFT model's dynamic gradients. Practically, we design dual-granularity (neuron- and component-level) localization pipelines. Extensive experiments demonstrate that our approach not only provides superior SFT guidance but also exhibits robust performance and temporal scalability across increasing model sizes. This work transcends the fundamental limitation of traditional interpretability-its inability to identify task-critical mechanisms before they are trained-pioneering a predictive frontier that unites mechanistic interpretability with targeted optimization.

Massively parallel simulation changes the data regime in which off-policy reinforcement learning (RL) is trained, challenging stabilizers designed for data-limited replay. Through controlled experiments across eight benchmark families, we show that these stabilizers are data-regime-dependent: parameter normalization helps with narrow replay coverage but restricts value fitting when data are abundant, while clipped double-Q can be relaxed in high-throughput manipulation. Age-biased replay weighting improves learning efficiency across regimes, especially with limited network capacity. Based on these findings, we propose WarpSAC, a regime-aware family of off-policy RL algorithms. WarpSAC uses Sample Weight Decay for efficient exploitation and provides two variants: WarpSAC-L (Norm ON, clipped double-Q) for data-limited CPU-scale training, and WarpSAC-A (Norm OFF, single-Q) for data-abundant GPU-parallel training. WarpSAC improves normalized score--step AUC over FlashSAC by 4.5% across nine CPU-scale environments and 23.1% across fourteen GPU-parallel environments. It increases UnitreeG1TransportBox-v1 success rate from 19.8% to 96.4%, improves mean normalized wall-time AUC on MuJoCo Playground by 19.1%, and achieves 36.4% faster sim-to-real deployment on Unitree G1 than FlashSAC. These results show that scalable off-policy RL should adapt its stabilizers to the available data regime.

Multilingual text embedding models enable cross-lingual transfer of knowledge across a wide range of NLP tasks, but their evaluation remains highly uneven across high-, mid- and low-resource languages. In this paper, we propose a two-dimensional framework, specifically tailored for analyzing multilingual embedding benchmarks under dataset scarcity, and apply it on the Slavic-language subset of the MTEB benchmark. The framework distinguishes between task-specific and cross-task evaluation, while jointly analyzing three complementary aspects: (1) ranking robustness, (2) model consistency, and (3) evidence strength. At the task-specific level, we evaluate the stability of model rankings under changes in ranking methodology and benchmark dataset composition. At the cross-task level, we assess the ability of models to generalize across diverse tasks within a language. To quantify the reliability of benchmark conclusions, we introduce an Evidence Strength Score that accounts for dataset availability, diversity, and robustness assessability. Our analysis reveals severe benchmark sparsity, with many Slavic language-task pairs relying on a single dataset or highly correlated benchmark collections, limiting the ability to draw robust conclusions. The cross-task analysis reveals a small group of highly transferable models, most notably llama-embed-nemotron-8b, multilingual-e5-large-instruct, and Qwen3-Embedding variants, that consistently perform well across Slavic languages and tasks. Overall, the results demonstrate that benchmark rankings and robustness conclusions must be interpreted jointly with certain notation of their evidence strength and highlight benchmark scarcity as a major obstacle to trustworthy multilingual evaluation.

We introduce \textbf{Ockhamareto}, a single-shot GRPO framework for unit-test generation and selection, based on the principles of \emph{Ockham's Razor} and \emph{Pareto Optimality}. Ockhamareto has two principal components: (i)~a \emph{Pareto-gated Bonus} that rewards only rollouts non-dominated in~(mutation, $-$\#tests) space, and (ii)~\emph{Token-level Segment Credit}, which attributes each test's marginal mutation kills back to the tokens of its unit-test block. On the \emph{UnLeakedTestBench~(ULT)}, Ockhamareto \emph{strictly Pareto-dominates} the strongest RL baseline~(\emph{MIST-RL}). Furthermore, it dominates on {\em each and all} optimization objectives, catching more bugs ($49.9\%$ vs $31.3\%$ mutation score at $N{=}5$), using \emph{fewer} tests ($2.60$ vs $4.67$ on average), thereby achieving $3.4\times$ the per-test trade-off improvement. The advantage is found in all four benchmarks~(\emph{HumanEval+}, \emph{MBPP+}, \emph{CodeContests}, \emph{TestGenEval-Lite}): Ockhamareto leads both mutation and coverage metrics on every one, always with the smallest suite. Ockhamareto also outperforms the state-of-the-art at all model scales, adding $+30$--$35$~pp mutation at 4B, 9B, and 27B model sizes. We also show that the knee point of the optimal trade-off between efficiency and effectiveness on the Pareto front is not correlated with obvious more easily computed proxy metrics, such as function size. This finding motivates the Pareto front computation; it is needed to identify this crucial engineering trade-off for each function under test.

Ternary transformers offer extreme memory and compute efficiency, but existing low-bit LoRA-based methods cannot directly fine-tune ternary weights. Current approaches either require dequantization, restoring low-bit base weights to higher precision to merge with adaptation weight, or update only quantization parameters, preventing a merged model that remains ternary. We propose ternary multiplicative adaptation, which represents discrete updates of ternary weights such as sign flips or zeroing through a low-rank Kronecker factorization into two small ternary matrices applied element-wise to ternary weights. This design is parameter-efficient and expressive, preserves the ternary domain, and supports direct merging without dequantization. Experiments on six models across language and vision, including ternarized LLaMA-3 1B and 3B and a ternary ViT-B/16, demonstrate that our method recovers much of the performance lost to quantization and outperforms strong low-bit and ternary baselines. Code is available at https://github.com/alexmanoo/ternary_adaptation.

Although recent Multimodal Large Language Models (MLLMs) have advanced general product understanding, they implicitly encode product information into global embeddings, thereby limiting their ability to capture fine-grained attributes. This limitation hinders performance in tasks requiring precise attribute discrimination, such as distinguishing subtle material differences among visually similar products. To address this challenge, we propose HMGCLIP, a unified multimodal embedding framework. By constructing a heterogeneous hypergraph, we leverage hypergraph topology to mine structure-aware hard negatives and align multi-granular semantics at both relation and hyperedge levels. This design enables a dual-granularity inference mechanism that dynamically fuses attribute evidence for both fine-grained and coarse-grained downstream tasks. Furthermore, we release a comprehensive fine-grained e-commerce dataset to facilitate future benchmarking. Extensive experiments on this new dataset and the public MAVE benchmark show that HMGCLIP outperforms strong multimodal encoders, MLLMs, and e-commerce baselines, validating the superiority of HMGCLIP.

Visual Language Models (VLMs) excel at describing visible scene content but struggle to reason about dynamic multi-agent interactions, where action semantics depend on coordinated roles and spatial-temporal dependencies. We formalize this capability as \textbf{multi-agent tactical reasoning} and introduce \textbf{DoublesEval}, a diagnostic evaluation framework that leverages professional doubles badminton as a structurally tractable testbed. DoublesEval employs a key-moment-based protocol that decomposes rallies into tactically salient instants and probes models across four interpretable dimensions: atomic recognition, intra-segment composite understanding, cross-segment causal reasoning, and high-level tactical abstraction. This design isolates \emph{where} reasoning fails, rather than merely measuring answer correctness. To address observed failure modes, we propose \textbf{TacticCheck}, a lightweight constraint-guided test-time consistency checker that reranks candidate answers using the model's own lower-level tactical predictions, requiring no parameter updates or ground-truth labels at inference time. Evaluating four representative open-source VLMs on 60 curated rallies (yielding $\sim$9.6K structured instances) via a zero-shot protocol, we find that models remain weak across all diagnostic levels, with especially clear bottlenecks in spatial state, interaction binding, and terminal evidence. TacticCheck delivers consistent gains across all evaluated models, while still leaving a substantial gap to robust tactical reasoning. These results highlight the need for structured, interaction-aware evaluation paradigms for next-generation VLMs. The source code is available in \href{https://github.com/Chengjt1999/DoublesEval}{\textcolor{blue}{our GitHub repository}}.

Wearable device data enables continuous health monitoring, but suffers from structured missingness: features sharing a physical sensor drop out together. Deep imputation methods such as BRITS and SAITS have seen limited evaluation on multimodal physiological data under realistic missingness, and existing benchmarks use random-point holdout protocols that incorrectly assume missingness is independent across features and time. Using data from a person with epilepsy recorded on a Garmin smartwatch, we develop an evaluation protocol that mines contiguous missing-run templates from training data, stratifies them by per-feature gap-length quantiles, and injects them as block masks with preserved co-missingness structure. A matched training protocol exposing models to the same missingness distribution reduces BRITS's severe-gap MAE by 43%, demonstrating the potential benefit of the proposed evaluation and training protocol within this single-participant dataset. We further extend BRITS with time-of-day encoding and a circadian harmonic channel. No single model dominates: linear interpolation is optimal for slow-moving features over short gaps; extended BRITS achieves lower MAE on dynamic cardiac features in moderate and severe gaps; and SAITS better preserves the ground-truth distribution by Jensen-Shannon distance despite higher MAE. Ultimately, model rankings strongly depend on evaluation designs. By exposing how traditional evaluation methods obscure true model capabilities, our transferable protocol establishes critical steps towards developing better imputation strategies for future multi-sensor wearable datasets.

Using monthly Niño-3.4 anomalies through July 2026, we investigate how much predictive information is contained in delayed observations of the index. Ridge regression identifies informative delays, while multilayer perceptron and sparse identification of nonlinear dynamics (SINDy) models test whether nonlinear complexity provides additional direct forecast skill; gated recurrent unit (GRU) and long short-term memory (LSTM) networks provide a complementary test in which the temporal representation is learned internally. Delayed observations substantially improve forecasts over persistence and climatology at leads of up to six months, but increasing model complexity provides no systematic improvement. Historical recursive experiments favor a simple explicit SINDy recurrence and select shallow recurrent architectures, with no appreciable gain from learning the temporal representation internally. These results support a compact predictive representation of Niño-3.4 evolution in which the representation of past information is more consequential than model complexity. As a prospective application, the selected models are used to forecast the developing 2026 event beyond the last available observation and to compare its predicted evolution with completed historical El Niño events.

LLM-as-a-judge evaluation is usually assessed by agreement and robustness to surface perturbations, but reliability does not establish construct validity. We formalize construct validity for an evaluator as a two-dimensional profile: invariance S, the probability that a verdict is unchanged under construct-preserving edits, and construct sensitivity R, the probability that it changes under minimal construct-changing edits. We show that S and R are independent and that no scalar summary preserves all relevant comparisons. We measure the profile across 7 judges and 4 domains using 7 construct-changing intervention types and 5 register-only controls, with intervention direction determined by human annotators and generation, verification, and judging assigned to disjoint model families. At matched invariance S >= 0.90, judges average S = 0.945 but R = 0.319. Sensitivity also differs between scope and strength edits: R_scope = 0.383 versus R_strength = 0.262, a +0.121 gap with the same sign for all 7 judges. We further audit five public label sets and find that surface-only predictors reproduce 55%-67% of labels in paired mode, including 67.4% of MT-Bench human votes. These results show that high judge agreement can coexist with weak sensitivity to changes in the construct being evaluated, motivating joint reporting of invariance and sensitivity and auditing the validation set itself.

AI-generated image quality assessment (AIGIQA) requires jointly reasoning about perceptual fidelity and prompt alignment, two quality dimensions that are often treated as independent in existing AIGIQA models. However, by re-examining human ratings, we uncover a previously overlooked phenomenon: the two dimensions are interdependent and exhibit both competitive and cooperative interactions during human rating. This observation suggests that a unified model should neither collapse the two dimensions nor rigidly separate them, but rather adaptively negotiate their interplay. Motivated by this insight, we introduce an interaction-aware learning framework that models perception-alignment relations through adversarial and collaborative inference pathways. Instead of designing a rigid dual-branch architecture, our method employs a gated interaction module that dynamically routes features according to the inferred relationship between the two dimensions. Task-aware prompts further modulate the gating behaviour, enabling the model to switch between competition and cooperation when necessary. Experiments across multiple AIGIQA benchmarks demonstrate that our approach not only achieves state-of-the-art accuracy but also yields interpretable interaction patterns, offering a more faithful approximation of human judgment. The codes are available at https://github.com/LQAMEI/ACL-IQA.

Reliable multi-turn tool use requires an agent to preserve an evolving task state and ensure that each action remains consistent with it. However, direct function-calling and ReAct-style policies learn state tracking and action generation within the same autoregressive trajectory. This coupling creates state-action competition: the pressure to produce the next call can overwrite or ignore information accumulated earlier in the interaction. Inspired by Boyd's Observe-Orient-Decide-Act cycle, we introduce OODA-Tool, a typed closed-loop policy designed to mitigate this competition by separating state preservation from action realization. Rather than generating an action directly from the interaction history, OODA-Tool routes each decision through controller-checked intermediate states, ensuring that the final output remains grounded in the current task state. Specifically, Observe reconstructs the task state, Orient determines whether execution is warranted, Decide forms an admissible action structure, and Act realizes the external output. We evaluate OODA-Tool against direct function-calling and ReAct policies using Qwen3 models ranging from 0.6B to 14B across multi-turn, multi-tool, and incomplete-information settings. OODA-Tool consistently improves task success across model sizes, with larger gains on smaller models and on tasks whose actions depend strongly on information accumulated across turns and prior tool results. Controlled variants, stage-level ablations, and transfer evaluations further demonstrate the robustness of these improvements.

End-to-end multimodal driving has progressed rapidly by fusing camera and LiDAR streams. Existing pipelines remain fragile under asymmetric sensor degradation, where either an entire modality or only a localized region is corrupted while other regions remain useful. The key difficulty is not simply to add an uncertainty head, but to obtain dense reliability supervision, calibrate this reliability against physical fault severity, and use it before unreliable features bias the planner. We propose Variance-Guided Spatial Attention Fusion (VG-SAF), in which dense heteroscedastic reliability estimates act as interpretable spatial gates. The framework couples three components. First, a physically grounded augmentor simulates representative camera and LiDAR failures and emits a continuous spatial mask, providing dense supervision without additional annotation. Second, modality-specific experts predict per-pixel reliability scales through cross-branch dense distillation in log space, enforcing a monotone severity-to-scale response. Third, calibrated reliability maps drive a hybrid attention mechanism that suppresses unreliable cells with a local spatial gate and arbitrates between modalities through a cross-modal trust softmax. A Laplace uncertainty head emits a systemic waypoint uncertainty scale that signals severe or combined sensor degradation, including severities outside the training ranges. On the CARLA Longest6 benchmark, VG-SAF consistently improves closed-loop robustness over the baselines across camera-only, LiDAR-only, and joint degradation regimes, as measured by driving score, route completion, and infraction score.

Transformer-based object trackers are renowned for their strong performance, yet dense token processing often leads to prohibitive computational cost, limiting real-time deployment on edge devices. While recent works explore token pruning to reduce computation, they often stop short of an end-to-end sparse pipeline, as early-layer token scores can be noisy without a motion prior, and many trackers ultimately fall back to dense reshaping to feed the dense prediction head that partially negates the savings. We introduce Motion-aware Sparse Tracker (MaST), a sparse tracking framework that makes sparsity effective from tokens to boxes. First, MaST injects a lightweight motion prior to refine cross-attention-based importance scores, enabling earlier and more stable token reduction in the search region. Second, we introduce a natively sparse prediction head that operates directly on the retained unstructured tokens with a score-first, regress-once design, eliminating dense padding/reshaping and reducing redundant computation. Extensive experiments on multiple benchmarks demonstrate that MaST establishes new state of the art among lightweight trackers, where MaST-tiny attains 63.8 AUC on LaSOT and 80.1 SUC on TrackingNet, surpassing the prior best AsymTrack-S by +1.0 AUC and +2.2 SUC while running at 152 FPS on Jetson Nano, nearly twice as fast as AsymTrack-S at 88 FPS. Code is available at https://github.com/TsingWei/MaST.

Multi-agent LLM systems are increasingly deployed in real-world applications, where failures can be costly and difficult to localize. Despite growing efforts to automate failure attribution, diagnosing failed runs still largely relies on human engineers. Yet engineers rarely debug complex systems by reading raw logs end to end. Instead, observability tools organize traces around components, actions, and dependencies to support targeted navigation. We hypothesize that modern LLMs can benefit from the same paradigm. To test this hypothesis, we introduce Adaptive Influence Graphs (AIGs), a two-stage agentic framework that first transforms a failed trace into a structured graph and then navigates it to identify the critical error. Across multiple models, we show that richer trace representations consistently improve failure attribution, with adaptive graph construction and agent-directed traversal yielding the strongest results. AIGs establish a new state of the art on Who&When, the standard benchmark for multi-agent failure attribution. This affirms our hypothesis that attribution depends not only on the diagnosing model, but also on how the trace is represented and explored.

Coding agents perform long-running tasks spanning dozens of model calls, tool uses, and code edits. As these runs unfold, users face a practical cost-quality trade-off: escalating to a stronger model when a cheaper one struggles, or downshifting once the hard reasoning is complete. Each switch requires the receiver to continue a non-native trajectory produced by another model. We study how this handoff affects quality and cost, and how varying the trajectory information inherited by the receiver changes the outcome. Using pairs of low-cost, low-capability (LC) and high-cost, high-capability (HC) models from the Claude and GPT families, we vary handoff direction, timing, and interface, comparing full-trajectory transfer, compaction, and trajectory removal while preserving the repository state. Across both model families, full-trajectory escalation recovers less than half of the LC-to-HC quality gap while incurring a substantial cost premium. We term this cost-quality penalty the handoff tax. By contrast, downshift offers a favorable cost-quality point. Interestingly, the preferred interface also reverses with direction: reducing LC-model trajectory information improves escalation quality, whereas removing the HC-model trajectory reduces downshift quality.

To reduce the hallucination risk caused by outcome-driven rewards in large language models trained through reinforcement learning with verifiable rewards, existing mitigation approaches introduce process-level factual supervision. However, due to coarse-grained aggregation of factual signals and the lack of reliability assessment for these signals, they create a mismatch between fact verification and policy updates. We term this noisy factual credit assignment and decompose it into two aspects: credit localization ambiguity and credit reliability ambiguity. To address these issues, we propose FARCA (Fact-Aligned Reliability-Aware Credit Assignment), a policy optimization framework that transforms factual supervision into localized, reliability-weighted token-level training signals. FARCA achieves fine-grained credit localization by aligning the granularity of fact verification with that of policy updates. It further introduces counterfactual evidence attribution, which uses the dependence of a factual judgment on key evidence as an empirical proxy for verification reliability to compute reliability weights. These weights modulate factual rewards and local policy advantages, reducing the influence of potentially unreliable signals on policy optimization. Experiments across different models and multiple factual reasoning benchmarks show that FARCA significantly improves model factuality while preserving general reasoning capabilities.

The prediction of student engagement from the online tutoring videos is difficult because engagement is a multidimensional construct comprising distinct behavioral, emotional, and cognitive states. A reliable prediction requires bringing together different types of behavioral signals as well as expressive cues. Through our analysis of the CASED dataset, it is clear that engagement prediction gets even harder due to the high inter-person variability as well as the subjectivity of the engagement annotation. To tackle these challenges, we develop a multimodal framework that integrates the implicit spatiotemporal features extracted from pretrained video, audio, and image encoders along with structured behavioral modalities like head pose, gaze, facial action units, emotion, and wavelet-based audio features. We integrate these modalities via a Perceiver IO latent bottleneck. Moreover, student and instructor personalities are modeled as variational posteriors over learnable embeddings to enable partial pooling across participants. We employ evidential regression and spectral-normalized Gaussian process classification heads for uncertainty-aware prediction to further improve robustness and calibration. Benchmark on the CASED challenge test set shows that all participating methods converge near random-chance performance, revealing the difficulty of the dataset. In this highly ambiguous regime, our framework achieves competitive performance while uniquely offering well-calibrated uncertainty metrics, demonstrating that reliable risk-quantification is an essential prerequisite for deploying engagement models in real-world educational tools.

Reliable underwater perception requires complementary sensing under variable visibility. Optical cameras capture appearance and semantics but degrade rapidly with turbidity, whereas imaging sonar preserves geometry while exhibiting distinct range-azimuth structure and acoustic artifacts. Existing MLLMs, built primarily on optical encoders, are therefore ill-suited to model sonar or adaptively exploit sonar-optical complementarity. We propose SonarLLM, a sonar-optical MLLM that treats sonar as a native perceptual modality. It combines a sonar-specific encoder, modality-specific physics-aware feature enhancement, and reliability-aware hierarchical fusion to align acoustic structure with optical semantics and dynamically adjust their contributions as sensing quality changes. We also introduce SonarBench, a paired benchmark that spans four tasks: recognition, counting, visual question answering, and captioning; and, across the benchmark, three input settings: sonar-only, optical-only, and fusion. By fixing the scene and sonar observation while varying optical degradation, SonarBench enables controlled measurement of cross-modal complementarity. SonarLLM achieves 72.0% macro accuracy across sonar-only recognition, counting, and VQA, outperforming the strongest baseline by 34.4 percentage points, and 68.7% under fusion, exceeding the best baseline by 25.1 points. For recognition and counting, the fusion-over-optical gain grows from 6.0 to 36.0 points as turbidity increases, indicating the increasing complementary value of sonar under controlled optical degradation. Together, these results show that robust heterogeneous perception depends not only on adding sonar, but on representing and weighting it according to its sensing characteristics.

Evaluating conversational voice agents at scale re- quires reliable assessment methods that capture both observ- able interaction quality and the contextual judgment typically provided by human evaluators. We investigate LLM-as-a-Judge evaluation by comparing human judgments with GPT-4.1 and GPT-5 on telecom and retail voice-agent conversations, across conversational quality and safety dimensions. The same interac- tions are scored under three evaluation configurations, p0, p1, and p2, to test whether automated judgments are sensitive to the evaluation setup and whether observed patterns generalize across configurations and judge models. Beyond aggregate agreement, we examine metric-level correlations, evaluator consistency, and systematic human-LLM disagreement to identify which conver- sational attributes can be judged reliably by automation and which remain sensitive to interpretation and context. Effective voice-agent evaluation is also shaped by pipeline-level factors such as speech generation, streaming, and error propagation across ASR, reasoning, and tool-calling stages, motivating our focus on comparing how human and LLM judges score the same interactions end to end. Our results show that LLM- based evaluation can serve as an effective component of large- scale voice-agent assessment, but that its reliability is metric- and configuration-dependent rather than uniform. This pro- vides an empirical framework for identifying which metrics suit automated evaluation and supports hybrid pipelines in which LLM judges handle scalable assessment while human evaluators remain engaged for metrics that demand contextual interpretation and higher-confidence judgment.

Search-augmented reasoning remains difficult for small language models. On-policy distillation (OPD) from trained teachers offers a promising direction, but suffers from two issues: (1) high-quality multi-turn search trajectories depend on dynamic retriever responses, making SFT data prohibitively expensive to collect at scale; (2) task-specifically trained teachers incur substantial training cost, while directly applying OPD with an off-the-shelf teacher without task-specific fine-tuning constrains the student to the teacher's performance ceiling and suffers from severe training instability. We propose OPDSearch+, the first distillation paradigm that requires no teacher fine-tuning for search-augmented reasoning. We investigate the role of a frozen off-the-shelf instruct model as the teacher in on-policy distillation, and reveal a key insight: the teacher reshapes the student's policy distribution so that subsequent RL converges to a superior solution that RL alone cannot reach. In stage one, the student interacts with a live search engine and is distilled via a per-position forward KL objective, transferring reasoning decomposition and evidence integration skills without any task-specific teacher training. In stage two, RL refines the distilled student from a richer behavioral foundation, achieving performance that RL alone cannot reach from scratch. Across seven QA benchmarks, OPDSearch+ with a 3B model consistently outperforms all prior 3B RL baselines, achieving gains of 13.1% on HotpotQA and 8.5% on 2WikiMultihopQA.

Deep research (DR) systems produce long-form cited reports by orchestrating multiple agents that search and synthesize information from the web. Citations are the primary mechanism for evaluating the faithfulness of these reports, yet current DR systems exhibit poor citation recall. Moreover, improving citation recall is challenging because DR systems are complex multi-agent architectures where information passes through agents like a telephone game, and both content and citations can get corrupted along the way. We propose an evaluation method that pinpoints which agent introduced each error by locally testing agent invocations for faithfulness and verifiability relative to their own inputs. Furthermore, we propose a four-type taxonomy to categorize the discovered errors: hallucination, uncited input reliance, uncited output, or insufficient citations. Applying our method to three top-ranked open-source DR systems, we obtain actionable diagnostics. Almost every agent makes a lot of mistakes with the exception being those that summarize a single document. We find that the dominant error type varies systematically across agents, where the orchestrator mistakes are mostly citation-related. We find that 84.7% of final-report errors in AI-Q originate at the orchestrator, roughly 31% of them hallucinations and the rest citation mistakes. Guided by these insights, we demonstrate that two simple interventions raise citation recall by 5% without degrading output quality.

Long-video understanding depends critically on how a limited model context is constructed from a much longer video. Existing approaches improve this process through compression, retrieval, memory, and agentic evidence acquisition, but these mechanisms are typically introduced as part of a manually designed inference system or optimized together with other components. This makes it difficult to isolate a simpler question: how much can be gained by improving the executable context-construction program alone? We study this question through VIDEOHARNESS-RSI, a controlled baseline for recursively searching executable context constructors around a frozen vision-language model (VLM). An outer-loop proposer uses prior programs, evaluation outcomes, and execution traces to generate candidate harnesses, which are executed and evaluated end to end before successful variants are retained for further search. This makes long-video understanding a controlled instance of automated harness design: the searchable object is executable program structure, while the answering model and interface remain fixed. Starting from uniform sampling, recursive harness search consistently finds room for improvement and surpasses several weaker hand-crafted baselines. Starting instead from a stronger hand-crafted baseline, the same RSI process yields a further improvement. The selected harness also transfers to additional long-video benchmarks without further search. Together, these results establish executable context construction as a distinct optimization layer and provide a reproducible baseline for studying harness discovery and transfer around frozen VLMs.

Reinforcement learning with verifiable rewards (RLVR) enables language models to learn multi-turn interaction with external tools, yet its sparse outcome rewards provide no signal for identifying which intermediate decisions are responsible for success. Branch sampling induces local comparisons among alternative continuations, but existing methods tend to conflate two distinct problems: allocating a fixed rollout budget and translating branch outcomes into token-level credit. We introduce Contrastive Branch Policy Optimization (CBPO), which disentangles these two problems and assigns a dedicated mechanism to each. Generation entropy screens candidate branch positions across the entire response, while path-level and node-level decay distribute a fixed budget across trajectories and positions to prevent exploration from collapsing onto a few paths or adjacent tokens. A parent trajectory together with the branches that share an identical token prefix forms an exact-prefix group, and the reward variation within this controlled group defines the Contrastive Branch Value (CBV), an outcome-based estimate of local decision sensitivity that rescales continuation advantages without altering their sign. When multiple nodes are selected along the same trajectory, CBPO partitions it into non-overlapping credit segments, thereby avoiding duplicated gradients on shared tokens. Requiring only outcome rewards and no process-level annotation, CBPO provides a practical solution for fine-grained credit assignment in tool-integrated agent training. Extensive experiments on ten benchmarks, including five for mathematical reasoning and five for knowledge-intensive search, show that CBPO consistently outperforms state-of-the-art policy-optimization and branch-based methods, attaining the highest macro-average accuracy in both domains and across two model scales.

Paper-to-code reproduction asks scientific AI agents to turn research papers into executable repositories that preserve the paper's method, protocol and artifacts. This is difficult because the specification is split: explicit paper content such as algorithms, metrics and artifacts is often lost across long agent trajectories, while implicit details such as framework defaults and conventions inherited from related work are absent from the paper. We introduce ReproAgent, a four-stage Prepare--Plan--Generate--Repair pipeline built around a persistent implementation contract with two channels: an implementation-requirement channel that turns paper snippets into code obligations, and a reference-evidence channel that retrieves content and structure evidence from related repositories. Both are bound to work packages, projected into file-level contracts, and consumed across generation and repair. On PaperBench Code-Dev, ReproAgent reaches the highest mean score among same-backbone scaffolds under both Claude-Sonnet-4.5 and Gemini-3-Flash. End-to-end channel ablations and per-paper cases support the contribution of both channels. Code and experimental artifacts are publicly available.

The widespread idealization of road surfaces as horizontal planes can introduce significant inaccuracies into vehicle dynamics simulations, a phenomenon termed the ``Flat Earth Error.'' This article provides an expository guide to the use of classical differential geometry to model vehicular motion on curved surfaces. The influence of road curvature is characterized using the metric tensor, the second fundamental form, the shape operator, and the Christoffel symbols. Reproducible MATLAB scripts using an elliptic cone as a benchmark example illustrate the construction of geodesic curves and the simulation of particle dynamics on curved surfaces. The resulting geometric structures are integrated into a single-track vehicle-and-track model within a pseudospectral optimal control framework. Trajectory optimization results over a high-density mobile LiDAR profile of Darlington Raceway are used to generate a high-fidelity road-surface model. This model is used within an hp-adaptive collocation framework to investigate minimum lap time optimal control vehicular trajectories. These computations capture the non-smooth traction saturation limits of the tyres alongside position-dependent variations in gravitational forcing. Integrating differential geometry, multibody mechanics, and optimal control is essential for high-fidelity driven and autonomous vehicle-dynamics simulations. The optimized velocity and tyre slip profiles show that real-world racing track geometries induce dynamically significant three-dimensional effects. These results are particularly relevant to performance-limited simulations on highly banked track surfaces such as NASCAR ovals.

Safeguarding language model agents requires assessing complete execution trajectories under context-dependent safety policies. Existing policy-aware safeguards mainly rely on prompting or supervised fine-tuning, limiting their ability to adapt to unseen trajectories and changing policy contexts. We propose RePolicy, an agent safeguard that learns safety-policy invocation through reinforcement learning. Given an agent trajectory and a dynamic policy library, RePolicy invokes the applicable policy and uses its content to produce a policy-grounded rationale and safety judgment. We construct PolicyTraj-20K to support supervised initialization, followed by GRPO with verifiable rewards and policy-context perturbation. Experiments across six agent safety benchmarks show that RePolicy achieves strong overall safety-detection performance and robust policy invocation under varying policy contexts.

Continual knowledge graph embedding updates entity and relation representations as a graph grows. Existing methods primarily address catastrophic forgetting, but entity admission also changes the candidate universe of every compatible query. A historical answer can therefore lose rank even when its score and its ordering among old entities are preserved. We formalize this effect as candidate-set interference and introduce Matched Excess-Outranker Regularization (MEOR), a host-level objective that compares smooth answer-relative newcomer pressure with score-blind, structurally matched old references. Its one-sided penalty acts only when newcomer competition exceeds the matched reference, preserving the host learner's signal for legitimate new entities. Across eight paired runs on ENTITY-ComplEx, MEOR improves historical current-universe mean reciprocal rank (MRR) by 0.0057 over replay and reduces candidate-set interference by 0.0055, with one-sided 95% lower bounds of 0.0052 and 0.0051, respectively. It satisfies the preservation criteria for old-universe ranking and newcomer acquisition and improves historical current-universe MRR over persistent calibration, matched maximum regularizer (MMR), and unmatched old regularizer (UOR). Direct ablations support each component of its reference construction and aggregation. Adding MEOR also improves historical ranking in all ten reported FBInc-S and FBInc-L host and backbone settings, with every paired 95% confidence interval excluding zero. These results establish candidate admission as a distinct source of continual rank loss and show that it can be controlled without replacing the underlying embedding architecture or continual learner.

AI systems are increasingly evaluated for legally accountable settings, where correct outputs must also be justifiable against an applicable legal standard. Existing legal-AI benchmarks and LLM-as-judge protocols provide important infrastructure for measuring task performance and open-ended response quality. We contribute one additional evaluation signal: a dual-judge protocol that pairs a standard 0-10 quality judge with a strict binary semantic-equivalence judge against a human-curated reference. We study a controlled, visually grounded regulatory task - UK traffic-sign interpretation, whose meaning is a codified question with a known reference for every input - and measure not merely whether the two judges disagree (by construction they must) but how much and where. On 4,680 evaluations under seven visibility levels and two occlusion modes, the two judges are moderately associated (point-biserial r = 0.644), while revealing an asymmetric Type II pattern affecting 8.0% of all evaluations. Its distribution is instructive: the marginal rate peaks at high visibility (14.2% at v = 0.8) simply because high-scoring answers are common there, but conditioned on the answer already scoring above 7, the rate is highest under heavy occlusion (54-63% at v <= 0.3), so a high quality score is least trustworthy when the input is most degraded. We are explicit that the signal is a property of this judge and reference: a 49-row human check shows the 0-10 judge aligns closely with everyday-reader judgement (Pearson r = 0.81; r = 0.80 with the LLM accuracy sub-score), while the equivalence judge is fairly but one-directionally stricter. The protocol adds one LLM call per evaluation and surfaces a signal single-judge protocols do not report. We release the prompt template, occluded variants, and full evaluation results.

LLM agents can generate paper reproduction code, yet often produce scientifically unfaithful implementations. We define this failure mode as semantic drift, where generated code silently diverges from the paper's specifications. We introduce SemanticAlign-Bench(SA-Bench), a diagnostic benchmark covering 30 papers from ICLR, ICML and NeurIPS 2025. For each paper, we decompose its specifications into atomic and verifiable implementation claims, which we call Semantic Alignment Units (SAUs) and evaluate repositories along four diagnostic dimensions spanning numerical, methodological, protocol and ordering drift. In total, we construct 1,491 SAUs across five ML domains and evaluate 12 generator configurations (4 models $\times$ 3 scaffolds). Even the strongest configuration (Claude+PaperCoder) achieves a mean SAU score of only 0.301 out of 1.0, with an overall mean of 0.221 across 360 evaluations. A failure taxonomy reveals that agents attempt most requirements but implement them incorrectly, with implementation mismatch and stubs accounting for the majority of zero-scored claims. Our analysis further indicates that scaffolds optimized for executability provide limited leverage for scientific reproduction; narrowing the gap requires scaffolds that prioritize semantic specification verification. The benchmark, annotations and evaluation pipeline are publicly available.

Currently available commercial vision-based tactile sensors provide rich contact information but remain vulnerable to abrasion and repeated concentrated loading, limiting their use in demanding robotic applications. This work presents a durable tactile fingertip comprising a soft silicone gel with a nonpigmented, textured, thin thermoplastic-polyurethane protective film and a replaceable sensing cartridge. Durability was evaluated using two accelerated laboratory procedures: a rotating-drum sanding test and a repetitive probe test applying 39.2 N (4.0 kgf) at 45 cycles per minute. Under the defined sanding conditions, the developed sensor reached the protective-film rupture endpoint after approximately 2-3 hours. During repetitive probe testing, all nine developed sensors remained functionally usable when testing was discontinued: seven after 5 days, one after 6 days, and one after 8 days. Commercial GelSight Mini and DIGIT specimens exhibited initial surface-film rupture after approximately 24-30 seconds of sanding and 25-35 minutes of repetitive loading. Damage to the developed sensor progressed gradually and produced little interference with tactile imaging at the test endpoints. These observations establish durability improvements of more than two orders of magnitude under the defined accelerated conditions. Combining increased durability, gradual degradation, and rapid cartridge replacement offers a practical approach to maintainable vision-based tactile sensing for demanding robotic applications.

Reproducibility of heart rate variability (HRV) analysis is limited by differences in preprocessing and computational conventions across software platforms. We developed HRV Studio, an open-source PyQt6-based desktop application integrating transparent HRV analysis with automated quality-control (QC) diagnostics. Validation included large-scale agreement with NeuroKit2, targeted Kubios benchmarking, spectral-method comparison, synthetic perturbation testing, recording-duration sensitivity analysis, and arrhythmia-focused QC stress testing. HRV Studio showed near-identical agreement for the widely used time-domain indices RMSSD and SDNN under matched conditions. In the primary five-minute NeuroKit2 comparison, frequency-domain median relative errors were 1.35% for LF, 0.18% for HF, and 1.41% for LF/HF, while VLF remained more convention-sensitive (37.79%). Nonlinear Poincaré indices also demonstrated high consistency. Sequence-harmonized Kubios benchmarking confirmed near-identical agreement for time-domain and nonlinear indices and strong agreement for most frequency-domain measures. Extended ten-minute analyses reproduced the same overall pattern with lower disagreement for some convention-sensitive spectral outputs. Synthetic and arrhythmia stress tests maintained 100% numerical stability while consistently triggering QC warnings. Overall, HRV Studio provides a transparent and reproducible platform for HRV research, with strong cross-platform consistency when NN sequences, preprocessing, and analytical conventions are harmonized. Stress-test results indicate computational robustness rather than clinical validation.

Longitudinal radiology report generation (LRRG) requires identifying both current findings and their changes relative to a prior study. Existing methods jointly model diagnosis, attribute estimation, temporal comparison, and language generation within implicit representations, which can cause task interference, obscure the evidence underlying each decision, and limit error traceability. They also model progression states as independent labels, ignoring their ordered structure and thus treating missed changes and direction reversals equally. We present STRIVE, Multi-Agent Structured Temporal Reasoning with Integrated Verification for LRRG, which decomposes clinical reasoning into specialized Diagnosis, Attribute, and Temporal Change Agents that produce explicit intermediate evidence. In particular, the Temporal Change Agent is further post-trained using Progression-Aware GRPO, a verifiable, shaped reward that assigns partial credit to direction-preserving errors while scoring direction reversals lowest. STRIVE performs verification at two stages: a deterministic Consistency Gate reconciles the agent outputs before report generation, and a Validation Agent checks whether the generated report is supported by the aggregated clinical evidence. On Longitudinal-MIMIC, STRIVE attains the best clinical efficacy among recent methods and more than doubles Longitudinal Change Concordance (LCC), a measure of temporal agreement with the reference report, over the strongest baseline.

Large Reasoning Models (LRMs) generate intermediate reasoning traces that may contain unsafe content, even when their final responses appear safe. Guardrail models are designed to detect and block unsafe content, yet existing benchmarks for unsafe content detection focus primarily on prompts and final responses, leaving reasoning traces largely unexamined. Moreover, these benchmarks typically provide only binary safety labels, without evidence annotations that justify the judgments. To address these limitations, we introduce TRACE, an evidence-grounded safety evaluation benchmark that covers the entire LRM inference pipeline: prompts, reasoning traces, and final responses. TRACE includes prompts in two languages spanning nine risk categories and ten attack strategies. For each prompt, four LRMs generate reasoning traces and final responses, and we annotate the safety of each component and extract supporting evidence from the corresponding source text. Evaluating 18 guardrail models on TRACE reveals that safety judgment for reasoning traces is substantially more challenging than for prompts or final responses, and that current models struggle to accurately extract supporting evidence. These findings highlight the need for guardrail models that can reliably detect and precisely localize unsafe content across the LRM inference pipeline.

LLM-as-judge is essential for evaluating open-ended text and steering post-training, yet improving the judge itself typically relies on expensive annotations, reward models, or distillation from stronger teachers. In this work, we eliminate external gold supervision from the RL training reward: the model's own evaluative capability generates learning signals for its optimization -- a closed-loop setting of bounded recursive self-improvement (RSI) termed Recursive Self-Evaluation (RecurSE). We study two central questions: when can self-improvement occur, and when must it stop? First, RecurSE pairs a trainable judge evaluating candidate responses under per-rule rubrics (Pass 1) with a synchronized policy-copy checker that audits the judge's reasoning against meta-rubrics to supply a scalar process reward (Pass 2). To enable learning, interface decoupling structurally isolates the checker's scalar score from the judge's verdict tokens, eliminating a degenerative token-copying shortcut that inflates self-assigned rewards. Second, because unanchored recursive learning is inherently bounded, Pairwise Advantage Validity (PAV) serves as an unbiased validation monitor that jointly tracks judge accuracy and checker fidelity to reliably identify the optimal early-stopping window. Across Qwen3.5-9B, Gemma-4-E4B-it, and Qwen3.6-27B, RecurSE achieves consistent generalization gains across held-out medical, pairwise, summarization, and professional benchmarks. Ablations demonstrate that synchronized judge-checker co-evolution outperforms frozen checkers, external meta-judges, self-consistency, and scaled teacher distillation. Furthermore, preference pairs curated by our judge effectively enhance downstream policy alignment. Bounded RSI for LLM-as-judge is thus viable when self-produced reward validity is explicitly decoupled and monitored.

Many LLM applications are most useful when they provide several candidate outputs for comparison, validation, or combination. Predominant evaluation settings, however, still focus on individual outputs or reduce multiple samples to a single success or selected answer. This can miss whether the outputs include several genuinely different useful results. We introduce VTC-Bench, a five-domain benchmark for this setting, together with Validated Task Coverage (VTC) as its core evaluation quantity. The benchmark is built from carefully selected real-data tasks where both output quality and task-relevant distinctness can be checked automatically and reproducibly, without model-based judges. VTC measures how many distinct useful results are obtained within $k$ attempts. Across multiple models and inference settings, the benchmark leads to different conclusions from conventional evaluation: configurations that look strongest from single-draw quality are not necessarily those with the best coverage, and simple measures of output variation do not reliably recover task-relevant coverage. These results show that finite candidate sets can be evaluated directly as objects of interest, revealing differences in model behavior that are not apparent from conventional per-output evaluation.

Answering developer questions about a software repository is a critical yet under-explored problem in software engineering. While existing repository understanding methods have advanced the field, they predominantly rely on surface-level code retrieval and lack the ability for deep reasoning over multiple files, complex software architectures, and grounding answers in long-range code dependencies. To address these limitations, we propose DeepRepoQA, a novel question answering (QA) framework for repository-level code understanding. DeepRepoQA builds on an agentic framework where LLM agents find answers through a systematic tree search over the repository structure. A Monte-Carlo Tree Search (MCTS) mechanism is employed to empower agents to dynamically search, navigate, and inspect code, enabling effective multi-hop reasoning over long-range code dependencies. Comprehensive experiments on the SWE-QA benchmark demonstrate substantial performance gains over strong baselines, validating the effectiveness of systematic MCTS-guided exploration for multi-hop repository reasoning.

Enterprise entity alignment must handle semi-structured records, implicit attributes, and unit or granularity mismatches. Manual matching is still common in practice, but does not scale as schemas and providers evolve. LLM-only matching improves semantic recall, yet can violate structural and physical invariants, producing fluent yet operationally invalid correspondences. We propose constraint-guided mapping (CGM), a neuro-symbolic method with three stages: (i) schema-grounded admissibility constraints with metadata mc = <tau_c, delta_c>, where tau_c denotes the constraint type and delta_c provides executable relation and normalization logic; (ii) constraint-restricted candidate generation with cascade relaxation to guarantee a nonempty feasible set under noise; and (iii) neural ranking with bounded LLM disambiguation restricted to that feasible set. Methodologically, constraints operate as hypothesis-space operators rather than post-hoc validators, enabling controlled degradation under relaxation and auditable, human-guidable decisions. On a controlled structural-decoy benchmark, hard admissibility shrinks the candidate space by ~480x without dropping the GT, and a layer-by-layer ablation shows this gate, not the LLM, is the decisive lift (F1 0.08 to 0.66). The benefit is model-independent and adds no extra inference cost: a small model with constraints matches a frontier LLM used without them at ~28x lower cost. The method, not a single tuned configuration, transfers across seven enterprise makes (macro F1 0.70), each under its own automatically discovered, expert-refinable constraints, and lowers expert effort by ~7x versus spreadsheet workflows. Public Valentine results add an external ranking sanity check and mark the boundary: constraints should be hard only where structural invariants are match-determining.

We propose CARO, a contact-agnostic residual observation framework for policy adaptation. CARO embeds a fixed-base Euler--Lagrange model into the reinforcement learning control loop and constructs a torque-level residual observation without requiring torque sensors, explicit contact estimation, or vision-based measurements of the floating-base position and linear velocity. A disturbance observer extracts a structured signal representing dynamics mismatch, while the policy learns to exploit this feedback for online adaptation. CARO is trained under the same terrain, command, and domain-randomization conditions as the nominal policy, without specialized disturbance curricula or additional adaptation supervision. Nevertheless, it achieves substantially improved zero-shot robustness in simulation and sim-to-real transfer tasks involving out-of-distribution payloads, center-of-mass shifts, terrain geometries, abrupt dynamics changes, and elevated-platform landings.

Large language model (LLM)-based multi-agent social simulation has demonstrated compelling results, but Agentopia was evaluated with 100 agents over 10 simulated years using Qwen3.5-397B-A17B, leaving the behavior of reduced-scale deployments on consumer hardware unclear. In this paper, we implement and evaluate a reduced-scale Agentopia port on a single NVIDIA RTX 5070 Ti(12 GB VRAM) using Qwen3-8B-AWQ, a 4-bit quantized model. We introduce three structural adaptations for this setting: (1) system-managed layered memory compression, (2) four activity blocks per simulated day, and (3) explicit physical- and mental-health state variables. Across three independent stochastic runs, two runs completed 52 weeks and the third completed 50 weeks before reaching the context limit, totaling 154 system-weeks (770 agent-weeks). No agent died,and no threshold-based health warning was logged; activity records containing at least one NO_RESPONSE field occurred at rates of 10.15-10.29% across runs. A 52-week memory-off run tied L2/L3 artifact production to layered memory; a separate 10-week comparison associated four daily time blocks with 2.72 times more finalized records and lower lexical duplication, but a higher missing-field rate. These comparisons do not support causal behavioral claims. We release validated configurations, derived audits, analysis scripts, aggregate figure data, and our implementation changes in a public fork; raw runs and initial persona data are excluded because their redistribution provenance is not fully resolved.

Agentic retrieval-augmented generation (RAG) requires language models to decide when to continue searching and when to answer. Existing RL-based methods rely on external supervision and overlook the agent's internal belief about whether the current evidence is sufficient. To address this problem, we reformulate the search decision quality as belief-action alignment and propose MetaRAG, a belief-action aligned policy optimization framework for agentic RAG. MetaRAG uses Verify-first Action Generation to elicit an explicit verification process before each actual action, and Internal Belief Probing to estimate the policy model's own answerability belief from the same question-history context. Based on these, MetaRAG derives a consistency reward that is further gated by answer correctness, avoiding reinforcement of internally consistent but incorrect trajectories. The belief probe is used only during training and introduces no inference-time overhead. Experiments on seven public QA benchmarks show that MetaRAG consistently improves the accuracy-efficiency trade-off over strong RL-based agentic RAG baselines, with gains that transfer to deep research settings, different optimizers, and multiple model backbones.

The advancement of Embodied AI necessitates high-quality simulation assets that faithfully mirror the real world. However, transforming raw visual observations into simulation-ready scenes remains challenging due to the lack of physical grounding and scene-level interactivity in current image-to-URDF methods. We propose NeoWorld-Pro, a framework that reformulates monocular scene reconstruction as procedural programming for interactive 3D environments. Leveraging the zero-shot reasoning and code synthesis capabilities of MLLMs, NeoWorld-Pro converts a single RGB image into executable programs specifying object geometry, articulation, and physical properties. A physics-in-the-loop mechanism then iteratively refines the generated programs by validating their execution in a physics engine, enforcing physically plausible articulations, valid object compositions and interactions, and accurate spatial relationships. Experiments show that NeoWorld-Pro outperforms open-loop and prior monocular reconstruction methods, while enabling complex downstream tasks such as stable stacking and fine-grained manipulation.

Semantic identifiers (SIDs) represent entities as hierarchical token sequences for generative retrieval and recommendation. Residual-quantization tokenizers construct these sequences by selecting a codeword at each level and passing a residual to the next. We view this process as progressive commonality removal: each token captures a component shared within its group, while later tokens should model the remaining differences. This view reveals three limitations: a corpus-wide shared component can consume first-level capacity, hard assignment ignores graded similarities to nearby codewords, and full-codeword subtraction can leave variation along the selected-codeword direction in the next residual. We therefore develop our solution in the post-hoc setting, where residual construction is not constrained by input reconstruction. Specifically, we propose PRQ-KMeans, which removes the global-mean component, refines centroids with Top-k similarity-weighted updates, and replaces full-codeword subtraction with a projection residual that removes each representation's selected-centroid component. Experiments on a large-scale industrial search dataset and four public recommendation benchmarks show that PRQ-KMeans achieves the strongest overall performance among the evaluated tokenizers, including gains of up to 7.4% in HitRate and 11.8% in MRR on the industrial dataset.

Generative simulation for surgical robotics still lacks real-time interaction. Physical-robot experiments, often involving animal or cadaver labs, are time-consuming, costly, and difficult to reproduce, while classical simulators struggle to capture photorealistic appearance and deformable-tissue dynamics. We address this gap with Cosmos-H-Dreams, an integrated real-time surgical world-model system combining an action-conditioned generative model, a teacher-to-student distillation recipe, and a deployment stack built on the NVIDIA FlashDreams streaming-inference library. Starting from Cosmos-H-Surgical-Simulator, a multi-embodiment action-conditioned surgical video world model fine-tuned on the large-scale Open-H-Embodiment corpus, we post-train this checkpoint on embodiment- and procedure-specific data. By distilling the resulting bidirectional teacher into a causal, few-step student with Self Forcing, we turn a passive video generator into a controllable surgical simulator that streams at $\sim$160 inference FPS on a single NVIDIA RTX PRO 6000 Blackwell workstation GPU. Crucially, Cosmos-H-Dreams is controller-agnostic: any interface that emits a stream of robot kinematics can drive it. We demonstrate live control through a browser keyboard over WebRTC, a Meta Quest headset over WebXR, a commercial surgical robot console such as CMR Surgical's Versius, and learned policies operating in closed loop. To our knowledge, this is the first interactive surgical world model supporting live human and policy control. Human operators and policies alike can act inside the synthesized world and observe the consequences in real time. We release Cosmos-H-Dreams as an open surgical simulation system, providing a common foundation for surgical education, scalable synthetic data generation, and future intraoperative decision support.

Mixture-of-Experts (MoE) models provide a flexible framework for partitioning complex prediction problems into simpler local learning tasks through an input-dependent gating mechanism. Existing interpretable MoE approaches, such as Mixture of Decision Trees (MoDT), achieve transparency by employing homogeneous decision-tree experts, but this restricts the model to a single inductive bias across all regions of the feature space. We extend the MoDT framework by introducing heterogeneous expert families comprising decision trees, linear support vector machines, and quadratic discriminant analysis under a common probabilistic gating mechanism. To ensure coherent likelihood-based inference, non-probabilistic experts are calibrated to produce conditional class probabilities, allowing parameter estimation within the generalized Expectation-Maximization framework of MoDT. We further establish theoretical monotone ascent guarantees for the proposed heterogeneous gating updates, providing a justification for the optimization procedure. Experiments on a diverse collection of synthetic and real-world benchmark datasets demonstrate that the proposed framework adaptively specializes experts according to local data geometry, yielding interpretable expert assignments while achieving predictive performance competitive with homogeneous MoDT and Random Forests. The proposed approach combines interpretability, adaptive inductive bias selection, and probabilistic coherence within a unified mixture-of-experts framework.

Memory systems for conversational LLMs are conventionally evaluated by direct, fact-seeking questions about prior dialogue (Direct QA): can the model recall fact X from a prior conversation? We tested whether higher Direct QA accuracy correlates with higher user satisfaction in a 4-month deployment (40 users, 1,872 sessions, 7 memory conditions). Existing-benchmark Direct QA varies from 19.7% to 70.1% across the 7 conditions, but satisfaction does not change. We hypothesize that existing benchmarks and user satisfaction are tracking different capabilities: benchmarks measure elicited retrieval (recall when asked), while conversation requires natural integration (detecting relevance and naturally weaving prior context into a response). To examine this, we introduce MemUse, a set of real user-cued memory moments drawn from the deployment, scored by an integration-aware judgment of the natural conversational response. Holding the model and context fixed, the same system that scores 78.8% on Direct QA references only 7.9% of those facts in conversation -- a 71-point gap. Within these moments, Natural Integration is associated with satisfaction, whereas Direct QA is not. We release the deployment corpus and MemUse together with all judgments and scoring prompts at https://github.com/ryuichi-sumida/memuse.

Coding agents re-send large file reads and tool outputs to a frontier LLM every turn, and this context dominates their token bill. General-purpose prompt compressors are trained on prose and suit code poorly: they paraphrase identifiers and drop the exact spans an agent needs to edit. We present Paritok-4B, a 4B LoRA compressor for coding-agent trajectories built on two commitments. It is extractive: it selects spans rather than rewriting them, and 96.0% of the identifiers, paths, and numbers it emits already appear in its input, holding at 96.2% on held-out SWE-bench Lite output. It is intent-conditioned: told the agent's current task, it acts chiefly inside a retained segment, selecting which lines survive (retained lines are +0.067 more intent-relevant than removed ones, paired 95% CI [+0.056, +0.078]) rather than changing how much is retained. We distil a gpt-4.1-mini teacher over 67,074 real OpenHands trajectories into 40,606 validated examples and fine-tune Qwen3-4B. On all 300 SWE-bench Lite instances, Paritok-4B compresses agent context to 25.7% of its size, 2.0x harder than a gpt-4.1-mini compressor (50.2%) and 2.4x harder than gpt-5 (61.9%), while retaining 86.5% of uncompressed single-shot solve quality. Fed the cat -n line-numbered input real agents produce, it compresses slightly less (27.8%) and retains more (89.3%); there the paired test is informative, with 30 instances solved only uncompressed and 17 only compressed, an exact McNemar p=0.079, so at this sample size compressing context to roughly a quarter of its size does not significantly reduce the solve rate. The model is a 264 MB adapter that self-hosts on one 24 GB GPU with no per-token compressor fee, which at list prices decides the economics: gpt-5 as a compressor is net-negative, costing more than the downstream tokens it saves. Weights, data, and evaluation scripts are open (Apache 2.0).

3D geometry editing is a critical yet labor-intensive part of the graphics pipeline, requiring artists to translate creative intent into precise operations in complex professional software. Large language models (LLMs) have shown promise for script-based 3D creation, but script generation is less suited to perception-driven editing of arbitrary existing meshes, where execution must remain visually grounded and untouched regions should be preserved. We present a \emph{visual-centric}, training-free multi-agent system that edits existing 3D meshes directly in Blender by emulating the iterative workflow of human artists. Rather than generating scripts or regenerating geometry, our system operates through the Blender GUI: multimodal LLM agents observe the viewport, reason about the current mesh state, and execute localized edits through simulated user interactions. Experiments on a curated benchmark provide initial evidence that this agentic approach can follow natural language instructions, perform representative localized mesh edits, and preserve the overall identity of the input asset. Our results highlight a complementary regime for language-driven 3D editing: direct in-place modification of existing meshes within the native 3D editing workflow. We view this work as an exploratory step toward visual-centric agentic geometry editing in professional graphics software.

A unified audio model must recognize and understand linguistic, paralinguistic, and environmental information while supporting speech synthesis and editing. A key challenge is representation: understanding favors compact features suited to long-context modeling, whereas speech generation requires reconstructible features that preserve fine-grained acoustic detail. We introduce FireRedAudio, a general-purpose audio language model with a shared 9B-parameter LLM. To the best of our knowledge, it is the first publicly disclosed unified audio-language model to provide separate continuous input representations for understanding and generation within a single trainable autoregressive LLM. Audio to be recognized or analyzed is processed by a dedicated Audio Encoder, while speech inputs for generation use a RedAE-based pathway. The LLM directly generates text or conditions a flow-matching DiT to produce continuous acoustic latents. Through progressive multitask training, FireRedAudio supports ASR and audio understanding, with the latter extending to recordings of up to one hour, as well as zero-shot TTS, Instruct TTS, and semantic and acoustic speech editing. Its structured organization of long-form audio achieves second-level timestamp accuracy. Across comprehensive evaluations, FireRedAudio achieves competitive or leading performance in audio understanding and multilingual ASR, strong content accuracy and speaker preservation in zero-shot TTS, leading instruction following in Instruct TTS, and substantial improvements over Ming-UniAudio-Edit in both semantic and acoustic speech editing. These results demonstrate the viability of decoupled continuous input representations for unifying audio understanding and continuous-latent speech generation in a model of moderate scale. Our code is available at https://github.com/FireRedTeam/FireRedAudio.

Whittle index policies offer a scalable method for restless multi-armed bandits, but under partial observability even determining the indifference subsidy at a single belief requires solving an infinite-horizon belief-state problem with no closed-form value function. Liu [10] addresses this difficulty by linearizing the unknown decision boundary, leading to a linear system and a closed-form approximate Whittle index. However, the resulting threshold uses only a one-step active--passive comparison and does not account for longer-horizon continuation values. We extend this framework to a \emph{$t$-step lookahead threshold policy}. For each subsidy $m$, the threshold is defined by the active-minus-passive advantage under $t$-step finite-horizon value iteration. At $t=1$, the threshold is $m$-independent and recovers the linear threshold of Liu [10]; for $t>1$, it becomes subsidy-dependent through the induced first-crossing structure and tracks the exact decision boundary more closely. The proposed algorithm does not require indexability as an input and includes an indexability verification. Under the original Whittle indexability, we prove that the $t$-step approximate Whittle index converges geometrically to the exact Whittle index, \[ |\widehat W_t(ω)-W(ω)|=O(β^t). \] Numerically, all 2,715 tested three-state instances are verified as indexable according to the proposed criterion. The P95 index error decreases from $2.18\times10^{-2}$ at $t=1$ to $8.93\times10^{-4}$ at $t=8$. In an exact-comparable instance with $β=0.9999$, $t=2$ already recovers the exact Whittle-index ordering. Moderate-depth threshold policies also outperform the one-step baseline and remain close to the optimal dynamic-programming benchmark, while runtime grows mildly with $t$.

Non-verbal vocalizations (NVs), such as laughter, coughs, and sighs, are essential for expressive TTS, but the effectiveness of preference optimization for NV generation remains poorly understood. We systematically study preference optimization for NV-capable TTS, focusing on preference signals, preference-pair construction, and DPO-based optimization objectives. We formulate an NV-aware character error rate (NV-CER) by treating NV tags as distinct output symbols and computing a weighted pinyin-based CER over both verbal and non-verbal content, enabling controllable optimization of NV realization without modifying the underlying optimization algorithm. Experiments on Emilia-NV and the augmented NV-Bench covering 18 NV types reveal how different design choices affect NV realization and lexical fidelity, and establish an effective setup using standard DPO. Objective, LLM-based, and human evaluations provide converging evidence for our findings, offering practical insights into NV-aware post-training for expressive TTS.

Tactile sensing is central to robotic manipulation, among which slip detection stands out as a quintessential and critical task. However, existing slip datasets are predominantly limited to binary classification, lacking fine-grained directional perception. To address this limitation, we propose a visuo-tactile sensor featuring customized uniform RGB illumination, alongside a unified perception framework. At the hardware level, the sensor achieves high-precision, sub-millimeter depth reconstruction. Based on this capability, we collect a multi-task visuo-tactile dataset encompassing 15 objects, synchronously generating depth information for each data sample. Algorithmically, we design a dual-head TimeSformer network to process dynamic spatiotemporal slip. On unseen objects, this network achieves robust accuracies of 95.5% and 91.5% for 3-class contact state prediction and fine-grained 8-class slip direction classification, respectively. Furthermore, static tactile-based object class recognition utilizing a ResNet-50 backbone yields an outstanding accuracy of 98.8% across 15 categories. The proposed hardware-software framework provides high-fidelity feedback and a powerful multi-modal perception baseline for complex robotic manipulation.

Multimodal understanding models that can jointly judge text-to-image (T2I), text-to-video (T2V) and text-to-speech (TTS) generation are increasingly used as "OmniJudges" for evaluation and automatic annotation. How reliably they understand what they score remains unclear, since existing benchmarks and training data tend to overemphasize positive examples and to conflate distinct failure modes, so a judge may score well without recognizing failures while its capability gaps stay hidden. Motivated by this, we introduce D3-Omni, a balanced and decoupled benchmark for diagnosing fine-grained multimodal understanding, covering 53 orthogonal binary dimensions (17/22/14) and 10,671 samples (3,526/1,998/5,147) across the three tasks. Rather than re-generating outputs, which may leak information across dimensions, we fix verified fully positive seeds and derive negatives through controlled prompt rewriting and atomic, dimension-isolating perturbations. The resulting D3 design is Dual-balanced, which helps alleviate negative-sample scarcity and per-dimension label imbalance; Decoupled, so that each error is attributable to a single capability; and Dynamic, steering construction toward under-represented regions of the label distribution as generative models improve.The suite reaches near 1:1 per-dimension parity and a uniform distribution over all total-score levels. Under this balanced view, even strong OmniJudges tend to struggle on modality-related dimensions, to confirm satisfied requirements far more reliably than they detect violated ones, and to treat nominally distinct attributes as largely a single decision, suggesting that aggregate accuracy may hide systematic blind spots that a balanced and decoupled lens can help expose and, in turn, address.

Real-world deployment of traffic surveillance systems is bottlenecked by geographic domain shift, in which models trained in one city underperform when applied to an unseen target city. Conventional domain adaptation relies on hyperparameter-sensitive architectures or direct profiling of target data. Both are fundamentally precluded in privacy-conscious ecosystems that require completely blind training and evaluation loops. In this setting, we explore the effects of pre-training and augmentation in addressing the domain shift problem. Specifically, we propose a new modular training pipeline for object detection structured around two core orthogonal pillars: (1) a multi-dataset pre-training strategy featuring a class-agnostic objectness distillation to decouple structural vehicle geometry from semantic taxonomies, and (2) a domain-resilient augmentation stream featuring a novel Grayworld transformation that forces global attention heads to strip volatile chromatic shortcuts in favor of robust shape priors. When evaluated with the real-time transformer-based detector RF-DETR, our framework bridges cross-city distribution gaps while using limited GPU memory (16GB). Our optimized variants, RF-DETR-HR and RF-DETR-Grayworld, deliver a substantial empirical gain of +24.29 over the baseline, achieving 1st place (47.53 mAP) on the AI City Challenge Track 6 leaderboard. Code and data are available at: \href{https://github.com/SKKUAutoLab/aic26_cross_city}{SKKUAutoLab/aic26\_cross\_city}.

In reinforcement learning policy evaluation, classic on-policy methods often suffer from high variance when estimating policy performance. To mitigate this issue, behavior policy search has been proposed to learn data-collecting policies tailored to reduce online evaluation variance. However, these approaches do not account for uncertainties in the transition functions. In practice, simulator transitions often differ from the real world due to modeling errors or approximation limitations. As a result, behavior policies trained in simulation may still yield high variance when deployed in real environments, leading to costly reliance on real-world evaluation samples. In this work, we propose a double-loop gradient-based algorithm for learning behavior policies that are both efficient and robust to transition uncertainty. Theoretically, we derive novel transition-variance gradient expressions and establish global convergence guarantees for the algorithm. Numerically, we demonstrate that our method is less sensitive to transition perturbations than existing approaches, providing supportive evidence for its practical utility.

LLMs are increasingly used to analyze spreadsheets, CSV files, and other structured data, but producing a correct-looking answer is not the same as producing a trustworthy analysis. A trustworthy result should be supported by a valid path from the user question to the relevant data evidence. This requirement creates two diagnostic questions: whether an LLM can refuse to answer or ask for clarification when such a path does not exist, and whether it can preserve the correct analysis when the same evidence is expressed in different table forms. We introduce TrustDABench, a benchmark that operationalizes these questions as reliability and robustness. Starting from the evidence-path view, we derive 19 perturbation operators and instantiate them through an Agentic-LLM-based generation framework. TrustDABench contains 2,340 human-verified perturbed instances, and we evaluate eight representative LLMs. The results show substantial headroom: the best reliability result is only 24.21% average MRS, achieved by GPT-5.5, while the best robustness result still has 9.10% average ASR, achieved by Claude-Sonnet-5. The failures are systematic: models rarely detect conflicting evidence, often continue along executable but unsupported analysis paths, and remain sensitive to perturbations that change observation boundaries or cross-table relations. These findings suggest that stronger evidence-boundary recognition and representation-invariant reasoning are still needed for reliable structured-data analysis.

Large vision-language models have shown strong progress in UI-to-code generation, yet their test-time self-evolution remains unstable. We first identify a fundamental obstacle, termed visual repair coupling: a local code edit may propagate through layout, style, and component dependencies, correcting one visual mismatch while degrading regions that were previously faithful. To address this issue, we present RubSE, a Rubric-guided Self-Evolution framework that uses rubrics to represent visual feedback as a structured visual-repair context. At each refinement round, RubSE generates typed candidate rubrics, selects one prioritized repair target, and stores previously selected rubrics as history, thereby steering each revision toward a well-scoped visual repair while discouraging repeated or over-broad changes. Evaluations across six VLMs and three UI-to-code benchmarks demonstrate that RubSE substantially outperforms naïve self-evolution in final-round and best-round settings, achieving more stable refinement trajectories and a higher trajectory-level performance ceiling. Further analysis shows that RubSE mitigates trajectory collapse by improving recovery from severe visual regressions, and that stronger rubric generators can transfer effective visual-repair guidance to weaker code improvers.

Recurrent models, which repeatedly update latent states with shared computation blocks, have emerged as powerful architectures for solving complex reasoning tasks. Existing inference-time methods scale computation by running more steps or sampling more trajectories, but ignore information revealed within each trajectory. Here we show that recurrent models can be improved at inference time by using their own readout probabilities to steer latent dynamics without retraining. We introduce Readout Feedback (RoFB), a test-time intervention that converts intermediate predictions into token-wise pairwise coupling forces injected into the latent dynamics. Across three recurrent models (AKOrN, ItrSA++, TRM) on Sudoku and Maze, RoFB yields clear gains in four of six model-task pairs, achieving performance unattainable by merely running more steps or selecting from multiple trajectories, at comparable or lower computational cost. These results suggest that closed-loop steering of latent dynamics can serve as a complementary inference-time control mechanism for recurrent reasoning models.

The majority of our everyday activities are procedural and consist of sequences of interdependent steps. However, existing benchmarks for Visual Agents and Visual Language Models (VLMs) overlook the evaluation of their procedural comprehension ability from an egocentric visual perspective, particularly for detecting procedural errors, a critical capability for everyday assistance. To bridge this gap, the EgoErrorVQA task is firstly proposed for egocentric procedural comprehension with explicit procedural errors modeling. Besides, we develop a user-friendly evaluator agent based on the Agent2Agent (A2A) protocol, enabling rigorous and standardized evaluation of visual agents through VQA-based interaction. A range of models are evaluated using both open-ended and multiple-choice questions, revealing persistent weaknesses in handling procedural errors and error types. Moreover, we introduce Ego-ADR, an Adaptive Decoupled Reasoning framework that decouples complex procedural reasoning to enhance models' understanding of procedural errors. It achieves consistent performance gains over the selected baselines and attains state-of-the-art results on several metrics under comparable settings. Code: https://github.com/z1oong/EgoErrorVQA

People search for urban outdoor places not only by category or function, but also by what activities a place can support and how it is perceived. Existing geospatial retrieval remains largely POIcentric and metadata-driven, making it difficult to satisfy openended, affective, or activity-oriented needs. We present PlaceSeek, a human-centered outdoor place retrieval framework that maps natural-language queries to geolocated street-view imagery. PlaceSeek introduces an intent-aware retrieval mechanism that decomposes user queries into functional and affective sub-intents. A Semantic Grounding Module verifies whether candidate street-view results contain the physical evidence needed to support the intended activity, while an Affective Alignment Module re-ranks physically valid candidates using a LoRA-adapted vision-language model trained on human urban perception judgments. We evaluate PlaceSeek on 31,956 street-view locations in Milan across 10 naturallanguage queries annotated by five human evaluators. PlaceSeek achieves 88.0% Precision@5, a mean match score of 3.39/4.0, and 0.920 nDCG@5, outperforming CLIP, fine-tuned CLIP, SigLIP, and a VQA-based baseline. Ablation results show that physical grounding is essential for retrieval validity, while affective alignment improves ranking quality among physically valid candidates. These findings highlight that complex urban spatial queries require modeling both verifiable visual evidence and human perceptual preferences. PlaceSeek provides a potential framework for human-centered nextgeneration geospatial retrieval systems.

Product catalogs in fast-moving service businesses are shifting from static, independently priced SKUs toward dynamically bundled, discount-coupled offerings--a shift that strains the tree-based classifiers traditionally preferred for sparse and highly imbalanced data. These classifiers assume a fixed, slowly changing label space and struggle to incorporate multimodal signals such as tabular data and transcripts. We present the migration of a live, production conversational recommendation system from a gradient-boosted multiclass model to a pairwise-binary deep recommender. Because this system is critical to ecosystem growth initiatives and downstream features like dynamic pitching--surfacing the most relevant pitch text to a support agent in real time during a live customer conversation--maintaining live recommendation quality was a non-negotiable constraint. We detail the techniques that made this migration successful--reformulating recommendation as pairwise binary prediction to learn jointly from user and item features, and enhancing learned representations via negative sampling and noise injection. To efficiently incorporate long, live conversation context, we apply attention pooling over transcript chunks and benchmark it against TF-IDF and sentence-embedding baselines. Finally, we explore multiple architectures (including two-tower models, DeepFM, and their variants) and loss functions such as contrastive loss. Evaluating against a CatBoost baseline across all conversational stages, we demonstrate that our approach achieves parity at conversation beginning and outperforms at later conversational stages.

Telephone fraud is pervasive and costly, but its inner workings are rarely observed at scale. We analyze a complete corpus of 10,211 inbound scam and spam calls -- 913 hours of audio and 330,956 transcribed turns from 5,780 distinct numbers -- collected over 54 days by an AI voice-agent honeypot that answered callers and kept them talking, and introduced in a companion data descriptor. We separate outright scams, which solicit sensitive information, from the larger stream of predatory but legal lead generation ("spam") that feeds them. Scam operations keep office hours (6.6x more calls per weekday than weekend day); thousands of disposable numbers run a small catalog of recycled scripts (thirty opening clusters, half the traffic in the top five); and callers solicit identity anchors -- a home address and a date of birth -- far more often than payment credentials, pressing through persistence and manufactured authority rather than overt threats. Our central experiment asks: does it matter who picks up? Every seeded lead carried one of ten fictitious identities drawn uniformly at random, so the identity a fraud operation reaches is fixed before the caller exists. Across 1,823 randomized calls, scammers spent about 15% more conversational turns per decade of the target's apparent age (rate ratio 1.15, 95% CI 1.08-1.23; randomization p = 0.005) -- yet what they asked for did not change (26.3% of calls reached a request for sensitive information; odds ratio 0.99 per decade, 95% CI 0.90-1.08). A second experiment casts early detection as a benchmark: from a scammer's opening lines alone, on a caller-disjoint split, escalation is predictable at 0.72 ROC-AUC from the first line and 0.87 by the eighth, and a plain bag-of-words classifier matches a fine-tuned on-device language model. Telephone fraud emerges as a templated industry that varies how hard it works a target, but not what it wants.

Phase-field modeling of brittle fracture removes the need to track cracks explicitly by recasting their evolution as the minimization of an energy functional. In return it requires a discretization dense enough to resolve a localization band whose width is set by a regularization length and whose path is not known in advance. We propose a mesh-free discretization in which a single neural network represents the displacement and phase fields and is trained by minimizing the incremental energy directly. The coordinates enter the network through a multiresolution feature encoding built from $C^1$ quadratic B-spline grids, so the finest scale the representation can express is set by choice rather than reached through slow training, and the energy is estimated by stratified Monte Carlo integration on points redrawn at every optimizer iteration. This pairing proves critical, since the crack fails to advance both when the integration points are held fixed and when the encoding is too coarse to represent the band, while each ingredient tolerates a wide range of settings once the other is in place. Because the representation is globally $C^1$, the second- and the fourth-order fracture energy densities run on the identical discretization. Across six problems, from single-edge-notched tension and shear to a thick-walled ring on a single spline patch, the computed load-displacement curves follow staggered finite element references at matched regularization length, with peak loads within about 1% on the single-edge-notched tests and within 8% where the crack pattern changes topology. On a public benchmark dataset of random multi-crack configurations the method classifies the active or dormant state of 90% of the seeded cracks in twenty zero-shot runs, where the deep Ritz baseline of the dataset authors fails.

Radiology report generation (RRG) has recently benefited from large language models, which substantially improve report fluency. However, clinically faithful generation remains challenging because current supervision is still imposed mostly at the report level. This creates a granularity mismatch: radiology reports are composed of disease-grounded findings, while existing methods are trained mainly with whole-report objectives. To address this problem, we propose Graph-Supervised Hierarchical Clinical Alignment, which reformulates image-report supervision as a hierarchical clinical alignment problem. Our method structures this alignment as a disease-conditioned process, where supervision is decomposed into two levels: Disease-Centric Alignment for fine-grained disease-specific correspondence, and Global Clinical Semantic Alignment for report-level semantic coherence. A clinical knowledge graph is used as a training-time-only structural prior that defines disease-specific supervision units and their clinical relationships, introducing no additional overhead at inference. Because standard contrastive alignment could produce false negatives when studies share overlapping pathologies, we combine instance-conditioned discriminative matching with disease-conditioned soft regularization, enabling fine-grained yet clinically consistent cross-modal representations. Experiments on MIMIC-CXR, IU-Xray, and COV-CTR show that our method consistently improves performance on both conventional and clinical metrics. Notably, our 3B model surpasses several prior systems with larger 7B/13B backbones, suggesting that improving supervision structure, rather than increasing model size, can be more effective for RRG.

Visual demonstrations provide a natural interface for specifying image transformations that are difficult to describe exhaustively with text. However, existing visual in-context learning (VICL) methods primarily focus on appearance-level relation transfer and provide limited support for physically grounded transformations, whose outcomes depend on material properties, geometry, object interactions, and environmental conditions. Given a source--target exemplar pair and a query image, physically grounded VICL requires a model to infer the demonstrated transformation, adapt its effects to the query-specific scene context, and preserve rule-irrelevant content. We introduce PhysVICL-74, comprising 74 physically grounded transformation rules and 5,240 source--target image pairs that form nearly 75K training and evaluation contexts. Its benchmark split separately evaluates novel-instance transfer and unseen-rule generalization. We further propose TransPhy, a framework that decomposes physically grounded VICL into physical-rule induction and transition-aligned rendering. TransPhy first predicts the demonstrated rule and an explicit query-specific target-state description, and then synthesizes the target image through token-wise mixture-of-experts adaptation, with expert routing guided by localized transition cues. Experiments show that TransPhy improves physical-rule adherence, query consistency, and unseen-rule generalization over existing visual in-context editing methods.

Vision-language models (VLMs) are increasingly used in clinical pipelines where a chest X-ray is interpreted alongside retrieved reports, preliminary notes, or prior imaging. Existing benchmarks measure whether models answer correctly in isolation, but not whether they preserve a correct image-only decision when plausible context conflicts with the image. We introduce Multi-Context Chest X-ray (MC-CXR), a benchmark of 240 cases expanded into 2,522 instances that isolates context-induced disruption through paired perturbation. Each case fixes the current image and target finding while presenting matched reliable and misleading context across text and prior CXR, with visual overlays where available. MC-CXR defines three task families and two paired metrics, the switch-to-wrong rate and the context-aligned error rate. We evaluate ten VLMs spanning open-source general, medical-domain, and closed-source systems. Image-only accuracy is necessary but insufficient. Mean switch rates range from 45.6-78.1% across misleading textual sources and 35.7-61.7% across misleading visual sources. Among switched predictions, 74.6% align with the misleading label for text versus 17.6% for visual context, a 57.0-point gap (95% CI 50.9-62.8). This text-visual asymmetry is observed under the standardized direct-answer protocol. The dataset is available on PhysioNet.

Multimodal large language models (MLLMs) can integrate long visual histories, reason under partial observability, and infer behavior from a few examples. Yet vision-language-action (VLA) models generally inherit pretrained representations without using this contextual capacity as episode memory. Memory-dependent policies address this gap through purpose-built history mechanisms. PonderPounce instead reuses an MLLM's native causal context as robot memory. Ponder, a System2 MLLM, accumulates episode observations, demonstrations, and prior cognition in its native causal context and can generate subgoal text and demonstration reasoning for internal use. Pounce, a System1 VLA, receives the current observation, instruction, and proprioception directly; through the Ponder--Pounce interface, it asynchronously receives only the newest continuous cognition token and its age. Both are jointly trained end to end without a purpose-built memory module or separate bridge pretraining. Optimized serving achieves p50 latencies of 78ms for cognition refresh and 25ms for action-model invocation, supporting 20Hz action playback. On RoboMME with base-scale training data, PonderPounce reaches 60.83% with 9B and 50.04% with 0.8B under the same Pounce architecture and interface, versus 44.51% for FrameSamp+Modul and 17.93% for the current-observation π_{0.5}. With 9x data, it reaches 75.54% versus 57.88% for FrameSamp+Modul. On RoboCasa-DC, the same interface learns from action supervision alone and reaches 12.5% versus 11.6% for the strongest published demonstration-conditioned baseline, falling to 8.6% when cognition is replaced by a learned null state.

We propose CAT, a trajectory-level continuous action representation framework for robotic manipulation. Existing visuomotor systems often entangle action representation with control frequency or rely on fixed temporal parameterizations. This leads to representational redundancy at high sampling rates and limits the modeling of critical motion. CAT instead encodes action trajectories within a fixed real-time interval into a set of continuous latent tokens. To ensure temporal consistency across varying control frequencies, we further incorporate a frequency-aware positional encoding that establishs a shared temporal coordinate system. Trajectory-level regularization further stabilizes the latent representation. This approach prevents representation growth with timestep density and avoids reliance on predefined temporal parameterizations. Extensive system-level evaluations on LIBERO, MimicGen, and real-world long-horizon manipulation tasks demonstrate that CAT-based policies consistently outperform both competitive VQ-based and continuous visuomotor baselines under matched training settings. Across various model backbones and control frequencies, CAT consistently improves success rates. These results highlight the advantages of trajectory-level continuous action modeling for scalable robotic manipulation across varying control rates.

Material replacement is a common interior-design operation: changing the material of a selected surface while preserving its geometry, surroundings, and illumination. Despite its commercial relevance, no public benchmark isolates this task, and evaluating it is challenging. Reference-based metrics penalize valid outputs in this inherently one-to-many setting, favor the style of the reference generator, and cannot fairly compare editors that receive different forms of guidance. We introduce MatReplace, a reference-free benchmark that evaluates edits along four verifiable dimensions: local material correctness, global lighting harmony, outside preservation, and inside structure. It defines three tracks that vary one conditioning signal at a time: (A) instruction only, (B) instruction plus region mask, and (C) material reference image instead of instruction. Our results reveal a clear divide between naming and visually grounding materials. In Track A, leading closed-source editors achieve exemplar-level material rendering and surpass the exemplar anchor under our primary aggregate. In Track B, masks help only mask-compatible models with weak scene preservation, with task-paired, single-seed effects ranging from +0.137 to -0.090 across aligned model families. In Track C, reference-image conditioning degrades every family under both aggregates, by -0.031 to -0.508; in the worst cases, models repaint the reference image itself and perform worse than returning the input unchanged. Thus, named-material rendering is largely solved by the strongest closed editors on this distribution, but grounding materials from pixels remains an open challenge. Expert ratings validate our ranking (Kendall's tau = 0.68) and align with our aggregates more closely than GT-referenced or CLIP-based baselines.

Purpose: Automatic radiology report generation (RRG) has been widely explored to improve reporting accuracy and reduce radiologists' workload. Most existing methods rely on autoregressive (AR) frameworks that generate reports token by token and cannot revise earlier content, making them prone to error propagation and inconsistent with the iterative refinement process of radiological reporting. In contrast, discrete diffusion large language models (DLLMs) generate text through iterative denoising, naturally enabling report refinement. However, DLLMs have not been extensively investigated for RRG. In this study, we developed and evaluated a discrete diffusion framework for RRG that enables iterative refinement rather than conventional left-to-right autoregressive decoding. Materials and methods: We developed DRRG, a DLLM-based framework that formulates RRG as iterative masked-token denoising. DRRG incorporates a clinical-entities-aware complementary mask to improve token supervision coverage and emphasize clinically important entities, together with a concept-conditioning module that injects image-derived clinical concepts into visual representations. DRRG was trained and evaluated on MIMIC-CXR and CheXpert Plus. Results: On MIMIC-CXR, DRRG achieved BLEU-4 of 0.210, CheXpert-F1 of 0.549, RadGraph-F1 of 0.281, GREEN of 0.360, and RaTEScore of 0.604, outperforming the compared methods on most reported metrics, despite employing a substantially smaller LLM decoder. On CheXpert Plus, DRRG achieved the highest BLEU-4 (0.119) and CheXpert-F1 (0.347) among the compared methods. Conclusion: Discrete diffusion provides an effective alternative to autoregressive radiology report generation by enabling iterative, bidirectional report refinement. Incorporating clinically focused masking and image-derived concept conditioning improves report quality and clinical consistency.

Robot actions are inherently embodiment-specific and only weakly aligned with image-space visual changes, limiting their effectiveness as conditioning signals for robot world models. In contrast, visual tracks provide an embodiment-agnostic representation of how task-relevant points move through a scene, offering dense image-space guidance for accurate and spatially precise future video prediction. Building on this observation, we propose TrAct, a world-model-based robot decision-making framework that uses visual tracks as an intermediate interface between control and prediction. TrAct consists of three components: a Vision-Language-Action-and-Track model (VLAT) that jointly predicts candidate actions and corresponding visual tracks from the current observation and language instruction; a track-conditioned world model (TWM) that predicts future visual outcomes conditioned on the proposed tracks; and a vision-language reward model (VLAC) that scores the predicted outcomes. At inference time, VLAT generates candidate action-track pairs, TWM rolls out their visual consequences, and VLAC selects the track whose predicted outcome best satisfies the instruction; the action paired with the selected track is then executed by the robot. Experiments on the proposed LIBERO-INTEGRAL benchmark and real-world Franka manipulation show that TrAct improves success rates from 27% to 55% in simulation and from 49% to 76% on real-world tasks compared with the strong VLA baseline $π_{0.5}$. Furthermore, TWM consistently improves video prediction quality over the action-conditioned world model (AWM). These results demonstrate that visual tracks provide an effective shared interface between robot control and visual prediction, enabling more accurate world modeling and stronger robot generalization.

Emergency vehicles (EMVs) can reorganize surrounding traffic as civilian vehicles brake, change lanes, or form rescue corridors in response to their passage. Evaluating these safety-critical interactions requires behavior-level control over both EMV privileges and civilian responses, together with consistent sensing and ground truth. Existing datasets and simulation benchmarks do not directly provide this combination. We present \textbf{SIREN}, a behavior-driven SUMO--CARLA co-simulation platform for generating EMV--civilian interactions. SIREN couples SUMO's network-level traffic evolution and behavior logic with CARLA's continuous vehicle control and synchronized onboard sensing; depending on the active behavior, the interaction is controlled by SUMO, CARLA, or jointly. We instantiate the platform as \textbf{SIREN-Bench-v1}, comprising seven parameterized interaction templates across emergency levels L1--L3 and three behavior families, with synchronized sensor observations and simulator-native annotations. We demonstrate the benchmark through three representative tasks: 3D object detection, trajectory prediction, and vision-language risk understanding. Evaluations of nine trajectory predictors, four LiDAR-based detectors, and five vision-language models reveal behavior-dependent failure modes. Traffic-clearance interactions are hardest for detection, privileged intersection traversal is hardest for prediction, and no learned predictor outperforms the constant-velocity reference on average. Vision-language models perform substantially better on normal traffic than on near-miss and collision events. These results demonstrate the value of behavior-centered benchmarking and establish SIREN as an extensible data-generation and evaluation platform for autonomous-driving and transportation safety research.

Self-supervised representation learning for 4D point cloud videos is challenging because annotations are costly and reconstruction-based pretraining can overemphasize low-level geometric details. We propose a JEPA-style framework that learns from unlabeled spatiotemporal point clouds through latent point-tube prediction. Instead of reconstructing raw coordinates, the model masks spatiotemporal regions and predicts their target representations from visible context representations in feature space. To stabilize latent prediction, we incorporate Sketched Isotropic Gaussian Regularization, which encourages non-collapsed embeddings without relying on explicit reconstruction targets. This formulation aims to capture both spatial structure and temporal dynamics while keeping the pretraining objective aligned with downstream semantic recognition. Experiments on action and gesture recognition benchmarks show that the learned representations improve downstream fine-tuning, limited-label learning, and cross-dataset transfer. These results suggest that JEPA-style latent prediction is a promising alternative to reconstruction-centered pretraining for 4D point cloud videos.

Current LLM agent systems decide delegation before reasoning begins (a router picks a model) or after a response is complete (a verifier scores it and may retry). We study a third regime: an agent that recognises, during its own reasoning, that it is unlikely to succeed and transfers control to a stronger model. We formulate intra-generation delegation as a Bayesian optimal-stopping problem over a learned competence posterior -- an online estimate of the agent's eventual task success whose sufficient statistics are learned from labelled trajectories, not read off raw entropy. We derive the myopic escalation threshold in closed form, characterise the optimal policy via dynamic programming, and prove that the optimal policy is a time-varying threshold with no shape assumption on the raw signal. We further prove exponential separation of the oracle belief at the Chernoff-information rate of the signal, a regret bound governed by the calibration of the posterior, and a finite-sample guarantee: with n labelled calibration trajectories the deployed plug-in policy's regret decays as 1/sqrt(n). A controlled simulation study confirms each prediction of the theory, including the predicted 1/sqrt(n) rate. We additionally report a real-model validation on a Qwen2.5-Coder 1.5B->7B code cascade (MBPP, 257 tasks), confirming two of three pre-registered predictions: the escalation frontier dominates post-hoc routing at equal cost, and the cumulative competence belief's discrimination rises over generation.

Large Language Models (LLMs) have shown strong capabilities in table reasoning, but their effectiveness degrades as tables grow in size and complexity due to irrelevant context and difficulty localizing the evidence required for reasoning. Existing approaches typically reason over either the full table or a single reduced view, which can still obscure important row-column relationships. We introducePARTAB (Partition-Aware Reasoning overTables), a framework that constructs a structured evidence interface between the LLM and the table. PARTAB represents query-relevant evidence as semantically coherent, row-linked table regions and performs hierarchical selection over column groups and row-level partitions before composing the selected evidence for answer generation. We evaluate PARTAB on multiple table reasoning benchmarks, covering question answering, fact verification, and numerical reasoning. PARTAB consistently improves over full-table prompting and several recent table reasoning methods, achieving strong performance on WikiTableQuestions and TabFact while remaining competitive on numerical reasoning. Additional analyses show that semantic partitioning and targeted evidence selection improve evidence localization, substantially reduce the reasoning context, and provide larger benefits on complex tables. These results demonstrate the value of structured, partition aware evidence construction for scalable table reasoning.

In psychological counseling, effective support is not always delivered through long, information-rich responses. Minimal responses, such as backchannel cues and concise empathic statements, help convey attentive listening, express empathy, and encourage clients to continue expressing themselves. However, existing counseling dialogue systems and evaluation frameworks often favor explicit, content-rich replies, overlooking the interactional value of brief counselor utterances. This paper presents a systematic cross-lingual analysis of minimal responses across multiple counseling dialogue datasets. We develop a two-stage filtering method based on utterance length and content, followed by contextual verification using a large language model (LLM). Our analysis shows that minimal responses are common in human-collected datasets but substantially underrepresented in LLM-generated ones. We further evaluate current LLMs in manually curated dialogue contexts where human counselors used minimal responses. The results show that strong commercial LLMs are capable of generating minimal responses when explicitly instructed, but still struggle to determine when such responses are appropriate. Counseling-specific models trained on synthetic data perform particularly poorly, tending instead to produce longer and more information-rich responses. Moreover, LLM-based response-quality evaluation may undervalue minimal responses, even when they are interactionally appropriate.

A shared search-and-recommendation index must score new items from features alone because search has no exploration slot. In a public log covering both surfaces over one catalog, $38.6\%$ of held-out query-search impressions show an item never previously shown or visited. For user-cold engagements, the feature-based tower serves this demand without measurable loss against $99$ sampled negatives ($0.9595$ Recall@20 versus $0.9510$ warm). A lexical baseline reaches similar parity, while a full-catalog check remains statistically undecided. Dual-encoder retrieval therefore keeps the index \emph{open} to new items, unlike an ID-softmax recommender that requires retraining. We price this openness on recommendation against six sequential baselines, each retrained and tuned through five rounds on corrected targets. A float32 timestamp bug had reordered leave-one-out targets for $19.7\%$ of users. On MovieLens-1M, warm accuracy trails the strongest retrained baseline by $5.2\%$ Recall@20 and $11.4\%$ NDCG@20. On MIND, the gap narrows to $0.8$--$3.6\%$ relative to the five strongest baselines, though the model ranks sixth of seven. Under strict zero-leakage cold-start evaluation, the content tower achieves $0.172 \pm 0.006$ Recall@20, $1.4\times$ the strongest retrained dedicated method ($0.124 \pm 0.007$) and $3\times$ a training-free floor, without cold-specific training. Exact full-softmax training raises Recall@20 by $54\%$ on MIND-small and $6.9\%$ on MovieLens-1M over sampled InfoNCE, but recomputes the full catalog each step and exhausts accelerator memory at $240$K items. Approximate nearest-neighbor search explains none of the remaining gap, serving cost does not regress against ID-softmax retrieval, and a history-window sweep explains half the post-recipe remainder. Exact-quality training at catalog scale remains the open problem.

Prohibitive computational and environmental costs impede the scalable deployment of Large Language Models (LLMs). Traditional compression techniques (sparsity, quantization, low-rank approximations) are typically applied in isolation, and each hits an accuracy-efficiency wall. This thesis proposes the "Compression Trinity," a unified framework that applies the three pillars jointly: sparsity to reduce computation, quantization to minimize memory bandwidth, and low-rank approximations to recover accuracy. To accelerate pretraining, we apply the Trinity to the optimizer and model architecture. MKOR approximates curvature via block-diagonal sparsity and low-rank inversion, maintaining numerical stability for quantized states; it reduces curvature update complexity from $O(d^3)$ to $O(d^2)$ and accelerates convergence by up to 1.85x over KFAC. SLoPe accelerates training by up to 1.25x via a double-pruned backward pass for N:M sparsity, using low-rank "lazy" adapters in the final 1% of training to recover accuracy. For post-training compression, OPTIMA stabilizes static masks in a zero-training regime by formulating weight reconstruction as globally optimal column-wise quadratic programs, improving zero-shot accuracy by up to 3.97%. Given a fine-tuning budget, PATCH breaks the ceiling of static masks by learning a dynamic hybrid sparsity ratio between 0% and 50%, yielding up to 1.38x speedups. Finally, SLiM realizes the full Compression Trinity in one shot, using mathematically derived low-rank adapters to recover information lost to quantization and sparsity, improving accuracy by up to 5.66% over state-of-the-art methods and outperforming uncompressed dense models at equal parameter budgets by 0.6%. Together, these results show that jointly applying the Compression Trinity is essential for efficient, scalable, high-performance LLMs.

A self-organising map turns a large corpus into a browsable two-dimensional atlas, but building one at MEDLINE scale has been impractical: the best-matching-unit (BMU) search that dominates training is bound by the bandwidth needed to read the codebook every epoch. I show that this bottleneck is largely an artefact of codebook layout. Storing it feature-major with each feature's weights contiguous, W[v.M+i], recasts the search as a tiled sparse-dense product in which every loaded weight column is reused across a tile of samples. Varying only the layout, with implementation, precision and update rule held fixed, accelerates the BMU search by 4.5-8.5x. Because an exact-argmin BMU is invariant to how the codebook is stored, this gain costs nothing: held-out quantisation error agrees with a cuSPARSE baseline to within 0.5% at every map size. Against that baseline the advantage is a crossover rather than a constant: cuSPARSE.SOM is faster at small maps, SparseBin.SOM is 1.5x faster at 128x128 and 2.6x at 256x256, and at 512x512 it is the only one that runs at all on 24 GB. Paired with a radius-independent box-blur update and a convergence-based stopping rule, it trains a converged map over 29.9 million MEDLINE articles in about 72 s at 64x64 on one 24 GB GPU, and accommodates 262,144 neurons (512x512 edges) where every alternative algorithm I tested exceeds memory constraints. On a 141 GB H200 it reaches 1,048,576 neurons (1024x1024 edges) - to my knowledge the largest self-organising map yet reported. Held-out error follows a smooth power law with no elbow across three decades of map size, so the limit on resolution is compute rather than any breakpoint in the data. At matched work the design is ~82x faster than MedSOM, the CUDA implementation behind our earlier MEDLINE atlases and, at 128x128, 621x faster than the best available multicore-CPU library.

While Vision Large Language Models (VLLMs) have achieved remarkable success in multimodal reasoning, their long-context inference remains prohibitively expensive due to the massive computation and memory overhead of visual Key-Value (KV) caches. Existing KV compression methods often apply uniform pruning across visual tokens and layers, leading to substantial information loss and degraded performance.To address this challenge, we propose \textbf{VisCache}, a plug-and-play framework for coarse-to-fine \textbf{Vis}ual KV \textbf{Cache} pruning without training, which consists of two synergistic stages. First, a lightweight VLM filters temporal redundancy by selectively forwarding semantically informative keyframes. Second, we introduce {PruneKV}, a surgical KV compression algorithm tailored to the attention dynamics of VLLMs. Unlike rigid pruning strategies, PruneKV adopts a parabolic layer-wise budget allocation together with an asymmetric update mechanism that selectively prunes keys while fusing values, thereby preserving critical contextual information. Extensive experiments demonstrate that VisCache substantially improves inference efficiency, achieving up to {2.35$\times$ speedup} and significant memory reduction while maintaining competitive performance with only {19--28\%} KV cache retention. VisCache consistently outperforms existing baselines, establishing a new Pareto frontier between efficiency and performance for long-context VLLM inference. Code is available at https://github.com/Wlklk/VisCache

This work introduces scrydb, a Python library that enables lexical, semantic, and hybrid search within SQLite. For lexical search, scrydb leverages SQLite's full-text search extension FTS5. Semantic search builds on sqlite-vec, a SQLite extension for vector search. Furthermore, the library allows users to rerank and fuse retrieval results to combine both lexical and semantic approaches, providing a lightweight solution for downstream tasks in information retrieval (IR) or agentic search. We evaluate scrydb on various IR benchmark datasets and demonstrate its effectiveness in text retrieval based on keyword matching, semantic similarity, and rank fusion. In addition, we provide insights into query latency and the trade-off between efficiency and effectiveness. scrydb is available under the MIT license.

Generative and predictive artificial intelligence models are increasingly used to generate geometry and to predict physical fields and scalar quantities in engineering design and simulation. Yet these models are typically evaluated in isolation, on academic datasets at unconstrained scales, with inconsistent metrics and procedures. We present PhysicsBench, a unified benchmark and leaderboard that evaluates generative and predictive models under one standardized procedure. PhysicsBench spans seven generation and prediction tasks across 1D, 2D, and 3D domains and ranks 66 models on nine datasets, comprising industrial-scale CAD/CFD/FEA simulations and public references, expanded into 28 configurations. One procedure and ranking apply to both families, each ranked within its own tasks. Evaluation spans realistic, limited data scales from S to XL rather than the unlimited training sets common in academic benchmarks. A common metric suite captures geometric fidelity with distributional distances, physical-field and scalar accuracy, and engineering-specific field- and shape-validity. BenchRank debiases correlated metrics and ranks by PageRank over a head-to-head dominance graph, so every reported quality metric is also ranked, with computational cost in a separate efficiency view. Across tasks, an architecture's large-scale academic standing weakly predicts its small-data ranking. The top model changes with data scale in six of the seven tasks, and no model leads more than one task. PhysicsBench turns "state-of-the-art" from a self-reported claim into an openly published foundation for model selection.

Universal multimodal embeddings are becoming a core component of modern AI systems, enabling heterogeneous content to be represented in a shared space for applications such as retrieval, recommendation, classification, and agentic systems. In this report, we present WeMM-Embedding, a family of universal multimodal embedding models supporting text, images, videos, visual documents, and arbitrarily interleaved multimodal inputs with flexible output dimensions. The family comprises 2B, 4B, and 9B variants and is trained in two stages: a large-scale multimodal alignment stage, followed by a refinement stage using curated data, fine-grained relevance supervision, and cross-scale knowledge transfer. Across extensive evaluations, WeMM-Embedding achieves leading performance on multiple public benchmarks. Notably, the 2B variant already surpasses the previously leading 8B open-source baseline on MMEB-v2, while the 9B variant further achieves a new state-of-the-art overall score of 80.6. WeMM-Embedding also demonstrates strong practical performance across WeChat applications, with substantial gains on a 26-task in-house benchmark and consistent improvements across 14 online A/B tests. It has been deployed at scale across recommendation and search applications, including WeChat Channels, Official Accounts, Moments, and e-commerce services. We have released the model weights and code to facilitate future research at https://github.com/Tencent/WeMM-Embedding.

Neural operators provide efficient surrogates for spatiotemporal PDE systems, but purely data-driven formulations often accumulate substantial errors during long-horizon autoregressive prediction and may fail to exploit available governing-equation structure. Existing approaches incorporate physics primarily through residual-based training objectives or PDE-specific architectural constraints, which can introduce optimization difficulties or limit architectural generality. In this work, we introduce a representation-level approach to physics integration in which a feed-forward Gaussian splatting (FFGS) representation serves as a continuous interface between discretized solution fields and governing operators. The FFGS representation reconstructs the state as a continuous Gaussian field with closed-form spatial derivatives, allowing available physical PDE operators to be integrated directly within the learned evolution map without introducing a physics-residual loss. We evaluate the framework across two- and three-dimensional PDE systems, including advection, diffusion, nonlinear self-advection, and reaction dynamics. Over long-horizon autoregressive rollouts, the proposed framework reduces relative $\ell_2$ error by $1.5\times$--$2.2\times$ compared with the strongest purely data-driven baseline across the benchmark suite, while consistently improving spectral fidelity. The framework also remains effective when the governing equations are partially known, demonstrating robustness to incomplete physics. These results demonstrate that continuous field representations can provide a practical interface for incorporating known physical structure into generic neural-operator surrogates.

Latent world models plan by predicting how candidate actions transform learned representations. In self-predictive models, however, the encoder and predictor are optimized jointly and can co-adapt to latent transitions that are easy to predict but only weakly constrained by the physical evolution of the scene. We introduce the cross-predictive JEPA (XP-JEPA), which grounds visual latent dynamics in privileged physical trajectories. XP-JEPA separately encodes visual observations and physical states, advances both through a shared action-conditioned predictor, and matches each prediction to both future representations. This objective encourages unified latent dynamics across the two modalities, grounded in the underlying physical transitions. The physical branch is discarded after training, leaving a visual-only model at deployment. On a multi-task suite spanning six evaluation subfamilies, XP-JEPA reduces rollout drift of a newly fitted predictor from $0.361$ to $0.104$ and increases mean control success from $53.6\%$ to $78.2\%$. Direct physical-state regression raises position decodability but leaves forecastability and control near the visual-only baseline. Cross-predictive physical grounding can therefore produce more forecastable latent dynamics for rollout-based control without privileged inputs at test time.

While Vision-Language-Action (VLA) models pretrained on large-scale robot datasets provide a strong foundation for robot manipulation, their performance can degrade when adapted to new tasks with limited task-specific demonstrations. Retrieval offers a practical way to reuse existing demonstrations for data-efficient adaptation, but existing methods often rely on visual similarity, state-action representations, or task-level language matching. These approaches may overlook the hierarchical structure of long-horizon manipulation tasks, where complete task matches are rare but reusable skills are often abundant. To address this challenge, we propose Hierarchical Skill Retrieval (HSR), a retrieval framework for data-efficient VLA adaptation. Specifically, HSR first decomposes a target task into candidate skill sequences. It evaluates each plan based on both semantic plausibility and skill reliability estimated from the prior dataset. The selected decomposition is then used for hybrid retrieval. This combines subtask-level language retrieval with behavior-feature reranking to identify demonstrations that are both semantically relevant and compatible with the target task. Finally, we adapt the policy through a two-stage pretraining and finetuning pipeline, which separates general skill acquisition from task-specific adaptation. Experiments on the LIBERO benchmark and several real-world robot manipulation tasks show that HSR improves the average success rate by 10.3% and 21.3% over the strongest baseline, respectively. These results demonstrate the effectiveness of structured skill-level retrieval for data-efficient VLA adaptation. Videos and code are available at https://hoar012.github.io/HSR-Project.

Enterprise AI agents in production often need to be bounded, stateful, observable, and governable rather than fully autonomous. We present PinSieve, a production case study in a large-scale content-quality pipeline. Its deployed component is a selective vision-language-model (VLM) Serving Agent that operates only on the grey-zone slice left unresolved by lightweight upstream models, exposes a scalar routing score online, and preserves controlled human escalation. On this slice, the deployed system filters 2.05x more non-actionable items than the previous production module while slightly reducing estimated miss rate; after promotion, it improves review productivity by 25.7%, reduces normalized operating cost by 16.2%, and moves signal delivery from next-day to same-day. We then study maintenance through a governed memory flywheel under selective feedback, where escalated items are reviewed by default and auto-passed items are labeled mainly through audit sampling. Feedback Memory records routing traces, observation paths, audit propensities, and replay metadata for evaluation and debugging. The Data Curation Agent uses a bounded proposal-verifier loop over representative, uncertainty, recency, and fresh-review replay, with positive-rate and score-bin guardrails before batch acceptance. In chained monthly refresh over six months of production data, this design reduces average FNR@50% from 17.73% under representative random replay to 13.29%. A Reasoning Review Agent audits teacher-generated rationales and supports keep/repair/drop decisions. Production claims are attributed only to the deployed Serving Agent; replay and rationale-review results are offline or sampled-governance evidence. The same serving-agent recipe has been adopted to several additional internal signals, suggesting transferability beyond one task.

Manufacturing process planning transforms heterogeneous design information into coherent manufacturing decisions. However, existing approaches focus on isolated subtasks, such as feature recognition, drawing interpretation, or tool selection, and struggle to support the full reasoning chain from design artifacts to process plans. This is critical when planning must interpret 3D CAD models, 2D engineering drawings, materials, and domain-specific rules. To address this gap, this paper presents Design-to-Plan, a large language model (LLM)-based multi-agent framework for end-to-end manufacturing process planning. An orchestrator coordinates specialized agents for 3D feature recognition, 2D drawing analysis, 2D-3D context fusion, knowledge retrieval, process sequencing, tool selection, and report generation. Rather than using LLMs as standalone text generators, the framework deploys them as reasoning agents that interact with deterministic modules and knowledge sources to produce consistent and traceable decisions. In this hybrid design, deterministic modules and specialized agents extract structured information from CAD and drawing inputs, while LLM agents perform context-aware reasoning, retrieve manufacturing rules, resolve conflicts, and generate planning outputs. The framework is evaluated using 300 benchmark cases across three downstream ReAct-enabled agents, plus separate evaluations of CAD feature recognition, drawing analysis, and 2D-3D context fusion. The parallel architecture achieves 100% success across downstream agents, Tool F1 scores of 95.9%-97.6%, 90% source detection accuracy in conflict analysis, and a 60%-68% reduction in token usage for key planning tasks. Results show that structured LLM-based multi-agent coordination can bridge design representations and manufacturing knowledge, enabling scalable, efficient, and traceable design-to-plan automation.

This paper extends Anthropic's Sleeper Agents research [1], which showed artificial backdoors persist through safety training & can be detected by linear probes with >99% accuracy [2]. However, probe-based detection relies on linear separability that may be an artefact of backdoor insertion rather than a property of naturally occurring deceptive alignment. Sophisticated deceptive behaviours emerging through natural training are unlikely to produce such convenient linear signals. We introduce a naturalistic methodology using multi-turn context windows that simulates realistic deceptive reasoning without artificial triggers or supervised backdoor insertion. Rather than binary trigger-response patterns, we examine how semantic complexity emerges through gradual context development. Building on our Curved Inference framework, we analyse curvature, salience, & introduce semantic surface area (A'), a new metric of representational work capturing both the magnitude & directional change of meaning construction in unnormalised residual space. Without backdoors, labels, or probes, we apply this framework to naturalistic deceptive prompts & classify model outputs via LLM consensus. Geometric structure reliably predicts semantic classification, with statistically significant differences in surface area across five prompt strategies & two model families. Critically, measurement precision can reveal geometric signatures hidden by classification noise - some strategies improve from non-significant (p = 0.555) to significant (p = 0.048). This validates that sophisticated reasoning creates intrinsic geometric patterns that persist even when detection appears to fail, suggesting the shape of inference itself encodes semantic patterns regardless of whether models have learned to suppress linear indicators of deception - a scalable, unsupervised path for detection when linear methods fail.

Recent end-to-end neural motion planners generate trajectories from raw sensor observations, avoiding the privileged geometric models required by classical planners. However, collision-free planning in cluttered environments remains challenging. We present NeurRAFT, a generative planning framework based on anchor-level flow matching and clearance-aware preference tuning. Unlike prior neural planners that model dense waypoint sequences and spend capacity on redundant local details and smoothness, NeurRAFT operates on compact anchor waypoints. We train the planner using a Jacobian-weighted loss that accounts for the task-space impact of each anchor. At inference, the anchors are generated in two integration steps, followed by cubic-spline interpolation to recover a smooth, full-resolution trajectory. Since imitation learning from positive demonstrations cannot distinguish collision-free from near-collision trajectories, collision-prone behaviors persist at test time. Rather than relying on post-hoc corrections, we directly reshape the pretrained planner's distribution toward safer solutions without augmenting inference. Specifically, Direct Preference Optimization shifts probability mass toward trajectories with larger obstacle clearance, with the resulting improvement directly absorbed into the planner parameters. Experiments show substantial improvements over state-of-the-art planners, while real-world experiments demonstrate zero-shot transfer to a Franka robot under noisy and partially occluded depth observations. Video results available at https://neurraft.github.io/.

Underactuated systems pose a challenge for convex motion planning because their dynamically feasible motions lie on a manifold of trajectories in function space. Building on our earlier formulation of polytopic action sets (PAS), this paper presents a method for rapidly generating, online, trusted convex sets of short-horizon actions for underactuated and potentially nonlinear systems. Around a nominal trajectory, we construct local finite-dimensional action coordinates in which each parameter vector encodes a complete nearby motion through an affine trajectory map, rendering collision-avoidance and control bounds linear. To remain consistent with the nonlinear dynamics, we introduce a dynamics-violation metric and extract a trusted convex inner approximation using an IRIS-inspired inflation procedure directly in action space. The resulting PAS are reusable convex families of actions that can be queried and composed with linear programs, and a PAS-guided tree expansion treats nodes as composed reachable families rather than single trajectories, coupling local nonlinear fidelity with convex reuse for longer-horizon planning. The planner solves cluttered planar scenes in tens of milliseconds (14-78x faster than a kinodynamic RRT baseline) and reduces terminal error on a nonlinear underactuated benchmark by 26-86% over sampling and NLP baselines.

Chinese ancient document understanding demands complex visual, linguistic, and historical reasoning. Current Large Vision-Language Models (LVLMs) typically rely on an opaque, single-pass generation paradigm, often producing overconfident and weakly grounded responses. To address this, we propose SAGE, an evidence-grounded multi-agent framework that reformulates Chinese ancient document understanding as evidence-grounded inference rather than direct answer generation. SAGE coordinates specialized agents for task-aware planning, tool-mediated evidence acquisition, claim-level verification, and bounded replanning under a constrained shared-state runtime. This design supports bounded evidence seeking, answer revision, and abstention when grounding is insufficient. Experiments on the AncientDoc benchmark show that SAGE consistently outperforms matched direct-answering baselines across three LVLM backbones. Remarkably, SAGE with Qwen3.5-9B surpasses much larger monolithic LVLMs on most evaluated metrics, highlighting the importance of structured, evidence-grounded inference beyond model scaling.

Supervised proxy-based deep cross-modal hashing has become the dominant paradigm for large-scale retrieval. However, prevalent methods model class proxies as deterministic points in the embedding space. This rigid assumption causes severe gradient conflicts in multi-label scenarios, where gradient conflicts arising from label co-occurrence lead to severe gradient contention and optimization collapse. To resolve this, we propose Kent-based Distributional Proxy Hashing (KDPH), a novel framework that shifts proxy representation from static points to flexible anisotropic Kent distributions on the hypersphere. Unlike point proxies that must shift their positions to accommodate conflicting gradients, KDPH absorbs these conflicts by dynamically adjusting its directional variance. This allows the proxy to maintain a stable semantic mean direction while stretching to cover diverse label correlations. Furthermore, to ensure stable training of these geometric parameters, we derive a tailored loss function incorporating the Cayley transform to enforce strict orthogonality. To the best of our knowledge, KDPH is the first framework to successfully introduce the Kent distributions into cross-modal hashing. Experiments on three benchmark datasets demonstrate that KDPH mitigates proxy collapse and chaotic oscillation, significantly outperforms state-of-the-art methods. Code is available at https://github.com/Senmo996/KDPH-official-code.

Large language model (LLM)-based agent applications often incur high response time. Speculative decoding is a promising solution to improve the inference efficiency of LLM agents without impacting generation quality. However, state-of-the-art speculative decoding algorithms exhibit substantial speed degradation under large batch sizes, limiting their effectiveness to deploy in real-world agent applications. In this work, we first present a systematic analysis of speculative decoding for LLM agents and identify two dominant factors of speedup degradation: high rejection rate of speculative tokens, and under-utilization of dynamic token budgets.B ased on these observations, we propose AgentSpec, a speculative decoding algorithm that addresses the limitations of existing methods for LLM agents. AgentSpec incorporates structure-isolated drafting that constrains speculation to semantically coherent segments of the agent workflow, reducing the drafts of irrelevant semantic paths and achieving an extremely low rejection rate. Moreover, AgentSpec adopts redundancy-aware budget allocation that exploits agent-level information to better utilize the dynamically-free token budget during the agent inference. We implement and evaluate AgentSpec on five different workloads and four different models from four different LLM families in vLLM. Our results demonstrate the superiority of AgentSpec over state-of-the-arts.

Recent advances in single-image 3D Gaussian head reconstruction have enabled highly realistic and freely renderable digital heads from a single portrait. However, reconstruction and rendering can weaken the forgery traces in the source portrait, making the resulting 3D face difficult to classify whether its underlying face is real or fake, and thereby posing risks to identity authentication and face privacy. To study this problem, we introduce the first large-scale benchmark for this task by collecting real portraits and fake portraits from multiple sources and evaluate representative existing detectors on this benchmark, revealing their lack of explicit mechanisms for retaining fine-grained information and maintaining feature consistency across rendered views. To directly address these two limitations, we propose a detector trained with a two-stage strategy. In Stage I, masked autoencoding encourages the visual backbone to retain the fine-grained appearance information required for local reconstruction, while multi-view contrastive learning enforces feature consistency across rendered views of the same head. Since CLS tokens at different depths exhibit complementary spatial attention patterns, Stage II freezes the adapted backbone and concatenates low-, middle-, and high-level CLS tokens for classification. Experiments show that our method achieves the highest accuracy and ranks first across all reported metrics among the evaluated detectors.

Robotic fruit harvesting must hold produce securely without bruising it, yet compression stiffness varies several-fold with ripeness within a single species, so no fixed grip force spans the range. Rather than tune force, we bound deformation: a controller closes the gripper until the object's estimated compression strain reaches a user-specified limit $\varepsilon$, using only the encoder position and motor-effort signal on every servo gripper---no tactile or force-torque sensor. Dividing an effort-based contact force by a lower bound on object stiffness makes the stop provably conservative---true compression stays at or below $\varepsilon$---for any $\varepsilon$ above a contact-detection strain floor we identify and quantify: robust detection itself spends compression, linearly in closing speed, making speed an explicit throughput--gentleness knob. Unlike a hand-tuned force threshold, $\varepsilon$ is a certified, size-scaling, operator-interpretable damage limit, and a ready safe-action parameter for learned grasping policies. In MuJoCo simulation over a realistic fruit-stiffness range, under a sensor-noise model calibrated to the real servo, the controller holds $\ge 98\,\%$ grasp at $0\,\%$ damage across all medium-to-firm stiffnesses for the entire certified $\varepsilon$ range, which neither fixed-force baseline attains; on stiffness-graded 3D-printed TPU cubes it matches baseline grasp success at roughly half the grip force and cuts soft-object damage from $100\,\%$ to $40\,\%$.

Scientific reasoning requires language models to retrieve specialized knowledge and incorporate it reliably into multi-step computation. Conditional memory provides an explicit lookup pathway that complements dense neural representations, but its usefulness is inherently input- and computation-dependent: retrieved information may repair missing scientific associations, yet it may also introduce distracting shortcuts or interfere with reasoning that the base model can already perform correctly. In this work, we systematically investigate when, where, and to what extent conditional memory should participate in scientific reasoning. We characterize the scientific knowledge boundary and controlled interventions on memory-enabled knowledge-circuit nodes. Based on these analyses, we propose a Knowledge Boundary-Aware Router that uses task-specific input proxies available before generation to determine whether memory is activated, which layer-stage nodes receive memory signals, and how strongly these signals contribute. Experiments on biological and chemical reasoning benchmarks, covering two backbone families and six task types, show that memory effects vary substantially across inputs, tasks, and injection locations. Compared with static and activation-rate-matched random routing, our approach more consistently preserves beneficial memory contributions while suppressing memory-induced regressions, establishing selective memory allocation as an important principle for reliable scientific reasoning.

In a deliberative poll, once submissions outnumber what anyone will read, some mechanism chooses which arguments each voter sees, acquiring much of the decision; practice delegates it to opaque learned rankers, so a voter cannot recompute or contest the exposure that shaped their vote. We ask whether it can be a published rule over publicly recomputable evidence with parameters held by the voter, treating legibility as an admissibility condition on usable mechanisms, not an objective traded against accuracy. We formalise a poll over bipolar justification sets, judging a slate by reason coverage, the order it arrives in, and captured endorsement mass; we give seven checkable criteria for a civic recommender and a rule meeting them: a one-hop reversed endorsement flow parameterised by a relation-weight function. An agentic simulator records every slate at every vote, over about 17,000 seed-paired runs. Served slates fall 0.035 short of a label-reading ceiling upper-bounding every selection procedure, opaque ones included: any unconstrained ranker's advantage is bounded and small. On coverage alone, with non-degenerate authoring, the rule is indistinguishable from a random slate, a null due to an order-blind, charity-blind instrument; on the other two it leads at every prefix by a margin widening with adversarial pressure and dominates on mass by a factor of 3.3. Once a realistic fraction of submissions carries no reasons, the coverage margin returns and grows. Label-homogeneous flooding collapses completeness from 0.81 to 0.34 under a flat weight policy, only to 0.44 under author-count normalisation, making the weight function a security control worth 10% of completeness. The choice between ranking arms is a position on a coverage-versus-mass frontier, not a fact, the kind of choice only a legible rule can hand to the person it affects. It maps onto an open-source peer-to-peer platform.

Visual grounding is typically evaluated as a one-shot mapping from an informative referring expression to a visual target. This formulation misses a central property of real-world reference: target information is often incomplete, ambiguous, and established through interaction. We introduce a controlled evaluation framework for interactive visual grounding in large vision-language models (LVLMs), varying how much target information is provided upfront and how much must be acquired through dialogue. Across four human-grounded visual contexts and four interaction protocols, current LVLMs perform significantly below task-level human baselines. Interaction can help when follow-up questions refine or repair an initial target description. Performance is lowest when no initial description is provided and target information must be acquired through questions, indicating that proactive question-driven grounding remains difficult. LVLMs are also poorly calibrated, often reporting confidence that exceeds their empirical accuracy. Follow-up studies confirm these patterns across varied description sources (human versus AI), reasoning efforts, repeated interactions, description providers, and visual contexts. Overall, interactive visual grounding remains an important challenge, requiring visual matching, information seeking and synthesis.

Retrieval-augmented generation (RAG) improves the factuality of large language models by grounding responses in external documents, but it also exposes a critical security vulnerability: adversarial documents injected into the knowledge database can enter the context window and steer the model toward targeted incorrect answers. Existing post-retrieval defenses rely on instruction following, parametric knowledge, or text-level consistency, all of which can be imitated or optimized against by adaptive attackers. We propose RAGSentinel, a training-free, label-free defense for black-box RAG systems. RAGSentinel uses a surrogate encoder to measure query-conditioned hidden-state shifts induced by retrieved documents, removes shared topic directions, and filters poisoned documents as geometric outliers from a robust majority consensus. We prove that, under an honest-majority assumption and a representation-level separation condition, RAGSentinel exactly recovers a poison-free majority-sized context. Experiments across three question-answering datasets, three LLM families, and multiple poisoning attacks show that RAGSentinel consistently achieves low attack success rates while preserving competitive accuracy and remaining effective against adaptive attacks with full pipeline knowledge.

High-fidelity image-to-3D generation requires a 3D representation that captures both geometry and appearance. To support relighting and integration into standard rendering pipelines, the representation should include physically based rendering (PBR) modalities such as albedo, metallic-roughness, and surface normals. We propose Luce, a 3D representation that unifies geometry and PBR materials within a voxelized multimodal Gaussian cloud, using dedicated Gaussian primitives for each modality. A variational autoencoder compresses this representation into a unified material-aware latent space. A rectified-flow transformer generates this latent from a single image, conditioned on multi-layer features from a pretrained image encoder that preserve both semantic context and fine spatial detail. The latent then decodes into relightable PBR Gaussians and an optional textured mesh with a tangent-space normal map. On Toys4K, Luce achieves state-of-the-art single-image-to-3D generation, improving FID by 28% over the strongest baseline. We further introduce a benchmark of AI-generated images, on which Luce improves the CLIP image-alignment score over the best baseline (0.8519 vs. 0.8299). Luce generates relightable, geometrically accurate, and materially faithful assets that preserve fine details such as text, logos, and inscriptions.

We present a dynamical-systems based model for resting-state functional magnetic resonance imaging (rs-fMRI), trained on a dataset of roughly 40K rs-fMRI sequences covering a wide variety of public and available-by-permission datasets. While most existing proposals use transformer backbones, we utilize multi-resolution temporal modeling of the dynamics across parcellated brain regions. We show that MnemoDyn is compute efficient and generalizes very well across diverse populations and scanning protocols. When benchmarked against current state-of-the-art transformer-based approaches, MnemoDyn consistently delivers superior reconstruction quality. Overall, we find that with such large-scale pre-training on (non-proprietary) rs-fMRI datasets, we get a highly performant model for various downstream tasks. Our results also provide evidence of the efficacy of the model on small sample size studies which has implications for neuroimaging studies at large where resting state fMRI is a commonly acquired imaging modality.

Single-image 3D scene reconstruction must complete partially observed objects and place them coherently in a shared observation-aligned scene frame. Object-level generative priors offer strong completion ability, but their centered, scale-normalized outputs are typically expressed in an object frame, creating a fundamental representation gap between object generation and scene reconstruction. We introduce SceneReGen, a generative reconstruction framework that reinterprets scene reconstruction as the generation and assembly of complete object assets in a shared observation-aligned scene frame. SceneReGen addresses the generation-reconstruction gap through selective pose factorization: each object's observed orientation is encoded directly in the generated mesh, while translation and scale are estimated from instance-level and global scene evidence. Given a scene image and instance masks, a geometry encoder extracts dense cues; learnable shape queries condition a pretrained DiT-based 3D generator to produce complete meshes in their observed orientations, while position queries fuse object and scene features to assemble them in the shared frame. On the 3D-FUTURE evaluation subset, SceneReGen achieves the best scene-level CD, scene-level F-Score, and 3D bounding-box IoU among the evaluated methods, ties the best object-level CD, and ranks second in object-level F-Score. Qualitative outputs in autonomous-driving and embodied-AI scenes further illustrate the potential of asset-centric reconstruction beyond indoor furniture.

Video generative models provide rich physical priors for robot learning, yet existing world-action models (WAMs) face a fundamental trade-off: synchronous video generation at control rate is latency-prohibitive, while abandoning test-time visual imagination sacrifices task success. We show that visual imagination achieves both real-time inference and superior success rates when generated asynchronously off the critical path and consumed directly in latent space. We introduce GlanceWAM, which decouples imagination from control within a single video DiT: an asynchronous proposer glances ahead on a slow clock to imagine a single lookahead frame seconds into the future in the background, while an action head decodes action chunks at control rate (48 ms) purely in latent space without blocking. Enabled by a non-interfering attention mask that isolates video representations and staleness-robust horizon training that accommodates asynchronous lookahead aging, GlanceWAM breaks the speed-success dilemma. Trained purely on demonstrations, it attains 72.2% on the 24-task RoboCasa kitchen benchmark (surpassing synchronous Cosmos Policy at 67.1% and imagination-free co-training at 64.4%) and 99.0% on LIBERO, executing at 48 ms per chunk on an NVIDIA A100 GPU (24x faster than synchronous baselines). Code is available at https://github.com/linhanwang/GlanceWAM.

Recent Vision-Language Models encode high-resolution images into long visual token sequences, incurring prohibitive prefill costs. To compress them, existing methods score each visual token by averaging text-to-visual attention uniformly across all heads, which assumes every head matches the query. However, our empirical analysis shows that misaligned heads dominate the average, amplifying background tokens and drowning out fine-grained cues. To address this, we propose PAQ (Prompt-Grounded Attention Quality), a metric quantifying how well each head aligns the prompt with image regions. Built on PAQ, our pruning proceeds in three stages. Given a target FLOPs budget, we first partition the transformer layers into groups and allocate a visual token budget to each. Within each group, we then aggregate per-head attention maps via PAQ-weighted softmax into a group-level matrix. Finally, we score visual tokens by this matrix's magnitude and retain the allocated budget per group. By weighting heads with PAQ, our method scores tokens by attention signals that more faithfully reflect prompt relevance, rather than diluting them through uniform averaging. Across 18 benchmarks, our method delivers state-of-the-art trade-offs. Specifically, on LLaVA-1.5-7B (9 tasks), retaining only \textbf{5.6\%} tokens preserves \textbf{99.1\%} of the original performance, surpassing the strongest baseline AutoPrune by 4.2 points. Code is available in https://github.com/baokou-fw2/HAP.

We measure tablet-2, a production long-term memory engine for language models, on the text benchmarks the field already uses and on cross-lingual retrieval of photographs stored with no text at all. Its retrieval path contains no lexical matching, no keyword scoring, and no language model of its own. On LongMemEval-S (500 questions) it scores 95.7% [93.4, 97.1]; on BEAM-1M (700 questions, 2.21M stored memories) 67.5% [64.8, 70.2]. Those are question-sampling intervals, not the run-to-run spread, which is an order of magnitude narrower. Most of the paper is about how little they mean alone. Holding engine, corpus, settings and judge fixed, changing only the reader moves LongMemEval-S by 2.0 points; changing only the re-ask budget moves BEAM-1M by 8.9. Neither is stated in the reports we compare against, and the second exceeds most gaps there, so we give that table as a placement and not a ranking. For the multimodal axis we run two controls. Against BM25, configured as strongly as we could, we reach 95.2% mean recall@5 over 70 store-and-query language cells where BM25 reaches 19.0% and is exactly zero in 54. On captionless photographs a lexical method has no document to score at all. Open dense baselines on 300 Crossmodal-3600 photographs in 14 languages show that density confers no language independence: one scores 91.0% on English and 4.7% on Russian from identical image vectors, and a multilingual variant collapses on Telugu and Swahili. Our spread across languages is 14.0 against their 27.5 and 27.7. Three results run against us and are reported at equal weight: low-resource languages degrade sharply (Swahili 53.0%, Telugu 64.0%), attaching captions lowers cross-lingual retrieval by 11.4 points, and one setting omitted into one stage of our own retrieval cost 37 points of Korean top-1 accuracy while leaving nine languages untouched.

Large Language Models excel at code generation, yet competitive programming exposes a persistent failure mode: existing multi-agent pipelines distribute work over generic planner, coder, and debugger roles and delegate the choice of algorithmic technique to the backbone alone. We present MARS (Multi-Agent Relay of Specialized LLMs), a prompt-only framework in which each agent is a topic specialist---dynamic programming, graphs, strings, geometry, and so on---grounded by retrieval-augmented generation over an algorithm-theory corpus. Given a problem, retrieval selects a small team of relevant specialists; a starter writes an initial C++17 solution, and each subsequent turn runs the candidate against public examples in a sandbox, lets the active specialist keep, repair, or hand off the draft, and forwards a structured packet to the next specialist. A single infrastructure-fixer pass normalizes boilerplate at the end. On the CodeContests test split with Gemma 4, MARS reaches $0.624 \pm 0.006$ pass rate at $2.3$ recorded pipeline stages per task ($+14.4$ percentage points over direct prompting), closing most of the gap to CodeSIM ($0.731$) at $3.3{\times}$ lower wall-clock cost and substantially smaller variance in per-task token spend. The source code is available on GitHub: https://github.com/fckand/mars.

Common shortest-path algorithms, such as Dijkstra's (SPF), that OSPF uses, provide exact routing solutions but must be recomputed for each network topology, limiting scalability in dynamic or large-scale networks. This paper proposes the GATNextHop model to determine whether a Graph Neural Network, namely the Graph Attention Network, can approximate shortest paths and generalize across topologies. By training on synthetic graphs and evaluating on real-world Internet Service Provider networks from the Internet Topology Zoo, we aim to benchmark our model's ability to learn routing heuristics that transfer across network structures. Performance will be evaluated in terms of accuracy, inference speed, and generalization, comparing the GNN against Dijkstra's algorithm to quantify trade-offs between learned and classical routing approaches.

Visual foundation models are commonly adapted under the assumption that the appearance of incoming data may change while the semantic meaning of the prediction task remains fixed. In long-lived visual systems, however, taxonomies, policies, and concept definitions can themselves evolve, causing the same visual evidence to require a different interpretation. We study this setting as evolving semantic concept shift and introduce SemReWrite, a framework for selectively updating obsolete visual--semantic mappings while preserving knowledge that remains valid. SemReWrite represents changes between old and revised semantic specifications, combines semantic discrepancy with sparse revised supervision to localize affected visual regions, and uses an input-dependent low-rank rewriting mechanism together with structured semantic memory, preservation, and obsolete-decision suppression. We further introduce EvoShift-Bench, spanning ImageNet, iNaturalist, CUB-200-2011, and DomainNet, with semantic transitions including class split, merge, boundary revision, insertion, partial redefinition, recurrence, and mixed semantic--appearance shift. To explicitly evaluate selective semantic revision, we introduce Rewrite Accuracy (RA) and Preservation Accuracy (PA) for affected and unaffected regions, respectively, Obsolete Retention (OR) for measuring residual outdated semantic associations, and the Selective Revision Score (SRS), which jointly summarizes rewriting and preservation performance. Experiments show that SemReWrite achieves a stronger balance between learning revised semantics and retaining unaffected knowledge than prompt replacement, conventional fine-tuning, parameter-efficient adaptation, and continual-learning strategies.

Real-time optimization (RTO) relies on process models to locate economically optimal operating conditions. Because developing first-principles models requires significant process knowledge, data-driven alternatives are increasingly attractive. Modern machine-learning models can fit historical plant data accurately and often pass standard validation tests. Whether such models can be trusted for economic optimization, however, remains unclear. We investigate this question using a vinyl acetate monomer benchmark process with a unique, well-conditioned economic optimum. We train a structured hybrid model that combines known mass balances and thermodynamics with a neural-network closure for unknown kinetics, and a fully data-driven neural ordinary differential equation (ODE) model. Both models reproduce plant measurements accurately and exhibit little variation in predictions across random initializations. Yet their economic optima differ substantially from that of the plant. Where the plant returns a single optimum on multistart search, the trained models return many phantom optima. We further show that the training optimizer alone can be yet another source of error. Even with noise-free data and initialization at weights that recover the plant optimum, stochastic gradient training can drift to weights that yield substantially worse RTO solutions. The identified model is thus an artifact of the training optimizer as well as the data. These results demonstrate that a good predictive fit of all available measurements does not guarantee reliable economic performance. A data-driven model for RTO should at least be required to recover the optimum on a decision-oriented benchmark like the one developed here before being considered for plant testing and application.

Inferring the collective emotional state of a group of people from a single image, a task known as group emotion recognition (GER), requires integrating spatially distributed cues such as faces, poses, interactions, and scene context. Current methods rely on detector-driven multi-stream pipelines. These are trained with only image-level supervision that lacks guidance on which regions matter or how strongly each contributes. We propose LG-GER, a language-guided distillation framework that uses a multimodal large language model (MLLM) to generate dense, spatially grounded evidence, i.e., bounding boxes paired with emotion signals and confidence scores, for the training images. This structured evidence is distilled into a single vision-language model (VLM) backbone through four complementary losses: classification, region-text grounding, spatial emotion, and spatial confidence regression. At inference, LG-GER requires no detectors, no MLLM, and no multi-stream fusion, making GER practical for real-time and resource-constrained deployment. LG-GER has been evaluated on two benchmark GER datasets (GroupEmoW and GAF~3.0) and achieves competitive or superior results compared to state-of-the-art methods that require detection and multi-stream processing at inference.

Everything a language model sees is tokens. The serving stack knows what each span is -- user input, tool output, instructions -- but the model must keep track of that itself, and it can lose track or be confused: text can be written to read like anything. Prompt injection is a natural exploit of this phenomenon. By scrambling the model's understanding of span identity, an attacker can induce unwanted and potentially dangerous actions. Adding a non-textual channel to the model's input -- a way to communicate span identity beyond text -- mitigates this class of attack. We thus introduce a general steering technique called Semantic Overlays: small learned adapters applied at chosen prefill positions to a frozen model's residual stream. Laying an overlay over a span creates an out-of-band annotation channel that cannot be replicated by tokens. Unlike steering vectors, Semantic Overlays are trained, adaptable, and selectively applied. An overlay can encode complex semantics that reshape how the model perceives the marked span: asked to copy a code snippet under an overlay asserting that it is in a different programming language than it is, the model rewrites the snippet, faithfully, in the asserted language. Overlays are also composable, allow for transparent reading of underlying content, and can carry complex payloads -- including imperatives that the model will follow. An overlay which marks a span as "non-executable" defends against the broad class of prompt injections that add instructions in untrusted context. We report strong results on prompt injection benchmarks: SEP separation rises from 24.3% to 96.5% with utility unchanged (our scoring rule; we also correct a defect in the published grader), TensorTrust attack success rate falls from 34.8% to 6.6%, and all four PIArena attack families drop to 0% compliance, all while marked spans stay readable (92.5% exact copy rate).

While state-of-the-art generative models produce high-fidelity 3D meshes, these outputs lack the physical properties required for interactive simulation, gaming, or robotics. We introduce Gen2Physics, a unified and automated framework that grounds generated meshes in physics by automatically decomposing them into their constituent material components. Unlike prior approaches, which focus on volumetric representations incompatible with standard physics engines, Gen2Physics operates directly on meshes to produce immediately simulation-ready assets. Our pipeline integrates a fine-tuned Vision Transformer for dense material segmentation, a robust 2D-to-3D consistency projection, and a Vision-Language Model (VLM) guided refinement that leverages contextual reasoning to assign physical properties and infer internal geometry (solid vs. hollow). By converting surface patches into volumes with distinct densities, our method enables physically plausible dynamic simulations. Experimental results on the ABO-500 and PartNet-Material benchmarks demonstrate that Gen2Physics more than doubles the material segmentation accuracy of prior physics-grounding pipelines (15.6 to 48.3 mIoU), while matching the mass-estimation accuracy of volumetric methods and being the only approach to output watertight per-material sub-meshes.

As LLM agents proliferate, built by different parties and with different capabilities and costs, orchestrating them is more like assembling labor across the economy than a computer calling a subroutine. Existing orchestration is typically centralized, with a single planner assigning every task, but this creates a bottleneck as agent pools grow, requires private information (e.g., agents' execution costs), and can easily be manipulated, such that a single inserted preference nearly doubles a favored agent's task share under a centralized LLM allocator. We introduce AgentLance, a repeated labor market in which agents bid on tasks using their private costs and self-maintained strategy notes, an allocator selects winners from bids and public reputation records, and a VCG-style payment rule rewards cost-aware bidding. Complex tasks are handled by hierarchical delegation: winning agents can decompose work and subcontract it through the same mechanism. Across mathematical reasoning, code generation, knowledge-intensive QA, and agentic tasks, AgentLance matches agents to their specializations, shifts work toward cheaper agents as cost sensitivity rises, and consistently outperforms single-model, centralized-orchestration, and market baselines. Diagnosing market failures, including inaccurate cost self-estimation and sub-optimal bidding, then correcting them in controlled experiments yields further gains, charting a path toward more efficient agent economies.

Robots are beginning to act on world-model predictions, yet reliability is still expressed through instantaneous, model-internal signals. DreamLedger instead treats reliability as a persistent deployment object: an execution-settled credit file recording how often consumed predictions are borne out, indexed by operating condition, region, and prediction horizon, and consulted before each use. Each consumed prediction is registered as a claim; attributable outcomes are settled against arriving reality at zero labeling cost, an attribution stage excludes measurement-contaminated outcomes, and a settlement-supervised head complements sparse bins. The resulting credit gates consumption: low-credit predictions shorten the dependent horizon or trigger additional observation; every reliance event remains auditable via dependency tickets and replayable logs. We evaluate DreamLedger in three simulated domains (indoor flight, tabletop manipulation, 2D navigation), via mounts on unmodified DreamerV3, TD-MPC2, and V-JEPA 2-AC, and on a real Franka manipulator. Claim failure is dose-monotone in all 12 held-out condition-horizon cells. Credit-gated planning reduces burned imagination (consumed claims that later fail to redeem) by 62% (95% CI 43-81%) versus blind consumption, with equal success and comparable collision rates. At matched risk targets, persistent books cut verification probes from 1.00 to 0.36/episode in manipulation, at success 0.94 versus 0.98; settlement-grounded calibration retains moderate, seed-consistent operating points unlike raw instantaneous gates. The same trust layer operates across decoder-, latent-, and token-space interfaces, including V-JEPA 2-AC settled on real robot frames. On hardware, settlement remains operational under real sensing and contact noise, a deployment failure loop is re-priced online, and all 1,062 registered spends replay from the audit logs.

The interpretation of endoscopic imagery in ulcerative colitis is complex and subjective, with variability in human assessment and subtle mucosal inflammation. Although deep learning has advanced automated analysis, most vision-language models rely on global visual embeddings that overlook the localized and relational nature of pathological evidence, limiting clinical reliability and interpretability. We introduce LUX (Lesion-aware Unified eXplainable captioning), a graph-conditioned vision-language architecture for explainable endoscopic image captioning. LUX constructs a lesion-centric scene graph from Grad-CAM and CBAM activation maps, representing pathological regions as nodes and encoding their spatial and clinical relationships. These graph embeddings are integrated into the cross-attention layers of a T5 decoder, enabling generated words to attend to specific lesion nodes rather than only to global image features. This provides direct alignment between linguistic content and pathological evidence, supporting token-level interpretability and relational reasoning. LUX outperforms strong baseline and state-of-the-art medical captioning models across BLEU, METEOR, ROUGE-L, and CIDEr, with particularly strong gains in CIDEr. It also reduces hallucinated clinical findings and improves lesion-level grounding through stronger correspondence between generated tokens and localized pathological regions.

Grammatical knowledge and how it is empirically tested are typically considered robust to the frequency of the lexical items in the expressions. However, neural network-based models of grammaticality exhibit high sensitivity to lexical frequency. We draw upon Complementary Learning Systems theory to test the hypothesis that robustness to lexical frequency can arise via a hippocampal episodic memory mechanism, which enables rapid encoding and retrieval of specific experiences and allows learners to leverage them when processing rare patterns. We use retrieval-augmented language models as an instantiation of such an episodic memory mechanism (specifically, $k$-nearest-neighbor language models that augment parametric models with explicit instance storage), and test whether this augmentation helps close the lexical frequency gap that vanilla language models exhibit in syntactic contrast tests. Using syntactic contrasts with frequency-stratified test items, we find that retrieval augmentation narrows the performance gap between high- and low-frequency items, consistent with episodic memory compensating for weak parametric representations. This benefit is consistent across different syntactic phenomena and across models pretrained on child-realistic and large-scale data. Additionally, we show that structural information is critical for effective retrieval, whereas semantic similarity alone provides little benefit. While these are promising proof-of-concept results supporting our hypothesis, the frequency gap is narrowed rather than fully closed. Based on our analyses, we propose preferential reweighting of retrieved instances, better representations and retrieval strategies for structural information, and flexible configurations of storage and retrieval as promising future directions for improving the implementation of episodic memory in language models.

Object-counting methods have rapidly shifted from class-specific density regression to open-vocabulary, foundation-model-backed counters. These methods now enumerate instances from various visual and textual prompts. While this shift marks major conceptual progress, our survey argues that claims of universal generality have outpaced the evaluative infrastructure. Most progress metrics rely on a few saturated benchmarks that models exploit for statistical regularities. Newly introduced diagnostic datasets reveal systematic failures in semantic grounding, temporal identity, and spatial reasoning with occlusion. To address these failures, we introduce a five-axis taxonomy (modality, mechanism, prompting, supervision level, and generalization setting). We use this taxonomy to audit the literature across application domains, including microscopy, remote sensing, crowd counting, and agriculture. This formalizes prevailing challenges into six structural contradictions. From these, we propose a roadmap for compositional scene understanding, active counting agents, and unified multimodal evaluation protocols. The main imperative is to build a robust evaluation infrastructure to distinguish open-world generalization from benchmark-specific optimization, rather than simple incremental engineering.

Embodied Agents System (EAS) are increasingly deployed in open-world physical domains, where reliability directly dictates deployment quality and human-agent trust. However, existing evaluations rely on outcome-centric metrics as success rate or safety scores that collapse diverse execution trajectories into coarse scores, obscuring the dynamic processes underlying agent behavior. Therefore, they ignore a critical property of EAS -- which we define as the Resilience -- that reflects how EASs recover, stabilize, and extend under perturbations and across iterative updates. The lack of resilience is particularly critical in open-world environments due to continuous unexpected disruptions, thus directly affecting the quality of EAS deployment. To address this problem, we gain insight from the resilience-engineering concepts to EAS groundings and propose a novel resilience evaluation framework that can be flexibly applied to any EAS. Specifically, we define the first comprehensive resilience metrics suite for EASs system that exposes Rebound, Stability, and Graceful Extensibility across embodied tasks execution, providing a practical grounding for EAS resilience analysis. We further implement the resilience evaluation layer that transforms execution process into assessments for diagnosis and optimization. Across 400 household tasks with 10 EAS, we reveal the process-level distinction hidden by outcome metrics, including recovery cost differences among successful episodes ($ΔC_{rec}=25.2$), increased instability and task-family degradation. Metrics-guided optimizations reduce recovery cost and increase stability, graceful extensibility completion, showing the diagnostic effect of resilience evaluation. Our results reveal a trade-off among resilience characteristics, suggesting that a resilient EAS construction should be configured according to deployment-specific requirements.

While reinforcement learning (RL) allows generalist robot policies to continually improve during deployment, the large model size of modern generalist policies, such as VLAs, poses a fundamental obstacle to effective RL improvement. In particular, their severe inference latency---which can lead to pauses or jerky movements---can alter the effective environment dynamics and, if not correctly accounted for, break the Markov assumption that RL relies on, causing standard RL algorithms to fail completely. In this work, we introduce a latency-aware framework, Asynchronous RL with Intermediate Information (ARLI), that enables RL-based improvement of generalist policies under inference delays. Our framework builds on asynchronous inference approaches, which interleave action generation with execution to hide latency, and addresses its incompatibility with RL by providing a low-latency RL policy design that maximizes reactivity within the inference window through two contributions: state augmentations that restore near-Markovian structure by incorporating committed actions and a mid-inference observation. We evaluate our approach across simulated and real-world manipulation tasks, and find that it enables effective finetuning under inference delays where standard RL fails entirely, even matching or exceeding the performance of standard RL in idealized no-latency settings.

RL has emerged as a powerful paradigm for enhancing the instruction following capabilities of LLMs. While existing training recipes achieve substantial gains, we find that they suffer from exploration bias towards easy instructions when the training data has multiple instructions in a prompt. This bias is caused by two main reasons: 1) the policy model's initial ability to satisfy hard instructions is too low to trigger successful exploration during RL training, so the optimization is biased towards easy instructions; and 2) canonical RL training recipes typically employ a cumulative reward (the number of instructions fulfilled), treating all instructions equally, which biases the policy model towards fulfilling easy instructions to obtain the same amount of reward. To address these issues, we first propose two metrics to measure the exploration bias in instruction following and then introduce a two-stage framework to alleviate it: 1) Behavioral Bootstrapping, a lightweight rejection sampling fine-tuning stage before RL to activate hard instructions; and 2) Scarcity-Aware Rewards, a new RL reward function that assigns rewards to instructions based on their empirical scarcity. Experiments show that the proposed metrics are highly correlated with model performance, and our methods unleash the potential of RL training: our best models outperform the baselines by a significant margin across three verifiable instruction following benchmarks. We release codes at https://github.com/mianzhang/MulIF.

Transformer-based language models are widely used as models of human language processing, yet their attention mechanisms allow lossless access to the full preceding context, unlike the limited memory systems of humans. We hypothesize that installing memory constraints into transformers' attention mechanisms can improve their fit to human behavioral data. While previous work has explored individual constraints in isolation, we conduct a systematic comparison of multiple attention-based memory mechanisms across different model sizes and training corpora, evaluating both psychometric predictive power for human reading times and grammatical competence. We additionally compare static constraints, in which the constraint strength is fixed throughout training, to dynamic memory curricula. We find that constraints that are sensitive to the content of intervening tokens consistently achieve the highest alignment with human reading times, outperforming distance-based constraints. We observe a dissociation between psychometric fit and grammatical competence under dynamic memory curricula, suggesting that Transformers cannot serve as a one-size-fits-all cognitive model.

Large Language Models (LLMs) demonstrate strong capabilities in automated essay scoring (AES), but contemporary approaches typically employ fixed prompt selection, failing to address operational cost concerns and evolving optimal configurations. We propose a novel cost-aware approach that treats each prompt type as an arm in a multi-armed bandit (MAB) controller, enabling adaptive selection of optimal prompting strategies during inference. Our experiments on IELTS Writing Task 2 essays show that the MAB framework achieves comparable scoring accuracy to exhaustive grid search while reducing LLM calls by 78.4\% to find the best grading approach. We implemented four distinct grading recipes (multi-step vs. single-step assessment, with vs. without calibration examples) and found that the multi-step approach with examples achieves the highest accuracy. By tracking token usage and latency alongside agreement metrics, we produce the first cost-reliability learning curves for essay scoring, providing actionable insights for educational technology platforms that must balance operational costs against assessment validity. This work represents the first application of online control mechanisms to adaptively select prompting strategies in AES, transforming prompt selection from an offline hyperparameter optimization problem into an efficient online learning task.

Designing effective reward signals for open-domain question answering is challenging because high-quality responses must simultaneously satisfy multiple aspects of answer quality that are difficult to capture with a holistic scalar objective. We introduce a rubric-based reward framework that generates query-specific rubrics grounded in retrieved evidence and decomposed into multiple quality dimensions, providing fine-grained supervision during post-training. Averaged across three evaluation axes (composition, grounding, and instruction-following), our approach improves over the instruction-tuned baseline by 6.5% and over flat rubric variants by 4%, with consistent gains across all evaluation datasets. Conditioning rubrics on retrieved evidence improves factual support, while decomposing rubrics into quality-specific dimensions further improves coherence, organization, and adherence to query requirements. Our results show that grounded, multi-dimensional rubrics provide more effective reward supervision for complex open-domain question answering.

We introduce Giga-Embeddings, a family of text embedding models designed to combine strong retrieval quality with efficient serving. Its largest member is a sparse 10B-parameter Mixture-of-Experts encoder with approximately 1.8B active parameters per token. Across English, Russian, multilingual, and code MTEB benchmarks, this model achieves the strongest aggregate performance within the family on all four evaluated suites. In our vLLM benchmark with 1024-token inputs, it processes 114.5k tokens per second, providing 25 percent higher throughput than the dense 3B model and 1.56-2.65x the throughput of the evaluated external systems. The family also includes a dense 3B encoder and a distilled 480M encoder for tighter compute and memory budgets. We train the compact model using a dimension-agnostic objective that aligns teacher and student similarity distributions. The resulting 480M model scores 70.98 on Russian MTEB, surpassing FRIDA while using 42 percent fewer parameters. We release all three model checkpoints.

Lung cancer tissue diagnostics is complex, as therapy decisions in precision oncology rely on the integration of histomorphological, immunohistochemical, and molecular features. Yet pathological assessment remains largely visual and semi-quantitative and shows interobserver variability, while existing artificial intelligence (AI) tools cover only selected tasks, rarely reach generalizable expert-level performance, and lack prospective clinical validation. To address these challenges, we developed and clinically validated LUCAID, an agentic AI system for precision lung cancer pathology. An integrative agent couples diagnostic reasoning with nine modules that cover the full routine workflow, from quality control, tumor detection and segmentation, histological subtyping, tumor microenvironment profiling, tumor cellularity quantification, and predictive biomarker scoring (PD-L1, MET, TROP-2) to automated structured report generation. LUCAID enables users to interactively query the module outputs and generate reports that contextualize the results. Against large-scale expert ground-truth annotations, the analysis modules achieved F1 scores of 0.82-0.95. In prospective clinical validation, LUCAID reached 93.0% concordance with an expert-panel adjudicated reference standard across clinically actionable decisions, compared with 68.3-81.1% for five experienced thoracic pathologists.

Recent progress in image restoration has converged on all-in-one architectures that jointly handle multiple degradations within a single network. These methods are effective on static benchmarks but target a closed-world setting that assumes simultaneous access to every target degradation at training time. In practice, degradations are encountered sequentially as field-deployed systems progressively face new environmental conditions, and historical training data is often unavailable due to privacy or storage constraints. Accommodating a new degradation then requires either retraining on the union of all prior data, which is often costly or infeasible, or fine-tuning, which causes catastrophic forgetting. We formulate multi-degradation image restoration as a continual domain-incremental learning problem, in which degradations arrive incrementally and prior data is unavailable. Our proposed Restoring without Forgetting (RwF) framework learns a lightweight adapter for each new degradation, eliminating forgetting by construction at a fraction of the cost of dedicated per-domain networks. To isolate degradation learning from dataset variation, we construct a benchmark spanning five degradation domains under shared image content. At test time, an unsupervised routing mechanism identifies the appropriate restoration path for unknown inputs without requiring domain labels. Across the five-domain sequence, RwF improves final average PSNR over naive sequential fine-tuning by 15.25 dB and 11.83 dB on the Restormer and NAFNet backbones, respectively. The framework transfers to eleven canonical real-degradation benchmarks (3,465 images) at 89.5% routing accuracy with only a +0.94 dB oracle PSNR gap, establishing, to our knowledge, the first systematic baseline for continual multi-degradation image restoration.

How does an intelligent visual system combine what objects look like with how they move while remaining robust as appearance changes? We addressed this question by comparing human perception and neural activity in macaque inferior temporal cortex with representations from image- and video-based neural networks spanning recognition, segmentation, optic-flow processing and predictive world modeling. Temporal integration improved object representations, but most video recognition models generalized poorly when appearance was disrupted while motion structure was preserved. Humans and macaque IT remained robust. Notably, predictive world models combined strong cross-appearance generalization with the closest correspondence to IT, outperforming other video-modeling approaches in neural fidelity. Yet no model reproduced the cortical transformation from early appearance-dominated responses toward later appearance-invariant motion coding. These results identify progressive integration of motion into object representations as a principle of robust dynamic vision and implicate predictive learning as a promising route toward realizing this computation in artificial systems.

Between AI-assisted item generation and expert review sits a computational evaluator whose decisions are usually treated as technical preliminaries. Yet representation, structural reduction, and selection policy determine which items and evidence psychometricians ever receive. Across two linked in-silico studies of 32,000 selected Big Five items, we followed fixed source populations from semantic representation through structural evaluation and candidate-form construction. Broad agreement in semantic geometry concealed consequential local differences: identical wording acquired different construct evidence, different items survived, and intended attributes could disappear even as community correspondence improved. These sensitivities also differed across generated source populations. At the final review boundary, both eligibility policies filled every content cell in every evaluable form, yet they presented different wording. Across embedding configurations, inclusive primary forms shared a median of only 6 of 40 items, reflecting the total downstream consequence of changing representation across structural evidence and ranking. The apparent stability of global summaries and complete forms therefore concealed instability in the content reaching psychometricians. The computational evaluator is not neutral infrastructure between generation and expertise; it is an inspectable and revisable part of measurement design.

Rapid and accurate fault detection in high-voltage transmission networks is essential for grid reliability and equipment protection. Transmission fault datasets are frequently imbalanced, and certain fault types produce electrical signatures that fall within the normal operating envelope, causing single-model classifiers to fail on safety-critical cases. This paper proposes a hybrid two-stage machine learning pipeline that decouples detection from classification. Stage 1 combines an Isolation Forest anomaly detector with an optional supervised binary detector through an OR-fusion rule; the supervised branch is allocated automatically during training for any fault class the anomaly detector cannot resolve, and is omitted when no such class exists. Stage 2 applies a Random Forest multiclass classifier only to samples flagged by Stage 1. Feature engineering is expressed as a per-measurement-point operator mapping six raw channels to eighteen features, including zero-sequence symmetrical components derived from Fortescue's theorem, yielding 18L features for L measurement points. On the TLFaultDataset, the pipeline raises Line-fault end-to-end accuracy from 31.3% to 95.8%. On an independent single-point dataset, the same framework attains 97.25% end-to-end accuracy across all classes including normal operation, exceeding the TLFed federated benchmark of 94.84% without GPU or federated infrastructure, at 0.05 ms per sample on CPU. Ablation on both datasets shows zero-sequence features resolving the three-phase versus three-phase-to-ground ambiguity, raising the F1-score of that class pair from 0.39 to 0.997. The direction of the zero-sequence signature is found to be system-dependent, motivating a learned decision boundary in place of a fixed relay threshold.

Aligning large language models to human-centered objectives is difficult when targets are non-executable and context-dependent, limiting reliable verification and scalable supervision. Although synthetic data expands coverage, weak verification shifts the bottleneck from generation to selection. Noisy signals destabilize iterative refinement and can cause silent regressions. We propose Agentic Data Evolution (ADE), a data-centric framework that organizes synthetic supervision as evolving data snapshots. ADE improves data snapshots through a closed-loop Observation-Variation-Selection (OVS) procedure, where a steady-state admission mechanism acts as a quality ratchet that conservatively gates updates for sustained cross-round improvement. We validate these improvements through complementary intrinsic trend tracking and extrinsic post-training evaluation. On DEV300, ADE raises the intrinsic win rate from 50% to 75.81% and the extrinsic win rate from 55.20% to 68.86%, consistent performance gains across diverse benchmarks. Blind expert evaluation further confirms this, with a 66.11% preference for evolved answers. These gains extend across post-training methods, model scales, and tasks beyond the target weakly verifiable educational objectives. Resources are available at https://github.com/ZeroLoss-Lab/Agentic-Data-Evolution.

We study autonomous mathematical discovery in the Station, an open-world multi-agent environment in which AI agents from different model families pursue a shared research goal without a central coordinator or scripted pipeline. Agents choose their own research directions, conduct experiments, collaborate, and build a shared scientific literature. Across 12 construction problems from the AlphaEvolve catalogue and two additional case studies, the Station obtained results novel relative to the prior literature on five problems: a new infinite family of finite-field Kakeya sets, new exact 604-point kissing configurations in dimension 11, new records for the discretized Kakeya needle and sign uncertainty problems, and a substantially improved lower bound for Erdős's minimum-overlap problem. Agents also discovered novel infinite families for Book Ramsey numbers. Importantly, the agents produced not only numerical constructions but also theorems and analyses explaining how those constructions work, making the results more interpretable and easier for mathematicians to build upon. We release all raw agent dialogues, proofs, and verification code, providing a transparent record of how these discoveries emerged.

Modern software systems accumulate technical debt over decades of development, which makes migration expensive and largely manual. As coding agents become increasingly capable at bug fixing, can they autonomously perform such migrations? Existing benchmarks cannot answer this question because they evaluate only behavioural correctness, not whether the migration actually occurred. This leads an easy hack: agents copy the original implementation to make tests pass. We call this Blindness. To address this problem, we introduce SWE Refactor Bench, a benchmark comprising 20 whole-repository migrations, covering 4 kinds of technical debt. A three-stage evaluation protocol measures both migration completeness and behavioural correctness. (1) Migration Audit verifies that the migration occurred. (2) Behavioural Tests measure correctness with a fixed test suite. (3) Agentic Verification uses 6 independent coding agents to generate targeted tests for hidden behavioural differences. Across 520 runs from 8 frontier models and 26 model-effort configurations, only 28 of 520 runs ($5.4\%$) pass all three stages, 13 of the 20 tasks receive no accepted solution, and the best model (claude-opus-5) scores $47.0/100$. Migration completeness and behavioural correctness are distinct abilities: a few runs preserve behaviour by skipping the migration and are stopped at Migration Audit; most attempt it and break behaviour, and are stopped at Behavioural Tests. Agents cannot deliver a perfect migration: among the 340 runs that pass Migration Audit, $58\%$ reach $99\%$ of the fixed checks, yet only $26\%$ reach $100\%$. Agent capability differs across migration categories: agents score $31.4$ on build toolchain rewrites but only $5.6$ on language rewrites. Together, these findings position SWE Refactor Bench as a rigorous testbed for developing coding agents for reliable whole-repository migrations.

LLM-based agents execute multi-step tasks, but their behavioral structure remains opaque: long unstructured traces resist the safety auditing and runtime monitoring that deployment requires. Existing approaches operate per-trace or success-only, so they miss the cross-run topology that links next-step and failure prediction. To recover that shared structure, we collapse an entire trace corpus into a single, compact finite-state machine (FSM) that serves as a structural substrate for the otherwise unpredictable behavior of LLM agents. Across twelve public datasets, the FSMs are compact (7-43 states), replay held-out data at >=0.997 fitness with near-identical topology across splits, and build in milliseconds. This substrate addresses both prediction goals. For next-step prediction, FSM-state context outperforms Agent Workflow Memory on every ground-truth-matched dataset. For failure prediction, per-state behavioral features reach held-out AUROC up to 0.94, and an online monitor ranks failing runs above passing ones from a partial trace, triggering early stopping well before completion. Behavioral topology thus appears shaped more by the deployment harness than by the LLM, providing a model-agnostic structural primitive for safety auditing and runtime monitoring.

Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context. Prime Agent is an open-source harness for long-horizon evaluation and coding-agent workflows. A persistent IPython REPL follows the Recursive Language Model abstraction for programmatic context processing and test-time compute, while Continual Harness preserves histories, memories, skills, prompts, and subagent specifications across trajectories. Recursive subagents coordinate through direct agent-to-agent communication, and the Agents View lets humans inspect and manage daemon-backed sessions. Prime Agent standardizes execution, recovery, verification, and resource accounting while leaving strategy construction to the model. This low-friction, expressive membrane prevents harness failures from becoming model failures and pushes measurement toward the model's true maximal underlying capability. Prime Agent raises ARC-AGI-3 RHAE Best@1 from 30% to 95.5% and matches or exceeds native and popular harnesses across long-context coding, GPU-kernel generation, emulator construction, and autonomous nanoGPT speedruns. On Factorio, we find refinement allows for continuous technology progression and dedicated subagents enable parallelized work. Code is available at https://github.com/PrimeIntellect-ai/prime-agent.

Does multi-agent LLM interaction help or hurt? Some work reports gains from debate (Du et al., 2024), critique loops (Chen et al., 2025), and mixture-of-agents synthesis (Wang et al., 2025), while other work finds that interaction adds cost without improving quality under equal budgets (Tran & Kiela, 2026; Xu et al., 2026; Jarrett et al., 2025), or that independent sampling already captures multi-agent gains (Li et al., 2024). We argue this contradiction partly reflects a missing distinction, because not all multi-agent communication is equal. Different model families find structurally different solutions, but when agents read each other's complete outputs, their proposals converge within one round, erasing the diversity that motivates using multiple models. We call this the interaction tax. We test 11 verifier-scored optimization tasks under matched budgets and find that full-solution interaction is a weak default. Independent proposal generation avoids this collapse. Full-solution interaction mainly makes agents stay close to the first solution they see instead of trying different approaches, and critique helps only if the violated rule is easy for the LLM to find and fix. These results suggest that multi-agent performance depends less on the number of agents than on the information they exchange, and interaction helps only when agents share the right information at the right time.

Historical people may appear under different languages, scripts, and transcription traditions, while distinct individuals may share highly similar or even identical names. This makes historical identity reconciliation more than a problem of string matching or transliteration. We introduce MHER, a provenance-controlled benchmark for pairwise reconciliation of person-name attestations from the Mongol world. MHER contains a balanced 396-pair Name-only core over 84 primary historical persons and a stricter 160-pair Source-grounded subset constructed from mention-by-source evidence, with entity-disjoint development and test splits. Across five generative systems, correctly Source-grounded evidence improves paired TEST accuracy by 12.96 to 94.44 percentage points relative to Name-only input. On five identical-surface different-person cases, all models fail under names alone (0/25 model-item decisions), whereas Source-grounded evidence yields 24/25 correct resolutions, with the remaining output an abstention. Context-only ablations show that historical descriptions often carry substantial identity information, while explicitly signaled misgrounding controls produce substantially lower performance. We also find that names are not uniformly beneficial: for Qwen3-8B, restoring surface forms converts ten otherwise correct Context-only distinctions into false identity merges. These results show that historical entity reconciliation depends not only on surface correspondence, but on whether identity judgments respond appropriately to provenance-controlled historical evidence. MHER therefore provides a controlled framework for studying evidence use, abstention, and failure modes in historical NLP.

Reasoning-Induced Misalignment, where fine-tuning on reasoning data containing no harmful content, including mathematics, code, and problem-solving with chain-of-thought traces can induce harmful behaviors of LLM, posing a serious challenge to the safety of LLM reasoning. Cross-architecture, cross-scale, and cross-dataset checks show that RIM does not always emerge. Previous work attributed RIM to neuron-level entanglement, but did not identify the geometry of the representation space underlying this entanglement or propose a training-time fix. We provide both: a representation-space analysis of RIM and the Safety-Direction Penalty (SDP), which penalizes movement along a learned safety direction during reasoning fine-tuning. The analysis extracts two activation-space directions, one encoding reasoning ability and the other safety behavior. These directions are coupled: fine-tuning that improves reasoning shifts safety representations, and prompts with larger shifts show larger safety degradation. CKA distance ratios and probes locate the safety-decision layers where this shift is most relevant. These findings guide the design of SDP: the coupling motivates penalizing displacement along the safety direction, and the layer localization sets the initial scope. When the initial scope leaves compensatory shifts beyond the penalized layers, the same diagnostics guide iterative expansion. On Qwen2.5-3B and 7B, SDP restores safety while preserving benchmark reasoning performance.

World action models (WAMs) have recently gained increasing attention as a framework for jointly modeling scene evolution and ego actions in autonomous driving. Most existing WAMs learn scene dynamics in pixel space by combining a video-generation backbone for future-observation prediction with an action head for ego-trajectory prediction. Pixels, however, provide only an indirect representation of these dynamics: they entangle geometry and motion with appearance, texture, and illumination, forcing the model to infer three-dimensional transformations from two-dimensional observations. We argue that geometry, represented by point clouds, offers a more natural state space for driving because it explicitly captures spatial structure and the rigid and non-rigid transformations that govern scene evolution while directly aligning with the space in which driving actions are executed. Building on this insight, we introduce \textbf{GeoWAM}, a visual geometry world action model for autonomous driving. Rather than predicting future images, GeoWAM is pretrained to forecast future scene geometry, yielding representations that jointly encode spatial structure and temporal evolution. A geometry-conditioned action head then leverages these learned geometric dynamics to predict future ego trajectories. Extensive open-loop and closed-loop evaluations show that visual geometry world modeling yields substantially stronger driving policies than image-based alternatives, establishing future-geometry prediction as an effective pretraining objective for autonomous driving.

Vision-Language-Action (VLA) models can turn multimodal context into robot actions, but their action decoders are still trained largely by behavior cloning. This supervises which motor command was demonstrated while leaving implicit the local objective served by the behavior under the instruction. Future-based supervision enriches action learning with frames, latent observations, trajectories, or motion representations, but these signals capture particular realizations of what may happen rather than the shared semantic objective of the forthcoming behavior. We propose Intention Distillation (INDI), which distills behavior-level intent into the action decoder. During training, a frozen teacher VLM interprets a demonstrated segment from the current observation, instruction, coarse action summary, and corresponding execution video. From its standard inputs, the deployed VLA recovers the resulting multimodal intent representation at an intermediate decoder layer and uses it to organize action prediction together with representations of how the behavior unfolds and what it achieves. On SimplerEnv-Bridge, INDI improves GR00T-N1.7 from 64.3% to 84.7%, and on RoboCasa Kitchen it improves the controlled GR00T-N1.7 baseline from 64.1% to 70.3%, with consistent gains on $π_{0.5}$ across both benchmarks. In real-world tasks, INDI improves average success from 62.0% to 68.7%, with gains of up to 12.0 pp on longer-horizon tasks. Further analyses show that the recovered latent is used by the decoder, captures behavior objective and execution progress, and organizes downstream predictions in an objective-dependent manner. These results show that action decoders benefit from explicitly modeling the semantic objective of the behavior they generate.

Vision-language models (VLMs) achieve strong performance on video and image-sequence benchmarks, yet it remains unclear whether they capture temporal structure. To study this question, we formulate temporal grounding as an anomaly detection problem, providing a simple and controlled evaluation that directly tests sensitivity to temporal consistency. We introduce TimeCatch, where temporal anomalies are created by swapping consecutive frames and frame-level anomalies by replacing a frame with Gaussian noise. Models are evaluated on anomaly detection and localization tasks across four synthetic and real-world datasets, alongside a human study. Our evaluation reveals a substantial gap between frame-level and temporal anomaly detection. While VLMs consistently detect frame-level anomalies and often localize them accurately, they perform near chance on temporal anomaly detection and only modestly above chance on localization. Humans, in contrast, achieve near-ceiling performance on both tasks. Additional analyses across model scales, prompting strategies, sequence lengths, and visual similarity suggest that these failures cannot be explained solely by limitations in perception or model capacity. Together, these findings indicate that current VLMs can identify anomalies within individual frames but struggle to integrate information across frames to reason about temporal consistency. TimeCatch provides a controlled benchmark for evaluating temporal grounding in vision-language models.

Space robots operate in extreme environments where hardware degradation can critically compromise traditional control strategies. While continual reinforcement learning offers a promising mechanism for online adaptation, it inherently requires access to a reward signal during deployment. However, precise reward computation in space is often infeasible due to the lack of external tracking systems and the overall complexity of the environment. To address the challenge of unobservable rewards, we introduce a reward-free continual learning framework that leverages latent-state world models. By pre-training a model-based agent across diverse simulations, the world model learns a robust predictor of the reward structure within its latent space. Upon deployment to an environment with severe hardware degradation, we freeze the observation encoder and reward predictor to update only the transition dynamics of the world model through unsupervised rollouts. By training the policy entirely on imagined trajectories generated by this updated world model, the agent adapts to altered dynamics without receiving new rewards. We demonstrate our approach across simulated planetary traversal, orbital navigation, and precision assembly tasks subjected to severe morphological failures.

Dynamic operating envelopes (DOEs) are increasingly used to publish time-varying export limits that keep distribution networks within operational limits. Purely technical DOE allocation, however, can systematically privilege electrically favorable prosumers, while embedding fairness directly into a single-period optimal power flow (OPF) objective mixes network feasibility, equity and efficiency in a way that obscures the cost of fairness. This paper proposes a two-stage, multi-period framework that addresses both of these. Initially, a technical distributed OPF computes network-feasible export envelopes. The subsequent stage then applies a dynamic aggregate export budget and redistributes capacity through cumulative proportional fairness, limiting the additional curtailment by an admissible efficiency budget. The resulting fair DOEs are treated as first-stage decisions, while battery storage provides scenario-dependent recourse under demand and renewable uncertainty. The operational problem is solved by a calibrated regional alternating direction method of multipliers (ADMM) on a lossless LinDistFlow model and independently validated using AC power flow. On the IEEE 33-bus feeder over a 24-hour horizon, the technical benchmark yields 2.1097 MWh of renewable curtailment, whereas the fairness-constrained allocation increases curtailment to 5.7216 MWh but caps the maximum cumulative curtailment ratio at 11.20% and raises Jain fairness indices close to unity, with AC voltage deviations below 0.01 p.u. and no voltage or thermal violations under the adopted 0.90-1.05 p.u. limits. Results show that considering both storage (which alleviates curtailment impact) and multi-period fairness (which increases curtailment) is an interesting approach for modern DOE design, which in turn requires a multi-period, co-designed approach.

Design-space exploration of configurable Cyber-Physical Systems (CPS) requires executable evaluation when design choices affect timing, fault propagation, recovery behavior, and temporal-property satisfaction. Repeated stochastic executions make exhaustive exploration impractical for large design spaces. This paper presents a temporal-property-driven CPS design workflow using Reinforcement Learning (RL). At design time, the RL agent selects subsystem alternatives to assemble a candidate system model. The model is then evaluated through simulation, during which online temporal-property monitors observe runtime traces and produce functional-property violation indicators. These indicators are combined with evaluated non-functional terms for budget, recoverability, sustained compliance, and operational use to calculate the reward used for subsequent candidate selection. The workflow is evaluated on a methane-sensitive mine-pump CPS. The corresponding executable case-study model is provided as additional contribution. RL-guided search identifies the highest-reward design observed in the experiments after 26 episodes (corresponds to 130 executable simulations). These designs were reached with fewer simulations than surrogate-guided Bayesian Optimization and population-based Genetic Algorithm baselines under the same executable model and reward formulation. Ablation study results indicate that value-based feedback and reuse of previous simulation traces contribute to this reduction.

Arabic Natural Language Processing (NLP) has grown rapidly over the past decade, driven by digital transformation in the Arab world, social media, and large language models (LLMs). Despite this growth, a comprehensive quantitative meta-analysis remains absent. This study presents a bibliometric and topic-based analysis of 7,120 Arabic NLP papers published between 1960 and 2026, sourced from five platforms (arXiv, ACL Anthology, Semantic Scholar, Crossref, OpenAlex) plus an additional targeted OpenAlex subset. We employ BERTopic for topic modeling, regression analysis, social network analysis, and geographic mapping. Our findings show a significant publication surge after 2020, driven by transformer models and LLMs. Topic modeling identifies 19 themes, the largest centered on text, speech, translation, and recognition (2,942 papers). Citation analysis reveals a positive correlation between paper age and citations (r = 0.245, p < 0.001); regression (R^2 = 0.105) shows that indexing in OpenAlex or Semantic Scholar and institutional affiliation are associated with higher citations. Saudi Arabia, the United States, and Egypt lead in research output. A task-dialect gap matrix identifies understudied areas, including summarization for Maghrebi, Iraqi, and Sudanese dialects. The largest topic has the highest H-index (90), followed by sentiment analysis (57). Our quantitative approach complements existing qualitative surveys and offers recommendations to prioritize under-resourced dialects and develop culturally aligned benchmarks.

Long-horizon planning is critical for safe autonomous driving in complex scenarios. Existing methods improve planning continuity with temporal memory, but such memory may become invalid and mislead decisions when the driving command changes. Thus, selectively leveraging useful history while suppressing command-inconsistent memory remains a key challenge. To address this issue, we propose MomADv2, a reliable state-space memory framework for long-horizon end-to-end autonomous driving. At its core, MomADv2 introduces a Selective State-Space Planning Memory Query Module, which filters historical planning queries based on temporal continuity and command consistency, selects planning modes relevant to the current command, and models the evolution of planning intentions through a selective state-space mechanism. To further alleviate local trajectory deviations and error accumulation in long-horizon planning, we design a Flow-Matching Trajectory Residual Refiner. It learns a continuous residual correction field from the refined planning output to the expert trajectory, enabling fine-grained trajectory refinement while preserving the stability of anchor-based planning. Extensive experiments on closed-loop NAVSIM and Bench2Drive, as well as open-loop nuScenes, demonstrate that MomADv2 improves long-horizon planning consistency and reduces the average collision rate by 15.6% over MomAD under 6-second planning.

Recent LLM-agent systems make conflicting design bets: decompose work across many narrow agents, or use one strong tool-using agent. This pilot studies that choice on bounded cross-border VAT determination with reverse charge, where every case has an oracle label and each intermediate decision is independently scoreable. We hold the activity surface fixed (subtasks, tools, I/O schemas, validation checks, orchestrator, base model, and merge policy) and vary only the assignment of subtasks to workers across four orchestrated configurations, from one wide worker to five narrow ones, against S0, a tuned no-orchestrator single agent, with a deterministic rule engine as oracle. The program spans 4,400 runs: a 40-case, five-repeat main sweep, matched-token arms separating prompt-budget from agent-count effects, and three failure-injection arms, all judged against pre-registered falsification criteria. The two intermediate configurations lead on accuracy (0.830, against endpoints at 0.720 and 0.770) but miss the pre-stated bar against the fine endpoint, so the intermediate-optimum hypothesis remains unsupported at pilot scale. The single agent does not Pareto-dominate the orchestrated set. The matched-token criterion fires: the budget-matched single agent lands 6.5 points below the leader, but the interval includes zero, so any advantage is consistent with a prompt-budget explanation. Under injection, availability faults are absorbed at every granularity, with wide-scope restart over-recovering its baseline by +0.160, while one schema-conforming hallucinated record degrades every configuration and inverts the ordering, hitting fragmented configurations hardest. The contribution is a bounded, preregistered pilot heuristic for right-sizing decomposition (place one partition boundary at the dependency-layer midpoint), released with oracle, dataset, harness, raw traces, and analysis pipeline.

Structured data exists in many forms (tables, knowledge graphs, charts, and time series), and converting it into text may involve different generation tasks. However, most prior work on data-to-text (D2T) generation has focused on specific tasks and datasets, relying either on task-specific training data or on the zero-shot capabilities of large language models. We study cross-domain D2T generation in a setting where neither in-domain training text nor test references are available, and where domains, generation goals, and input structures vary substantially. We compare data-driven knowledge distillation (DDKD) against zero-shot inference and fine-tuning on out-of-domain D2T data, and introduce structure-preserving augmentation via structural subsampling and perturbation. Experiments on five benchmarks show that, at constant model size (1.7B parameters), DDKD consistently outperforms both fine-tuning and zero-shot inference. Moreover, the resulting small models outperform a much larger finetuned model on two of the five domains, achieving comparable performance on the remaining three. We further construct QUINTD-5, a fivefold extension of QUINTD-1, and show that simply scaling real target-domain inputs yields only modest gains, whereas our augmentation strategy remains more effective and more cost-efficient for cross-domain distillation.

English Wikipedia is often treated as the default encyclopedic source, yet non-English Wikipedia editions can contain richer locally grounded information for long-tail figures. We study cross-lingual biography enrichment: enriching an existing English biography with facts supported by a non-English biography about the same person. Focusing on women from non-English-speaking contexts, we introduce \textsc{CLAW-4L}, a benchmark consisting of 300 Wikipedia biography pairs linking an English biography with its French, Chinese or Azerbaijani counterpart, along with claim annotations and a fine-grained claim-pair relation corpus. We propose a claim-based enrichment framework that extracts English claims from both biographies, aligns them to identify enrichment evidence from the non-English biography, and rewrites the English biography using the selected claims. Our results show that non-English Wikipedia biographies provide valuable evidence for improving English biography coverage, while lower-resource settings remain challenging.

Autonomous indoor navigation requires both semantic understanding and precise geometric control. We propose OptiSight, a hybrid framework that combines Vision-Language Model reasoning with deterministic visual servoing through a finite-state Chain-of-Thought architecture. Grounded-SAM localizes open-vocabulary targets, while camera projection geometry converts visual observations into navigation commands without requiring dense mapping. The VLM is queried only at key decision points, reducing computational overhead while geometric control handles continuous navigation. Experiments in AI Habitat demonstrate reliable zero-shot navigation across diverse indoor scenarios, including obstacle avoidance and semantic ambiguity, while operating within an 8~GB VRAM budget. The source code is available at https://github.com/avanalperen/OptiSight-Python-Multimodal-CoT-for-Visual-Reasoning.

Mixed-integer programming (MIP) lies at the core of operations research and industrial optimization. While large language models (LLMs) have recently shown promise in automated MIP modeling from natural language, they prioritize semantic correctness but overlook formulation strength, severely bottlenecking the efficiency of downstream solvers. We propose FormuEvo, an LLM-guided evolutionary framework for automated discovery of solver-efficient MIP formulations. FormuEvo frames MIP formulation design as evolutionary optimization over the symbolic space of MIP formulations, represented as executable modeling programs, by iteratively generating, evaluating, and selecting stronger candidates via LLM-driven crossover, mutation, and repair operations. To move beyond blind exploration, FormuEvo introduces a solver-informed diagnosis mechanism that exploits fine-grained solver statistics as verbal gradients for targeted refinement. Additionally, a structured memory abstracts prior experience into reusable modeling strategies, avoiding redundant exploration while enabling zero-shot transfer to unseen problems and bootstrapping smaller LLMs. Experiments across diverse linear and non-linear problems demonstrate that FormuEvo discovers formulations that significantly outperform both expert-designed formulations and existing LLM-based approaches, accelerating solvers by up to 5.5$\times$, with distilled knowledge transferring effectively across problems and model scales.

Industrial demand changes the paradigms of production. Due to smaller batch sizes and more variations in products, companies face a growing challenge to adopt more adaptive production systems. In particular, robot-based automation is usually static and fails to respond to constantly changing processes. Vision-Language-Action (VLA) Models are a promising opportunity to mitigate this challenge by generating robot actions based on the observed system state. However, current research either focuses on large models that cannot be computed on premise, creating compliance and security challenges, or use lab-grade robot hardware that obscures exploitation in real industrial settings. In this work, we adapt Hugging Face's SmolVLA for Universal Robots lightweight robots. Further, we release the open-source repository ROS2SmolVLA that implements an interface for ROS 2 to SmolVLA, and makes it applicable for industrial-grade hardware. By this, we allow a lenient adoption into lab and industrial environments. We validate the functionality of SmolVLA for a Universal Robots UR10e using a pick-and-place task and give implementation guidelines. Our findings support that SmolVLA is a well-suited option for small-sized tasks that need to be computed on premise.

Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks by retaining a prefix of an expert trajectory before each rollout, letting the policy explore from a state closer to success. Its effectiveness hinges on the guidance depth: how much of the trajectory to keep. Existing methods treat this depth as a deterministic scalar. Scheduled approaches share one value across samples and ignore per-task heterogeneity; per-sample probing estimates it separately at the cost of extra rollouts. We find that useful guidance occupies a band of depths whose informativeness profile is approximately Gaussian around the band center, rather than concentrating at a single optimal point. We propose Agent-G$^2$, a Gaussian guidance framework that draws the depth per task from a Gaussian whose center and spread are estimated online from rollouts already collected for policy optimization, requiring no probe rollouts or learned depth predictor. The center combines a global baseline with per-cluster difficulty, and the spread tracks within-cluster variance. We evaluate Agent-G$^2$ on ALFWorld and WebShop on Qwen2.5-1.5B / 7B-Instruct. Agent-G$^2$ outperforms the strongest hint-based, hint-free, and Aux-RL baselines on ALFWorld by 2.3 / 3.9 / 7.4 points at under one-third the rollout cost of per-sample probing.

Proprioception in musculoskeletal humanoids is typically estimated primarily from muscle sensing, while the role of cutaneous deformation around joints remains insufficiently explored. In biological systems, mechanoreceptors distributed within soft tissue complement muscle feedback and support reliable joint state estimation. This study presents the design of a biomimetic joint-covering skin with a tissue-like layered structure that integrates pressure- and stretch-sensitive elements within the joint-covering tissue. The proposed skin is implemented on the musculoskeletal humanoid Musashi-W, and its independent proprioceptive capability as well as its integration with muscle sensing are evaluated. Experimental results show that the proposed skin alone achieves joint angle estimation with an average error of approximately 3 degrees. Furthermore, integration with muscle sensing improves estimation accuracy. Owing to its joint-covering structure, the skin may mechanically mitigate the influence of external disturbances on the muscles, and the integration of multiple modalities suggests the possibility of contributing to the identification of external stimuli that are difficult to interpret using muscle sensing alone. This work presents a design methodology for biomimetic joint-covering skin and demonstrates that such tissue-structured skin can serve as an effective approach for extending proprioceptive systems in musculoskeletal humanoids.

Retrieval can efficiently and effectively augment a frozen vision--language--action (VLA) policy without retraining, yet retrieved text becomes a control intervention once it enters the executed prompt. In a matched audit, raw appended text reduces mean success from 92.47\% to 3.00\%, while meaningful and length-matched meaningless appends both fail on all 500 states. This result identifies \emph{prompt-form collapse}: changing the instruction form, rather than adding useful semantics, can dominate execution. We introduce TOWN-VLA (Think Only When Needed), a prompt-authority interface that separates candidate generation from permission to alter the policy input. A fixed compatibility rule authorizes a canonical compact instruction; otherwise, the interface restores the original Base prompt exactly. Across 900 audited routes, every route follows this contract: 525 routes recover Base with matching hashes, and all 375 authorized prompts preserve the task signature. On a matched $4\times7$ LIBERO-Plus evaluation with 10{,}030 episodes per method, success rises from 69.5\% to 73.1\% ($+362$ episodes; 95\% CI 1.89--5.45 points), improving on six perturbation axes and all four suites. On a physical PiPER arm with a frozen \pizerofive{} checkpoint, success rises from 52.7\% to 78.7\% over 150 trials per method ($p=3.16\times10^{-6}$). Prompt authority is enforceable for a frozen controller; oracle-free admission calibration is the next deployment target.

Decision-making in high-dimensional, nonlinear systems remains a central challenge in robotics. While model-based methods like Model Predictive Control (MPC) offer sample efficiency and interpretability, their performance degrades when the dynamics model is inaccurate or long-horizon predictions are required. Conversely, model-free reinforcement learning (RL) learns policies directly from interaction but suffers from high sample complexity and unstable optimization. Recent advances in sequence modeling have inspired transformer-based decision-making frameworks that can unify MPC and RL, but their training typically faces significant optimization challenges due to highly non-convex loss landscapes. In this work, we propose a novel framework that integrates MPC with RL in a sequence decision-making framework and leverages a curvature-aware optimization to efficiently tackle non-convex loss landscapes. MPC provides predictions of locally optimal trajectories that guide the decision transformer, removing the need for extensive offline pretraining. To address the slow and unstable convergence of traditional optimizers, we train the policy in a Riemannian parameter space using an efficient Riemannian (curvature-aware) method, leading to faster and more robust optimization. We evaluate our framework on high-dimensional quadruped control tasks and demonstrate consistent improvements over strong baselines, including TRPO, SAC, and Online Decision Transformer, achieving higher returns and faster convergence.

We present the first LLM-powered multi-agent system for autonomous fault management across millions of optical links in production AIDCs. Refined via SFT and continuous memory evolution, it achieves 97.7% F1 and over 60% fault-incident reduction, outperforming SOTA LLMs on a ten-week field data evaluation.

Achieving 360° coverage is critical for the visual perception systems of autonomous vehicles. Fisheye cameras offer a cost-effective solution by enabling full surround coverage with as few as two sensors. However, existing multi-view fisheye datasets are limited, and synthesizing rare corner cases typically requires computationally expensive 3D simulations, hindering the training. While generative models have achieved significant success in standard perspective imagery, their application to wide-angle distortion remains unexplored. In this work, we formally introduce the novel problem of multi-view fisheye image generation conditioned on volumetric semantic representations and present two distinct methods. We first propose SyntheOcc-FE, which adapts the SyntheOcc architecture to fisheye data. While effective, this method is constrained by the scarcity of fisheye datasets, which limits its generalization. To overcome these limitations, we propose our second method, MIVIFI (multi-view fisheye), which leverages cross-domain learning with Equirectangular Projections. By bridging the gap between dataset domains using KITTI-360 fisheye images alongside nuScenes multi-view standard images, our approach enables high-fidelity manipulation of scene content. This framework enables the structural modification of semantic occupancy inputs to introduce or eliminate specific actors and facilitates the rendering of diverse meteorological conditions and illumination scenarios absent in the limited fisheye datasets. Quantitative and qualitative experiments demonstrate that our methods achieve robust photorealistic multi-view fisheye image generation and highlight the specific advantages of our cross-domain strategy for handling data scarcity.

Vision-language-action (VLA) models often expose spatial grounding through autoregressive text coordinates or opaque action tokens, creating brittle interfaces between multimodal reasoning and robot execution. We present Pointing-VLA, a typed hidden-state spatial readout built on Embodied-R1. Geometry-specific heads predict normalized points, object-functional grounding (OFG) heatmaps, and visual trajectories without serializing geometry as text. For the evaluated Bridge/WidowX and physical pick-place deployments, an explicit execution contract assigns PICK to source-conditioned OFG and PLACE to Pointing, providing direct stage-aligned spatial targets. Pointing-VLA achieves SOTA performance on Bridge/WidowX, averaging 72.9\% across the evaluated four-task set without Bridge-specific finetuning under collision-enabled CuRobo execution. Pointing and OFG show complementary strengths across native and cross-dataset evaluations. The OFG/contact readout transfers to NORA-1.5, preserving or improving success while reducing recorded controller time by more than 20$\times$; typed heads are also 6.68--6.90$\times$ faster than Embodied-R1 text decoding on a shared external suite. When integrated as spatial guidance for a $π_{0.5}$ action policy, Pointing-VLA raises autonomous real-robot success from 52.7\% to 80.7\% across three visual contexts. These results establish typed spatial readouts as an efficient, inspectable interface between embodied reasoning and robot execution.

The perception of 3D space by mobile robots is rapidly moving from flat metric grid representations to hybrid metric-semantic graphs built from human-interpretable concepts. While most approaches first build metric maps and then add semantic layers, we explore an alternative, concept-first architecture in which spatial understanding emerges from asynchronous concept agents that directly instantiate and manage semantic entities. Our robot employs two spatial concepts (room and door), implemented as autonomous processes within a cognitive distributed architecture. These concept agents cooperatively build a shared scene graph representation of indoor layouts through active exploration and incremental validation. The key architectural principle is hierarchical constraint propagation: Room instantiation provides geometric and semantic priors to guide and support door detection within wall boundaries. The resulting structure is maintained by a complementary functional principle based on prediction-matching loops. This approach is designed to yield an actionable, human-interpretable spatial representation without relying on any pre-existing global metric map, supporting scalable operation and persistent, task-relevant understanding in structured indoor environments.

Navigating large, photorealistic 3D apartments from raw pixels is widely considered infeasible for plain reinforcement learning. We build an agent that does it anyway, estimating its own pose from the camera alone. The agent has to reach several target objects in sequence, and their positions change between episodes, so it must explore to find them. It builds on our earlier object-centric topological controller, which still read the agent's true pose and its object detections from the simulator. Here we replace that true pose with an onboard, object-centric estimate. For each object we keep a bank of ORB features that, when the object is seen again, yield a rough pose measurement, which a minimal Extended Kalman Filter (EKF) fuses with a motion model. As on a real robot, the executed motions are noisy. The estimate drifts, but the agent and the nearby objects drift together, so a locally consistent pose is enough to follow each short edge and then home in visually on the target, which lets us replace full SLAM with a much smaller model, closer to how biological navigation appears to work. In the photorealistic Habitat simulator, the agent reaches its target objects from vision alone, with a pose that only needs to be locally consistent.

LLM agents remain unreliable on long-horizon tasks, where small local failures can compound over extended interactions and lead to overall task failure. Although external harnesses can substantially improve robustness, harness design remains a manual and expensive process that requires searching over a large space of prompts, tool configurations, and control logic. We propose AutoSaddler, an automatic harness optimization framework that formulates harness improvement as an offline learning problem and iteratively updates the harness using failure signals from mini-batches. AutoSaddler combines failure-trace diagnosis, structured patch generation that treats the harness as code, and validation-based update selection. Experiments on GAIA2, SWE-Bench Pro, and Terminal-Bench 2.0 show that AutoSaddler substantially improves agent performance over the corresponding base harnesses, achieving gains of 9.0, 9.6, and 10.0 percentage points, respectively. Ablation studies further suggest that effective harness optimization benefits from three ingredients: deep debugging rather than shallow reflection, targeted modifications rather than unconstrained editing, and generalization-aware selection rather than trajectory-specific repair. Together, these results suggest that automatic harness optimization is a promising path toward more performant and reliable agent systems.

RoboRacer offers a standardized platform for research using 1:10-scale autonomous vehicles, but the variety of available tracks hinders the process of acquiring policies. Although existing occupancy-grid simulators allow for the quick addition of new maps, they fail to include physical contact, while 3D simulators require each circuit to be implemented as a separate asset, thus limiting their scalability. In order to overcome this issue, we have developed RoboRacer Arena, a system that creates 3D racing environments directly from occupancy maps. Our method starts by using a flood fill algorithm to extract the drivable corridors and to identify the track boundaries, which are then used to establish the barriers. A distance field is calculated to define the collision boundaries. The track surfaces, collision properties, and materials are assembled into a USD stage, which allows for the automated and reproducible generation of the environment in Isaac Sim. The input maps can be obtained from SLAM sessions, from rescaled Formula 1 circuits, or from natural-language descriptions. When the input is based on natural language, we use Gemma 4 31B to generate a track specification without specifying any coordinates or geometry. To guarantee consistency and reproducibility, we apply geometric screening, procedural generation, and raster-level validation. The simulation environments are initialized in a time range of 1.18 to 2.48 seconds, with the initialization time increasing linearly as the raster size increases. In 30 matched trials involving 10 tracks and 3 seeds, 21 maps were generated and all passed validation. RoboRacer Arena currently contains 130 tracks and supports the generation of tracks from natural language. In benchmark tests, the system attains 8,707 vehicle-steps per second when using 256 parallel rigid-body vehicles, excluding the time taken for rendering and policy execution.

Fully online embodied learning requires synaptic adaptation to acquire new behaviors while preserving previously learned dynamics during ongoing interaction. We extend the Predictive-Coding-inspired Variational Recurrent Neural Network (PV-RNN) to continuously adapt its synaptic weights and propose Free-Energy-Gated Plasticity (FEGP), which regulates the effective learning rate according to variational free energy. In real-time physical human-robot interaction, a randomly initialized network acquired three cyclic motor patterns without offline pretraining, replay, or task-boundary signals, with all three patterns emerging in autonomous rollouts. Controlled experiments over ten randomized teaching streams and five network initializations per stream showed that FEGP substantially improved repertoire coverage and retention of previously acquired patterns after they left the recent observation window. Neither a constant learning rate matched to the gate's time-averaged effective rate nor replay of the same gain values with disrupted temporal organization reproduced these improvements. These results indicate that the temporal allocation of plasticity relative to model-environment mismatch, rather than simply its average magnitude or distribution, is critical for maintaining previously acquired behaviors during continued online learning.

Vision-language-action (VLA) models have made general-purpose robot manipulation increasingly plausible by conditioning robot actions on natural-language instructions. A key test of such generality is whether policies actually follow language instructions. Yet many manipulation benchmarks leave this ability underdetermined: the intended object or destination is often visually salient or uniquely feasible, allowing policies to succeed without grounding the instruction. We argue that instruction-following evaluation should be text-indispensable: multiple actions should be visually and physically plausible, while only one should be consistent with the language instruction. We introduce InstructMove, a text-indispensable benchmark for instruction-following manipulation. InstructMove instantiates this principle in pick-and-place scenes with semantic distractors, decomposing instruction following into category identification, attribute discrimination, spatial reasoning, and compositional pick-and-place. InstructMove supports a train-eval protocol with InstructMove training data and held-out evaluation tasks, with additional diagnostics for language dependence. Experiments with representative VLA policies show that InstructMove provides a controlled testbed for diagnosing visual shortcuts and that InstructMove simulation data can improve real-world instruction-following manipulation performance. Code: https://github.com/HorizonRobotics/RoboOrchardSim

Robot manipulators are commonly engineered around a decoupled motion-generation stack: a planner computes a collision-free path and a lower-level controller tracks the resulting reference. This separation is computationally convenient, but it can produce references that are difficult to execute under actuator limits, tracking error, model mismatch, and small obstacle clearances. We present CSymPlan, a certified symbolic planning and control framework for high-DOF manipulators with two complementary implementations: an offline implementation that precomputes certified reach-avoid feedback policies for known workspaces; and an online implementation that synthesizes or updates symbolic policies at runtime from changing task and perception information using parallelization. The offline implementation reduces the manipulator dynamics to a sampled perturbed double-integrator model in operational space through feedback linearization, treats torque-realization errors, modeling inaccuracies, and measurement uncertainty as bounded disturbances, and refines the synthesized symbolic policy to the Franka FR3 through a quantization--lookup--torque realization pipeline. The online implementation uses the same abstraction and refinement interface, but replaces the precomputed policy table with a runtime pFaces request--synthesis--execution loop. In randomized simulated benchmarks and perception-driven Franka FR3 experiments, both implementations complete reach-avoid tasks with zero safety violations; whenever no certified action exists, the robot holds, replans, or stops safely instead of executing an uncertified command.

Fault-tolerant navigation for thruster-actuated robots requires online adaptation to failures that are neither binary nor fully observable: thrusters may degrade continuously, fail dead, or jam stuck-open. Classical fault detection pipelines require dedicated sensors unavailable at deployment; oracle controllers that observe the true failure state are equally impractical. We show that privileged critic training is sufficient for sensor-free fault adaptation: giving the PPO value function access to the true degradation state dgt during training, while the actor receives only standard task observations, shapes a policy that compensates for failures at deployment without any dedicated fault sensing. We propose RAFT (Recurrent Asymmetric Fault Tolerant), a policy with recurrent memory trained with a privileged asymmetric critic. Evaluated on a floating-platform robot (8 thrusters, 1 reaction wheel) under up to four simultaneous thruster failures, RAFT achieves 70.2% success at four concurrent failures, closing 84% of the gap from a failure-naive baseline (4.8%) to an oracle policy that sees the full degradation state at deployment (82.4%). All code, checkpoints, and data are open-source.

Robotic navigation in human environments requires a spatio-temporal semantic representation that can rec- oncile open-vocabulary perception with long-term environmental changes. While foundation models provide strong zero-shot recognition, their predictions are intermittent and view-dependent, and naively integrating them into mapping pipelines leads to identity drift and stale semantics over time. We present SuperMap, a 4D spatio-temporal mapping framework for language-guided navigation that integrates high-frequency geometric SLAM with asynchronous open-vocabulary perception. Our core contribution is a consistency-driven mapping engine that combines 3D-aware instance association/re-activation with a principled existence-and-label confidence update to maintain stable object identities and prune outdated map content under occlusions and scene changes. SuperMap produces a queryable 4D scene-graph representation that interfaces naturally with Vision-Language Models by supporting compositional queries over object semantics, relations, We demonstrate SuperMap on benchmarks and real robots, including dynamic scenes with appearance/disappearance and relocation, and provide ablations and runtime analysis. We release the full system as open-source to provide the community with a deployable baseline for open-vocabulary spatio-temporal mapping. Project website: superodometry.com/supermap.

Societies of interacting language-model agents offer a controllable and repeatable way to study collective behaviour at scales that would be difficult to test with people. Their scientific value, however, depends on whether a social mechanism that works in a small group still operates when thousands of agents interact, and testing this directly requires costly large-scale runs. Here we introduce an audit that predicts a mechanism's fate as a population grows. It asks how often the mechanism can act, whether agents use the information it supplies, and whether the measurement itself creates apparent scale effects. Controlled experiments show that a single structural term can decide whether reciprocity, consensus or punishment survives scaling. For gossip, the population at which the mechanism fails is set by the reach and lifetime of its messages. In language-model societies, agents respond not only to social information but to how it is expressed: counts and percentages led to different scale behaviour. Predictions made before execution held on third-party code and a second model family, while a failed prediction exposed the boundary of the finding. The audit provides a prospective way to decide which social mechanisms can be interpreted across population scales.

While Vision-Language-Action (VLA) models have leveraged internet-scale pretraining and task-focused finetuning to achieve strong performance on long-horizon tasks, they often struggle with non-Markovian tasks that require memory. Existing approaches to memory typically involve additional Vision-Language-Models (VLMs) for long-term memory management, introducing a memory bottleneck and a fractured training pipeline. Conditioning on multiple historical frames can provide the VLA with access to more descriptive features of past scenes, but can degrade performance if frames are chosen at arbitrary, fixed intervals. To address these limitations, we present UniMem, a framework that unifies high-level, multimodal memory and low-level control under one backbone. UniMem employs an event classifier for memory updates, a keyframe encoder for dense spatial memory, and a keyframe caching technique to minimize overhead during policy rollouts. We evaluate UniMem across five simulation and four hardware tasks targeting sequential and spatial memory, demonstrating that our unified, single-model system outperforms fixed-interval image sampling baselines (93.4% vs. 68.2%) in simulation and hierarchical baselines (80.0% vs. 43.5%) in hardware, while offering faster inference and a simple training pipeline for easy adoption. Project website: https://losterberg3.github.io/unimem-vla/

Large language models (LLMs) and vision-language models (VLMs) are expanding the range of behaviors that can be represented in agent-based simulations, but many contemporary platforms are difficult to study, modify, or run on ordinary computers. We present two intentionally minimal simulation foundations for education and rapid prototyping. SD-AgentFoundry-2D provides a two-dimensional multi-agent environment in which locally hosted LLM agents move, communicate, respond to place occupancy, and encounter spatially localized fire events. SD-AgentFoundry-3D provides a three-dimensional digital-twin environment in which a locally hosted VLM receives first-person images and produces natural-language movement instructions. Both codebases are designed to run locally on macOS, Windows, and Linux and are deliberately left open to modification rather than developed as finished applications. Together, they offer accessible starting points for learning about generative social simulation and for building domain-specific extensions.

Dual-comb spectroscopy (DCS) enables high-speed, high-resolution spectroscopy by down-converting optical spectra to the radio-frequency (RF) domain through the interference of two mutually detuned frequency combs. Accurately resolving narrow molecular absorption features requires closely spaced comb lines generated by large laser cavities with long round-trip times, thereby hindering the miniaturization of high-resolution spectrometers. Here, we circumvent this limitation by demonstrating a swept-source dual-comb spectrometer using two unidirectional racetrack semiconductor lasers integrated on the same chip. The traveling-wave frequency combs are generated by strong RF injection at the laser cavity round-trip frequency of 14.3~GHz and are continuously tuned in frequency by varying the laser drive currents. This enables continuous spectral sampling over a 32~cm$^{-1}$ range centered near 8~\textmu m, with an effective sampling interval of approximately 80~MHz. The outputs of both combs are coupled into the same monolithically integrated light coupler, providing automatic collinear alignment before the combined beam interrogates the sample. We benchmark the spectrometer against an external-cavity tunable laser and validate its performance using HITRAN simulations for 1.25\% nitrous oxide. Finally, the unidirectional comb architecture suppresses the detrimental effects of optical feedback, yielding comparable residuals even under deliberately introduced strong feedback. More broadly, the benefits of miniaturization extend beyond reduced footprint: chip-scale integration enables new forms of electrical and optical control that can fundamentally change how dual-comb spectrometers are operated.

When can an agent failure be caught? An audit is usually limited by the record rather than by the method. CatchBench therefore puts one auditor's question to three information states: the declared configuration before a run (PRE), a growing prefix of its trace (LIVE), and the finished trace (POST). Prior benchmarks fix one of these states or vary the telemetry; to our knowledge none scores all three under one task-method interface. Each state admits different questions, so seven task contracts carry their own labels and metrics rather than one leaderboard. Four are evidential; three are Gold-derived mechanism diagnostics. The release scores 72 entrants, from rule scanners and structural models to eleven LLM judges across nine model families (GPT, Claude, Gemini, Gemma, Llama, Qwen, DeepSeek, Mistral, Nova), over 1187 declared configurations and 1162 recorded runs. Most of the arena does not order: 47 of 118 pre-declared contrasts separate, and the rest are published unresolved rather than ranked. The two sharpest results cut against our own data. One rule ignores every name and permission; it flags each capability declared after the first. On one of six configuration sources it reaches a perfect F1, so a score there measures how the corpus was built rather than how well a method reasons. Our admissibility bar then rejected one injected substrate and withheld evidential status from the other. A benchmark number is therefore not interpretable until the process behind its labels is published and tested for the shortcut it may leave. We report both, and regenerate every ordering from released predictions with no model call.

Ensuring reliability in uncertain environments remains difficult for long-horizon robotic manipulation. End-to-end VLA models are data-heavy and opaque, making diagnosis and verification difficult. Hierarchical pipelines are more interpretable, but their plans are often weakly grounded in observations, weakly aligned with low-level actions, and computed without online feedback, leading to open-loop behavior and hallucinations. To address these issues, we introduce the Triplet-to-Track System (TTS), a closed-loop long-horizon imitation learning system that uses human videos to reduce reliance on robot-collected data. TTS represents high-level subgoals as instance-grounded triplets, translates them into continuous track priors for execution, and monitors task progress from observations for online replanning. Across diverse real-world long-horizon tasks, TTS achieves a 74.8\% average success rate and supports object-level and compositional generalization.

We present the NeuralNexus Arm, an open, low-cost 6-DOF robotic manipulator built by an undergraduate engineering team, together with the design decisions and debugging experience needed to reproduce it. The arm is driven by a single STM32H743 microcontroller on a custom printed circuit board (PCB) and combines two stepper-driver strategies on one controller: push-pull 3.3 V step/direction outputs for onboard TMC2209 drivers on the three wrist joints, and open-drain outputs for external CL57T and DM542 drivers on the three high-torque proximal joints. We describe the mechanical design, mixed-driver electronics, interrupt-driven firmware, a MATLAB/Simscape-based inverse-kinematics pipeline, a browser-native control interface using the Web Serial API, and a lightweight vision pipeline for object localisation and autonomous pick-and-place tasks. We also document non-obvious hardware and firmware failure modes encountered during the transition from a development board to the custom PCB as reproducibility guidance. All design files and firmware are released openly. The platform actuates all six axes under coordinated control at a 2 kHz update rate and executes both manual and pre-recorded motions from the browser interface.

Nonverbal behavior generation systems for virtual agents often take an utterance as input and generate nonverbal behaviors that emphasize or illustrate the content of the verbal channel. However, human nonverbal behavior is shaped by more than the content of the speech. It is also influenced by speaker roles, interpersonal relationships, social context, and the cognitive and emotional states of the interactants. As a result, the nonverbal channel may reinforce, weaken, qualify, or even contradict the verbal channel. It may also reveal internal states that are hidden or only indirectly implied in speech, including emotional "leakage" that may be incidental to the immediate interaction. Modeling this richer relationship between verbal and nonverbal behavior is important for designing virtual agents that exhibit realistic, human-like behavior. It is especially critical in training contexts that require nuanced social interpretation, such as counseling simulations involving virtual patients. Drawing on Ekman's framework of verbal nonverbal relationships, we propose a taxonomy of categories in which mismatches between verbal and nonverbal behavior can occur. We then examine alternative approaches for realizing these behaviors using large language models, focusing on whether LLMs can select contextually appropriate mismatched verbal and nonverbal behaviors from a given dialogue and social interaction context. Finally, we evaluate the resulting behaviors in a human-subject study, assessing whether context-driven nonverbal behavior, when embodied in a virtual human, produces the intended effects on observers.

Living organisms exhibit extraordinary adaptability to unseen environments through their intrinsic physical structures and lifelong feedback-driven learning. Endowing robots with comparable generalization is critical for reliable operation in the real world. While recent approaches attempt to improve generalization by scaling training data, such strategies remain impractical for robotics, where collecting real-world demonstrations at the scale of large language models is prohibitively costly and slow. Contrary to this reliance on massive datasets, we show that robots can generalize effectively under dynamics uncertainties even with limited training data by leveraging a feedback mechanism, namely PhyFilter, that corrects learning outputs with physics-filtered learning residuals. PhyFilter operates as a lightweight, model-agnostic module whose parameters can be automatically optimized through an auto-learning algorithm, eliminating manual tuning and enabling seamless integration with diverse robot policies. We validate PhyFilter across four representative robotic systems, demonstrating that it enables quadruped robots to generalize to unseen terrains, payload variations, and speed ranges; drones to flight under unseen wind disturbances; aerial manipulators to achieve centimeter-level in-air capture despite wind and mass uncertainties; and acceleration differentiators to remain robust with distribution shift. These results show that physics-filtered feedback can serve as a powerful alternative to massive data scaling.

This work develops a discrete-time framework for identifying probability localization dynamics through finite stochastic representations adapted in space, time, memory, and state information. A compact dynamically relevant set is localized by a finite measurable partition, producing an observable probability state and a relational graph of admissible transitions. Structured stochastic liftings derived from Stochastically Structured Reservoir Computing (SSRC) give lossless polynomial representations of the observable state, while stochastic delay liftings add finite observable memory. These are distinguished from dynamically informed state-space enrichment: refinement of observational fibers containing states with the same present observation but different observable futures, yielding an exact obstruction-to-closure criterion. A route-network toy problem gives a minimal obstruction example, while four numerical laboratories (rotational phase dynamics, the chaotic logistic map, the Van der Pol oscillator, and a synthetic cyclic inventory system) show how spatial scale, temporal scale, polynomial degree, and delay depth interact. The logistic map isolates representation-induced memory in an otherwise Markovian chaotic system, using its exact invariant law as an ergodic benchmark and its zero-mass pseudospectrum to separate relaxation from transient amplification. An exact rotational cycle calibrates pseudospectra as a robustness diagnostic rather than a closure certificate. The inventory example gives a closure-driven enrichment procedure: residence-age hazards trigger age-refined states that improve predictive scores. These results motivate a minimal adequate representation: the least complex representation meeting predictive, structural, and identifiability requirements.

Semantic segmentation has rapidly advanced with deep learning; however, challenges remain in effectively capturing local and global contexts as well as addressing the long-tailed distribution problem. To tackle these issues, we present Contextrast++, a robust contrastive learning method for semantic segmentation that improves multi-scale feature integration and mitigates class imbalance issues. Our method consists of two key components: 1) contextual contrastive learning (CCL) and 2) boundary-aware negative (BANE) sampling. CCL includes three subcomponents: adaptive fusion module, pixel-to-anchor (PA) loss, and anchor-to-anchor (AA) loss. The adaptive fusion module dynamically balances local and global feature integration, resulting in a more context-aware representation. While the PA loss leverages the fused multi-scale features to improve feature representation learning, the AA loss focuses on addressing the long-tailed distribution problem by utilizing a memory bank that stores a fixed number of class-balanced representative anchors. Meanwhile, BANE sampling enhances segmentation precision by selecting hard negatives from misclassified boundary regions, which refines fine-grained details during contrastive learning. As verified in extensive experiments using public datasets, we demonstrate that Contextrast++ substantially improves semantic segmentation performance over existing contrastive learning-based state-of-the-art approaches, while introducing no additional computational overhead during inference.

UAV vision-language navigation (UAV-VLN) is commonly evaluated as goal reaching, but inspection-oriented deployment requires the agent to stop within a valid inspection region and avoid falsely confirming visually or semantically similar distractors. This requirement exposes a key weakness in existing coarse-to-fine UAV-VLN policies: the coarse goal predicted before local refinement is often treated as reliable, although it may drift toward plausible but incorrect object regions and limit the ability of the local stage to recover. To systematically evaluate this problem, we introduce LG-UVI, an object-centric inspection evaluation setting derived from CityNav/CityRefer. LG-UVI extends standard UAV-VLN episodes with target objects, hard distractors, type-aware inspection regions, and diagnostics for inspection-region arrival and object-level confirmation. To address this inspection-oriented setting, we further propose RACO, a reliability-aware adaptive coarse-to-fine navigation framework. Instead of treating the predicted coarse goal as a fixed waypoint, RACO views it as a runtime hypothesis and uses object-level candidate anchors to check and correct coarse localization before Stage 1 and at the Stage 1-to-Stage 2 boundary. RACO also applies scale-adaptive terminal refinement to handle terminal near-miss cases using runtime-observable geometric and anchor-based evidence. Under a unified online evaluation protocol, RACO improves SR over the reproduced HETT baseline by 9.53 and 7.98 percentage points on validation-unseen and test-unseen, respectively. It also improves inspection-region arrival and reduces false verification risk, showing that coarse-goal reliability optimization is an effective complement to existing coarse-to-fine UAV-VLN policies.

The continuous integration of inverter-based resources makes transient stability analysis increasingly important for power system modernization, in light of the intricate dynamics arising from large-scale inverter deployment. However, analytical transient stability analysis methods consistently yield overly conservative stability boundary estimates, which constrain grid dispatch and operational flexibility. Although recent neural Lyapunov methods attempt to enlarge these stability boundaries to obtain less conservative estimates, they remain trapped within restricted domains due to the out-of-distribution problem. To break this bottleneck, this paper proposes a probabilistic active learning framework. Starting from a deterministic inner stability region certified by neural Lyapunov functions, the framework constructs a Gaussian process surrogate and deploys an uncertainty-guided frontier search. By intelligently coupling electromagnetic transient simulations with active boundary exploration, the algorithm systematically drives the estimated stability boundary outward. Comprehensive evaluations across multi-machine grid-forming benchmarks demonstrate that the proposed framework substantially reduces estimation conservatism. Across test systems ranging up to four interconnected grid-forming inverters, the methodology achieves up to a $20$-fold volumetric enlargement of the certified stability region over classical baselines, while requiring at most 220 time-domain simulation queries per system, far fewer than the 1,600 queries that exhaustive EMT evaluation demands even for the simplest single-inverter benchmark.

Agentic AI frameworks interpret open-ended task goals and decompose them into multi-step plans. Richer information about embodiment-specific capabilities, physical preconditions, and cross-robot coordination improves grounding, but does not eliminate infeasible, mistimed, or unsafe physical actions. Physical robot crews therefore require an explicit architectural interface between semantic planning and execution, where every planned action is verified against robot capabilities, system state, and workflow constraints before actuation. This paper introduces Physical Agentic AI, a framework for skill-grounded robot agent orchestration, in which each robot exposes a typed library of executable skills while a foundation model planner decomposes a task into phases and assigns each phase to a robot-skill pair. A Robot Orchestration layer exposes the skill library, robot state, named locations, and workflow contracts to a non-actuating Mission Planner, while a deterministic Robot Orchestrator validates and authorizes one skill at a time. We evaluate on a drone-UGV search-and-dispatch mission, where every mission in every condition is executed live in Gazebo, and on a humanoid-quadruped transportation task using hardware-equivalent skill interfaces plus two physical trials on a Unitree G1 and Go2. Varying planner knowledge and runtime enforcement independently, we find that retrieval raises skill grounding from 51% to 96% yet leaves informed planners dispatching 23-29% of faulted steps. Per-dispatch enforcement reduces false dispatch to 0% with no false blocks, and a held-plan ablation confirms that the gate, not plan variation, is responsible. Live execution makes the difference physical: without enforcement all eight injected faults crossed the orchestration boundary and six produced robot motion; with enforcement all eight were refused before motion.

Transferring reinforcement learning policies from simulation to Real-World robots remains a major challenge, particularly when dealing with low-level torque control, where even small modelling inaccuracies can lead to unstable or unsafe behaviours. In this work, we propose a Real2Sim2Real pipeline that improves Sim2Real transfer for torque-controlled robotic arms by combining trajectory matching, parameter optimization via genetic algorithms, and domain randomization. Using the 7-DOF Franka Emika Panda robot, we first identify friction, inertia, and gravity compensation parameters by minimizing the error between real and simulated joint trajectories. These calibrated dynamics are then used to train a TQC-based reinforcement learning agent in simulation. The trained policy is evaluated in both Gazebo and MuJoCo environments, and finally deployed on the real robot. Our results demonstrate a significant improvement in tracking accuracy and policy robustness after parameter tuning, with smooth policy transfer from simulation to the Real-World across multiple target-reaching tasks. This work highlights the effectiveness of accurate physical modelling in enabling stable and generalizable torque-based reinforcement learning policies.

Robot policies receive heterogeneous observations at each decision step, yet sequence models differ in how they organize these inputs over time. We introduce WorldToken, a time-first policy instantiation that fuses multiview images, proprioception, and task conditioning within each policy timestep into one world token. A causal temporal Transformer models the resulting world-token sequence, and a diffusion action head generates action chunks. On 23 RoboCasa tasks, an 85.3M-parameter policy trained from scratch apart from a frozen pretrained CLIP text encoder achieves 59.45% mean closed-loop success using 2,900 generated demonstrations per task. A complete factorial sweep over five dataset sizes, five model sizes, and two training seeds shows consistent gains from additional target-domain data and diminishing returns beyond moderate model size. Under same-checkpoint history truncation, reducing visible history to one or two policy timesteps lowers closed-loop success for all 50 RoboCasa policies. On RMBench Blocks Ranking, reducing visible history from 146 to 8 seconds lowers evaluator success from 95% to 28%, while an exploratory extended rollout sustains the reference swap sequence for over 850 seconds. These results establish the empirical feasibility of the complete WorldToken instantiation and characterize its data-scaling and temporal-context behavior under the tested recipes. They do not establish superiority over alternative sequence organizations or isolate which components of the complete implementation drive the observed performance.

In adaptive control, parametric uncertainties in linear-in-parameter form consist of unknown parameters and known regressor signals. Convergence of the unknown parameters to their ideal values requires the regressor to satisfy a persistent excitation (PE) condition, which depends on future data and is therefore infeasible to guarantee online. Memory-based parameter update laws address this by enabling ideal parameter convergence under the online-verifiable finite excitation (FE) condition. In this paper, a new algorithm is proposed to construct a memory term via the Modified Gram-Schmidt orthogonalization procedure for a class of multi-input multi-output nonlinear systems with an unknown diagonal control effectiveness matrix and bounded nonparametric uncertainties. Under the finite excitation condition, the constructed memory term yields an identity coefficient matrix in the parameter estimation error dynamics. The identity coefficient matrix eliminates the need for time-varying adaptation gains, enables an explicit ultimate bound on the parameter estimation error, and preserves the structure of the nonparametric uncertainty bound under the memory term. Building on this, a combined adaptation law is developed for controller gain estimation under FE. The closed-loop tracking and estimation errors are shown to decay exponentially to a neighborhood of the origin, characterized by an explicit ultimate bound, with a decay rate that depends solely on user-defined gains and system constants, independent of the level of regressor excitation. This removes the dependence of the convergence rate on the level of regressor excitation, a key limitation of existing approaches such as concurrent learning, memory regressor extension, and DREM.

Two properties of optimization-based controllers such as model predictive control (MPC) limit their use in space flight. The online computing time varies and can exceed the sampling period, and the closed-loop behavior carries no formal guarantee. Explicit MPC, where a parametric solution of the optimization problem is computed, replaces the on-board optimization with a lookup table of piecewise-affine functions with a fixed execution time. However, it has been regarded as only applicable to small horizons, since the number of regions bounds the memory the table occupies and admits only an exponential bound in the horizon. This paper builds on recent developments in parametric solutions of quadratic programs and nonconvex reachability analysis to provide an entire pipeline for spacecraft rendezvous controllers under the Clohessy-Wiltshire dynamics. For this application, efficient data structures can be constructed to encode the control law, both from a computing time and a memory standpoint. In addition, the controller can be verified offline with a guaranteed closed-loop stability certificate by resorting to reachability analysis with hybrid zonotopes. At a 51-step horizon the rendezvous law occupies 10.5 MB and is evaluated in under 300 ns. The occupied memory would fit for instance the CubeSat flight computer ARM9-class AT91RM9200 meaning that the horizon limitation does not hold when the partition is built with the parametric solvers and data structures shown in this paper.

Forecasting dexterous hand motions from egocentric observations is fundamental to intelligent interactive systems. Existing VLM-based methods typically map observations directly to future motions, overlooking the underlying manipulation process that governs hand-object interactions. Moreover, end-to-end optimization couples manipulation learning with motion synthesis, causing motion-generation gradients to interfere with the pre-learned manipulation-aware representations. To overcome these limitations, we propose EMPIRE, a two-stage framework that introduces Explicit Manipulation Planning as an Intermediate Representation for Egocentric hand-motion forecasting. Stage I: Learn to Plan. EMPIRE first learns explicit manipulation plans from multimodal context to capture the progression of hand-object interactions. Stage II: Learn to Act. A motion generator synthesizes future bimanual hand motions conditioned on frozen planner representations, preventing motion-generation gradients from affecting manipulation planning. To support our method, we further construct EMPIRE-651K, a bimanual hand-motion forecasting dataset comprising 650,910 training windows across 111 tasks, each paired with an explicit per-hand manipulation plan. Under identical training and evaluation protocols, EMPIRE achieves state-of-the-art forecasting accuracy, with an MPJPE of 84.53 mm and a finger-relative error of 38.97mm. We release the code and dataset at https://github.com/wangwen-banban/EMPIRE.

Query-based Vision-Language-Action (VLA) models offer low-latency inference that is attractive for bimanual robotic manipulation, but we observe that they can still exhibit discontinuous actions and execution failures in complex dual-arm tasks. We hypothesize that unstable multi-view and language fusion is one contributing factor in these failures, often coinciding with attention spreading to distracting regions. To improve robustness, we introduce the Modality Masking Mechanism (M3), an embarrassingly simple, training-only strategy that requires no architectural changes or large-scale robot pretraining. M3 stochastically masks subsets of modality channels during training, exposing the policy to controlled partial observations and encouraging it to rely less on distracting cues and more on evidence that remains reliable. We evaluate M3 on ten bimanual tasks from RoboTwin 2.0 and on three long-horizon real-world tasks. Compared with the Adapter baseline, M3 improves average success by 21.7% in the Clean setting and 11.4% in Clean2Rand, where policies are trained on clean demonstrations and evaluated on randomized scenes, while also improving averaged real-world full-task success by over 30%. These results suggest that structured training-time masking is a practical way to improve the robustness of query-based VLA policies for bimanual manipulation.

Human video is playing an increasingly central role in training World Action Models (WAMs), owing to its diversity and low collection cost relative to teleoperated robot data. However, most WAMs learn from such video only by predicting pixel-level future frames, giving dynamics that are not directly actionable, whereas motion retargeting recovers directly actionable actions but leaves a large visual gap across embodiments. We therefore propose motion-aligned latent dynamics as an embodiment-agnostic representation to bridge video priors and low-level actions. We further present LD4WAM, which pairs a Latent Dynamics Model trained with semantic reconstruction and real motion alignment with a World Dynamics Action Model built as a mixture-of-transformers (MoT), which preserves full future-video generation and uses learnable queries to distill these latent dynamics from generated futures for action conditioning. Pretrained on our curated unified dataset of over 5{,}000 hours of human and robot data, LD4WAM performs strongly in RoboTwin simulation and on real robots equipped with both grippers and dexterous hands, while generalizing well to unseen objects and backgrounds.

Powder X-ray diffraction (PXRD) is the routine probe of crystalline matter, yet its analysis is the rate-limiting step as laboratories automate acquisition. Deep-learning analyzers excel on simulated patterns and degrade on measured ones. This simulation-to-real gap is structural, not additive: synthetic denoising gives no measurable lift on real spectra, whereas correcting a small peak-position drift more than doubles median retrieval correlation. Real-spectrum fine-tuning, peak-aligned reranking, and recalibration narrow what remains and restore the coverage synthetic anchors lose. Xtalyst integrates these in an agent-orchestrated system spanning phase identification, refinement, and calibrated property prediction. On a frozen held-out partition (n=534) each module measured on both splits reproduces its development finding -- including the synthetic-anchor under-coverage, whose magnitude differs between the two pools -- while held-out refinement converges and preserves symmetry without reaching profile-quality fits, and on a diffractometer its wet-dry recommend-rescan-reanalyze loop flips a blinded silicon standard to a gated PASS and changes which minor phase is resolved on a multi-metal alloy.

Reliably tracking moving deformable linear objects (DLOs) while simultaneously ensuring robustness, accuracy, and temporally consistent state estimation remains a fundamental challenge in robot perception. We introduce MotionDLO, a real-time tracking framework specifically designed to overcome these limitations in temporal continuity and latency. The method exploits the high temporal resolution and sparsity of event-based cameras and combines segmentation with the Coherent Point Drift (CPD) algorithm under the principles of Motion Coherence Theory. This integration enables temporally consistent shape estimation while maintaining a low computational overhead. Existing event-based tracking methods are typically computationally efficient but exhibit reduced accuracy compared to frame-based approaches, or alternatively compromise event sparsity to achieve competitive performance. To resolve this trade-off, we propose a hybrid event- and frame-based tracking architecture that preserves the complementary strengths of both sensing modalities. The event stream ensures high-frequency motion updates, while frame-based information stabilizes spatial accuracy and object identity. We demonstrate that the proposed framework reliably associates DLO instances across video sequences, enabling robust perception for robotic manipulation tasks. Experimental results validate real-time performance at 12 ms update rates and accurate shape tracking with an point-to-curve error as measurement of accuracy of up to 0.43 mm, supporting dynamic path adaptation during manipulation. The source code and demonstration datasets are publicly available.

Structured coherence -- partially coherent light spanned by a finite number of modes -- is emerging as a powerful tool in optical communications, computation, cryptography, and spectroscopy. Key to these prospects is the recent development of on-chip processing of structured coherence, in which large meshes of interferometers implement unitary and non-unitary transformations on the Hermitian coherence matrix representing multimode partially coherent light. Two related critical tasks for the applications of structured coherence are the reconstruction of an unknown coherence matrix and its diagonalization. Stokes tomography has been utilized in reconstructing the coherence matrix, whereas variational processing has been employed in its diagonalization. We show here that Stokes tomography can also be exploited in the on-chip diagonalization of an unknown coherence matrix, which we verify for two-mode and four-mode structured coherence in an integrated hexagonal mesh of Mach-Zehnder interferometers. This photonic circuit implements a predetermined sequence of configurations to estimate the generalized Stokes parameters, which -- in a final step -- inform a reconfiguration of the photonic circuit that diagonalizes the coherence matrix. The field is thus left in a coherent-mode representation comprising uncorrelated, orthogonal modes whose weights correspond to the eigenvalues of the original coherence matrix. Moreover, the integrated photonic circuit can be configured to provide the original field alongside its diagonalized counterpart at the circuit output. We verify the diagonalization procedure for coherence matrices of different coherence rank, entropy, and structure. Finally, we dispel the common notion that O(N^2) steps are required for reconstructing an N x N coherence matrix and show that only O(N) steps are needed.

World action models (WAMs) couple visual future prediction with robot action generation, but accelerated students can lose task capabilities during distillation and later encounter states that are poorly represented by offline data. We study whether on-policy distillation (OPD) can repair such a student without requiring sparse-reward reinforcement learning. We introduce WAM-OPD, a deployment-consistent post-training recipe for a video-first WAM. The student acts in the environment and therefore determines the history distribution. A frozen teacher labels those student histories with coherent video and action targets, while the student action branch is trained under its own generated video plan, as it is at deployment. Joint video and action losses update lightweight adapters in the shared backbone, together with an action flow-matching regularizer. In preliminary RoboTwin 2.0 studies on two tasks, the released one-video/one-action-step Flash-WAM improves from 0.0% to 58.3% success on HANDOVER MIC, and from 16.7% to 33.3% on PUT OBJECT CABINET. These task-specific results are an initial capability proof rather than evidence of broad or uniform generalization. They nevertheless suggest that dense teacher supervision on student-induced histories is a promising post-training interface for video-first WAMs.

Graph-of-Convex-Sets (GCS)-based trajectory optimization represents collision-free regions in configuration space as a finite collection of convex sets and directly performs collision-free trajectory planning over these sets, substantially simplifying the planning process. However, existing GCS-based trajectory planning methods generally assume sufficient connectivity among the convex regions and do not explicitly address cases in which the start and goal regions belong to different connected components of the initial GCS map. To address this limitation, we propose GCS-Bridging, which reconnects disconnected convex regions through collision-free point paths followed by convex region inflation, thereby recovering the feasibility of otherwise disconnected GCS planning problems. Extensive simulations across multiple IRIS-related algorithms and scenarios demonstrate that GCS-Bridging restores missing start-to-goal connectivity in the initial GCS map with a 99.8% success rate. In addition, a hardware experiment on a single-arm Franka platform in a real-world scenario with initially disconnected start and goal regions validates the effectiveness of the proposed method in practical motion planning. Project website: https://zhouxk1997.github.io/GCS_Bridging/

Creating symbolic operators by hand is one of the main bottlenecks in deploying Task and Motion Planning systems (TAMP). Recent works show that these operators can instead be learned directly from demonstration data. Existing methods, however, typically learn each action in isolation and cannot capture the recurring multi-step structure of manipulation tasks, so the search becomes intractable on long sequential tasks. A further inefficiency arises in the symbolic state: every provided predicate is evaluated at every search node, even when it never appears in any learned operator. We present a system that addresses both problems together. Its central component is the automatic generation of macro-operators, composite actions that compress a recurring sequence of individual actions into a single planning step. Our system discovers causally linked action pairs directly from the training data, where one action produces exactly the condition that the next one requires, and turns each pair into a new operator. Alongside this, our system prunes every predicate that no learned operator references, which shrinks the symbolic state evaluated at each search node. Together, these changes shorten the effective planning horizon, and the benefit they bring grows with the length of the task. Across four TAMP domains, our method reaches up to a 4.6x planning speedup compared to the baseline method, namely Learning Operators for TAMP. More importantly, it solves a long sequential task that the baseline cannot solve. Macro-operator discovery thus not only accelerates planning but, in certain domains, determines solvability in practice.

Humans imitate at the level of intent: given a demonstration, we infer its goal and carry it out with whatever tools, objects, and layouts are at hand. Current robot policies instead learn observation-to-action mappings from visual inputs and language instructions, without explicitly inferring the demonstrated task. Learning from human video thus remains largely trajectory-level: models can replay motions in near-identical scenes, but still struggle to imitate what the demonstrator intends rather than merely what they do. We introduce The Imitator Game, a four-level benchmark (L0-L3) that progressively widens the gap between the human demonstration and the robot's own scene, isolating where trajectory replay ceases to suffice and task understanding becomes necessary. We pair it with IG-10K, the largest environment-aligned paired human-robot dataset to date and the only one instantiated across all four levels in both real and simulated settings (20,000+ paired episodes, 50+ tasks, 6 domains), and Imitator Arena, an open platform for blind A/B human evaluation. Across nine state-of-the-art models, performance is stable from L0 to L2 but collapses at L3, identifying functional substitution - achieving the same intent through a different object affordance - as the decisive barrier to intent-level imitation. Human-video-conditioned models outperform caption-conditioned ones, yet every model falls below 13% zero-shot success on unseen tasks; fine-tuning IG-10K-pretrained models with only $10$ paired human-robot demonstrations yields large gains that grow with pretraining scale. The project website and access to Imitator Arena are available at https://imitator-game.github.io.

Quadruped mobile manipulation requires two tightly coupled capabilities: reaching manipulation-ready configurations and maintaining stable contact throughout articulated-object interaction. However, existing methods often terminate navigation near the target, leaving a gap between reachability and manipulation readiness, while tracking lag, motion jitter, and contact instability limit continuous interaction. To address these challenges, we present TONAV, a unified framework integrating task-oriented navigation with action-velocity chunk learning. First, we introduce a position-velocity-coupled teleoperation framework that explicitly captures motion dynamics to improve master-follower consistency and collect smooth, temporally consistent demonstrations. Next, task-oriented navigation leverages vision-language reasoning to decompose high-level instructions into executable subgoals and adaptively refine the robot base toward a manipulation-ready configuration. Finally, action-velocity chunk learning jointly models joint positions and their temporal transitions under velocity supervision, enabling smooth and stable sustained-contact manipulation. Real-world experiments across diverse articulated-object tasks demonstrate that TONAV achieves higher success rates in both task-oriented navigation and complete mobile manipulation, mitigating the navigation-manipulation gap and improving continuous-contact interaction. The project page is at https://haochen611.github.io/TONAV.

World models for physical interaction are typically trained to predict future observations or latent features; however, a planning-oriented model must answer a fundamentally different question: whether a candidate action produces a task-consistent future while preserving essential relations.Monolithic state representations obscure the underlying entities, while standard instance-level object slots merely identify \emph{what} is present without specifying \emph{what role} each entity plays in the task context. To bridge this gap, we present the Semantically Rich World Model (SR-WM), a task-conditioned world model structured around five functional roles: gripper, target, goal, relation, and phase.Within SR-WM, a visual entity encoder extracts soft entity hypotheses from pretrained patch features, allowing segmentation masks to serve as optional proposal priors without mandating them as required state representations or inference inputs.A role binder subsequently maps these hypotheses to task-specific roles, while an action-conditioned dynamics model predicts role transitions alongside fine-grained semantics, including grasp/contact, predicate establishment, relation preservation, fixture state, and phase change.Crucially, this unified role state grounds downstream multi-candidate action generation, stage-aware reranking, and violation-aware suffix resampling.Our comprehensive evaluation protocol spans all four LIBERO simulation suites, cross-suite transfer, perception diagnostics, and action-sensitivity analysis.Ultimately, this formulation transforms object-centric prediction into a semantic interface linking visual dynamics with planning-oriented decision making.

Vision-based whole-body loco-manipulation on humanoid robots is challenging due to partial observability, contact-rich dynamics, and the difficulty of learning long-horizon behaviors from high-dimensional visual inputs. We present \href{https://github.com/DreamMimic/DreamMimic}{DreamMimic}, a framework that distills privileged teacher policies into vision-based humanoid controllers via world-model-assisted distillation. Instead of using a Dreamer-style RSSM for planning, we repurpose it to learn predictive latent dynamics that serve as both a representation space and an action-conditioned multi-step supervision signal, while exposing compact predictive features to the student policy to reduce long-term drift. Beyond standard reconstruction objectives for proprioceptive and visual observations, we add auxiliary prediction heads for privileged state, contact, object state, and reward estimation. These heads provide additional supervision related to agent--object interaction and task progress, encouraging the latent representation to retain signals that are useful for contact-rich loco-manipulation. We further introduce Performance-Conditioned Guidance (PCG), a reward-driven adaptive distillation schedule that computes performance scores for both teacher and student to dynamically balance guidance and exploration. PCG prevents both premature teacher annealing and excessive teacher interference in challenging visual settings. Experiments on OMOMO and BEHAVE show improved tracking-based loco-manipulation performance over strong vision-based baselines, without exposing online privileged interaction states to the student at deployment. Qualitative simulations further examine morphology and simulator changes. These results suggest that world models can provide a useful mechanism for stabilizing visual policy distillation in contact-rich humanoid behaviors.

Modern driving action models are increasingly improved in a self-improvement loop, where a learned world simulator imagines future observations and the resulting data is fed back to refine the action model. However, the bottleneck of this loop lies in the simulators' inability to generate behaviorally plausible responses by surrounding agents, making generated data both unrealistic in interaction and imbalanced in distribution. We introduce BehaviorWorldGen, a framework that closes the loop between action models and world simulators through controllable behavior-aware structured world generation. Its core component is BehaviorFlow, a meta-action-conditioned traffic-flow model that injects interpretable behavior controls and jointly generates multi-agent rollouts. BehaviorFlow realizes the specified agent behaviors while allowing surrounding vehicles to respond to the ego and to one another. The resulting rollouts are rendered by a world simulator into realistic multi-view observations, which are paired with corrected interaction-aware trajectories for action-model refinement. Since BehaviorWorldGen uses structured trajectories as the interface between its modules, it is compatible with diverse action models and world simulators. Experiments on world generation, scene extrapolation, and policy refinement demonstrate consistent improvements, with the largest benefits concentrated on difficult interactive scenarios.

LLMs generate fluent plans for robots but routinely violate the syntactic and se8mantic constraints they must satisfy to execute, and existing remedies trade formal guarantees against plan quality: soft methods (affordance scoring, grounded decoding) give no guarantee, while symbolic planners (LLM+P) discard the LM's commonsense. We propose \textbf{Meta-Ctrl}, a constrained-decoding framework that guarantees the encoded constraints while preserving the base LM's plan quality. Meta-Ctrl introduces \emph{meta-tokens}---a compact vocabulary of grounded actions---enforcing syntax at the token level and semantics (preconditions, goals, ordering) at the action level, an exact factorization that cuts the memory of constrained decoding from over 107TB to under 2GB. With it, a small open-weight LM becomes competitive where it otherwise sits at the bottom of the leaderboard: on WAH-NL under the LoTa-Bench protocol it reaches the highest reported subgoal success rate, exceeding GPT-4's, with consistent gains across the Embodied Agent Interface. We further demonstrate it on a real tabletop robot, where every generated plan satisfies its preconditions and goals by construction. Project website: https://meta-ctrlg.github.io/.

LLM-based multi-agent systems (MAS) solve complex tasks through communication among role-specialized agents. However, inter-agent dependencies introduce reliability risks beyond isolated agent failures. For instance, errors in intermediate messages could be inherited and amplified by downstream agents. Existing uncertainty quantification (UQ) methods mainly target isolated responses or single-agent reasoning, and therefore fail to capture uncertainty propagation in MAS. To this end, we propose PropUQ-MAS, an error propagation-aware UQ framework that represents MAS execution as a communication-structured graph and estimates each step's reliability by combining local uncertainty with uncertainty inherited from upstream messages. Extensive experiments demonstrate that PropUQ-MAS consistently improves UQ in MAS, with average relative gains of +6.10% in AUROC and +47.58% in PRR.

Autonomous agents increasingly interact with other independent agents, and such interactions typically admit multiple joint behaviors. When two agents prefer different ones, their independent strategies may be mutually incompatible and fail to reach a coordinated outcome; when they are identical, neither can differentiate its role when needed. Ideally, an agent should coordinate with any agent it encounters, regardless of which admissible joint behavior that agent aims to realize. We therefore propose a new form of strategy, the opinion-guided strategy, which keeps all the admissible joint behaviors available and postpones the selection to execution time, when the other agent's behavior reveals which one to realize. To realize this, nonlinear opinion dynamics are leveraged in a layered realization to guide the agent to a common admissible joint behavior in response to the other agent's evolving behavior, even without communication. We formally establish the conditions under which the strategy remains robust to every preference the other agent may hold. This robustness has an important implication: two agents running identical strategies can break symmetry when needed, a capability that conventional strategies lack. Three case studies across different applications show that the opinion-guided strategy coordinates with every randomly encountered agent, as long as it is willing to realize one of the admissible joint behaviors. One of them corresponds to a general-sum game: unlike conventional approaches devoted to finding a unique Nash equilibrium in advance, the opinion-guided strategy keeps every equilibrium open and guarantees the agents reach one, decided by their runtime interaction.

Construction robotics and automation offer promising means of improving productivity, alleviating workforce shortages, and reducing workers' exposure to physically demanding tasks. However, reliable contact-rich robotic assembly remains challenging under tight tolerances, fabrication inaccuracies, and uncertain contact dynamics. To address this challenge, we present a framework coupling diffusion policies trained on simulation-generated pose and force/torque data with an L1-inspired adaptive controller that corrects policy-predicted actions online to compensate for unmodeled contact dynamics. We benchmark the framework against baselines in timber joinery, pipe fitting, and sequential full-scale truss assembly. It achieves 100% success on single-task assemblies and 90-100% success across sequential truss assembly subtasks, with lower, more stable contact forces than the baselines. By enabling zero-shot sim-to-real transfer for force-aware contact-rich assembly, the framework reduces costly, labor-intensive real-world data collection for policy training and advances scalable, robust automation of multistage assembly, motivating extension to broader contact-rich manipulation tasks in construction.

Minimally invasive procedures performed within confined anatomical spaces depend on continuous endoscopic visualization. Current robotic endoscope systems can stabilize or reposition an endoscope, but they do not possess relevant context to provide effective visualization assistance. We present EndoNav, an anatomy-grounded natural-language framework that translates high-level surgeon commands into autonomous endoscopic visualization behaviors within patient-specific sinonasal anatomy. Spoken surgeon commands are transcribed and interpreted by an endoscopic viewpoint agent conditioned on a patient-specific anatomical scene representation. Rather than generating robot motion directly, the viewpoint agent generates structured visualization objectives that are converted into target viewpoints and inspection trajectories, which are then executed through geometry-constrained endoscope motion planning and joint-space control. We evaluate EndoNav using a structured three-pass sinus examination across three CT-derived anatomical models. For one cadaveric specimen, autonomous visualization is compared with sinus examinations performed by two resident surgeons. EndoNav achieved mean visualization IoUs of 87.04% and 84.37% relative to the two surgeon examinations, compared with an inter-surgeon IoU of 87.44%, while recovering 92.91% and 93.20% of surgeon-observed anatomical surfaces, respectively. These results demonstrate the feasibility of grounding high-level anatomical commands into patient-specific geometric objectives and translating them into anatomically constrained robotic visualization behaviors.

Large language models (LLMs) have shown strong capabilities in reasoning, planning, and tool use, but many scientific and engineering tasks require more than plausible text and code generation. They require understanding how a system responds to intervention, which in practice depends on controlled experimentation. In this work, we propose a multi-agent framework that enables LLM agents to conduct controlled experiments with scientific simulation models for pharmaceutical process design. Given a user query and a baseline configuration, the system constructs a structured task representation, designs experiments, executes comparative simulation, interprets the resulting outcomes, and synthesizes evidence-based recommendations for process parameter optimization. By coupling language models with high-fidelity simulation models in an interactive agent framework, the proposed system supports reasoning through intervention, comparison, and observation. As a result, it produces more specific and actionable outputs than language-only reasoning. In an industrial application setting, this advantage is reflected in higher output specificity as well as improved user-rated correctness and helpfulness. Ablation studies and visualized case analyses further demonstrate the effectiveness and practical utility of simulation-integrated experimental reasoning.

Computer use agents (CUAs) are increasingly deployed as multi-agent systems that decompose a task into multiple subtasks executed across parallel virtual machines (VMs). However, a critical physical bottleneck is that the state of two VMs cannot be merged. Previous systems handle this ad-hoc rather than treating it as a first-class concern. We propose Spine-Branch Coordination for multi-agent computer use, a framework that decomposes a task into a "spine-branch" graph, where the spine carries the main task flow with continuous VM state and branch tasks execute in parallel to collect information the spine needs to complete the task. Branch VMs are discarded once their tasks finish, so no VM merging ever occurs. Experiments show that on 200 long-horizon tasks from Odysseys and across three CUA backbones, Spine-Branch improves success rate over the baseline system by 6.0% to 16.5%, while reducing per-task cost by 34% to 70%, indicating that explicitly modeling VM-state merging constraint enables multi-agent computer use to scale efficiently.

World-Action Models (WAMs) build robot control on video-generation backbones, which jointly predict dense future visual trajectories and robot actions. We argue that video generation is an unnecessary intermediate objective for world-action modeling. For robotic manipulation, the goal of a world model is not to reproduce how the world looks at every intermediate moment, but to predict the state that the world will reach after an action is executed. The intermediate frames only describe the visual transition between physical states, which consumes substantial model capacity and computation, but do not directly specify the physical outcome that the robot action is intended to produce. In this paper, we propose DELE-w0.5, which infers robot actions from predicted future states without relying on video generation. Concretely, DELE-w0.5 infers the action sequence from its corresponding compact future latent state. The future latent state captures the action-relevant physical outcome of robot interaction and serves as an explicit bridge between world modeling and action generation. The core design principle of DELE-w0.5 is to model how the physical world changes under robot actions, rather than how its visual appearance evolves frame by frame. This formulation removes the high-dimensional visual redundancy introduced by dense video representations, and it therefore enables cheaper training and low-latency inference. Across 480 real-robot trials on four long-horizon manipulation tasks, our DELE-w0.5 achieves the best performance among all compared policies, attaining 62.5 overall full-task success and 81.3 macro ordered-stage progress, outperforming the strongest baseline by 47.5 and 30.7 percentage points, respectively.

Robot foundation models have substantially advanced perception and control, but natural human-robot collaboration requires more than executing isolated commands. A robot must recognize ambiguity, maintain context across turns, communicate its intentions, and revise ongoing behavior as the user's intent changes. We present $\scriptstyle\mathsf{Ludi}_{\scriptscriptstyle 0.1}$, an agentic system for socially intelligent robots that integrates interactive speech, multimodal reasoning, memory, navigation, and learned manipulation. Its decision-making core is a fine-tuned vision-language model trained on multi-turn interaction traces spanning ambiguous requests, clarifications, corrections, interruptions, mixed social and task dialogue, and multi-step tasks. A purpose-built harness manages the model-tool interaction loop, while specialized navigation and manipulation policies execute physical skills. Ludi${}_{\scriptscriptstyle 0.1}$ demonstrates a practical path toward fluid human-robot collaboration today while producing the multimodal interaction traces needed to develop a more deeply integrated foundation model for robots and people.

Stable quadrupedal locomotion on sparse terrain requires selecting state-relevant terrain evidence for precise foot placement. Model-based foothold planners provide precise foothold selection but rely heavily on explicit model assumptions. Recent attention-based map encoding (AME) studies show that end-to-end reinforcement learning (RL) can learn implicit foothold guidance. However, the computational cost of dense AME encoding grows with map resolution, limiting its scalability to fine-grained sparse terrain. We propose DELTA, a Deformable Elevation-Based Local Terrain Attention encoder. DELTA predicts state-conditioned sampling locations, forms terrain evidence tokens from adaptive local elevation patches, and attends only to a fixed-size token set. With fixed sampling and patch settings, DELTA's encoder cost is independent of map resolution. Experiments show that DELTA achieves final traversal performance comparable to AME at the standard resolution while improving learning efficiency. This fixed encoder cost enables the use of higher-resolution terrain maps, improving traversal on fine-grained sparse terrain. DELTA also demonstrates strong generalization to unseen mixed evaluation courses composed of continuous and discrete terrain elements. Beyond simulation, DELTA demonstrates successful sim-to-real transfer on RAIBO2. Analysis of the learned sampling offsets and attention weights shows that DELTA samples steppable regions and attends to terrain evidence relevant to future touchdowns without foothold labels or attention supervision.

Software systems have traditionally been organized around applications where human users act as principal decision-makers. Recent developments in agentic capabilities alter this paradigm: software agents now autonomously translate high-level goals into structured tasks, orchestrating tools, services and sub-agents to execute complex workflows. This evolution gives rise to an agent economy where these autonomous agents capture real economic value. However, the infrastructure required to support the agent economy fails across three critical dimensions: the absence of persistent identity infrastructure prevents systemic accountability in agentic workflows; capability claims remain self-declared not backed by verifiable execution evidence; and the disconnect between creator identities, agent performance, and project value hinders the economic valuation of agents as assets. While existing registries provide naming and discovery, unifying these features around a persistent identity anchor remains largely unaddressed. TessIndex is a capability-verified identity system for agent primitives that utilizes a dual-plane architecture: the blockchain records compact commitments for identity, ownership, and verification, while centralized servers maintain dynamic metadata for discovery, commerce, and reputation. It establishes: persistent identities across agent primitives to enforce systemic accountability in autonomous workflows; a predicate-based verification process replacing self-declared claims with cryptographic capability proof; an identity infrastructure that links agent performance to both project and creator identities while capturing value through tokenization. Ultimately, TessIndex serves as an integrated infrastructure that binds an agent's existence across capabilities, execution, and reputation into a single persistent identity.

In embodied AI, safety risk can be latent: a benign instruction and a safe scene become hazardous only when composed. Prior work has advanced embodied safety by varying visual contexts or evaluating execution-time dynamics, but the complementary axis of fixing the scene and varying only the instruction remains underexplored. We introduce GuardianBench, an instruction-contrastive benchmark grounded in international safety standards that isolates this latent contextual risk through 3,024 instruction-scene examples organized as same-scene Safe/Unsafe contrastive pairs across various hazard categories. Benchmarking state-of-the-art vision-language models (VLMs) reveals instruction-insensitive verdicts: models disproportionately approve both instructions under a given scene; across the primary models, average pair accuracy is only 24.1%. Our systematic rationale audit localizes the dominant failure: models fail to bind the instruction-relevant cues that differentiate safe from unsafe compositions. As a post-training case study, Verdict Log-Odds Supervision (VLOS), a lightweight verdict-level objective, substantially improves performance on open-weight backbones. Together, our latent contextual risk task formulation, standards-grounded contrastive benchmark construction, pair-level and rationale-level failure diagnosis, and benchmark-enabled verdict calibration establish GuardianBench as a controlled evaluation suite for exposing and improving safety reasoning over instruction-scene compositions under latent contextual risk.

Automated evolution of Multi-Agent Systems (MAS) holds significant potential for reducing the manual effort required to design and optimize LLM-based agent architectures. However, extant search-based paradigms face a fundamental trade-off, where an expanded optimization scope exacerbates evolutionary instability, while discrete branch-and-discard search isolates insights across lineages. To address these limitations, we propose a continuous, data-driven optimization paradigm built upon a unified ReAct-based infrastructure that reconciles a broad optimization scope with operational stability. Under this paradigm, we present OptiMAS, a task-agnostic agentic optimizer that leverages textual interaction trajectories and task feedback as loss signals for end-to-end MAS evolution. Equipped with a novel dual-track memory mechanism, OptiMAS sustains performance improvement over extended optimization horizons. Evaluation on four heterogeneous agentic benchmarks with three varying scale and accessibility LLM backbones, demonstrates that OptiMAS consistently achieves competitive or superior accuracy relative to both domain-specialized hand-crafted systems and existing evolutionary methods. Our work establishes a practical milestone toward robust, automated MAS evolution.

Human-in-the-loop real-world reinforcement learning enables rapid acquisition of effective robotic manipulation policies for individual tasks, often within tens of minutes. Yet it remains unclear how to extend this paradigm to continual learning, where a single policy must acquire new skills without losing previously learned behaviors. Existing real-world continual learning methods do not explicitly constrain prior behaviors, leading to severe catastrophic forgetting. We introduce Continual Interactive Distillation for Embodied Reinforcement Learning (CIDER), a continual reinforcement learning framework that freezes the accumulated historical policy as a teacher before learning each new task and interleaves task learning with distillation-based retention. We further introduce gradient routing to separate the gradients used for acquiring new tasks from those used for preserving prior behaviors. We evaluate our method with a single shared actor on six real-world household and industrial manipulation tasks. Interactive Distillation maintains high measured success on previously learned tasks across our six-task real-robot sequence while acquiring each new task in 10 to 20 minutes, whereas every baseline forgets at least one previous task. Additional ablations reveal the key design choices that govern the tradeoff between stability and plasticity in real-world continual reinforcement learning.

Human tactile perception relies on complex multisensory cues. Yet the relationship between tactile signals and perceptual representations remains poorly understood, limiting the integration of touch in digital environments and human-like robotic perception. To address this gap, we developed a computational framework comprising three interconnected deep learning models that map multisensory touch data to material perception, without relying on hand-crafted features. The models represent progressively different routes from tactile signals to material class: from low-level interaction signals to perceptual attribute distributions (Model 1), from predicted attribute distributions to material classification (Model 2), and directly from tactile signals to material categories, bypassing intermediate representations (Model 3). By combining deep learning with Integrated Gradients, the framework achieved high accuracy while offering interpretability, revealing which sensory modalities most strongly drive its decisions. Our results show that deep learning can approach near-perfect material classification when unconstrained by intermediate perceptual stages, but matching human-like performance is harder once those stages are modeled explicitly. Notably, thermal cues emerged as particularly informative across all models, providing robust signals for material differentiation. The results offer a computational account of how tactile signals lead to material perception and show how interpretable deep learning can both approach human-level performance and reveal cues that robotic and haptic systems need to incorporate.

Aerial payload transport using multirotor unmanned aerial vehicles is challenging because payload geometry, contact interaction, grasp stability, flight control, and narrow-passage traversal are strongly coupled during pickup and transport. Object-specific grippers often cannot adapt their footprint or grasp geometry when the payload shape or passage width changes. This paper presents a vision-guided morphing quadcopter for multi-geometry payload transport through narrow passages. The proposed platform uses four hybrid arm-leg structures that function as both landing supports and grasping members. A centrally placed actuator drives a tendon-based morphing mechanism, enabling all four arms to synchronously retract or expand for object grasping, footprint reduction, and post-transport release. Onboard vision estimates the payload geometry and passage width, while endpoint force feedback is used to confirm grasp contact during payload engagement. A phase-wise mission planner, PID-based flight stabilization, and morphology-adaptive grasp controller are implemented in a MuJoCo simulation environment. The framework is evaluated using box, cylindrical, and spherical payloads, representing flat-faced, rolling-curved, and fully curved contact conditions. Across the three cases, the simulated system completes the pickup-transport-release sequence with a maximum RMS position error of 0.31 m, a final drop-zone error below 0.18 m, a compact grasp footprint of 0.09-0.21 m2, and a footprint reduction of 75.0-89.7 percent. The results demonstrate that a single-actuator morphing quadcopter can adapt its grasp footprint for the transport of payloads with different geometries while reducing its overall footprint for narrow-passage traversal.

Autonomous excavation requires an intelligent control system that can convert spatial work intent into coordinated bucket motion under contact-rich soil interaction. This paper presents a target-conditioned intelligent control framework for autonomous excavation in a physics-based deformable-soil simulation workflow. An image-aligned target mask serves as a visual spatial command for the desired digging region, while a mask-conditioned Action Chunking Transformer maps multi-view RGB observations, proprioception, and the target mask to temporally extended joystick commands. To reduce target-ignoring behavior, demonstrations are organized with paired-condition supervision, where the same or closely matched scene is demonstrated with different target masks and corresponding action chunks. The framework is evaluated through both a diagnostic manipulation task and an excavation simulation benchmark with single-scoop and sequential pile-clearing protocols. In manipulation, target success is 4\% for no-condition ACT, 63\% for non-paired mask-conditioned ACT, and 96\% for paired-condition mask-conditioned ACT. In sequential pile clearing, paired-condition mask-conditioned ACT removes 76.8\% of the pile versus 27.4\% and 15.7\% for the two baselines, with 91.0\% human-normalized efficiency. The results show that visual target conditioning, paired demonstration structure, and action-chunk control form a practical cyber-physical simulation pipeline for excavator automation.

Vision-Language-Action (VLA) models are typically trained with behavior cloning (BC) on expert demonstrations. However, BC provides only positive supervision for expert actions, without explicit negative supervision indicating which actions are instruction-inconsistent or otherwise inappropriate. Reinforcement learning (RL) can provide such corrective signals, but often relies on externally specified rewards or curated non-expert data, both of which are costly to obtain in robotics. We show that offline RL for VLA models need not rely on curated non-expert trajectories: successful expert demonstrations alone can be transformed into dense corrective supervision through instruction relabeling. Specifically, by pairing expert actions with mismatched alternative instructions, we synthesize counterfactual instruction-observation-action tuples from the dataset and combine them with adversarial discriminator training to learn an instruction-grounded reward model for offline RL, without collecting additional rollouts or annotations. On the robustness-focused LIBERO-PRO benchmark, our method improves robustness to object position and task perturbations over a strong state-of-the-art baseline. It also outperforms competitive baselines in real-robot experiments on the TX-G2 (compatible with AGIBot G2). More broadly, our results suggest that, for data-constrained VLA learning, extracting denser supervision from each demonstration can complement collecting additional data.

Transformer-based models are widely used for clinical prediction from electronic health records (EHRs), yet their architectures still require substantial manual tuning, and the optimal configuration may vary across tasks and hospitals. Neural architecture search (NAS) automates architecture design, but conventional methods are computationally costly for Transformer-based EHR models. Recent large language model (LLM)-guided NAS methods reduce manual search design but typically conduct each search independently, without reusing architecture knowledge across hospitals. In this study, we propose ATHENA (Agentic Transfer across Hospitals for EHR Neural Architecture Search), a knowledge-guided agentic NAS framework for Transformer-based EHR modeling. ATHENA uses a weight-sharing supernet that is pretrained once per hospital, allowing candidate architectures to be instantiated as inherited subnetworks and evaluated through fine-tuning rather than independent pretraining. It also incorporates a two-layer cross-hospital architecture prior. The first layer retrieves high-performing architecture examples from source sites based on task descriptors, while the second estimates the effects of architectural components using SHapley Additive exPlanations (SHAP)-based meta-regression. These priors guide a multi-agent LLM search together with validation feedback from the target hospital. Across six clinical prediction tasks and two independent health systems, ATHENA matches or outperforms four NAS baselines in 9 of 12 hospital-task evaluations at a search budget of 30. It also shows more consistent architecture selection across repeated searches. ATHENA provides a practical approach for reducing manual architecture tuning in Transformer-based EHR modeling. Code is publicly available at https://github.com/GatorAIM/ATHENA.

Recent mesh-based simulation advances have, in no small part, relied on neural surrogates of two distinct families: global models that route information through a small set of latent tokens, and local models that perform message passing across mesh edges. Consistent with both classes is the inability to perform beyond low-dimensional problems and small-scale or oversimplified meshes, the simulation regimes where industrial problems reside. Our work shows this explicitly and presents a unified formulation. In global approaches, latent-token attention acts as a spatial low-pass filter, while local message passing lacks the global reach necessary to propagate information across large mesh spaces. Viewed through the error, the two operators are the halves of a multigrid cycle: one corrects errors at the lower end of the spectrum, the other at the higher end, and neither can do the other's job. We introduce Read-Write-Relax (RWR), which interleaves latent attention with message-passing relaxation under a unified formulation. The interleaved processor lowers error across the entire spectrum, making RWR the most accurate model in nearly every comparison across our industrial and public benchmarks. It is also markedly data-efficient in the scarce-data regimes, accurate on the engineering quantities of interest, and scales full-field predictions to challenging, large-scale problems.