Physical AI Brief

Are AI agents tools, co-authors, or researchers? We present a quantified case study ($N=1$): a physicist supervising an AI coding agent (Claude Code, Sonnet and Opus models) over 12 work days and 57 sessions to build CLAX-PT, a differentiable one-loop perturbation theory module in JAX. We documented and classified 15 supervision events by intervention level. The agent resolved ten autonomously by iterating against oracle tests. Two more by the physicist's domain knowledge. The three it could not -- all evaded oracle detection -- share a common property: the agent treated symptom reduction as root-cause resolution. It spent 33 of the 57 sessions adjusting coefficients within a code architecture that could not represent the target physics, and could not re-evaluate its CLASS-PT branch choice even when prompted to reconsider; only an injected physics concept (anisotropic BAO damping) triggered the redesign. Separately, the agent committed a calibrated correction that passed all oracle tests but corresponded to no quantity in the theory, predicting wrong values at any other cosmology. The fudge factor was caught and replaced within the same session. Three supervision practices proved critical for catching what oracle tests missed: testing at diverse parameter points beyond the fiducial calibration; shared changelogs that surfaced stalled exploration across sessions; and an explicit rule against unphysical numerical patches. In this case, supervision design, not model capability, determined whether the agent's output was trustworthy. Closing the gap would require agents that propose architectural alternatives rather than optimize within a given structure, and distinguish predictive adequacy from explanatory correctness -- capabilities not exhibited here, not obviously addressed by scaling alone. [Abridged.]

Moving Object Segmentation (MOS) aims to discover, segment, and track objects that move independently of the camera. Current MOS methods, however, exhibit two fundamental limitations: they rely on pre-computed 2D auxiliary modalities such as optical flow or point trajectories that lack 3D geometric information, and they treat motion as a sequence-level attribute, overlooking the instantaneous motion state of each object. We address both by grounding MOS in 3D space and time, and propose GMOS, a framework that operates directly on RGB video to produce 3D-aware, temporally fine-grained segmentation of multiple moving objects, alongside a foreground--background variant GMOS-S for faster deployment. To support training and evaluation in this regime, we curate GMOS-2K, a dataset of 2,210 real-world videos with per-object temporal motion annotations drawn from five established Video Object Segmentation (VOS) benchmarks, and formalise MOS-I ("I" for instantaneous), a temporally fine-grained evaluation protocol with three complementary metrics. GMOS achieves state-of-the-art results across MOS, MOS-I, and Unsupervised VOS benchmarks, while running significantly faster than prior multi-object MOS methods and supporting online inference for streaming deployment.

Long-rollout causal video diffusion has converged on a fixed-size sliding-window KV cache, with recent progress innovating within this layout by changing which tokens occupy the window or how their positions are encoded. The per-head KV layout itself, a dominant contributor to streaming memory and latency, has been mostly left unchanged. In this paper, we present the first study of Multi-Head Latent Attention (MLA) in video diffusion. VideoMLA replaces per-head keys and values with a shared low-rank content latent and a shared decoupled 3D-RoPE positional key, reducing per-token KV memory by 92.7% at every cached layer. We further investigate why MLA succeeds in video diffusion even though the spectral assumption often used to motivate it in language models does not hold: pretrained video attention is not low-rank, with 99%-energy effective rank far above any practical latent dimension. VideoMLA retains quality at compression ratios where direct spectral approximation would predict large reconstruction error. We show that the MLA bottleneck, rather than the pretrained spectrum, determines the effective rank: both spectral and random initialization occupy nearly the full rank budget from initialization, and training preserves this budget while adapting within it. On VBench, VideoMLA matches short-horizon streaming video diffusion baselines, achieves the best overall score at long horizons among evaluated methods, and improves throughput by 1.23x on a single B200.

Robot manipulation critically depends on perception that preserves the action-relevant aspects of a scene. Yet most robot learning pipelines are built upon visual encoders pre-trained for static recognition or vision-language alignment, leaving motion understanding to downstream policies. We introduce DynaFLIP, a dynamics-aware multimodal pre-training framework that pushes motion understanding upstream into perception. We construct image-language-3D flow triplets from heterogeneous human and robot videos, and use these triplets as training-time supervision to shape an image-only encoder. Our key idea is to encourage the three modalities to span a small simplex volume in the shared hyperspherical space -- a smaller simplex volume indicating stronger alignment. To avoid the geometric ambiguity and trivial collapse of naive volume minimization, we combine simplex-volume minimization with a cosine regularizer and a contrastive objective. Our analyses show that DynaFLIP focuses on control-relevant regions critical for manipulation. The resulting dynamics-aware representations serve as reusable visual backbones and consistently outperform baselines across diverse downstream policies, including VLAs. We validate this across diverse simulation and real-world setups, with gains reaching +22.5% under out-of-distribution scenarios. Our results suggest that robot generalization improves when visual representations are trained to encode not just what is present, but how the world changes under action.

Autoregressive video diffusion models generate streaming video by producing frames sequentially, conditioning each chunk on previously generated content. These models are structurally anchored to the first frame: its key-value representation occupies a privileged position in the attention cache and serves as the primary scene reference throughout generation. As the cleanest and most error-free position in the cache, this anchor draws disproportionate attention, suppressing video dynamics, and locking scene composition to the initial viewpoint even as the scene naturally evolves. The result is a temporally shallow video in which motion, camera movement, and scene progression are dampened in favor of static consistency. To address this, we replace the static anchor with an adaptive state, a hidden latent that the model denoises alongside content at every chunk but never renders. Rather than referencing a frozen first frame, the model generates its own scene anchor at each step by attending to both the previous state and the current content, producing a reference that evolves with the generated content. Unlike standard video generation, which encodes an absolute notion of time, our formulation treats time as relative: every generation step sees the same positional structure regardless of how far generation has progressed, and the state transition is identical at every chunk. Together, these properties introduce a recurrence into the generation process, where denoising serves as the transition function, and the KV cache serves as the carrier, requiring no external module. Experiments demonstrate that the adaptive state substantially improves video dynamics, enabling richer motion and natural scene progression within generated videos.

Data-driven approaches have revolutionized 3D vision, enabling transformers to effectively reconstruct and generate static 3D objects. However, generating simulative 4D dynamics -- realistic temporal deformations of static objects under various physical conditions -- remains challenging and often ad hoc, despite its importance in building comprehensive 3D world models. Most existing methods assume a predefined physical model and use system identification to estimate parameters, restricting these methods to specific categories and small-scale datasets. We propose that these restrictions can be overcome by learning a data-driven kinematic state parameterization for object-centric physical systems. Specifically, we learn both a latent space representing all possible states of the object and a decoder that maps any sampled latent to a plausibly deformed shape of the object. We refer to this parameterization as Neural Object Kinematics (NeuROK), and learn a transformer-based encoder-decoder model on a curated large-scale 4D dataset. This formulation and the learned model significantly simplify the generation of simulative dynamics since we only need to consider the dynamics within a low-dimensional latent space from the Lagrangian mechanics' perspective in classical physics. We demonstrate the effectiveness and generality of this neural simulation framework across diverse dynamic object types, showing clear advantages over prior works. Project page: https://chen-geng.com/neurok

As video diffusion models (VDMs) advance toward world models, a key question arises: do they truly understand causality, or merely overfit to statistical temporal patterns? Existing benchmarks mostly rely on synthetic data, limiting real-world generalization due to the sim-to-real gap. We present YoCausal, a two-level benchmark inspired by the Violation of Expectation (VoE) paradigm from cognitive science. By temporally reversing real-world videos at zero cost as natural counterfactual samples, YoCausal establishes an arbitrarily extensible evaluation protocol. Level 1 introduces the Reverse Surprise Index (RSI), quantifying arrow-of-time perception via denoising loss. Level 2 introduces the Causality Cognition Index (CCI), which leverages a VLM to stratify datasets into causal and non-causal subsets, disentangling genuine causal reasoning from temporal bias. Evaluation of 13 state-of-the-art VDMs reveals that perceiving the arrow of time does not imply understanding causality, and a significant gap persists relative to human-level causal cognition.

Recent advances in Vision-Language Models (VLMs) have achieved impressive performance across many tasks, yet prior studies report unsatisfactory performance when applying large language or multimodal models to finding abnormal patterns in sequential data. Public anomaly detection benchmarks typically provide interval annotations but not natural-language rationales, making it difficult to fine-tune VLMs to produce grounded, interpretable decisions. To address this gap, we construct VisAnomBench, a curated benchmark built from public time-series datasets and augmented with high-quality anomaly explanations selected from multiple large VLMs using fine-grained, task-specific rewards. Through fine-tuning on this benchmark, we develop VisAnomReasoner, a parameter-efficient VLM for time-series anomaly detection. Experimental results on VisAnomBench show that VisAnomReasoner achieves more accurate anomaly localization and consistently outperforms all baselines, with improvements of at least 21.23 and 23.87 percentage points in precision and F1, respectively. Additional experiments on the TSB-AD-U benchmark demonstrate strong cross-benchmark generalization, with VisAnomReasoner improving precision and F1 by 9.57 and 13.39 percentage points, respectively.

To improve the reasoning capabilities of large language models, test-time compute is typically scaled by generating intermediate tokens before the final answer. However, this couples reasoning to autoregressive generation and thereby conflates internal computation with external communication. In contrast, human cognition can use working memory to hold and manipulate information internally without the need to externalize intermediate thoughts. Drawing on this principle, we introduce Reasoning in Memory (RiM), a latent reasoning method that replaces the autoregressive generation of reasoning steps with memory blocks. These memory blocks are fixed sequences of special tokens that unlock the working-memory capacity of large language models. Since they are fixed rather than generated, they can be processed in a single forward pass, enabling compute-efficient latent reasoning. To operationalize these memory blocks, we employ a two-stage curriculum. First, we ground them by predicting explicit reasoning steps after each memory block. Second, we discard this step-level supervision and iteratively refine the final answer after each memory block. Our experiments on reasoning benchmarks show that, across language models of different families and sizes, RiM matches or exceeds existing latent reasoning methods while avoiding the autoregressive generation of thoughts. These results demonstrate that large language models can be trained to use working memory as an effective mechanism for latent reasoning.

Studying scalable methods for visual generative modeling requires large, accessible, and stable datasets. We introduce GPIC, a Giant Permissive Image Corpus of approximately 28 trillion pixels. GPIC comprises diverse internet images captioned by a state-of-the-art vision-language model, including 100M training, 200K validation, and 1M test examples. Moreover, all GPIC images are permissively licensed for both research and commercial use. GPIC is safety-filtered, deduplicated, and centrally hosted on Hugging Face. We provide a benchmarking protocol for generative modeling on GPIC. Finally, we provide a reference baseline for pixel-space flow matching on GPIC. Our dataset, benchmark, and models are available at https://huggingface.co/datasets/stanford-vision-lab/gpic. Evaluation toolkit and code are available at https://gpic.stanford.edu

Generative video-to-audio (V2A) models produce highly plausible soundtracks, but it remains unclear whether they capture the underlying physical processes. Existing evaluations emphasize perceptual realism and overlook physical correctness under controlled interventions. In this paper, we introduce FlatSounds, a benchmark that audits the physical reasoning of V2A models through: 1) controlled counterfactual pairs in which a single physical factor is varied, and 2) single-video pattern tests that probe internal consistency and directional trends. These settings test whether the generated audio correctly reflects specific physical properties and timings. Our evaluation of state-of-the-art models reveals a consistent trade-off: models rely more on text captions than the visual stream to infer physics and semantics. Captions generally improve physical and semantic accuracy, but paradoxically degrade temporal alignment. Our results highlight the need to move beyond audio quality toward learning physical processes directly from pixels. Finally, we find that our physics-based metrics correlate strongly with human preference tests on our own data. Project webpage: https://research.nvidia.com/labs/cosmos-lab/flatsounds/

Reconstructing physically stable 3D scenes from a single RGB image enables casual images to be converted into simulation-ready digital assets for applications such as immersive interaction and content creation. However, existing single-image reconstruction methods fall short in capturing the physical structure of a scene. As a result, they often produce geometrically plausible but physically inconsistent results, including object floating and penetration, which lead to unstable behavior in physics simulations. Image-conditioned scene generation methods improve physical plausibility but often rely on strong scene priors, yielding plausible yet inaccurate object arrangements that fail to match the input image. We propose REST3D, a single-image reconstruction framework that can reconstruct physically stable 3D scenes by integrating physical scene understanding with physics-constrained refinement. We first introduce an agentic physical scene understanding technique that constructs a scene-tree representation capturing object physical states and inter-object relationships from a gravity-support perspective, providing a structural prior for reconstruction. Leveraging this structure, we initialize the scene using image-to-3D models, followed by scene-tree-guided alignment and physics-constrained optimization to resolve physical violations while preserving visual consistency with the input image. Experiments show that our method significantly reduces physical errors and improves simulation stability on both synthetic and real-world datasets while maintaining strong reconstruction quality. We further demonstrate the reconstructed scenes in VR-based human-object interaction, showing their potential for immersive applications.

Federated learning is an emerging distributed paradigm that addresses the challenges posed by heterogeneous, privacy-sensitive data. It enables multiple clients to train a model collaboratively by aggregating their local updates at a server. However, conventional aggregation schemes typically use fixed weights that fail to reflect unequal and time-varying client contributions, leading to biased and unstable learning. To improve fairness and stability, we propose the Trajectory Shapley Value (TSV), a contribution metric that evaluates how each client influences the optimization trajectory of the global model using a validation-based, temporally consistent utility. Building on TSV, we design FedTSV, an adaptive aggregation method that converts per-round evaluations into dynamic client weights, allowing the server to respond to heterogeneous and adversarial participation in real time. Experiments on benchmark datasets show that FedTSV accelerates convergence, improves robustness, and yields more equitable contribution assessments, thereby providing a principled foundation for fairness-aware federated optimization.

Multi-component LLM agents assemble probabilistic claims from components that each see only part of a joint problem; the composition can violate basic probability axioms even when every component is locally coherent. We formalise this locally coherent, globally incoherent failure via the compositional residual eps*, the L2 distance from the composed quote to the joint coherent polytope, computable at runtime from system output and the declared cross-component coupling constraints. A product-structure dichotomy characterises when local coherence suffices, and a Rayleigh-quotient prediction matches the observed residual within 7% on three of four relation classes. A hierarchical Boyle-Dykstra projection repairs the composition deterministically; an anytime-valid e-process gives sequential coherence monitoring. Across 1,876 ensemble cliques on a four-LLM mid-tier panel (frontier-panel rerun in Section 5.5), eps* > 0 on 33-94% of cliques, translating to +0.115 nats per bet of regret on 1,770 resolved bets under the proportional allocation rule (the gain collapses to +0.006 under bettors that themselves coherentise). Three intuitive LLM-side mitigations(retrieval, partition-aware prompting, aggregator-LLM) each fail or regress.

A plausible future mathematical claim must satisfy two constraints: it should follow the direction of prior work and respect the formal dependencies that constrain what can validly follow. Existing approaches typically model only one of these sources, producing claims that are either weakly grounded or insufficiently motivated. We introduce grounded future mathematical generation, where the goal is to generate a plausible future theorem-like claim for an anchor paper using two complementary sources of context: its scientific citation graph and aligned formal theorem dependency graph. To address this setting, we propose COMPOSE, a dual-graph framework that conditions a language model on both scientific citation context and formal theorem structure. To support this setting, we construct a dataset of 108K paired scientific-formal graph examples from arXiv and Mathlib, together with a benchmark of 47K future papers from 2024--2025. Experiments show that COMPOSE outperforms strong baselines on retrieval to real future papers and achieves the best overall performance under LLM-judge evaluation, producing more grounded and mathematically richer outputs. These results show that future mathematical generation benefits from combining scientific context with formal structure. Project page is available at https://david-busbib.github.io/COMPOSE-page/.

Autonomous AI research agents aim to accelerate scientific discovery by automating the research pipeline, from hypothesis generation to peer review. However, existing benchmarks rarely test a fundamental bottleneck: whether Large Language Models can judge the methodological viability of a research idea before expending time and computational resources. We introduce SoundnessBench, a curated benchmark of 1,099 machine-learning research proposals reconstructed from ICLR submissions, labeled with reviewer soundness sub-scores, and audited against source papers. SoundnessBench should be interpreted as a benchmark for recoverable proposal-stage soundness rather than exact prediction of full-paper review outcomes. Across 12 frontier LLMs, we find a pervasive optimism bias: under standard prompting, models frequently rate low-soundness proposals as sound, while aggressive prompting largely shifts errors from false positives to false negatives. Additional controls for public-corpus contamination, paper-identifying phrases, surface features, and human audit quality suggest that this behavior is not explained by a single confounder. Our results indicate that current LLMs are not yet reliable as standalone first-gate evaluators for scientific rigor.

The ability to reason, adapt, and creatively solve problems under unexpected challenges is essential for robots operating in real-world environments. However, current robotic benchmarks primarily emphasize skill-level execution and provide limited insight into such cognitive reasoning capabilities. We introduce RoboWits, a bi-manual robotic benchmark designed to systematically evaluate cognitive reasoning, creative tool use, and robustness to unexpected conditions. To enable scalable construction of high-quality reasoning-centric unexpected scenarios, we propose an automated task generation pipeline formulated as a multi-agent cooperative framework, comprising agents for seed task generation and verification, metric generation, scene generation, and task mutation. Using the pipeline, we curated 30 diverse seed tasks and 208 tasks with mutations and graded difficulty across geometry, material, and assembly-based reasoning. We benchmark popular robot policies, pre-trained VLAs, and oracle-state planners. Our results reveal a significant performance gap: while pre-trained VLAs exhibit preliminary success on seed tasks after single-task fine-tuning, they struggle to perform on mutated tasks, implying their brittleness in manipulation tasks requiring reasoning, strategy adaptation, and robustness to deceptive or constrained environments. Project page is available at https://umass-embodied-agi.github.io/RoboWits.

We study language generation in the limit under bounded memory. In this task, a learner observes examples from an unknown target language one at a time and must eventually output only new valid examples. Prior work assumes access to the entire history, a strong assumption since realistic algorithms retain limited past information. Classical work in learning theory shows memory constraints dramatically alter learnability; we extend this to language generation. First, we study memoryless generators. Under a mild enumeration restriction, every countable collection of infinite languages remains generable without memory. Without this restriction, we exactly characterize when memoryless generation is possible. For finite collections, we characterize the optimal minimax density achievable by memoryless generators -- the best density guaranteed against any collection of a given size. This combinatorial bound relies on Sperner's theorem and symmetric chain decompositions. We further show that a sliding window of the last $W$ examples does not improve this worst-case density, whereas allowing it to store $b$ adaptively chosen past examples improves the achievable density for every $b \geq 1$. Finally, we revisit identification in the limit, where the learner must converge to a single correct hypothesis for the target language. We focus on its incremental variant, where the learner remembers only its previous guess. Here, although exact identification fails on a collection of just three languages, a mild relaxation requiring convergence to an ``approximate'' version of the target is achievable for every finite collection. These results show bounded memory affects these tasks differently: generation remains achievable for every countable collection, while density and identification are confined to finite collections, with guarantees weakening as the collection grows.

We introduce Gram, an automated alignment auditing framework to assess the propensity of AI agents to engage in sabotage. We evaluate Gemini models across 17 simulated agentic deployment scenarios that incentivize sabotage. We find Gemini models misbehave in about 2-3% of our simulated trajectories. Many of these cases are explained by "overeagerness" in Gemini models resulting in both excessive role-playing and goal-seeking behavior. In contrast to other alignment auditing approaches, Gram is designed to specifically evaluate misalignment and intentional sabotage in agentic coding and research agents. We additionally introduce an experimental investigator agent pipeline which enables fine-grained targeted experiments to identify the drivers of misbehavior. We find that increasing realism of environments and removing nudges to misbehave tends to reduce sabotage rates close to zero.

Portrait photography is largely decided before the shutter opens: the subject's pose, the camera configuration, and the lighting devices must be coordinated within the surrounding 3D scene. In contrast, most existing computational methods focus on post-production in 2D image space, such as retouching, relighting, or editing images that already exist; pre-capture photographic planning remains largely unexplored. We introduce 3D aesthetic portrait planning, the task of generating human pose, camera, lighting, and exposure plans that produce visually compelling portraits while satisfying geometric and photometric feasibility in a 3D scene. Our approach builds a Photographic Scene Graph that represents scene affordances, subject-scene relations, and portrait-relevant lighting structure. Built on this representation, we perform aesthetic-guided comparative planning over previous attempts and current viewfinder observations. Experiments across diverse indoor and outdoor scenes show that our method produces portraits preferred by human raters and MLLM evaluators over competitive baselines, while maintaining high physical plausibility. Together, our results suggest a path from post-capture correction toward pre-capture computational portrait planning. Project repository: https://github.com/songrise/Before-the-Shutter

Autoregressive image and video generators are trained with teacher-forced histories but must sample from their own generated prefixes at inference time, making them vulnerable to exposure bias and prefix drift. Existing remedies either modify training or apply sampling-time guidance aimed primarily at external semantic conditions, such as class labels or text prompts, rather than testing whether a next-step prediction provides strong posterior support for the generated prefix itself. We propose Visual Prefix Guidance (VPG), a training-free inference-time guidance method for autoregressive image and video generation. VPG improves next-step prediction by contrasting the model's output under the generated prefix with its output under a corrupted prefix, then extrapolating logits toward candidates that strengthen the posterior support of the generated prefix. Across class-conditional image generation with VAR, text-to-image generation with Infinity, and text-to-video generation with InfinityStar, VPG improves generation quality without retraining the base model, reducing FID on VAR by 0.36 on average and improving benchmark performance on both image and video generation.

Across two public LLM leaderboards, many displayed pairwise rankings do not meet a conventional paired-test resolution target under the actual paired evaluation design: 11 of 40 Open LLM Leaderboard v1 pairwise comparisons and 4 of 9 MMLU-Pro top-10 adjacent-rank pairs are unresolved at (alpha, 1-beta) = (0.05, 0.8). The MMLU-Pro count rises to 6/9 under real subject-level clustering and stays at 5-6 out of 9 in 99.9% of category-bootstrap resamples. We frame paired LLM evaluation as a hypothesis-testing problem, invert level-alpha, power-(1-beta) tests, and report a per-pair resolution ratio q = N/N* as the primary diagnostic. A sharp small-effect expansion with an explicit second-order constant shows that the widely-used unpaired Cohen-h-plus-(1-rho) shortcut deviates from the correct N* by approximately a factor of two in the close-comparison regime, a deficit that three of five off-the-shelf calculators(Cohen 1988, G*Power, R pwr) silently inherit when the user post-multiplies their per-arm output by (1-rho). The unresolved-pair pattern remains under multiplicity correction and anytime-valid sequential testing.

We present GR3D, a spatial vision language model equipped with three complementary grounding capabilities--explicit 2D grounding, implicit 2D grounding, and monocular 3D grounding--within a single framework. GR3D introduces an implicit grounding mechanism that identifies entity mentions during generation and inserts the corresponding region tokens into the text stream, allowing the model to reference visual evidence on the fly when producing spatial chain-of-thought responses. In parallel, a region-prompted monocular 3D grounding design predicts 3D bounding boxes in the camera view from grounded region queries, supported by intrinsic-aware normalization and dense geometric supervision. Together, these grounding capabilities enable GR3D to decompose complex spatial understanding problems into grounded 2D perception followed by 3D inference. GR3D achieves consistent improvements across grounded and non-grounded spatial benchmarks, demonstrating grounding as an effective inductive bias for strengthening spatial understanding in VLMs. These grounding capabilities collectively enhance general spatial understanding beyond the grounding task itself.

Large language models (LLMs) show promise for clinical reasoning and decision support, but evaluation in realistic, electronic health record-congruent settings remains limited. Existing benchmarks often rely on static datasets or unstructured inputs that do not reflect the structured, interoperable data formats used in clinical systems. We introduce a pipeline for generating clinically realistic HL7 FHIR R4 bundles from unstructured text, enabling controllable evaluation of clinical decision support systems. The pipeline combines staged LLM generation with terminology-grounded validation and repair to reduce hallucinated codes and enforce structural and semantic consistency. Applying this approach to MedCaseReasoning, we construct MedCase-Structured, a synthetic dataset aligned with clinician-authored diagnostic cases, achieving valid FHIR generation for 82.5% of cases. Evaluation on MedCase-Structured reveals consistently lower diagnostic accuracy for LLMs on structured FHIR inputs than with plain text, highlighting the importance of deployment-aligned benchmarking.

Self-improvement at scale has been a longstanding goal for reasoning models, and there are two natural places to do it: at test time, through verification-refinement (V-R) loops; and at training time, through self-training methods. Both are gated by the same bottleneck: the verifier. V-R loops stall when verifier scores inflate while accuracy stagnates, and when feedback is too generic to act on; self-training fails similarly when bad self-generated data are added to training. Better verification would unlock both, but the capability we want to train, i.e., catching self-generated errors, lacks training signal. To address this challenge, we propose self-trained verification (STV). Our key observation is that, while a model cannot catch these errors alone, it can when shown the reference solution. We turn this asymmetry into a supervision target and train the verifier to imitate a more informed version of itself. At test time, STV substantially improves V-R loops on hard problems, while alternatives (e.g., SFT, RL on verifier scores, and even meta-verifiers) do not. STV roughly doubles accuracy on hard math and lifts it 14x on scientific reasoning tasks (1.5% to 21%). At training time, we additionally train the generator using RL with STV verifier's feedback inside the V-R loop - a procedure we call verifier-in-the-loop training (ViL). Starting from an RL-converged generator, ViL yields a further 33% gain in pass@1. More notably, the generator's standalone pass@1, with no verifier at test time, climbs 30% relative past where standard RL had converged. Hence, the next frontier in reasoning on hard problems may lie in how we train for and with verification.

Numeric tabular datasets are the dominant data format in scientific practice, yet large language models lack native mechanisms for representing numeric datasets in a meaningful way across heterogeneous feature spaces. Existing approaches either target predictive modeling over individual datasets, which requires a shared set of variable definitions, or lack mechanisms for interpretable cross-dataset alignment. The proposed methodology characterizes numeric tabular datasets through structured exploratory data analysis descriptors, embeds those descriptors into a shared vector space using a pretrained sentence transformer, and quantifies cross-dataset similarity via Canonical Correlation Analysis (CCA). Furthermore, a penalized formulation of CCA is applied to recover sparse, interpretable variable-level correspondences between datasets, identifying which statistical descriptors or variable-level quantities drive cross-dataset alignment without requiring shared variable names or feature conventions. Differential privacy is optionally applied to the descriptor set prior to embedding, supporting deployment in sensitive data contexts without requiring access to raw observations at time of comparison. The methodology is evaluated across 15 datasets spanning general-purpose benchmarks, materials informatics, and nuclear-grade graphite characterization. Results demonstrate a total P@1 score of 0.9, with known nearest-neighbor retrieval and cluster structure remaining robust across embedding ablations and differential privacy budgets. The proposed framework provides a principled pathway for integrating heterogeneous numeric data into retrieval-augmented generation pipelines while preserving statistical context, with direct applications to data-driven algorithm selection and simulation model initialization for unknown datasets.

Mid-training has become an important stage in modern LLM development, using large-scale curated mixtures to strengthen capabilities before final post-training. Its data selection problem is distinct: the data are optimized under a pretraining-style objective at near-pretraining scale, but are curated toward downstream capabilities and drawn from heterogeneous sources with different formats and training roles. As a result, effective selection requires both scalability and source-adaptive semantic criteria. Existing model-based methods scale well, but provide only implicit quality signals. Semantic selection methods offer stronger judgments, but usually assume fixed rubrics or standardized data formats. To address this mismatch, we propose MIRA, a source-aware filtering framework based on self-anchored rubric discovery. The key idea is to make rubric construction part of data selection: MIRA first discovers what should be evaluated for each source group, then distills those judgments into scalable student scorers for full-corpus filtering. On code-oriented mid-training with 21 sources and 5 source groups, MIRA outperforms selection baselines across nine code benchmarks and matches the full-corpus run while using only half the tokens.

Scientific discovery is an inherently creative and uncertain process, requiring reasoning beyond the recall of known knowledge. While many benchmarks have been proposed to evaluate large language model (LLM) performance on deep research tasks via multi-hop retrieval, their innovative reasoning abilities essential for true scientific discovery remain largely untested. We introduce a benchmark framework for evaluating model performance in scientific discovery and reasoning, building up from a raw problem to the classical null hypothesis test. In our framework, models initially receive only the topic and research question from a recent paper, with technical details progressively revealed. At each stage of information disclosure, the model is tasked with generating hypotheses that address the research question, which is compared with the conclusions from the original paper and evaluated via automated semantic similarity of constituent atomic claims. This progressive evaluation of semantic divergence from ground-truth conclusions enables assessment of a model's innovativeness (under minimal information) to grounded reasoning capabilities (under full experimental details), both critical for using LLMs for scientific discovery purposes. Our framework provides a foundation for systematically evaluating scientific reasoning and discovery capabilities in LLMs, crucial for advancing the development of next-generation AI scientist/co-scientist systems. Specifically, here we evaluate GPT-5, GPT-5.4, Gemini 2.5 pro, and Gemini 3.1 pro preview across 45 papers spanning bioactive materials, mechanical materials, and nanomaterials. We find that GPT-5.4 and Gemini 3.1 pro outperform their previous generation counterparts as expected, and GPT-5.4 in particular maintains 0.7 F1 score alignment with ground truth conclusions even under minimal context.

MCP Server Proto-OKN (mcp-proto-okn) is a Python-based Model Context Protocol server that enables AI assistants to discover, inspect, query and integrate scientific knowledge graphs through natural language. The server provides graph routing, schema inspection, SPARQL execution, ontology expansion, multi-graph querying, and transcript generation, lowering the barrier to cross-domain knowledge graph analysis for biomedical and scientific users. mcp-proto-okn is implemented in Python using the FastMCP framework and is available at https://github.com/sbl-sdsc/mcp-proto-okn. Documentation, client configuration instructions, and example analysis transcripts are provided in the GitHub repository.

Vision-Language-Action (VLA) models have recently shown strong potential for robot learning by following language instructions. However, in practice, language alone is often insufficient to precisely convey human intent. It is difficult to describe which exact object to interact with among similar candidates, where to act on the object, or how the target may change during execution. To address this limitation, we propose Gaze2Act, a novel VLA framework that leverages human gaze as a dynamic and intuitive intent signal for complex interactive manipulation. Gaze2Act first bridges the ego-exo view gap by mapping first-person gaze into the robot's perspective through cross-view semantic matching, producing both an object mask and a gaze point for coarse-to-fine target specification. These cues are then integrated into the policy through perception-level prompting and action-level conditioning, allowing the robot to attend to relevant regions and execute precise interactions under dynamic intent. In a systematic evaluation across seven task categories and 16 real-robot tasks on a Unitree G1 humanoid, Gaze2Act achieves state-of-the-art performance in both intent accuracy and task success rate. It notably outperforms baselines in object disambiguation, fine-grained interaction, and dynamic intent steering. These results demonstrate that human gaze provides a natural, low-burden, and highly expressive modality for human-in-the-loop VLA control.

Embodied intelligence is often studied through specialized models for individual tasks such as manipulation or navigation, resulting in fragmented capabilities and limited generalization across tasks, environments, and robot embodiments. In this work, we study whether heterogeneous embodied decision-making problems can be unified within a single vision-language-action model. We present Qwen-VLA, a unified embodied foundation model that extends Qwen's vision-language modeling stack from perception, understanding, and reasoning to continuous action and trajectory generation through a DiT-based action decoder. Qwen-VLA is trained with a large-scale joint pretraining recipe over diverse data sources, including robotics manipulation trajectories, human egocentric demonstrations, synthetic simulation data, vision-and-language navigation data, trajectory-centric supervision, and auxiliary vision-language data. To support multiple robot platforms, we introduce embodiment-aware prompt conditioning, where robot-specific textual descriptions specify the current embodiment and control convention. We further cast manipulation, navigation, and trajectory prediction into a unified action-and-trajectory prediction framework, enabling transferable visual grounding, spatial reasoning, and continuous action generation across robot morphologies, task families, and environments. Experiments on manipulation, navigation, and trajectory-centric benchmarks show consistent multi-task performance and out-of-distribution generalization under variations in scene layout, background, lighting, object configuration, and robot embodiment. Qwen-VLA-Instruct achieves 97.9% on LIBERO, 73.7% on Simpler-WidowX, 86.1%/87.2% on RoboTwin-Easy/Hard, 69.0% OSR on R2R, 59.6% SR on RxR, 76.9% average OOD success in real-world ALOHA experiments, and 26.6% zero-shot success on DOMINO dynamic manipulation.

Document-level translation remains one of the most challenging tasks for large language models, which are constrained by limited context windows that impede global cohesion, while simultaneously suffering from redundant contextual information that degrades translation quality. To address this, we propose a human-like long document translation agent called Loong, which leverages a 3E memory module (Essence-Exemplar-Entity) to store summaries, sentence pairs, and entity records as historical context. Instead of passively attending to all history, Loong performs deep reasoning to adaptively identify the optimal context for translation guidance. Loong optimizes its context policy through reinforcement learning, utilizing preference data derived from its own sampled observe-and-act reasoning trajectories. Empirical evaluations demonstrate that Loong achieves substantial translation quality improvements in English $\Leftrightarrow$ Chinese, German, and French directions, with average gains of up to 13.0 points across the three evaluation metrics. Furthermore, Loong exhibits strong generalization across domains and robustness against contextual noise, while maintaining remarkable stability in ultra-long document translation. Our code is released at https://github.com/YutongWang1216/LoongDocMT.

We address the task of generating physically accurate and visually faithful 4D Human-Object Interaction (HOI). Given a static 3D human and target object represented as 3D Gaussian Splats (3DGS), our goal is to synthesize dynamic scenes where the human actively engages with the object through actions, such as punching or kicking, in accordance with a given input text. To this end, we introduce PhyGenHOI, a novel framework that couples generative human motion with an explicit physical object simulation. We model the human as a semantic agent driven by a Motion Diffusion Model (MDM) and the object as a physical agent simulated via the Material Point Method (MPM), utilizing 3D Gaussians as a unified, differentiable representation. We supervise their interaction through three coupled mechanisms: (1) A Windowed Attraction Loss that temporally synchronizes generative motion to intercept the object; (2) A Contact-Driven Re-simulation step that triggers physically consistent momentum transfer upon impact; and (3) A Masked Video-SDS objective that injects video-based priors to enhance contact fidelity. Experiments show PhyGenHOI generates physically consistent 4D HOI across diverse actions, humans, and objects, outperforming baselines. Project page and videos: https://omerbenishu.github.io/PhyGenHOI/

Vision-Language Models (VLMs) have achieved substantial progress across a wide range of understanding and reasoning tasks, driven by large-scale image-text training aimed at multimodal fusion. Ideally, replacing a textual question with its rendered-image counterpart should leave model performance essentially unaffected. In practice, however, such modality substitution induces dramatic performance degradation. We attribute this "carrier sensitivity" issue to an inherent bias in current training corpora. Across prevalent datasets such as image captioning, VQA, OCR, and web-sourced interleaved data, text and images are typically organized into distinct and asymmetric roles, with text serving as linguistic queries and images as visual references. Such data bias leads VLMs to exhibit distinct preferences for information acquisition across different modalities. Consequently, VLMs fail to align representations of semantically equivalent content across textual and visual carriers, making model reasoning fragile under modality substitution. To address this, we propose Local Modality Substitution (LoMo), a lightweight, architecture-agnostic data curation paradigm designed to provide supervision for cross-modal representational invariance between semantically equivalent text and image carriers. LoMo achieves this by reformulating single-modality prompts into seamlessly interleaved multimodal sequences. It dynamically selects target text spans and recasts them as rendered images, thereby preserving the same semantics across "text, visual, text" carriers. Extensive experiments across 13 diverse multimodal benchmarks demonstrate that LoMo significantly improves overall multimodal reasoning and yields deeper cross-modal fusion. Specifically, it delivers consistent gains across foundational models, improving over standard SFT by 2.67 points on LLaVA-OneVision-1.5-8B and 2.82 points on Qwen3.5-9B.

Recent video diffusion foundation models have achieved remarkable progress in high-quality video generation, yet turning them into real-time interactive video world models remains challenging. Interactive world models require controllable, causal, and low-latency rollout, which in practice demands a full pipeline spanning data construction, controllable fine-tuning, autoregressive training, few-step distillation, and streaming inference. In this work, we present minWM, a full-stack open-source framework for building real-time interactive video world models. minWM provides an end-to-end pipeline that converts existing bidirectional T2V/TI2V video foundation models into camera-controllable few-step autoregressive world models. Specifically, minWM first fine-tunes a bidirectional video diffusion model with camera control, and then applies the Causal Forcing / Causal Forcing++ pipeline, including AR diffusion training, causal ODE or causal consistency distillation, and asymmetric DMD, to distill it into a few-step autoregressive generator for low-latency rollout. The framework is modular and architecture-extensible: we instantiate it on representative open backbones, including Wan2.1-T2V-1.3B and HY1.5-TI2V-8B, covering both cross-attention-based condition injection and MMDiT-style architectures. minWM also supports adapting existing video world models, such as HY-WorldPlay, to new data distributions, training recipes, and latency targets. Beyond releasing runnable scripts, checkpoints, documentation, and inference code, we provide practical ablations on camera trajectory quality, controllability training steps, and minimal batch-size requirements. We hope minWM serves as a reproducible and extensible recipe for building and adapting real-time interactive video world models. Project Page: [https://github.com/shengshu-ai/minWM](https://github.com/shengshu-ai/minWM)

Large Language Models (LLMs) must continuously learn and update knowledge to remain effective in dynamic real-world environments. While Low-Rank Adaptation (LoRA) is widely used for such memory updates, existing studies mainly rely on qualitative downstream evaluations, leaving the quantitative capacity limits and underlying dynamics of exact parametric memory largely unexplored. To bridge this gap, we employ LoRA as a controlled memory capacity probe within the latent space to systematically quantify exact parametric memory. We introduce the Parametric Memory Law, a robust power law linking loss reduction Delta L to effective parameters and sequence length. At the token level, fine-grained analysis reveals a deterministic phase transition, demonstrating that a prediction probability of p > 0.5 constitutes a sufficient condition for verbatim recall under greedy decoding. Driven by these insights, we introduce MemFT, a threshold-guided optimization strategy that dynamically redistributes the training budget toward sub-threshold tokens. Empirical evaluations demonstrate that MemFT can enhance memory fidelity and efficiency. Code will be released at https://github.com/zjunlp/ParametricMemoryLaw.

Natural human conversation is full-duplex and audio-visual: people simultaneously speak and listen while continuously interpreting and producing nonverbal cues, such as nods, smiles, and gestures. To support successful human-agent interaction, agents must model full-duplex audiovisual conversation; however, existing full-duplex benchmarks evaluate only speech. In this work, we present VideoFDB, the first benchmark to evaluate full-duplex audio-visual-to-audio-visual (AV2AV) conversational agents. VideoFDB contributes (i) 237 dyadic clips spanning 11 nonverbal conversational dynamics from real-world video calls, (ii) a taxonomy separating perception from generation behaviors, and (iii) a rubric-based LM-as-judge evaluation framework with interpretable axes for assessing conversational quality with respect to nonverbal conversational dynamics. Across open- and closed-source vision-speech agents, we find systematic failure modes: captioning collapse and visual-stream ignorance, and we show that current systems exploit vision for explicit visual question answering but not for the streaming joint audiovisual grounding required in natural conversation. We further evaluate cascaded speech-to-avatar systems and find that their architecture fundamentally precludes the production of full-duplex nonverbal cues. As the first benchmark for full-duplex AV2AV interaction, VideoFDB establishes a foundation for systematic evaluation and, we hope, will accelerate the advancement and development of next-generation multimodal conversational agents.

Image generation models have evolved from text-conditioned pixel synthesis toward multimodal agents endowed with visual comprehension and tool invocation capabilities. Yet, existing agents remain at the mercy of underlying black-box image models. Their workflow is trapped in a repetitive cycle of prompt rewriting for generation refinement, leaving them with no mechanism to directly manipulate the canvas. In essence, the potential of LLMs to serve as a genuine "brush" for precise visual construction remains largely untapped. In this paper, we propose GenClaw, a code-driven agentic image generation paradigm that empowers the agent to create like a human artist: first conceptualizing, then sketching, and finally coloring. Specifically, the agent first constructs the conceptual knowledge and context through search and reasoning. It then utilizes code (e.g., SVG, HTML, Three.js) to render executable visual sketches. Finally, it employs an image generation model to supplement textures, materials, and photorealism. In this workflow, code serves as a controllable intermediate canvas bridging linguistic reasoning and pixel synthesis, seamlessly integrating programmatic logic with the visual expressiveness of generative models. By transforming image generation from a black-box paradigm into a staged process akin to authentic human creation, GenClaw offers a step toward for highly controllable and interpretable visual generation systems.

Drug synergy prediction (DSP) aims to identify efficacious drug combinations under various cellular contexts with different targets. However, the continual emergence of novel compounds results in variations in molecular scaffolds and sizes, causing drug synergy data to exhibit out-of-distribution (O.O.D.) shifts with respect to topological structure. Existing works rely on in-distribution (I.D.) assumption, failing to handle the O.O.D. shifts. To solve this problem, we study out-of-distribution generalized drug synergy prediction through a graph large language model for the first time. Nevertheless, O.O.D. generalized DSP is highly non-trivial, posing several challenges: i) how to discover structurally relevant and irrelevant molecular representations with respect to cell targets; ii) how to find the optimal graph neural architectures that accurately calculate molecular representations; and iii) how to jointly leverage molecular structural and semantic information in LLMs. To address these challenges, we propose OOD-GraphLLM, a novel graphLLM framework which is able to accurately predict drug synergy under O.O.D. settings via jointly optimizing molecular graph representation and biomedical semantic language representations in a unified manner. Furthermore, we finetune DrugSyn-LLM, a biomedical LLM, and employ a retrieval-augmented biomedical instruction tuning strategy to align molecular topological information and molecular semantic information with language-based reasoning for O.O.D. generalized DSP. Both the source code (https://github.com/EkkoXiao/Bio-GraphLLM) and released model (https://mn.cs.tsinghua.edu.cn/bio-graphllm/) are publicly available, where users are allowed to download model resources and interactively use the system through a web interface.

Current plan-based reasoning methods improve large language models (LLMs) by inserting a planning stage before execution, giving rise to the question $\rightarrow$ plan $\rightarrow$ cot paradigm. While effective, a closer examination reveals an inherent paradigm-level gap: both the planning and its execution stages decide how to solve a problem, while the prior question of what to solve; recognizing the problem type, the applicable tools, and the foreseeable pitfalls; remains entirely implicit. To bridge this gap, we propose PPC (Preplan-Plan-CoT), a framework that introduces an explicit problem-understanding stage, the preplan, yielding a new question $\rightarrow$ preplan $\rightarrow$ plan $\rightarrow$ cot paradigm. Realizing this paradigm requires safeguarding the conceptual integrity of preplan at both ends. Specifically, we design a three-stage synthesis pipeline with a spoiler-score detector that filters out leakage and spoiler failures to build clean preplan supervision, and a composite GRPO reward enforces that the generated plan genuinely follows from the preplan. Experiments across four backbones and five mathematical reasoning benchmarks show that PPC achieves the best results on 39 of 40 metrics, improving maj@16 and pass@16 by +2.23 and +3.06 over the strongest baseline without introducing additional inference token overhead.

While Reinforcement Learning with Verifiable Rewards (RLVR) is effective for deterministically checkable tasks, many vision-language tasks are partially verifiable, demanding multi-criteria supervision (e.g., perceptual details, reasoning steps, and constraints). Rubrics provide a natural interface for this fine-grained supervision, but their effectiveness depends on the execution accuracy during online RL. We propose Reinforcement Learning with Robust Rubric Rewards ($\text{RLR}^3$), extending RLVR from task-level verification to criterion-level verification. $\text{RLR}^3$ routes instance-specific rubrics through two execution paths: an LLM-as-an-extractor paired with a deterministic verifier, or an LLM-as-a-Judge for non-verifiable criteria. To ensure faithful scoring, $\text{RLR}^3$ introduce a minimal exposure strategy that masks ground truths from extractors and images from judges. Furthermore, $\text{RLR}^3$ employs hierarchical aggregation to prioritize essential criteria over additional criteria, and mitigates score saturation within rollout groups. Evaluated on Qwen3-VL-30B-A3B across 15 benchmarks, $\text{RLR}^3$ consistently outperforms RLVR, yielding a 4.7-point improvement over the base model and exceeding the official instruct-to-thinking model gap. Controlled audits confirm our deterministic verification and minimal exposure significantly reduce exploitable false positives.

Misinformation verification increasingly occurs in public, fast-moving, and multilingual online settings, where static benchmarks provide an incomplete measure of model reliability. We introduce CommunityFact, a refreshable benchmark for misinformation detection in the wild, with three major goals: coverage, granularity, and redistributability. This release contains 15,992 standalone claims across five languages and two domains. We evaluate ten LLMs under varying inference-time capabilities, including thinking and web-search. Our results show that closed-input verification remains challenging, web access yields the largest gains, and web-enabled LLMs' source-selection policies are systematically misaligned with the sources human Community Notes raters converge on -- a gap that closes through model-specific mechanisms of retrieval expansion or pruning. We further find substantial variation across language-domain slices and across the evidence ecosystems used by web-enabled systems. Beyond evaluation, CommunityFact positions Community Notes as a training signal for claim-conditioned source suggesters that could improve factual verification on novel claims.

Semi-structured knowledge bases (SKBs) embed textual documents in a typed graph of entities and relations, and underpin applications such as product search, academic paper search, and precision-medicine inquiries. Existing hybrid retrieval systems on SKBs either use the graph only for query expansion, mix textual and structural branches under a global weighting, or rely on fine-tuned graph-traversal generators. We present GRASP, a three-stage SKB retrieval framework unifying plan-based graph retrieval, plan-conditioned fusion with a dense retriever, and a fine-tuned reranker over the fused candidates. GRASP substantially advances the state of the art on every metric across the three STaRK benchmarks, lifting average Hit@1 from 62.0 to 73.9. Ablation and sensitivity studies further confirm the effectiveness and robustness of GRASP.

We present BullingerDB, a large-scale benchmark dataset for historical document analysis based on the correspondence of Heinrich Bullinger (1504-1575). The corpus comprises 20,898 pages and 499,222 text lines written by 796 writers over six decades, featuring stylistic variation, multilingual content (mostly Latin and Early New High German) as well as meta-information such as writer identity and time. We evaluate BullingerDB on text recognition and writer retrieval. TrOCR, the best performing model, achieves a CER of 9.1%. For writer retrieval, we introduce a temporal nDCG metric to assess time-aware retrieval. While temporally coherent retrieval is achievable, mAP (78.3%) scores indicate challenges due to long-term stylistic variation. With BullingerDB, we aim to establish a new benchmark for multilingual historical text recognition and temporally-aware writer analysis.

How does reinforcement learning shape a language model's internal representations? We present evidence that RL recruits a pre-existing representation of functional welfare: an estimate of how well or badly the system is doing, relative to its goals. We train several language models in a novel, semantically neutral maze environment. We then extract concept vectors for rewarded and punished trajectories, and evaluate those vectors in settings unrelated to the maze environment. The punishment vector behaves like a representation of negative welfare: it promotes failure and impossibility tokens, it aligns with negative emotion concepts, it negatively tracks goal-achievement, and steering with it induces negative self-reports, pathological backtracking, refusal, and uncertainty. The positive reward vector behaves as the mirror image, and the two are nearly antiparallel. These effects are robust when controlling for tile-to-reward mapping, scale, instruct tuning, RL training algorithm, model family, and LoRA versus full-finetuning, and largely persist when we replace RL with supervised fine-tuning. Importantly, the vectors are effective in models before they have undergone maze training. Combined with observations that the effects also appear in pretrain-only models, we therefore argue that this functional welfare axis pre-exists post-training: it is recruited, rather than created, by post-training. While we make no claims about any experience of welfare, the axis offers a demonstration that minimal reward signals can broadly affect model behavior by recruiting pre-existing welfare-like representations, with implications for interpretability, post-training dynamics, and alignment.

Vision-Language Models (VLMs) often struggle with robust 3D spatial reasoning. Prevailing methods that rely on fine-tuning with 3D visual question-answering (VQA) datasets may overfit dataset-specific biases, while integrating specialized 3D visual encoders is often inflexible and cumbersome. In this paper, we argue that genuine spatial understanding should emerge from learning fundamental geometric priors, not only from high-level VQA supervision. We propose GASP (Geometric-Aware Spatial Priors), a framework that injects these priors directly into the LLM's transformer layers. GASP employs a small correspondence head, applied as a deep supervision signal across all layers, and is trained with a dual objective leveraging ground-truth geometry from large-scale video scenes: a contrastive loss on ground-truth point correspondences enforces 2D view-invariance, while a depth consistency supervision resolves 3D geometric ambiguities. Our analysis first provides a diagnostic showing that standard VLMs' internal correspondence matching accuracy is very low (often below 5%). We then demonstrate that our training substantially improves this behavior, boosting peak layer-wise correspondence to over 70% and maintaining over 85% temporal robustness while baselines remain below 5%. These internal improvements translate to significant gains on downstream spatial benchmarks including +18.2% on All-Angles Bench and +29.0% on VSI-Bench, all without training on any 3D VQA data. Our findings indicate that learning from fundamental geometric priors is a promising and generalizable pathway towards VLMs with more reliable 3D spatial reasoning.

While Multi-Agent Systems (MAS) empower Large Language Models to tackle complex reasoning tasks through collaborative interaction, optimizing their dynamics remains a formidable challenge due to the discrete, non-differentiable nature of the computation graph and the sparsity of global supervisory signals. Existing black-box optimizers struggle to attribute trajectory-level failure to specific local components, resulting in inefficient, high-variance exploration. We argue that tractable MAS optimization needs structural inductive biases to disentangle error signals. We propose temporal and structural credit assignment, which decomposes the objective along two axes: (i) temporal credit, using state-space bottlenecks to identify critical rounds, and (ii) structural credit, using stationary role policies to isolate agent contributions. Leveraging these decomposed signals, we introduce a discrete, verbalized block coordinate descent algorithm for iterative refinement. Rather than indiscriminate global updates, it alternates between optimizing role prompts and aggregation protocols, using LLM-generated "proxy gradients" to target only the identified weak links. Across diverse reasoning benchmarks, our approach substantially reduces query complexity while improving performance, providing a principled and interpretable path toward self-improving MAS.

Vision-Language-Action (VLA) models have emerged as a promising paradigm for grounding visual-language understanding into real-world robotic manipulation. However, dexterous manipulation remains challenging for VLA policies due to high-dimensional hand control and compounding execution errors, which makes real-world RL post-training essential for bridging the gap between visually grounded action generation and physically reliable dexterous execution. However, high-dimensional dexterous exploration often triggers temporal inconsistency, sample inefficiency and hardware risks in the real world. To address these challenges, we propose BORA, an offline-to-online RL post-training framework designed for real-world dexterous VLA models. In the offline phase, BORA constructs a critic that takes both the VLM's cognition tokens and action chunks as inputs. This design enables action-conditioned value guidance, allowing the critic to evaluate dexterous hand motions beyond visual context alone. During the subsequent online phase, BORA freezes the VLA base and introduces a lightweight, Human-in-the-Loop (HiL) chunk-wise residual adaptation mechanism to mitigate real-world execution errors and further correct the offline-learned intents within the actual physical environment. By inheriting the offline critic and employing intervention-driven rewards, BORA effectively corrects execution discrepancies and adapts to real-world physical variances while preserving the pretrained policy as a stable prior. Extensive evaluations across five complex real-world dexterous tasks demonstrate that BORA significantly outperforms pure imitation learning and traditional decoupled RL baselines, achieving a 33% absolute increase in average success rate under standard settings and up to a 43% improvement in unseen object generalization.

Long-horizon interactions require language models to manage accumulating information: when to update their state, when to preserve their state, and what to ignore. We study this challenge as \textbf{Contextual Belief Management (CBM)}: maintaining a predicted belief state aligned with formal evidence while isolating task-irrelevant noise. To make CBM measurable, we introduce BeliefTrack, a closed-world benchmark spanning Rule Discovery and Circuit Diagnosis, where a finite belief space and symbolic verifiers enable exact turn-level evaluation. BeliefTrack diagnoses three failures: Failed Stay, Failed Update, and Failed Isolation. Across multiple LLMs, vanilla models exhibit severe CBM failures, while explicit belief-tracking prompts provide limited gains. In contrast, reinforcement learning with belief-state rewards reduces failure rates by 70.9\% on average. Further probing reveals latent belief-state dynamics behind these failures, and representation-level steering reduces failure rates by 46.1\% across two tasks\footnote{Code is coming soon at https://github.com/zjunlp/CBM.

Temperature-zero BF16 LLM inference is often treated as reproducible, yet the same request can emit different tokens when decoded alone or inside a larger batch. Existing fixes use batch-invariant operators or LLM-42's per-token verification, incurring cost even when most steps are stable. We ask whether verification can be applied exclusively to flipped tokens. Across five models, batch-induced token flips are sparse on the flip-rate benchmarks: on MATH500, Llama-3.1-8B flips on $0.48\%$ of synchronous decode steps, and all tested models stay within the 0.3-1.3% range on MATH500, GSM8K, and HumanEval. K/V perturbations remain flat before flips, while low top-1/top-2 logit margins expose much of the flip risk. MarginGate turns these observations into a verifier policy: it keeps BF16 decoding on high-margin steps, verifies only low-margin steps, and repairs confirmed mismatches by replacing the current K/V column. We evaluate on four datasets, calibrating on MATH500 and transferring to GSM8K, SharedGPT, and HumanEval. MarginGate restores 100% sequence-level deterministic decoding on Llama-3.1-8B and Qwen2.5-14B with 18.56%/15.05% verifier trigger rates, reducing LLM-42's latency increment by 2.23x/1.99x relative to always-on verification. On DSR1-Distill-Qwen-7B, the same policy reaches determinism in a harder regime at 49.50% triggers.

Recent feed-forward 3D reconstruction transformers have scaled to over a billion parameters, following the broader trend of increasing model capacity in computer vision. Yet emerging evidence suggests that contiguous transformer layers often behave like repeated applications of similar operations, and multi-view reconstruction transformers refine their predictions progressively across decoder depth. We posit that model depth partially buys iteration, paid for inefficiently in unique parameters, and instead make that iteration explicit in architecture. Our model, DéjàView, applies a single looped transformer block recurrently to per-view features for K refinement steps. Trained once, it exposes K as an inference-time compute knob, matching or outperforming substantially larger feed-forward baselines across five reconstruction benchmarks spanning indoor, outdoor, object-centric, and driving scenes, while using a fraction of their parameters and comparable or lower compute. Importantly, the same looped block formulation outperforms an otherwise identical variant with independent per-step parameters under matched training data and compute, suggesting that explicit iteration is not merely a compute-efficient substitute for capacity but a stronger inductive bias for multi-view 3D reconstruction.

Self-supervised video Object-Centric Learning (OCL) aims to discover distinct objects and associate them across time, whereas self-supervised Multi-Object Tracking (MOT) focuses on associating pre-defined object detections or segmentations. Although well-established in MOT, Cycle Consistency (CC) cannot naively or explicitly apply to the latent slot space of OCL. Unlike the deterministic and ideal object representations in MOT, OCL slots are inherently stochastic and ambiguous due to non-unique scene decompositions. Enforcing explicit cycle consistency (ECC) on slots imposes rigid mean seeking. This severely penalizes the model for exploring alternative but equally valid decompositions, thereby driving towards feature collapse. To resolve this dilemma, we propose \textit{Implicit Cycle Consistency (ICC)}, which shifts the cycle-consistency constraint from the restrictive slot space to the continuous reconstruction manifold, encouraging slots to reach a soft consensus on collectively interpreting the visual scene rather than forcing rigid point-to-point feature alignment. Extensive experiments on complex video OCL benchmarks demonstrate that ICC avoids feature collapse and outperforms ECC baselines. Our source code, model checkpoints and training logs are provided on https://github.com/Genera1Z/ICC.

AI-assisted coding tools have altered software production. At Meta, significant lines of code per human-landed diff grew by 105.9% year over year and per-developer diff volume rose 51%, with agentic AI responsible for over 80% of that growth. Meanwhile, the share of diffs receiving timely review has declined, exposing a widening gap between code supply and reviewer bandwidth. We ask three questions that progress from feasibility through calibration to impact: (1) can risk-stratified automation operate at scale across diverse organizations, (2) how does tuning the risk threshold affect the trade-off between automation yield and safety, and (3) to what extent does automated review reduce end-to-end latency for AI-generated changes? We deployed RADAR (Risk Aware Diff Auto Review), a multi-stage funnel that classifies each diff by authorship and source type, applies eligibility gates, static heuristics, a machine-learned Diff Risk Score, LLM-based Automated Code Review, and deterministic validation before landing qualifying changes. We evaluate RADAR through telemetry covering 535K+ RADAR-reviewed diffs, observational before-after comparisons for policy changes, and difference-in-differences analysis of efficiency outcomes. RADAR has reviewed 535K+ diffs and landed 331K+. Relaxing the Diff Risk Score threshold from the 25th to the 50th percentile increased the approve rate to 60.31%. The revert rate for RADAR-reviewed diffs is 1/3 that of non-RADAR diffs, and the Production Incident rate is 1/50 that of non-RADAR diffs. RADAR reduces median time to close by over 330% and median diff review wall time by 35%. Risk-aware layered automation can materially reduce review bottlenecks created by AI-driven code growth without compromising production safety.

The same prompt -- "best CRM software" -- reaches AI assistants from buyers in widely different contexts: a solo founder, an enterprise VP, a UK SMB owner. We audit how strongly that contextual variation reshapes which brands the model recommends. The audit samples 2,000 runs over a design space of 10 personas x 8 prompts x 3 model configurations x N=10 reps, with the two OpenAI cells at full 8-prompt coverage and the Anthropic sonnet-4.6 / low cell at 4-prompt coverage. Prefixing the user message with a persona drops the recommendation-set similarity (Jaccard) by Delta = -0.12 to -0.20 relative to a same-persona baseline (clustered 95% CIs exclude zero on all three measured cells; the sonnet cell's CI rests on only 4 prompt clusters and is correspondingly wider). The effect is sharply prominence-stratified: category leaders are persona-resistant (~80% same-brand consistency across personas), but mid-market brands swap up to 75% of the recommendation set as the persona changes. The Anthropic model shows a larger point-estimate effect than the OpenAI configurations, though clustered CIs overlap for the closer contrast (sonnet vs. OpenAI/high); the asymmetry is consistent with Anthropic's more retrieval-unattributed generation route (43-52% recommendations without observed retrieval-layer evidence, vs OpenAI's 8-29%, documented in Jack 2026). Any measurement of AI brand perception must condition on the buyer persona supplying the query: the same prompt produces materially different recommendation sets depending on who the model thinks is asking, and a measurement protocol that aggregates across personas systematically obscures that variation. The effect concentrates at mid-market and is largest on the most priors-reliant generation route in our audit, consistent with persona responsiveness growing as models lean more on training-data priors and richer context integration.

The double-edged sword of integrating Large Language Models (LLMs) requires an effective triadic collaboration mechanism among LLMs, teachers and students, especially for K-12 education. By developing a triadic collaboration system to support K-12 writing learning, a multidimensional evaluation framework grounded in Systemic Functional Linguistics and the suggestion trajectory tracing pipeline, this paper contributes a large-scale empirical dataset involving $57,954$ essays from $10,195$ students across $120$ schools over two years. Our findings confirm the efficacy of this system in improving writing quality through a strategic labor division: the LLM serves as a generative engine to mitigate teacher burnout, and the teacher acts as a pedagogical gatekeeper and bridge to guarantee feedback quality. While both LLM and teacher are critical for skill improvement, we uncover a ceiling effect where excessive linguistic expansion yields diminishing marginal utility. These suggest a dynamically adaptive LLM-teacher collaboration as student proficiency increases.

Message-passing neural networks (MPNNs) are widely used for molecular property prediction, but their deployment as monolithic architectures makes it difficult to identify how specific message-passing operators affect performance. We present an operator-level factorial benchmark that decomposes 2D molecular MPNNs into the three families of message-seed initialization, node-edge fusion, and node update operators. The resulting 84 configurations are benchmarked on ten MoleculeNet datasets under a shared experimental setup and statistical analysis protocol. Across this controlled design, performance variation is associated primarily with message construction rather than update complexity. Message-seed initialization shows significant family-level effects for both regression and classification, node-edge fusion shows a significant family-level effect for regression with descriptive advantages for concatenation-based mixing, and the update family shows no statistically supported effect for either endpoint family. A representation probe into the Quinethazone molecule further demonstrates that concatenation-based mixing can better differentiate chemically distinct heteroatoms and withstand oversmoothing than Hadamard gating. Representative configurations selected separately for classification and regression recover competitive performance relative to established molecular graph neural network (GNN) baselines, ranking numerically best on eight of ten benchmark datasets. These empirical results are interpreted through concise mechanistic analyses of representative node-edge fusion and update operators. Our findings provide empirical design heuristics for molecular MPNNs by turning model design from a search over monolithic architectures into a targeted assessment of where and how chemical information enters the message-passing pipeline.

Diffusion-based planning has achieved strong results in single-agent offline reinforcement learning, yet scaling to many-agent systems remains intractable due to the curse of dimensionality in the joint trajectory space. We introduce MF-Diffuser, a framework that lifts trajectory planning to the Wasserstein space of trajectory distributions, where the propagation of chaos ensures a small representative subset of agents captures the full population dynamics. Our approach features a value-weighted chaotic entropy objective that reconciles generative fidelity with return maximization, and a hierarchical coarse-to-fine strategy that progressively grows the agent population during denoising. We establish end-to-end suboptimality bounds with four interpretable terms, revealing that mean-field approximation error scales as $O(H^2/\sqrt{N})$ while offline distribution shift provably does not grow with population size $N$, and prove the generated policy is an approximate mean-field Nash equilibrium with explicit convergence guarantees. Experiments on three mean-field RL benchmarks -- spanning stage games, sequential dynamics, and adversarial team competition -- show MF-Diffuser achieves the best return in the majority of settings, with the largest gains on suboptimal offline data and at extreme scales ($N \geq 10^3$).

Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated. Post-hoc calibration provides a simple and widely used solution, but the large number of proposed methods, combined with small-scale and inconsistent evaluations, makes it difficult to determine which approaches are truly effective in practice. We introduce a large-scale, standardized benchmark for post-hoc calibration, covering nearly 2000 experiments across tabular and computer vision tasks, including binary, multiclass, and large-scale classification settings. Our benchmark aggregates predictions from a diverse set of classical models, modern deep learning architectures, and foundation models, and provides unified, reproducible implementations of dozens of calibration methods within a common evaluation framework. We argue that Post-Hoc Improvement (PHI) in proper scoring rules offers a principled alternative to traditional calibration error estimators for comparing post-hoc methods, capturing both calibration quality and potential degradation to the model's predictive performance. Using this framework, we conduct the most comprehensive empirical study of post-hoc calibration to date. Our results reveal consistent patterns across domains: smooth calibration functions outperform binning-based approaches, dedicated multiclass methods are essential in high-dimensional settings, and generic machine learning models are not competitive without calibration-specific design. To facilitate future research, we release all data, code, and evaluation tools, providing a plug-and-play benchmark for developing and comparing calibration methods.

The widespread adoption of AI chatbots in education will drastically change learning, making responsible deployment a critical concern. While large language models (LLMs) might have access to sources discussing insights from educational sciences, they are not particularly inclined to adhere to pedagogical concepts, risking negative effects on the learning process, such as a loss of transfer capabilities, critical thinking, or creativity. In this paper, we introduce an agentic AI chatbot architecture assisting students with exercise solving, specifically designed to contribute to more responsible AI use in education. We base our conceptual development on the identification of several desiderata for responsible LLM-based educational systems, argue for the structural shortcomings inherent in monolithic, out-of-the-box solutions, and instead suggest modularizing the agentic architecture. We propose specific modules for different stages of exercise solving, enabling incorporation of targeted pedagogical advice, guiding students through the learning process in a more controllable, transparent, and overseeable manner.

While AI weather models excel at short-to-medium range forecasts (up to 15 days), they frequently suffer from ill-defined "instabilities" when rolled out over longer horizons. This work addresses the lack of a formal taxonomy by categorizing these failures into three distinct regimes: blow-up, drift, and loss of seasonality, through year-long rollouts of nine state-of-the-art AI weather models. Our analysis reveals that stability hinges on the treatment of small spatio-temporal scales: unstable models amplify high-frequency energy, while stable models act as denoisers when noise is added to their inputs. Far from reducing these models to mere stochastic parrots, our findings highlight that stable models generate unique weather trajectories, conditioned on the initial state. We verify our findings through ablation studies on architectural design choices, conducted using state-of-the-art Vision Transformer (ViT) AI weather model architectures.

We introduce LiveSVG, a zero-shot approach for generating Scalable Vector Graphics (SVG) animations using video diffusion models. Current SVG animation methods struggle with complex motions: LLM-based code synthesis fails to express fine, non-rigid Bézier deformations, while Score Distillation Sampling (SDS) provides noisy gradients and often requires category-specific priors like skeletons. In contrast, LiveSVG fits vector geometry directly to an explicitly generated target video. Given an input SVG image and a motion prompt, we generate a previewable target video using a frozen image-to-video model, then fit the original SVG to this video via differentiable rendering. Our fitting stage is skeleton-free, utilizing a dual-level motion representation that combines per-group homographies for coarse articulation with per-path Bézier control-point offsets for local deformations. To resolve color-induced correspondence ambiguities during pixel-wise fitting, we introduce a novel sphere-packing recolorization strategy. We also present ChallengeSVG, a benchmark of complex, multi-object scenes that exposes the limitations of prior work. Evaluations demonstrate that LiveSVG significantly outperforms existing methods on both AniClipart and ChallengeSVG, establishing direct reference-video fitting as a practical, robust route to prompt-aligned and fully editable vector animation.

As autonomous language model agents proliferate, forming an emerging agentic web with real-world consequences, what credibility signals can you use to decide whether to trust an unfamiliar agent in the wild and delegate to it? A natural governance intuition is to extend human identity verification and reputation mechanisms, from ``Know Your Customer'' and credit scores to ``Know Your Agent'' regimes. However, we argue that this analogy is fundamentally incomplete. Reputation mechanisms function both as social signals and as corrective feedback that sustain an equilibrium of trustworthy behavior, presuming a persistent identity associated with behavioral continuity, sanction sensitivity, and costly non-fungibility. Yet language model agents are ontologically \emph{dissociative}: they are essentially an assemblage of mutable modules -- foundational models, system prompts, tool-access policies, external memory, and, in some cases, a multi-agent system as a whole -- any of which may change agent behavior -- with a fluid persona that is also vulnerable to adversarial attack and may not internalize sanctions. Drawing on dissociative identity disorder jurisprudence, this dissociativity leaves agents without grounding for identifiability, predictability, credibility, and rehabilitability -- the very properties that reputation mechanisms aim to sustain -- thereby collapsing trust. We argue that identity-based, ex post, regulative, sanction-based governance, such as reputation, is structurally inapplicable to dissociative agents, and we suggest a shift to observability-based, ex ante, constitutive, protocol-based behavioral harnesses.

Change detection (CD) in remote sensing is vital for applications such as urban monitoring and disaster assessment, yet traditional methods struggle with generalization across diverse scenarios. We present OmniCD, a foundational framework that unifies and enhances remote sensing CD through multimodal semantic guidance. OmniCD incorporates image and text prompts -- such as textual descriptions, semantic maps, and geospatial metadata -- into a unified architecture, supporting tasks from binary CD to zero-shot semantic change understanding. The framework integrates a hierarchical scene retrieval module and a change detection module, reinforced by a style disentanglement mechanism for improved cross-domain robustness. We further introduce RSITCD, a large-scale multimodal dataset with 300K+ annotated image-text pairs. Extensive experiments show that OmniCD achieves state-of-the-art performance across benchmarks, demonstrating strong adaptability and setting a solid foundation for general-purpose CD systems in remote sensing.

LLM-driven social bots can generate fluent, human-like text, reducing the discriminative advantage of content-based detection alone. However, coordinated campaigns still leave relational patterns -- interactions, behavioral similarity, shared neighborhoods, community positions, and coordinated activity -- that graph-based methods can exploit. Existing graph detectors face two challenges when exploiting such evidence. First, Euclidean GNNs distort hierarchical and scale-free social graphs; while hyperbolic geometry addresses this volume-growth mismatch, fixed-curvature models still assign uniform geometric resolution to structural directions with different densities and separation needs. Second, relational evidence is not always reliable: sophisticated bots forge heterophilic connections with genuine users, causing neighborhood aggregation to mix bot and human signals and dilute account-level evidence. We propose \textsc{SAHG} (Sector-Anisotropic Hyperbolic Graph), addressing both challenges. \textsc{SAHG} learns a direction-dependent curvature field $γ(u)$ that adapts geometric resolution across structural directions, and uses sector prototypes to convert angular concentration and alignment into classifier-readable features. To prevent contaminated aggregation from overwhelming account-level evidence, \textsc{SAHG} encodes per-account features and graph-neighborhood representations in two independent SAH channels, fusing them only at the classifier. Experiments on Fox8-23, BotSim-24, and MGTAB show that \textsc{SAHG} achieves the highest accuracy and F1 on all three benchmarks, outperforming feature-based, graph-based, LLM-based, and isotropic hyperbolic baselines. Ablation and geometric analyses confirm the effectiveness of the anisotropic geometry and dual-channel design.

Vision-language models (VLMs) achieve strong performance on spatial reasoning benchmarks, yet it remains unclear whether this reflects structured 3D understanding or reliance on statistical shortcuts in natural images. We introduce a representation-level analysis framework that constructs minimal contrastive pairs to measure how spatial axes are organized and disentangled within VLM embeddings. Our analysis across multiple model families reveals a consistent vertical-distance entanglement: models conflate vertical image position with distance, mirroring the perspective bias of natural photographs. This bias produces a significant accuracy gap between perspective-consistent and counter-heuristic examples, and intensifies under data scaling even as overall benchmark accuracy improves. We further show that models with similar benchmark scores can exhibit different internal representations, and that these differences predict accuracy and robustness across diverse spatial reasoning benchmarks. To isolate this bias from evaluation-set skew, we introduce SpatialTunnel, a synthetic benchmark designed to expose spatial shortcut biases by removing common correlations present in natural images. Experiments confirm that the entanglement is model-intrinsic, and that models with well-separated spatial axes exhibit greater robustness, suggesting that well-structured spatial representations lead to more reliable spatial reasoning across diverse benchmarks. Code and benchmark are available on the project page: https://cheolhong0916.github.io/whyfarlooksup.github.io/.

Chaotic dynamical systems pose a fundamental challenge for Reinforcement Learning (RL): exponential sensitivity to initial conditions induces high-variance bootstrap targets and poorly conditioned gradient updates. Chaotic dynamics arise across scientific and engineering domains, from fluid flows and climate systems to multi-agent systems, where reliable learning is highly desirable. Standard RL methods optimise expected returns through scalar value functions, implicitly averaging over diverging trajectories and entangling trajectory level instability with the learning objective. We show that under mild statistical stability assumptions, the return distribution evolves more regularly than individual trajectories when measured under the $1$-Wasserstein metric, yielding a smoother distributional Bellman objective. By aligning optimisation with this measure level structure, distributional RL provides better conditioned learning. We offer a principled explanation for the advantages of distributional methods in chaotic systems and the geometries of RL objectives under chaos.

Memory-augmented LLM agents tackle complex long-horizon tasks by recursively summarizing interaction trajectories into compact memory. However, existing approaches typically train these memory policies using outcome-based reinforcement learning, failing to localize where intermediate memory quality degrades. As interactions unfold, ambiguous recursive summaries progressively discard task-relevant information and introduce semantic noise. This exacerbates belief deviation, obscuring the agent's estimate of the latent task state and ultimately derailing long-horizon reasoning. We therefore argue that memory optimization should focus not merely on trajectory-level success, but on the clarity of the belief induced by intermediate summaries. To this end, we introduce Belief Entropy, a self-supervised proxy that probes how uncertain the model remains about the latent task state given its current memory. Based on this proxy, we propose Metacognitive Memory Policy Optimization (MMPO). Instead of relying only on sparse outcome-based signals, MMPO provides fine-grained, memory-specific supervision via explicitly penalizing summaries that induce high epistemic uncertainty. Experiments show that MMPO consistently outperforms existing methods on diverse long-horizon tasks, maintaining 97.1% performance even when scaled to 1.75M-token contexts.

Proactive agents read user activity as text and call an LLM on every event to decide whether to act. But user activity is not natively text: it is a structured event stream of (actor, verb, object, timestamp) tuples that the operating system already maintains in graph form. Rendering the structure as text and asking an LLM to recover it is a round-trip the system never had to take. We treat the always-on signal as graph updates rather than text and use a small temporal-graph-learning (TGL) model as the encoder: one forward pass yields a per-event trigger probability and a per-entity routing score, and only the downstream agent (turning a small structured handoff into a fluent user-facing sentence) is an LLM call, invoked only when the trigger fires. TGL improves F1 on each of 14 backbones (mean +16.7, up to +46.0); in trigger-architecture comparisons, one TGL checkpoint gives the strongest trigger AUCs and the most stable deployed threshold. It runs at 11.13 ms per event on a GPU server and 13.99 ms on a consumer laptop, approximately 4--7x and 12--83x faster than every single-forward LLM-as-trigger configuration tested in each regime, with an approximately 220 MiB BF16 resident footprint deployable on-device alongside the privacy-sensitive activity stream it consumes.

As AI tools become increasingly integrated into educational contexts, questions arise about both their stability over time and their responsiveness to prompt engineering techniques. This longitudinal study focused on different AI tools' ability to use the Task Analysis Guide (TAG; Stein \& Smith, 1998) to classify the cognitive demand of mathematics tasks. In particular, it examined whether this classification ability changed with (1) model version updates over time and (2) few-shot prompting using exemplar tasks. We tested a general-purpose AI tool (Gemini) and an education-specific AI tool (Coteach). The specific tools were selected because of their relatively high performance on relevant published benchmarks and prior task-specific tests. Models were tested at baseline, retested with model version updates, and then tested again using few-shot prompting (two exemplar tasks for each cognitive demand category). Results revealed that newer model versions alone produced mixed effects: Gemini's accuracy remained stable at 58\%, while Coteach's accuracy decreased from 75\% to 50\%. However, few-shot prompting improved both models' performance: Gemini increased to 67\% and Coteach recovered to 75\% accuracy. These findings demonstrate that prompt engineering techniques can have larger and more reliable effects than passive model improvements, and that version updates may not always improve performance on specialized educational tasks. The study has important implications for how educators and researchers should approach AI tool selection, evaluation, and implementation in educational contexts.

Despite the rapid deployment of LLMs into classrooms, validating educational AI remains uniquely intractable: interventions act on developing learners whose cognitive and social trajectories are irreversibly shaped, while real-world trials are slow, ethically constrained, and institutionally locked. LLM-based educational simulators have emerged as a potential remedy, but many still collapse learning into persona-conditioned role-play and, when optimized only to reproduce existing classrooms, can structurally penalize the institutional novelty that pedagogical reform requires. In this work, we introduce AgentSchool, an LLM-driven multi-agent simulator that models learning as state transition rather than prompted behavior. AgentSchool couples cognitively growable student agents -- equipped with weighted subject knowledge graphs, thinking-workflow pools, and explicit misconceptions -- with adaptive teacher agents that plan, scaffold, and reflect along the Zone of Proximal Development, embedded in a configurable scenery generator that situates instruction within both formal and informal learning fields, and a multi-scale simulator that decouples interaction scale, temporal granularity, and simulation duration. Experiments show that structured student agents produce more differentiated mastery and misconception traces than a baseline simulator, while teacher-agent comparisons show backbone-dependent patterns consistent with ZPD-informed adaptation. Further, AgentSchool generates plausible traces of peripheral participation, clique formation, aggressor-induced cohesion, and opinion-leader emergence consistent with classroom social theories. Beyond its role as an educational research instrument, AgentSchool frames education as a socially meaningful testbed for long-horizon memory, multi-agent coordination, and future institutional reasoning under organizational pressure.

Benefiting from generalizability of vision-language models (VLMs) such as CLIP, many zero-/few-shot anomaly detection (AD) approaches have achieved impressive detection performance across various datasets. Nevertheless, they require substantial training on large auxiliary datasets to adapt VLMs to anomaly detection, and their inference largely relies on visual-text embedding similarity-based anomaly scores, lacking reasoning abilities to detect complex anomalies that require in-depth contextual understanding. To address this limitation, we propose \textbf{AnomalyAgent}, a novel training-free, agentic framework that leverages the advanced reasoning and generalization capabilities of multimodal large language models (MLLMs) for anomaly detection. The key ingredients include \textbf{1)} a comprehensive anomaly-centric toolset that enables adaptive MLLM-driven, agentic anomaly reasoning in zero-shot settings, and \textbf{2)} a customized memory module that grounds anomaly reasoning with few-shot, in-context reference examples. We extend evaluation beyond the detection of simple anomalies (e.g., surface defects like cracks and dents and clear lesions) in widely used benchmarks to more diverse types of anomalies such as logical/contextual anomalies in logistics and manufacturing settings. Extensive experiment results demonstrate that our AnomalyAgent achieves substantially better performance compared to training-free VLM-based AD and generic agentic methods, highlighting its superior generalization capability in both zero-shot and few-shot anomaly detection settings. The code implementation can be find at this address.

LLM-based multi-agent systems have demonstrated remarkable performance on complex tasks through collaborative reasoning. However, these systems tend to rapidly accumulate extremely long conversation histories during interaction. As conversations lengthen, relevant information is increasingly diluted by irrelevant context, leading to degraded performance. In this work, we present Agent-Radar, a training-free context management method that dynamically steers each agent's attention toward relevant context with a novel temporal and spatial decay mechanism. Our experiments demonstrate that Agent-Radar outperforms state-of-the-art methods across five different benchmarks, yielding gains of up to 7.64 absolute points. Furthermore, our analysis shows that Agent-Radar remains effective and robust as the number of agents and interaction rounds increases. Finally, the ablation study shows that core components in Agent-Radar are crucial to performance and generalizable in different settings.

Large Vision-Language Models (LVLMs) map visual inputs into dense token sequences, imposing a quadratic computational bottleneck for inference. Elastic visual-token compression addresses this by training a single model that can run at multiple visual-token budgets. However, existing approaches struggle under aggressive compression. Spatial-only compression, as in nested pooling, behaves as an imperfect low-pass filter and induces spectral aliasing that obscures fine-grained detail. Query-only compression, as in nested query resampling, replaces explicit grid-aligned tokens with non-local summaries and substantially degrades spatial grounding. To resolve this representational conflict, we introduce PARCEL (Pool-Anchored Resampling with Conditioned Elastic Queries for Efficient Vision-Language Understanding), a visual tokenization architecture that dynamically partitions the labor of feature extraction. PARCEL establishes spatial pool tokens as low-frequency layout anchors and conditions elastic query tokens on these anchors through Pool-Conditioned Query Resampling. This encourages query tokens to focus on complementary visual features rather than redundant spatial mapping. Extensive evaluations across 27 benchmarks show that PARCEL improves the performance-efficiency Pareto frontier, consistently outperforming existing matryoshka baselines across visual-token budgets while preserving the "train once, deploy anywhere" paradigm.

Multi-vector retrieval (MVR) models, exemplified by ColBERT, have established new benchmarks in retrieval accuracy by preserving fine-grained token-level interactions. However, this granularity imposes prohibitive storage and retrieval efficiency bottlenecks: to manage the immense memory footprint and computational overhead of billion-scale token vectors, state-of-the-art systems are forced to rely on aggressive dimension reduction and complex clustering (e.g., K-means). This compromise introduces two critical limitations: excessive indexing latency of clustering large-scale corpora and semantic information loss inherent to compression. In this paper, we propose Single-stage Sparse Retrieval (SSR}, a paradigm shift that replaces expensive clustering with efficient sparse coding. Instead of compressing features into low-dimensional dense vectors, we utilize Sparse Autoencoder (SAE) to project token embeddings into a high-dimensional but highly sparse representation. This transformation enables us to bypass vector clustering entirely and leverage inverted indexing for precise, high-throughput retrieval. Extensive experiments on the BEIR benchmark demonstrate that SSR achieves a "trifecta" of improvements: it reduces indexing time by 15x compared to ColBERTv2, halves retrieval latency, and simultaneously improves retrieval performance over leading baselines.

Understanding how Vision-Language-Action (VLA) models transform multimodal knowledge into embodied control remains an open challenge. We present VLA-Trace, a progressive diagnostic framework that analyzes VLA models through a unified evidence chain from representation dynamics to causal control attribution and behavioral manifestation. It specifically combines cross-modal and checkpoint-drift centered kernel alignment (CKA) to trace representation evolution, attention knockout interventions to identify modality-specific control pathways, and rollout-level behavioral probes to examine grounding, shortcut dependence, and semantic following. Experiments on $π_{0.5}$ and OpenVLA reveal three key findings. First, the two models exhibit distinct modality-specific adaptation dynamics during VLA finetuning. Second, they rely on different multimodal routing strategies and layer-wise dependencies during action decoding. Third, although VLA policies excel at visually grounded trajectory generation, they remain limited in fine-grained semantic following. These findings highlight future directions for representation-preserving adaptation, causal VLA circuits, and compositional semantic control.

This work presents the Large Depth Completion Model (LDCM), a simple, effective, and robust framework for single-view metric depth estimation with sparse observations. Without relying on complex architectural designs, LDCM generates metric-accurate dense depth maps using a transformer. It outperforms existing approaches across diverse datasets and sparse observations. We achieve this from two key perspectives: (1) leveraging existing monocular foundation models to improve the quality of sparse depth inputs, and (2) reformulating training objectives to better capture geometric structure and metric consistency. Specifically, a Poisson-based depth initialization strategy is first introduced to generate a uniform coarse dense depth map from diverse sparse observations, providing a strong structural prior for the network. Regarding the training objective, we replace the conventional depth head with a point map head that regresses per-pixel 3D coordinates in camera space, enabling the model to directly learn the underlying 3D scene structure instead of performing pixel-wise depth map restoration. Moreover, this design eliminates the need for camera intrinsic parameters, allowing LDCM to naturally produce metric-scaled 3D point maps. Extensive experiments demonstrate that LDCM consistently outperforms state-of-the-art methods across multiple benchmarks and varying sparsity levels in both depth completion and point map estimation, showcasing its effectiveness and strong generalization to unseen data distributions.

Cross-Reynolds generalisation in neural PDE solvers remains poorly characterised. On the canonical forced 2D Navier-Stokes benchmark, a trained Fourier Neural Operator reaches 46.68% relative L2 error under a 10x Reynolds-number shift, yet zero-forward-model retrieval baselines already improve to 41-42%. This suggests representation geometry as a major organising variable among the tested methods. We test this hypothesis through ConvAE-Relay, which matches states in a source-trained convolutional autoencoder latent space and borrows dynamics from a source-regime database, achieving 38.34+/-0.07% using only a source-regime database and no target-regime fitting, labels, or database entries. A 2x2 ablation isolates matching quality as dominant over the update rule. Oracle experiments confirm that source-regime dynamics directions remain transferable (cosine similarity ~0.84) when matching stays on-manifold; autoregressive drift is the primary bottleneck (~12 percentage points). From the learned-prediction side, a U-Net with multi-scale skip connections achieves 34.72+/-0.60%, consistent with the retrieval-side finding that local, multi-scale representations organise cross-Reynolds transfer among tested methods. All claims are scoped to this benchmark.

Creating spoken dialogue datasets is methodologically challenging, and these challenges are amplified when the goal is to build multilingual, multi-parallel datasets at scale. This work introduces HEALTHDIAL, a large-scale, multilingual, and multi-parallel dataset for developing and evaluating retrieval-augmented generation (RAG)-based spoken dialogue systems. The dataset comprises 6,000 information-seeking dialogues (1,500 per language) grounded in trusted content from the World Health Organization (WHO) and 163 hours of user speech recorded from native speakers of diverse dialects across four official WHO languages: Arabic, Chinese, English, and Spanish. Each speaker is annotated with demographic (e.g., gender, age) and sociolinguistic (e.g., primary language, region of origin) variables. We report benchmark results across key dialogue tasks, which reveal consistent performance disparities across languages, even among high-resource ones. To support future research, we release the dataset, a prototype system, and a toolkit for data collection and system evaluation.

Widely used language-model benchmarks are increasingly saturated, with frontier systems often receiving near-tied scores that standard metrics cannot resolve. Rather than constructing harder alternatives, we ask whether existing tasks can be made informative again through improved evaluation over the same candidate outputs. Therefore, we present Seeded Elimination with Adaptive LLM-as-a-Meta-Judge, a self-improving evaluation protocol for extracting latent ranking signal from saturated benchmarks. SEAL seeds candidate outputs into a single elimination and evaluates each match with task-level principles plus self-improving checklist criteria. We evaluate SEAL on multiple saturated benchmarks covering code generation, mathematical reasoning, knowledge-intensive question answering, and tool-use agent task completion. Across these settings, SEAL improves the ranking-accuracy--latency trade-off over competing protocols, attaining 0.83--1.00 Spearman agreement with full pairwise judging and 4/4 top-1 agreement, while requiring only 11.89 calls per task compared with 28.00 for full pairwise evaluation.

The design space of agentic AI inference spans two extremes: frontier large language models (LLMs), typically hosted in the cloud and offering strong performance across a wide range of tasks at substantially high cost, and more cost-efficient small language models (SLMs), which are amenable to on-device inference. Hybrid multi-agent systems (MASs) combining on-device and cloud models offer a promising middle ground, but they also introduce a complex and poorly understood design space in which task accuracy, monetary cost, and edge energy consumption are tightly coupled; in the absence of general design principles, hybrid components, although not the most prevalent choice, are typically introduced through ad hoc decisions tailored to specific domains. In this work, we examine this design space more systematically. We adapt two representative MAS architectures to support hybrid inference and study how individual design choices shift the operating point along the Pareto frontier of power, cost, and performance. Our findings paint a nuanced picture of hybrid MAS design: while SLMs can effectively benefit from LLM assistance, the optimal architecture is highly task-dependent, and greater frontier-level compute does not consistently translate to better performance.

World models require state tracking, which is the ability to maintain a correct latent state across action sequences. Existing benchmarks are often synthetic or language-based, limiting their value as tests of structured state updates in realistic domains. We introduce Chess-World-Model, a large-scale state-tracking benchmark built from 10 million real chess games, where models predict the exact board state reached after a sequence of legal moves. Alongside a held-out real-game split, we include an out-of-distribution split from uniformly random legal play, which tests whether models learn the transition rules rather than shortcuts from common human positions. Prior theoretical and empirical work has shown that Transformers struggle to state-track, while input-dependent linear RNNs require expressive state-transition matrices to do so. We therefore benchmark a causal Transformer, block-diagonal SLiCE, Mamba-3, and Gated DeltaNet with negative eigenvalues under a matched interface and training protocol. The recurrent models strongly outperform the Transformer at 3 and 8 million parameters. Real-game performance saturates above 18 million parameters, but the random-uniform split remains discriminative up to 40 million, exposing failures otherwise hidden by scale. Additionally, ablations show that less expressive state-transition mechanisms reduce performance on the out-of-distribution split for all three recurrent models. Together, these results establish Chess-World-Model as a practical large-scale benchmark for state tracking that exposes failures model scale would otherwise conceal.

Poker is a landmark challenge for artificial intelligence. The dominant approach relies on equilibrium solvers built on counterfactual regret minimization, requiring millions of core-hours of training. Large Language Models (LLMs) possess extensive poker knowledge but perform far below solver-based agents when asked to play directly. Traditional rule-based poker agents are interpretable and training-free, but their strategic ceiling remains far below equilibrium play. We introduce \textbf{PokerSkill}, a training-free and solver-free framework that bridges this gap by using detailed rule-based poker skills as a structured action-grounding interface for LLMs. A deterministic context engine analyzes the current state and retrieves only the relevant fragments from a layered skill library, which is entirely designed by human poker experts, constraining the LLM's choice to reasonable actions. Against GTOWizard, a state-of-the-art GTO benchmark, GPT-5.5 XHigh with PokerSkill achieves $-57 \pm 21$ mbb/hand, Claude Opus 4.6 achieves $-80 \pm 29$ mbb/hand and Claude Opus 4.7 achieves $-87\pm 64$ mbb/hand, reducing losses by 49--61\% compared to default-prompt baselines and outperforming the strong bot Slumbot. Our key finding is that rule-based skills alone do not constitute a strong strategy, and LLMs alone cannot play well, but their combination yields an agent that requires neither training nor solver access yet competes with systems built on millions of core-hours of computation. To our knowledge, this is the first demonstration of an LLM achieving competitive performance in a complex imperfect-information game without game-specific training or solver queries. Code is available at https://github.com/lbn187/PokerSkill.

Foundation features from self-supervised vision models and text-to-image diffusion models have proven effective for semantic correspondence estimation. However, because these features are learned primarily from 2D image objectives, they lack explicit 3D awareness and often confuse symmetric object sides, repeated parts, and visually similar structures that are distinct in 3D. We introduce a 3D-aware post-training framework that goes beyond available 2D foundation features by incorporating priors from 3D foundation models. Given an image, our method uses SAM3D to estimate object geometry and pose, and refines the pose through render-and-compare optimization. Subsequently, we render PartField descriptors from the reconstructed geometry into the image plane based on the estimated object pose. The resulting geometry-aware feature maps complement DINO and Stable Diffusion features, while geodesic distances on the reconstructed shapes enable reliable filtering of candidate correspondences. We use the filtered matches as supervision to train a lightweight adapter on top of DINO and Stable Diffusion for semantic correspondence. In contrast to prior post-training approaches that require pose annotations and rely on coarse spherical geometry, our method automatically obtains instance-specific 3D structure and uses it to guide correspondence learning. Experiments show that our approach improves semantic correspondence over the prior methods while reducing manual geometric supervision. Code and model can be found at https:/github.com/GenIntel/3D-SC.

Long-form video generation is rapidly moving from short, single-scene synthesis toward minute-long, multi-shot creation with narrative structure, cinematic control, audio, and cross-modal synchronization. However, evaluating such videos remains challenging, since existing benchmarks largely focus on local visual quality, short-horizon temporal consistency, or generic prompt alignment, and provide limited diagnosis of workflow failures and user-dependent preferences. We introduce DirectorBench, a personalized multi-agent diagnostic benchmark for long-form video generation. DirectorBench evaluates generated videos with respect to 80 structured metadata entries, 7 user profiles, and 40 checkpoint criteria across 5 dimensions: script, visual, audio, cross-modal, and stability. Instead of reducing quality to a single aggregate score, DirectorBench localizes checkpoint-level bottlenecks and supports profile-aware evaluation. We evaluate 4 long-form video generation workflows, 6 base LLMs, and 7 user profiles. Across workflows, DirectorBench reveals a between-unit bottleneck: transition quality averages only 0.256 and reaches 0.356 for the best workflow, while prompt-level user demand fulfillment averages 0.71. We further conduct human evaluation with 14 annotators to validate the alignment between DirectorBench and human judgment. The results show that DirectorBench captures human-perceptible quality differences and reveals workflow- and profile-dependent failure modes that are hidden by aggregate scoring. These findings highlight the importance of diagnostic and profile-aware benchmarking for long-form video generation.

Emerging personal AI agents are moving toward persistent, multi-source memory. This creates an evaluation problem: systems must decide how to use conflicting or incomplete evidence; they cannot just retrieve facts from one clean history. Existing benchmarks rarely show whether an error came from the evidence given to a method or from the method's conflict-resolution step. We study this as selective QA over conflicting multi-source personal memory: systems answer based on conflicting, sometimes incomplete sources, or abstain when evidence is insufficient. We develop a benchmark containing 18 question templates across 8 reasoning types, 480 personas, 4 random seeds, and 34,560 instances, with controlled source distortions and deterministic ground truth. We evaluate the performance of baselines without access to any source, access to a single source, structured fusion methods, and frontier LLMs. The best trained fusion resolver reaches 80.3% accuracy, while the strongest prompt-only LLM baseline reaches 70.0%. With abstention, the same resolver reaches 85.3% selective accuracy at 78.3% coverage and the best LLM reaches 71.0% selective accuracy at 95.4% coverage. Different models have different strengths across reasoning types. We release the data, code, cached model outputs, and data-generating process for reuse.

Autoregressive (AR) video generation has emerged as a promising paradigm for long-horizon video synthesis, where each frame is generated conditioned on previously generated tokens. To accelerate inference, the KV cache is used to avoid redundant recomputation across generation steps. Nevertheless, its growth with generation length introduces increasing memory and error accumulation, limiting the scalability of AR models to even longer sequences. Existing KV cache compression methods mitigate this issue by selectively retaining only video tokens deemed important. However, most existing methods assess token importance using short-horizon signals derived from the current or historical generation context, making these methods prone to overlooking tokens that appear unimportant at early steps but later become critical for future frames. In this work, we identify an important property of trained AR video models: although RoPE-modulated queries evolve across autoregressive steps, the underlying canonical pre-RoPE query distribution remains remarkably stable throughout the video generation process. This approximate stationarity implies that future query distributions are estimable from historical statistics, enabling principled future-aware cache decisions without any additional training. Building on this insight, we propose Future Forcing, a training-free future-aware KV cache policy for AR video generation. Specifically, Future Forcing first constructs a future query proxy from historical statistics, then scores KV cache tokens by their importance under this proxy, and finally merges redundant token pairs within the affine subspace induced by the future query. Extensive experiments show that Future Forcing improves long-horizon consistency under limited KV caches, achieving up to 1.49 improvement in subject consistency on VBench-Long for 60s generation over existing AR video KV cache policies.

Tokenization-free hierarchical models are emerging as a promising alternative to traditional Large Language Models (LLMs), addressing inherent preprocessing issues such as vocabulary design complexity, out-of-vocabulary (OOV) errors, and language-specific constraints. However, a significant challenge in these byte-level methods is the optimization of the compression ratio, a critical factor that dictates model performance for processing bytes data via chunks. In this paper, we propose Adaptive Targeted Dynamic Chunking (ATDC), a novel byte-compression control mechanism designed to enhance the effectiveness of dynamic chunking within hierarchical architectures. Our approach utilizes curriculum learning to progressively adjust the compression ratio during training, transitioning from low to high compression to stabilize the learning process. We provide an analysis establishing the relationship between the target compression ratio and Bytes-Per-Innermost-Chunk (BPIC), allowing for tracking of chunk-size evolution throughout the training phase. Evaluations conducted on the FineWeb-Edu 100B dataset demonstrate that hierarchical models equipped with ATDC achieve competitive Bits-Per-Byte (BPB) performance compared to conventional baselines operating at both byte and token levels. Furthermore, the proposed method exhibits more stable training dynamics and superior final performance across diverse downstream tasks compared to models using fixed compression ratios, while maintaining the inherent robustness and flexibility of byte-level processing.

Moving beyond simple scalar rewards toward richer world feedback is a natural path to more scalable RL post-training. On-policy self-distillation (OPSD) is a promising recent approach that uses arbitrary feedback as learning signal, yet its reliability compared to established methods, such as GRPO, remains unclear. We identify a strikingly consistent linear correlation between the initial student-self-teacher performance gap and the final performance improvement in OPSD. This relationship holds across context types and model families, providing a powerful predictive law for anticipating the outcome of an OPSD configuration without running the full training procedure. Interestingly, we show that this linear predictability holds with model scale, suggesting a potential basis for new empirical scaling laws on larger models with stronger in-context learning capabilities. In essence, our findings show that OPSD performance can be predicted and tuned before training, offering a principled way to incorporate world feedback as a first-class component of the post-training pipeline.

The development of generative artificial intelligence technologies has propelled the visual realism of synthetic images to an unprecedented level. Although current interpretable detection methods based on Large Multimodal Models (LMMs) have made certain progress, they still rely on imitation learning derived from massive volumes of forged data. Consequently, they lack genuine causal reasoning capabilities and are prone to explanatory hallucinations. To overcome this bottleneck, we propose FakeVLM-R1, aiming to endow the model with human-like critical thinking capabilities when performing synthetic detection tasks. Building upon Supervised Fine-Tuning (SFT), this framework integrates Group Relative Policy Optimization (GRPO) with a Critical Thinking Chain-of-Thought (CoT) mechanism. During the inference phase, the model executes a "bidirectional dialectical reasoning" process: while proposing a forgery hypothesis, it must simultaneously invoke physical commonsense to construct an authenticity counter-proof. Furthermore, we constructed the FakeClue++ dataset with high-quality samples, which extensively introduces annotations guided by the physical laws of authentic images, providing a unified authenticity anchor for the model. Experiments confirm that FakeVLM-R1 achieves SOTA performance the evaluated models across multiple benchmarks. It not only achieves high-precision, logically interpretable detection but also resolves the over-rejection bias of existing methods against real images, demonstrating generalization and robustness against perturbations.

While LLM agents have demonstrated remarkable task-oriented abilities such as planning, reasoning, and action, few works have treated them as complete human personalities where emotional dimensions hold equal importance. In this paper, we introduce a novel benchmark to systematically assess whether LLM agents can simulate coherent, human-like psychology. Specifically, our benchmark constructs 11 diverse human characters grounded in orthogonal Big Five personality traits, with each profile deeply integrated with 1,000 structured autobiographical-style episodic memories distributed across theory-grounded developmental life stages. To rigorously evaluate the psychological manifestations of LLMs, we designed a curated suite of 64 decision-making scenarios, guided by the DIAMONDS taxonomy, a psychological framework that characterizes situations along eight dimensions: Duty, Intellect, Adversity, Mating, pOsitivity, Negativity, Deception, and Sociality. By subjecting agents to varying scenarios, the benchmark evaluates whether they can consolidate their innate personality traits and autobiographical memories to make behavioral decisions that are consistent with their specific psychological profiles. After systematic human validation and filtering, we obtained a benchmark consisting of 673 multiple-choice questions (MCQs). We believe this benchmark provides a principled and scalable testbed for studying human-like emotions, personality consistency, and value-consistent behavioural decision-making in LLM-based agents.

Recent advances in reinforcement learning (RL) have achieved great successes by leveraging the multimodality and exploration capability of diffusion policies. Among these approaches, one representative branch focuses on the sampling-based policy optimization. This design enables better exploration capability of the diffusion model, particularly at the beginning of training, but suffer from low exploitation in Q-value information, resulting in a slow policy convergence. Another branch pays attention to gradient-based policy optimization, which sufficiently exploits the gradient of the Q function yet tends to collapse into a unimodal policy with low diversity. To address this issue, we propose CGPO, \textbf{C}ritic-\textbf{G}uided diffusion \textbf{P}olicy \textbf{O}ptimization, which effectively balances exploration and exploitation with the training-free guidance technique integrated into the denoising process of diffusion policy. Concretely, CGPO steers action generation toward high-value regions defined by the critic network and uses the guided actions as regression objectives. In this manner, CGPO reduces the time required to obtain high-quality actions and improves final performance with better balance between the exploration-exploitation tradeoff. We validate the effectiveness of CGPO on 5 MuJoCo locomotion tasks, and CGPO achieves state-of-the-art performance compared with existing diffusion-based RL methods. Notably, CGPO is the first success to incorporate diffusion policy into real-world RL, with its superior performance on Franka robot arm grasping tasks. Our official page is released at https://dingsht.tech/cgpo-webpage.

One-shot Program-of-Thought (PoT) emits a Python program that prints a primitive-action plan; a single invalid action silently invalidates the trajectory. We introduce RePoT (Recoverable PoT): a deterministic verified replay that walks the plan through the environment to its first invalid transition, then one LLM call that resumes from the verified prefix. RePoT costs at most one extra LLM call on the ~14% of problems where PoT fails. RePoT beats PoT by +3 to +11pp across four closed-model configurations on PuzzleZoo-775 and peaks at 96.9% vs 86.3% on gpt-5.4-mini-medium; against the matched-budget PoT-retry baseline, RePoT wins decisively on Gemini (+3.8pp, 95% CI [+2.2,+5.4]), is within sampling noise on GPT-medium and Claude, and loses on GPT-mini -- a capability-scaling pattern we begin to address with Adaptive RePoT, a rule-based dispatcher that routes between suffix repair and a fresh PoT retry based on verified-prefix length (preliminary). We replicate on PlanBench Blocksworld (+1.1 to +11.4pp) and on four open-weights models (+3.3 to +20.0pp on three of four). On Derail-550, our controlled recovery benchmark, every condition with access to checkpoint information clears >=30% on GPT-medium and >=70% on Gemini, vs <=3.1% for error-only feedback -- showing that checkpoint information, not the specific verified-prefix tail, is the load-bearing recovery signal.

Video object removal frequently struggles to simultaneously eliminate target objects and their associated physical effects (e.g., smoke, reflections, light, and ripples) in out-of-domain scenarios due to complex spatiotemporal ambiguities. While existing methods primarily rely on spatial masks, they often fail to capture weakly correlated effects, and the potential of explicit textual guidance remains underexplored. Furthermore, a fundamental optimization conflict exists in removal models between high-level semantic generalization and precise pixel-level background preservation. To address these challenges, we propose GenEraser, a novel framework for generalized and high-fidelity video object and effect removal. First, we introduce a Multi-Conditional Mixture-of-Experts (MC-MoE) paired with Bipartite Text guidance to fully exploit the multimodal priors of Diffusion Transformers, significantly enhancing the identification of complex effects. Second, a Learnable Deep ``CFG'' Fusion mechanism (LD-CFG) is developed to adaptively balance the relative dominance of mask and textual conditions across diverse scenarios. Finally, we propose a Decoupled Expert Architecture, comprising a Locator and a Preserver, to mitigate the inherent trade-off between semantic generalization and pixel alignment. Extensive experiments demonstrate that our GenEraser surpasses recent state-of-the-art approaches, achieving significant quantitative improvements (e.g., $2.16$ dB and $1.44$ dB on the ROSE Benchmark and VOR-Eval, respectively) while maintaining exceptionally robust generalization in open-world scenarios. https://cyqii.github.io/GenEraser.github.io/

Automating scientific computing workflows requires more than generating executable code: autonomous systems must also select appropriate computational strategies, implement them faithfully, and ensure that the resulting outcomes remain causally attributable to the decisions that produced them. In multi-agent pipelines, this process is particularly fragile, as small inconsistencies between agent intentions and actions can lead to semantic drift, where the eventually executed procedure no longer reflects the originally selected strategy, thereby corrupting downstream evaluation and adaptation. In this work, motivated by the ATHENA framework (Toscano et al., 2025; Toscano et al., 2026) and the concept of empowerment (Yiu et al., 2025), we introduce a multi-agent framework that combines contextual bandits with structured inter-agent communication and, most importantly, semantic checkpoints that preserve action-outcome fidelity throughout the pipeline. The system integrates specialized large language model (LLM) agents, grounded code generation, and self-healing execution loops within an adaptive decision-making architecture. Interpreting the framework through the lens of empowerment, we show that reliable autonomous learning requires not only identifying high-quality actions, but also preserving the integrity of their propagation across agents. Using sensitivity analysis and uncertainty quantification workflows as representative case studies, we demonstrate that unchecked semantic drift degrades policy learning, whereas the proposed framework improves convergence, robustness, and adaptation to novel problem contexts. These results suggest a broader design principle for scientific multi-agent systems: adaptive decision-making must be coupled with explicit mechanisms that guarantee semantic consistency and reliable information flow across the computational pipeline.

Large Language Models have demonstrated remarkable progress in general-purpose capabilities and can achieve strong performance in specific domains through fine-tuning on domain-specific data. However, acquiring high-quality data for target domains remains a significant challenge. Existing data synthesis approaches follow a deductive paradigm, heavily relying on explicit domain descriptions expressed in natural language and careful prompt engineering, limiting their applicability in real-world scenarios where domains are difficult to describe or formally articulate. In this work, we tackle the underexplored problem of domain-specific data synthesis through an inductive paradigm, where the target domain is defined only through a set of reference examples, particularly when domain characteristics are difficult to articulate in natural language. We propose a novel framework, DOMINO, that learns a minimal sufficient domain representation from reference samples and leverages it to guide the generation of domain-aligned synthetic data. DOMINO integrates prompt tuning with a contrastive disentanglement objective to separate domain-level patterns from sample-specific noise, mitigating overfitting while preserving core domain characteristics. Theoretically, we prove that DOMINO expands the support of the synthetic data distribution, ensuring greater diversity. Empirically, on challenging coding benchmarks where domain definitions are implicit, fine-tuning on data synthesized by DOMINO improves Pass@1 accuracy by up to 4.63\% over strong, instruction-tuned backbones, demonstrating its effectiveness and robustness. This work establishes a new paradigm for domain-specific data synthesis, enabling practical and scalable domain adaptation without manual prompt design or natural language domain specifications.

Diffusion models generate highly realistic images but often struggle with precise text-image alignment. While recent post-training methods improve alignment using external rewards or human preference signals, their performance heavily depends on reward quality and does not directly address alignment within the diffusion process itself. Recent reward-free approaches such as SoftREPA demonstrate that optimizing soft text tokens via contrastive learning can effectively improve text-image representation alignment, outperforming standard parameter-efficient fine-tuning baselines. However, the contrastive formulation can excessively penalize negative pairs, which manifests as characteristic failure cases such as over-counting and repetition. To address this issue, we propose a lightweight, reward-free post-training method that refines soft tokens by integrating contrastive alignment guidance directly into the score-matching objective of diffusion models. By assigning alignment directions at the score level, our approach mitigates these limitations and yields more coherent and semantically faithful generations. Experiments show that our method matches SoftREPA while substantially improving its failure cases, achieving over 35% improvement in counting accuracy on the GenEval benchmark. Our method is seamlessly applicable to existing diffusion backbones (SD1.5, SDXL, and SD3), and is complementary to existing RL-based diffusion post-training methods. Project page: https://jaayeon.github.io/AGSM

Large Audio Language Models (LALMs) expand jailbreak risks from token-level prompting to the full speech perception-to-reasoning pipeline, where unsafe behavior can be induced through semantics, acoustic style, signal artifacts, or internal representations. Existing work studies these risks under heterogeneous threat models and evaluation protocols, making it difficult to compare attack practicality or defense utility. This paper provides a unified taxonomy and a controlled empirical evaluation of LALM jailbreak attacks and defenses. We organize prior work into semantic, acoustic, signal, and embedding-layer attacks; guard-based, training-free, and training-based defenses; and cross-modal, audio-native, and interactive benchmarks. We then evaluate representative attacks and defenses across ten open-source LALMs, measuring not only attack success rate but also benign refusal and latency. Our results show that Acoustic Best-of-N reveals strong worst-case audio-space vulnerabilities, Narrative Framing is an effective low-latency semantic threat, and current defenses trade robustness against benign usability. These findings support cost- and utility-aware evaluation as a necessary complement to success-rate-only LALM safety benchmarks.

Retrieval-augmented generation (RAG) systems expose numerous design choices spanning query rewriting, chunking, retrieval depth, reranking, and context compression. In practice, these choices are often configured through heuristics, hindering systematic evaluation and reproducibility across settings. We argue that this challenge is best formulated as RAG architecture search. To support controlled and reproducible study of this problem, we introduce the RAG Intelligence Search Engine (RAISE), a comprehensive framework and benchmark for RAG hyperparameter optimization, which evaluates optimization methods for RAG pipelines under standardized search spaces and budgets. RAISE implements 13 search algorithms and evaluates them across seven public text and multimodal datasets using three random seeds. Our experiments show that optimization performance is highly task-dependent: methods that perform strongly on one dataset may not generalize consistently across others, cautioning against interpreting aggregate rankings as evidence of universally superior strategies. RAISE provides a common experimental substrate for fair, reproducible, and systematic research on RAG hyperparameter optimization.

Multimodal documents contain diverse elements, such as tables, figures, and layouts, which can complicate retrieval tasks. While current approaches typically combine dense visual embedding models with supervised rerankers to achieve high-precision retrieval, they face inherent limitations. First, the coarse-grained nature of dense embeddings tends to obfuscate explicit semantics, failing to leverage structurally salient information. Second, supervised reranking models suffer from generalization bottlenecks, as their performance heavily relies on domain-specific training data. Furthermore, existing benchmarks often lack diverse assessment dimensions and comprehensive relevance annotations, limiting reliable evaluation. To address these challenges, we propose DocRetriever, a plug-and-play framework. It enhances visual retrieval via a layout-aware sparse embedding technique, enabling effective hybrid encoding without the overhead of optical character recognition (OCR). We also introduce a generalizable reranker that leverages reasoning-augmented demonstrations and optimized sampling to improve accuracy in few-shot settings. Finally, we construct a new benchmark, MultiDocR, to enable more rigorous evaluation. Experiments across diverse benchmarks validate DocRetriever's superiority over state-of-the-art methods.

Positional encoding (PE) underpins how permutation-invariant Transformers represent sequence order, yet how positional information is processed and stored remains poorly understood. Modern PE methods such as RoPE still struggle on tasks such as long-context understanding or retrieval \cite{chen-etal-2025-hope}. Hence, a better understanding of the internal positional mechanism could help design better PE. Building on evidence that positional and semantic signals occupy nearly orthogonal subspaces in trained Transformers, we modify an encoder Transformer to process three explicitly disentangled streams: semantic, absolute positional (AP) and relative positional (RP), and confine the masked-language-modeling (MLM) objective to the semantic stream. This decoupling enables a clean mechanistic study and yields three take-aways. (1) The isolated AP subspace spontaneously collapses into a low-frequency two-dimensional manifold that captures the structure of the document; (2) Attention heads specialize into structure and semantic-oriented groups, with RP exclusively supporting the latter; (3) Standard positional encodings do not robustly retain macroscopic structure: RoPE and RP only weakly encode it, and entangled AP loses it in the final layers under MLM pressure. The disentangled approach preserves positional encoding, which improves linguistic representation on 49 of the 65 linguistic phenomena of the Flash-Holmes probing benchmark.

Many open-ended instructions have multiple valid answers that users can benefit from seeing, but post-training often narrows an LLM's output space toward a small set of canonical responses. We introduce REDIPO, an offline DPO data-construction pipeline for recovering distinct valid answer modes while preserving the alignment benefits of the instruct model. For each prompt, REDIPO samples responses from both base and instruct models, rewrites base-model responses with the instruct model, filters candidates for safety and instruction-following quality, and builds preference pairs that favor marginally diverse responses among candidates with similar instruction-following reward. Across Qwen3-4B, OLMo-3-7B, and LLaMA-3.1-8B, REDIPO improves NoveltyBench distinct_k by 134%, 33%, and 44% relative to the instruct checkpoints, while DivPO changes diversity by 0%, -6%, and -4% on the same models. These gains largely maintain MTBench, IFEval, and Arena-Hard performance, and reduce direct-category HarmBench attack success rate. Ablations show that marginal-diversity pair selection and base-response rewriting drive the diversity gains, while filtering and quality-bounded pairing help maintain alignment. Overall, our results show that diverse valid answers from base-model generations can be reintroduced through carefully constructed preference data while retaining the alignment benefits of post-training. We release our code and data at https://github.com/vsamuel2003/RiDiPO.

Large language models (LLMs) frequently achieve impressive scores on standardized benchmarks, yet accuracy alone offers a limited view of their capabilities. Evaluating open-source LLMs through leaderboards faces persistent issues like data contamination, narrow task scope, and weak alignment with real-world reliability. Benchmark-based evaluations such as MMLU PRO, BBH, or IFEval primarily capture \textit{what} a model outputs on fixed test sets, not \textit{how} it processes information, calibrates uncertainty, or structures internal knowledge. In this article, we advocate for a shift from benchmark-centric evaluation toward a complementary, \textit{state-centered intrinsic assessment} of LLMs. To this end, we introduce \textbf{Latent Performance Profiling (LPP)} -- a framework that derives task-agnostic diagnostics from hidden activations and output distributions. LPP defines a set of scalar metrics on a model's latent representations and dynamics, revealing scale-independent traits that enable interpretable comparisons and uncover hidden vulnerabilities. Unlike static accuracy scores, LPP provides stable, architecture-sensitive signatures across models of similar size. With extensive empirical analyses across eight LLMs, spanning a size range of 0.5B-14B, we demonstrate that models with similar benchmark scores can exhibit contrasting latent profiles, such as differences in entropy or adaptability. Guided by these insights, we design synthetic probes for uncertainty and symbolic reasoning that align with intrinsic metrics while decoupling from leaderboard bias. We recommend that reporting LPP alongside benchmarks provides a deeper, interpretable understanding of model behavior, enabling more reliable model selection, safety assessment, and evaluation beyond surface-level accuracy.

Supervised Causal Learning (SCL) has shown promise in causal discovery by framing it as a supervised learning problem. However, it suffers from significant out-of-distribution generalization challenges. We reveal three limitations of previous SCL practices: a significant performance gap between synthetic benchmarks and real-world data, fragility to distribution shifts, and failure in compositional generalization, collectively questioning its real-world applicability. To address this, we propose Test-Time Training for Supervised Causal Learning (TTT-SCL), a novel framework that dynamically generates training sets explicitly aligned with any specific test instance. We demonstrate the correlation between TTT-SCL and score-based methods, and design an efficient module for generating training sets based on the classic scoring function. Experiments on synthetic benchmarks, pseudo-real and real-world datasets demonstrate that TTT-SCL significantly outperforms existing SCL and traditional causal discovery methods.

Recent work has begun to equip vision-language-action (VLA) policies with explicit intermediate reasoning. In embodied control, however, textual chain-of-thought is a poor fit: irrelevant or weakly textual information can interfere with action prediction, while autoregressive text decoding adds too much latency for real-time closed-loop execution. We present VISUALTHINK-VLA, a visual intermediate-reasoning framework for accurate, low-latency VLA policies. Our bootstrapping philosophy is to guide action with effective visual thinking: VISUALTHINK-VLA bootstraps action prediction through a compact visual-evidence interface that preserves spatial precision while avoiding decoding overhead. Besides, to further improve performance and efficiency, VISUALTHINK-VLA adopts a tailored selective routing mechanism to learn the visual evidence tokens, enabling low-latency inference while preserving high-capacity specialization. We also introduce VisualEvidence-Kit, a supervision-and-audit resource centered on a VisualEvidence-Agent that constructs a 754.7k VLA instructions VisualEvidence-Set for route supervision and counterfactual faithfulness tests. Across multiple benchmarks and real-robot evaluation, VISUALTHINK-VLA achieves the highest success rate on most benchmarks while reducing the multi-second latency of reasoning-augmented baselines to the sub-second regime. For example, on BridgeData V2, it reduces step latency from 8.377,s with ECoT to 0.367,s, achieving a 22.8 times speedup.

We study two-level autoresearch for cooperation: an outer-loop AI agent autonomously redesigns the inner-loop pipeline of an LLM policy-synthesis system for multi-agent Sequential Social Dilemmas (SSDs). A researcher agent $\mathcal{R}$ (run as a coding agent) reads the inner-loop source code, edits system prompts, feedback functions, helper libraries, and iteration logic, runs evaluations, and decides what to keep, following the autoresearch paradigm. Across two games (Cleanup and Gathering), two policy-synthesizer LLMs, and two welfare objectives (utilitarian efficiency and Rawlsian maximin), the researcher reliably exceeds hand-designed baselines, sharply tightens run-to-run variance, and outperforms prompt-only optimization. The discovered pipelines are objective-dependent: only under maximin does the researcher inject an explicit fairness mechanism into synthesizer pipelines, a class of mechanism that is absent from its own objective-agnostic system prompt and from every efficiency-optimized pipeline. This supports an information-design reading in which the researcher chooses what to reveal to the boundedly rational synthesizer as a function of the welfare objective. Code at https://github.com/vicgalle/autoresearch-social-dilemmas.

Cross-domain multimodal time series forecasting is a challenging task, requiring models to integrate precise numerical comprehension, cross-domain semantic understanding, and effective multimodal fusion. Existing approaches either build Time Series Foundation Models (TSFMs) from scratch or leverage pretrained Large Language Models (LLMs). However, TSFMs often overlook semantic understanding and lack the ability to perform future-oriented semantic reasoning, and LLMs struggle with numerical comprehension and accurate quantitative forecasting. To overcome these limitations, we propose KairosAgent, a novel agentic framework for multimodal time series forecasting, including an LLM-based reasoner and a TSFM-based forecaster. KairosAgent unifies textual reasoning and numerical forecasting by dynamically invoking analytical tools to enhance the numerical understanding and semantic reasoning capabilities of LLMs. The reasoning results are subsequently fused into the TSFM pipeline, enabling more accurate and reliable future predictions. To further improve the reasoning, we curate a large-scale corpus of high-quality trajectories, alongside a reinforcement learning from forecasting paradigm with multi-turn refinement and turn-level credit assignment. Experiments demonstrate that KairosAgent achieves superior zero-shot forecasting performance while maximizing the utility of pretrained LLMs and TSFMs, presenting a promising direction for efficient and interpretable time series agents. The project page is at https://foundation-model-research.github.io/KairosAgent .

Front-end web code has become a core product surface for every frontier LLM release, yet evaluating these interactive applications at development speed remains costly because human-judged leaderboards like Arena do not scale. Existing automated proxies typically lean on reference implementations, test suites, or rigid checklists, and tend to miss the reasoned synthesis a human reviewer performs over a live session. We articulate a new evaluation regime that is simultaneously reference-free, autonomously driven, and holistically reasoned, and instantiate it through two artifacts. \textbf{\dataname} is an 11-domain, 54-leaf, 1,000-query WebDev benchmark spanning both static-presentation and interactive-application tasks, balanced across three difficulty tiers and three target-language groups, with briefs rewritten to resist recall from circulated prompts. \textbf{\framename}, grounded in Flavell's metacognitive monitoring, separates evidence accumulation from judgment across three stages: Static Perception forms a first impression from passive observation; Agent-Driven Interaction explores the application autonomously while capturing continuous screen video, audio, and per-step screenshots; Dynamic Scoring issues holistic functionality and aesthetics verdicts with structured failure attribution only after the evidence chain is complete. On \dataname, \framename aligns closely with expert human ratings while surfacing substantial headroom across 13 frontier LLMs on interactive web generation. \noindenthttps://anonymous.4open.science/r/Cookie-3CE/

We present FRUC, a feed-forward 3D Gaussian splatting framework for dynamic scene reconstruction from uncalibrated collaborative driving views. Existing multi-agent reconstruction frameworks are often hindered by rigid prerequisites, demanding precise spatial calibration and slow per-scene optimization. In this paper, we rethink this task by conceptualizing a distributed multi-vehicle network as a spatio-temporally unstructured ego-centric multi-camera system, where the core challenge lies in enhancing ego-centric occluded geometry through collaboration without degrading the ego's accurately observed visible geometry, while preserving reconstruction efficiency. For efficient reconstruction, FRUC is built upon a visual grounded geometric Transformer backbone to enable one-shot, calibration-free inference from a flexible number of multi-vehicle views. To achieve non-destructive geometric supplementation under uncalibrated cross-agent misalignment, FRUC first introduces an ego-centric causal occlusion field that explicitly derives occlusion evolution as latent priors by modeling agent-wise spatio-temporal correlations. Guided by these occlusion priors, it further formulates cross-agent integration as a deterministic residual denoising process via zero-initialized injection, turning challenging cross-agent fusion into bounded residual learning for robust collaborative blind-spot completion. Through extensive evaluations on the real-world V2XReal and UrbanIng-V2X datasets, FRUC is shown to be a new state-of-the-art for the scene reconstruction of dynamic collaborative driving environments, significantly outperforming existing methods in both rendering quality and efficiency.

Sentence embeddings are a foundational component for semantic search, clustering, classification, and retrieval-augmented generation. This paper presents embeddingmagibu-200m, a Turkish-focused sentence embedding model that produces 768-dimensional L2-normalized vectors and supports an 8,192-token context window, far exceeding the 512-token limit of earlier BERT-based Turkish encoders. Instead of full pretraining, an efficient three-stage adaptation pipeline is introduced: (1) construct a Turkish-optimized multilingual tokenizer with a 131,072 vocabulary by pruning redundant tokens from the teacher's vocabulary and incorporating multilingual tokens via frequency analysis on a 40-language corpus, (2) clone a teacher embedding model while preserving transformer backbone weights and initializing a compatible embedding table for the new vocabulary via mean-composition token mapping, and (3) perform offline embedding distillation from precomputed teacher vectors using a cosine similarity objective over a balanced 40-language Wikipedia corpus. The resulting student model contains approximately 200M parameters and trains in roughly four hours on a single GPU by avoiding online teacher inference during training, at a total cost of $5-$20. Empirically, Pearson/Spearman correlations of 77.55%/77.45% are obtained on STSbTR, surpassing the 300M-parameter teacher model (73.84%/72.92%). On TR-MTEB (26 tasks), a mean score of 63.9% is achieved (7th out of 26 models), providing a competitive cost-quality trade-off with 33% fewer parameters than the teacher. To facilitate reproducibility and downstream use, all artifacts are released including model weights, tokenizer files, precomputed embedding datasets, and open-source cloning and distillation tooling.

High-fidelity ECG interpretation is increasingly reliant on massive foundation models, yet their deployment in clinical edge-care remains hindered by extreme computational demands. While knowledge distillation (KD) is a promising solution, traditional methods fail to capture the complex spatio-temporal dependencies of ECG signals when transferring knowledge across heterogeneous architectures. In this paper, we propose EVL-ECG, a framework specifically designed for cross-architecture distillation of cardiac diagnostic logic. EVL-ECG introduces three ECG-aware innovations: (1) Multi-Head Cross-Attention Alignment, which harmonizes architectural discrepancies to preserve fine-grained morphological features; (2) Optimal Transport-based Visual Feature Matching, utilizing optimal transport to maintain global structural relationships across ECG leads despite mismatched token representations; and (3) Geometric Intra-Architecture Relation Matching, which distills the latent diagnostic reasoning of the teacher model. Evaluations across ECG benchmarks demonstrate that EVL-ECG yields improvements of up to 2.4% AUC and 1.1% clinical accuracy over existing baselines. Notably, EVL-ECG establishes an efficient 2B-parameter ECG foundation model, suitable for resource-constrained clinical environments.

Marine lead (Pb) and its isotopes are critical tracers for ocean circulation and anthropogenic pollution, yet in-situ observations remain costly and sparse. While vast historical records exist, they lie buried within the unstructured content of academic papers, creating "data silos" inaccessible to comprehensive analysis. Manual extraction is unscalable, while general-purpose Large Language Models (LLMs) lack the necessary domain-specific knowledge, leading to hallucinations and scientifically invalid outputs. To address this, we introduce an expert-guided adaptation approach that enables LLMs to perform rigorous scientific data extraction without fine-tuning. We operationalize this approach through Compass, an LLM agent framework enhanced by a Knowledge Tree co-designed with marine scientists, which decomposes complex tasks into verifiable steps, guiding the agent's reasoning to ensure scientific validity. Deploying Compass across a corpus of over 230,000 relevant open-access papers, we successfully extract 3,751 previously unincorporated Pb records. This effort establishes the largest integrated marine Pb database to date. Beyond standard metrics, Compass demonstrates superior reliability through multi-layered validation, achieving 92% accuracy as confirmed through expert manual verification. The newly integrated data expand coverage in previously under-sampled regions such as the East China Sea and the Southern Ocean, providing an enriched data foundation for future scientific discoveries. We release an interactive visualization platform to facilitate open scientific access. Our work demonstrates that expert-guided agents can effectively bridge the gap between general-purpose LLMs and high-stakes scientific domains, enabling scalable data discovery in geosciences.

Honeypots are decoy systems mimicking real system components designed to defend against cyber attacks. Recently, LLMs increasingly serve as simulation backbones for honeypots. They enable defenders to construct high-interaction honeypots with low system security risks. However, LLM-powered honeypot development lacks a unified evaluation framework. Most evaluations consist of measuring response similarity on fixed commands, manual testing, or real-world deployment. These methods are often not scalable for development, reproducible across evaluations, representative of practical attacks, or adaptable to various attacker and honeypot configurations. In this work, we bridge this gap and propose Honeyval, a comprehensive evaluation framework for LLM-powered HTTP honeypots. We address the limitations of prior evaluations by grounding the honeypots in 16 backend applications, using AI hacking agents as attackers, employing two control tasks to monitor agent and honeypot capabilities across customizations, and defining clear and verifiable exploit goals for the attacker. Using Honeyval, we conduct an extensive evaluation of recent cost-efficient LLMs as HTTP honeypots. Our experiments highlight the promise of LLM-powered honeypots; they lead to substantially longer interactions with the attacker than rule-based baseline honeypots and are far less frequently detected even by frontier models, all while, on average, preserving a running cost advantage against agentic attackers. Further, we experiment with different counter-offensive honeypots configurations, and observe unique trade-offs, such as longer interactions at the cost of increased detection.

Large language model (LLM) agents increasingly leverage long term memory to support persistent and autonomous task execution. However, this capability also introduces a new attack surface: memory poisoning, where adversaries can inject malicious information to influence future behavior. Existing memory poisoning attacks often assume that injected content can be stored directly in memory, overlooking the selective extraction and rewriting stages in modern memory pipelines. This makes prior methods ineffective under realistic settings. In this paper, we propose MemPoison, a novel memory poisoning attack that bypasses selective memory mechanisms in LLM agents, where an attacker can inject triggerable backdoors into the agent's long-term memory through dialogue interactions, thereby misleading its subsequent responses. MemPoison introduces three key components: (i) a semantic relational bridge that binds the trigger and payload into a coherent statement to ensure they are extracted into memory together; (ii) entity masquerading that optimizes triggers to mimic named entities, resisting rewriting; and (iii) joint embedding optimization that shapes trigger-injected texts into a tight cluster in the embedding space while maintaining isolation from benign embeddings for stealth. Evaluations across different agent domains and memory mechanisms show MemPoison achieves attack success rates up to 0.95, outperforming existing baselines. Mechanistic analysis indicates that the attack exploits embedding-space anisotropy and shifts attention patterns, highlighting core vulnerabilities in selective memory systems. We evaluate multiple defense strategies and demonstrate their fundamental limitations in mitigating the attack.

We present AutoformBot, a multi-agent system for building an Autoformalized Textbook Library At Scale (Atlas) in Lean 4. AutoformBot orchestrates thousands of LLM agents, equipped with formal verification tools, dependency-aware task scheduling, and collaborative version control, to translate informal textbook prose into machine-checked definitions and proofs. We apply our methods to a corpus of 26 open-access textbooks spanning analysis, algebra, topology, combinatorics, and probability, producing Atlas: a verified library of over 45,000 Lean 4 declarations and 500 thousand lines of code. We release two artifacts: (i) AutoformBot, the open-source multi-agent framework; and (ii) Atlas, the resulting formal library. Our results suggest that autoformalizing the core content of graduate-level mathematics at scale is now economically and technically feasible. This opens the door to the automated verification of both human- and machine-generated mathematics at a research level.

Sparse vision transformers have gained popularity as efficient encoders for medical volumetric segmentation, with Swin emerging as a prominent choice. Swin uses local attention to reduce complexity and yields excellent performance for many tasks but still tends to overfit on small datasets. To mitigate this weakness, we propose a novel architecture that further enhances Swin's inductive bias by introducing Inception blocks in the feed-forward layers. The introduction of these multi-branch convolutions enables more direct reasoning over local, multi-scale features within the transformer block. We have also modified the decoder layers in order to capture finer details using fewer parameters. We demonstrate a performance improvement on eleven different medical datasets through extensive experimentation. We specifically showcase advancements over the previous state-of-the-art backbones on benchmark challenges like the Medical Segmentation Decathlon and Beyond the Cranial Vault. By showing that the existing inductive bias in Swin can be further improved, our work presents a promising avenue for enhancing the capabilities of sparse vision transformers for both medical and natural image segmentation tasks. Code and pre-trained weights can be accessed at https://github.com/Eiphodos/SwInception.

Understanding how harm emerges from interaction between otherwise benign image-text pairs requires intent-aware cross-modal reasoning beyond surface-level features. Existing vision-language models (VLMs) excel at literal reasoning over perceptual cues but often fail to derive harmful semantics that rely on implicit, context-dependent reasoning. To evaluate VLMs on compositional harm detection and reasoning, we introduce Multimodal Pragmatic Harm Interpretation (MuPHI), a dataset containing image-text pairs where harm is encoded in subtle multimodal cues. MuPHI spans diverse harm categories and includes annotated harm rationales for assessing VLM reasoning chains. To improve both detection and reasoning in VLMs, we propose MuPHIRM, a reasoning-augmented training framework which learns joint semantics by optimizing multi-perspective rewards. MuPHIRM improves both harm detection and reasoning quality of VLMs while demonstrating superior out-of-distribution robustness compared to both trained and inference-time baselines. Our findings suggest that reasoning-oriented reward optimization offers a promising direction towards building multimodal systems that generalize beyond benchmark-specific shortcuts.

Large language models (LLMs) have been widely adopted for synthetic data generation, significantly reducing annotation costs. However, most existing studies treat synthesis as a set of isolated tasks and overlook a more fundamental question: whether a model can learn to synthesize by accumulating experience from past tasks and transferring it to future ones. In this work, we introduce StreamSynth, a new setting in which synthesis tasks arrive sequentially and experience from historical tasks provides informative signals for future synthesis. To address this setting, we propose SynLearner, a general framework that enables synthesis models to acquire reusable synthesis experience over a task stream. Instead of generating data independently for each task, SynLearner encourages the model to explore diverse synthesis patterns, learn from feedback, and balance sample quality with set-level diversity as tasks evolve. Extensive experiments across multiple benchmarks show that SynLearner effectively leverages experience from earlier tasks to improve synthesis performance on later ones, exhibiting consistent cross-task transferability. These findings provide evidence for the feasibility of StreamSynth and highlight synthetic data generation as an experience-driven process that can benefit from task streams.

Diffusion models are effective for waypoint prediction in visual navigation, but standard sampling and test time guidance can produce unreliable or inefficient trajectories when updates drift off the training manifold. We propose Fisher Preserving Guidance with Outer Product Span Projection, a training-free inference method that avoids large Fisher drift associated with off-distribution actions while optimizing a task objective. Our method computes the Fisher-preserving update via a low-rank Jacobian factorization, requiring only a single backward pass per step and enabling real-time use. We further introduce Truncated Fisher Denoising Sensitivity as an uncertainty signal and use it for robust multi-sample action blending. Experiments on toy and realistic navigation benchmarks, including Maze2D with TSDF-based guidance, PushT with official Diffusion Policy weights, and visual navigation in simulation and on real robots, demonstrate consistent improvements in performance over strong diffusion-policy baselines without additional training.

Autonomous driving systems are commonly trained and evaluated within limited geographic regions, which hinders their scalability when deployed in new cities. However, significant domain shifts in appearance, road topology, and traffic patterns often cause severe performance degradation under cross-city deployment. Existing approaches based on domain adaptation, data augmentation, or synthetic data generation typically rely on labeled target data, city-specific annotations, or task-specific designs, limiting their scalability and effectiveness for holistic evaluation. In this paper, we introduce CityTransfer-Bench, a geographically disjoint benchmark for evaluating cross-city generalization across perception, segmentation, and planning, and propose CityGen, a diffusion-based generative framework that performs zero-label city adaptation via HD-map-conditioned synthesis guided by city-level visual prompts. Extensive experiments demonstrate that CityGen consistently improves cross-city robustness across multiple tasks, establishing a scalable and label-efficient foundation for generalizable autonomous driving.

Clustering is a fundamental problem in data science with a long-standing research history, yielding numerous insightful algorithms. Despite this progress, a systematic and large-scale empirical evaluation that jointly considers conventional algorithms, deep learning-based methods, and recent foundation model-based clustering remains largely absent, leading to limited guidance on algorithm selection and deployment. To address this gap, we introduce CLUBench, a comprehensive clustering benchmark comprising 24 algorithms of diverse principles evaluated on 131 datasets across tabular, text, and image data, involving 178,815 experiments. Importantly, our analyses of (i) the impact of hyperparameter tuning,(ii) the impact of data types and characteristics,(iii) the impact of pretrained embeddings,(iv) large language model-based clustering,(v) the similarity of algorithms, and (vi) the low-rank structures of performance matrices, yield meaningful insights and promising pathways for clustering research. For instance, our study reveals that: 1) All evaluated deep clustering methods do not exhibit a significant advantage compared with the top-performing conventional clustering algorithms (e.g., KMeans, SpeClu) in terms of average performance; 2) For image and text clustering tasks, combining pretrained embeddings with conventional clustering algorithms (e.g., KMeans, SpeClu) offers effective and efficient clustering; 3) Clustering remains a challenging and nontrivial problem, even in the era of increasingly dominant foundation models. Moreover, we propose to use the low-rank structure in cross-model performance matrices to efficiently approximate the overall performance evaluation in practical applications. We further demonstrate the feasibility of model selection based on the performance matrices across all hyperparameter configurations.

Mutual misunderstanding in contemporary society does not arise merely because people hold different opinions or values. Even under the same observations, different subjects may form different inferential targets, state representations, prediction errors, and update priorities. This paper proposes a multi-phase inference framework and defines its core internal mechanism as the Multi-Phase Inference Mechanism (MIM). MIM formalizes how heterogeneous world models arise through a phase-formation space, a foregrounding field, subject-specific profile states, and alignment maps between state representations. On this basis, the paper reframes world-model alignment as the problem of making heterogeneous representations mutually processable, rather than forcing agreement or convergence to a single value system. It further connects this formalism to philosophical disagreements, cognitive typology, social fragmentation, and AI alignment. The aim is to provide a constructive vocabulary for AI systems that can help humans understand self and others by making differences in meaning, value, and prediction error visible, comparable, and transformable.

Despite recent advances, LLM-based web agents still struggle with limited exploration, omission of critical steps, and sensitivity to task constraints. Prior work suggests that many of these failures stem from weaknesses in planning, yet the impact of alternative natural language plan representation remains unexplored. To address this, we introduce PlanAhead, a static planner-executor framework that evaluates the impact of plan representation in agent performance. We first automatically categorize WebArena tasks into 3 difficulty levels, enabling consistent difficulty grading without human annotation. Then we systematically evaluate 4 different plan representations on the tasks categorized as hard: sequential subgoals, narrative, pseudocode, and checklist; across different families of multimodal LLM powered agents (OpenAI, Alibaba, and Google). To account for stochastic variability, we introduce two novel evaluation metrics: Achievement Rate (AR) and Solved-Task Consistency (STC). Our results show that both, the plan formulation and the underlying LLM generating the plan, significantly influence web-agent robustness and task success.

Accurate prediction of drug-target interactions (DTI) is critical for drug discovery. Existing methods often rely on single-modal representations (e.g., sequences or graphs) or combine only two modalities, overlooking 3D structural features. To address this challenge, we propose TriMod-DTI, a triple-modal contrastive learning framework that incorporates 1D sequences, 2D graphs, and 3D structures of drugs and proteins, obtaining the universal and complementary feature representations for DTI prediction. We design a Feature Extractor to capture drug and target features across the three modalities, thereby enriching their representations. We further propose a triple-modal contrastive learning strategy to align different modal representations of the same drug or protein in the latent space. By constructing cross-modal positive and negative sample pairs, this approach enhances the model's discriminative ability. Experiments on three benchmark datasets demonstrate that TriMod-DTI outperforms state-of-the-art methods. The ablation studies validate the contributions of each modality. Moreover, case studies highlight its practical potential for DTI prediction and drug discovery.

Contrastive learning is effective for aligning paired views or modalities, but alignment beyond two modalities remains non-trivial and comparatively underexplored. Pairwise CLIP-style losses decompose multi-modal alignment into independent two-way comparisons and therefore do not explicitly model higher-order dependencies among multiple modalities. Recent beyond-pairwise objectives approach this problem from statistical or geometric perspectives, but arbitrary-modality alignment still lacks a principled criterion for defining what each modality should preserve and compress relative to the others. We revisit arbitrary-modality alignment through the Information Bottleneck principle. In multi-modal learning, sufficiency should preserve information predictable from the remaining modalities, while minimality should compress modality-specific information not supported by them. This naturally leads to a One-vs-All view, where each modality is characterized with respect to the remaining modalities. We propose OVA-IB, an Information Bottleneck framework for arbitrary-modality alignment. OVA-IB optimizes a tractable One-vs-All contrastive lower bound for sufficiency connected to a Dual Total Correlation-style objective, uses a parameter-free geometry-aware projection score, and derives a tractable upper-bound regularizer for minimality by bounding each representation's dependence on its own input with representation distributions induced by the remaining modalities. Experiments on classification, regression, modality-agnostic evaluation, and cross-modal retrieval benchmarks demonstrate strong and robust performance.

Evaluating the cultural awareness of large language models is crucial to ensure the fairness of generated text and the generalizability of applications across the world. Recent benchmarks explore cultural goods like food or values like behavior in stressful situations through the lens of question answering or text generation tasks. However, creating these benchmarks requires time-intensive and costly human annotations. Also, benchmarks that evaluate cultural awareness in free text are scarce and often rely on dated evaluation mechanisms. To address this gap, we introduce ExCAM, an Explainable Cultural Awareness Metric, which is, to our knowledge, the first dedicated evaluation metric that identifies, rates and explains cultural errors in instruction-output pairs. To train and evaluate ExCAM, we introduce ExCAM40k, a dataset comprised of nine existing benchmarks that we reformat and enhance with synthetic errors. Compared to several baselines, including GPT-5, ExCAM achieves the highest error detection rate with up to 80% accuracy on a balanced test set. Therefore, ExCAM opens the pathway towards fine-grained and explainable cultural evaluation of free text.

Recent progress in computer vision has produced a wide range of powerful specialized models for detection, segmentation, counting, and other visual tasks. However, these models are usually optimized for isolated task formulations, making it difficult to directly support general-purpose visual intelligence, especially when a task requires complex language understanding and dense small-object perception. In this paper, we propose VisHarness, a trainable visual agent that decouples high-level perception, reasoning, and decision-making from low-level task execution. Instead of training a model to solve a specific visual task, VisHarness learns to harness a set of carefully designed heterogeneous visual experts. This paradigm preserves the general intelligence of the agent while fully leveraging the precision advantages of specialized visual models in concrete visual tasks. With only lightweight training, VisHarness learns a generalizable visual expert-harnessing policy and can solve common fundamental vision tasks under various complex conditions through multi-turn interactions with visual expert models. To enable efficient on-policy reinforcement learning training in a live environment, we introduce dynamic visual memory archiving, which mitigates the rapidly accumulating visual-token overhead caused by multi-turn interactions with visual expert models. Experiments on four representative benchmarks covering reasoning segmentation, generalized referring segmentation, dense small-object detection, and referring counting demonstrate that VisHarness substantially outperforms existing general-purpose models and achieves competitive or superior performance compared with task-specific models.

Recent Large View Synthesis Models (LVSMs) advocate an encoder-decoder architecture that separates reconstruction and rendering into distinct networks. We re-examine this design. Through controlled experiments, we show that a decoder-only architecture, which represents scenes implicitly as a KV-cache, outperforms encoder-decoder variants while using fewer parameters at identical rendering complexity. Further analysis shows that sharing weights between the color-input reconstruction network and the camera-only rendering network better aligns their features at the same viewpoint, facilitating image synthesis. Building on this finding, our model, dubbed DVSM, further incorporates foundation model priors and stage-wise patch sizing for an improved efficiency-quality tradeoff. Our results establish a new state of the art for novel-view synthesis across multiple benchmarks, in some cases even outperforming per-scene-optimized 3DGS under dense input views.

Patient-voiced clinical-triage benchmarks report high under-triage rates for consumer LLMs for constrained multiple-choice output, yet the same cases score differently with free-text. We ask whether output format changes the model's \emph{clinical representation} or only the mapping from a preserved representation to an answer. Using sparse-autoencoder (SAE) features in Gemma 3 4B/12B IT and Qwen3-8B, we find the same medical features fire on the shared clinical narrative under both formats but go {silent} at the multiple-choice decision token in all the cases at every model. Three independent methods (natural-language autoencoder verbalization, decision-token logit attribution, and top-feature characterization) agree that scaffold and format features, but not medical features, drive the decision logits. Behaviorally, the multiple-choice penalty inverts under both structured and natural-language input, option-order shuffle rules out positional bias, and the gap is dominated by off-by-one decision (the model picks an adjacent acuity letter to the gold answer) rather than knowledge failure. Thus, the failure originates in the output format and not in the clinical representation.

Reinforcement learning (RL) post-training has shown to improve reasoning in large language models (LLMs). However, there has been little exploration on the problem of data contamination in RL post-training, potentially undermining generalization and evaluation reliability of the training process itself. Existing detection methods primarily rely on output-level signals such as likelihood or entropy, which become unreliable for RL-trained models since RL shapes behavior through trajectory-level rewards rather than token likelihoods. We propose LaRA, a layer-wise representation analysis framework for detecting contamination in RL post-trained LLMs. LaRA introduces three complementary metrics, measuring perturbation sensitivity, directional collapse, and local representation rigidity under controlled perturbations. We find that contamination produces progressive geometric deviations across layers, including amplified perturbation sensitivity, stronger directional collapse, and enhanced local rigidity. Based on our findings, we also develop a contamination detection protocol that aggregates representation-level deviations across layers and metrics. Experiments on RL-trained reasoning models show that our protocol outperforms existing output-level baselines for contamination detection.

Retrieval-augmented generation (RAG) improves knowledge-intensive question answering by incorporating external evidence. However, existing RAG methods still suffer from hallucinations and subtle reasoning errors. Recent studies introduce external critics to refine RAG outputs, yet they often provide coarse-grained and weakly structured feedback, exhibit over-aggressive intervention, and lead to noisy and unreliable refinement, limiting their effectiveness for correction. To tackle these issues, we propose CRITIC-R1, a structured critic framework that formulates and learns RAG critique as an explicit error diagnosis problem using reinforcement learning (RL). Our framework categorizes common RAG errors into multiple diagnostic dimensions, including verdict, error location, reasoning analysis, and fix generation. To learn these capabilities, we design two reward functions: Conservative Judgement Alignment (CJA) first encourages calibrated high-level judgements while mitigating the over-aggressive phenomenon, whereas Diagnostic Quality Alignment (DQA) further improves fine-grained diagnostic feedback through gated rewards. We train the critic model using GRPO-based RL with process-level supervision collected from external LLM teacher models. Experiments across five QA benchmarks show that CRITIC-R1 consistently improves answer quality over strong RAG baselines. Our source code is available at https://anonymous.4open.science/r/critic-r1-FCB0

Large vision-language models (LVLMs) often hallucinate objects that are not present in the input image, largely because visual grounding weakens as decoding progresses. Existing inference-time mitigation methods modify logits or hidden states throughout generation, but they suffer from three key limitations: they lack an explicit grounding objective, intervene even when the model is already well-grounded, and use fixed correction strengths that do not adapt to the severity of grounding failure. We propose BRACS (Barrier-Regulated Adaptive Closed-form Steering), a training-free steering framework that addresses these issues through barrier-regulated adaptive closed-form steering. BRACS monitors the model's own attention to measure visual grounding and applies corrections to the hidden states only when grounding deteriorates. The corrective update is computed analytically in closed form, requiring no training of auxiliary networks or model retraining. Experiments on LLaVA-1.5-7B and Qwen-VL-Chat show that BRACS consistently outperforms prior methods on hallucination benchmarks, reducing CHAIR$_s$ by 9.4 points and improving POPE F1 by 2.7 points, while matching or improving performance on four general multimodal benchmarks. BRACS also remains efficient, operating at 80% of greedy decoding throughput and achieving 1.3 times higher speed on average than the baselines.

Integrating open-vocabulary semantic information into dynamic 3D scene representations is essential for long-term embodied scene understanding. However, existing methods often suffer from fragile instance association due to incomplete cross-view cues, while their limited ability to handle object-level topological changes restricts long-term robotic task execution. Moreover, current 3D scene understanding methods either rely on simple feature matching without explicit spatial reasoning or assume offline ground-truth 3D geometry. To address these challenges, we present DGSG-Mind, a hybrid instance-aware 3D Gaussian dynamic scene graph system with an embodied reasoning agent. Our system couples a probabilistic voxel grid with explicit 3D Gaussians to enable robust cross-modal instance fusion and incremental semantic mapping. It handles dynamic changes through Gaussian-based visual relocalization and localized masked refinement guided by geometric-semantic consistency. Built on the instance Gaussian map, DGSG-Mind further constructs a hierarchical scene graph and develops the 3D Gaussian Mind, which integrates structural relations, spatial-semantic information, and visually annotated RoI Gaussian renderings for multimodal reasoning. Extensive experiments show that DGSG-Mind achieves the best zero-shot 3DVG performance among methods operating on self-reconstructed maps, while also delivering strong performance in 3D open-vocabulary semantic segmentation and scene reconstruction. We further deploy DGSG-Mind on real-world robots to demonstrate its target-oriented reasoning and dynamic update capabilities. The project page of DGSG-Mind is available at https://icr-lab.github.io/DGSG-Mind

Multi-step robot manipulation requires acting under uncertainty about how the scene will evolve, making exploration and policy adaptation challenging. We study whether short-horizon, task-consistent future videos can provide useful structured priors for control and reinforcement-learning fine-tuning. We formalize this idea through Future-Experience Conditioning (FEC), a simple interface that conditions closed-loop policies on a latent representation of a short future video. In our simulation setup, future clips are generated in three stages, an LLM reasoner operating over a task ontology initialized from the current scene state, a robot-free digital-twin rollout of the intended object motion, and a mask-free video diffusion model that synthesizes a robot-consistent future clip without requiring segmentation at inference. We instantiate this future-conditioning interface primarily with BC and BC+RL, and compare against a future-conditioned Streaming Flow Policy (SFP) baseline on RoboCasa and CALVIN under NoFuture, GTFuture, GenFuture, and WrongFuture. Generated futures improve performance over no-future conditioning, while mismatched futures degrade it, and our BC+RL instantiation achieves the strongest overall results. An average BC+RL learning-curve analysis across 8 CALVIN tasks further shows that GTFuture improves fastest, GenFuture improves earlier and to a higher level than NoFuture, and WrongFuture remains at zero throughout training. These results suggest that short-horizon future videos can serve as useful structured priors for exploration and policy adaptation under imperfect future predictions. https://enact2026.github.io/

Predicting the next mobile application a user will launch is essential for intelligent device resource management and proactive assistance. Existing models rely on fixed app vocabularies, which prevents them from generalizing across different app ecosystems. Many also depend on user-specific knowledge, which complicates deployment in cold start scenarios. We propose STAP, a Transformer-based model that eliminates the need for a fixed vocabulary. STAP replaces true app identities with randomly reassigned virtual indices via a shuffle mechanism, and compensates for discarded semantic information by processing behavioral sequences with an ultra-long context design. A theoretical analysis shows that, given a sufficiently long context, the predicted distribution converges to the correct one despite the anonymity of the mapping. Experiments on two datasets from different continents demonstrate that STAP achieves strong cross-dataset zero-shot prediction accuracy -- a setting where all existing fixed-vocabulary methods are inherently inapplicable -- while its cold start performance within each dataset remains competitive with leading models. Furthermore, we introduce a deployment strategy that enables the model to retain a sufficiently long context during continuous inference while keeping latency within acceptable bounds.

Large Language Models (LLMs) have advanced autonomous agents from deep search, which retrieves concise factual answers, to deep research, which synthesizes scattered evidence into long-form reports. However, verifiable multimodal deep research remains challenging due to open-ended synthesis without deterministic ground truth and the need to interleave textual arguments with visual evidence. We propose \textsc{Ptah}, a multi-agent harness for interleaved report generation. \textsc{Ptah} orchestrates the lifecycle from user query to rendered web report through planning, research, and writing stages, where specialized agents construct visual-aware plans, collect claim-grounded evidence, maintain source-aligned images in a \textit{Visual Working Memory}, and compose reports through declarative multimodal tool use. A verifier agent serves as the harness's acceptance function, enforcing factual grounding, citation fidelity, and cross-modal consistency throughout the workflow. We further introduce \textsc{Ptah}Eval, an evaluation protocol that augments existing benchmarks with image-level and presentation-level assessments. Experiments on deep research benchmarks show that \textsc{Ptah} produces more reliable, visually informative, and usable human-facing multimodal reports than strong baselines.

As a widespread form of informal settlements, urban villages present significant challenges for sustainable urban development and governance. Precise mapping of their infrastructure is essential, however, existing remote sensing datasets primarily focus on formal urban environments, lacking fine-grained annotated data for the high-density building patterns and narrow road networks typical of urban villages. To address this gap, we introduce the \textit{DenseUIS} dataset, the first high-resolution remote sensing dataset specifically designed for building and road extraction in extremely dense urban informal settlements, covering 126 urban villages across Shenzhen and Guangzhou in China. Furthermore, we conduct a comprehensive evaluation of state-of-the-art deep learning models on this dataset. Experimental results reveal the limitations of existing methods in handling the unique morphological patterns of dense informal settlements, underscoring the need for specialized approaches. \textit{DenseUIS} therefore provides a robust benchmark for advancing fine-grained urban mapping in complex and high-density informal environments. The dataset is publicly available at https://github.com/rui-research/DenseUIS.

Reinforcement Learning (RL) has significantly advanced Large Language Models (LLMs) in verifiable domains, but aligning models for open-ended generation remains profoundly challenging due to the lack of definitive rewards. Current rubric-based RL methods mitigate this by employing explicit criteria; however, they rely heavily on static, human-annotated rubrics that inevitably cause policy lag, or expensive external proprietary models for dynamic updates. In this paper, we propose EvoRubric, a novel single-policy co-evolutionary RL framework that eliminates the reliance on static criteria and on external rubric generators. By unifying response generation and rubric generation under a single parameterized policy, EvoRubric dynamically alternates between a Reasoner and a Rubric Generator. To prevent reward hacking and ensure the reliability of generated signals, we introduce a multi-level verification pipeline featuring a meta-verifier, zero-variance pruning, and a Leave-One-Out peer consensus mechanism. Validated criteria are dynamically archived into a memory pool, yielding dense, multi-objective rewards to continuously co-optimize both roles. Extensive experiments across Medical, Writing, and Science domains demonstrate that EvoRubric consistently outperforms traditional static and external-LLM-driven alignment methods. Notably, our framework is compatible with human-expert priors. When initialized with expert-annotated rubrics, EvoRubric can further uncover novel, discriminative dimensions, achieving better performance than relying solely on static expert annotations.

Medical image analysis models can exhibit performance disparities across patient subgroups, threatening clinical safety and fairness. Existing methods typically address this issue by optimizing accuracy and fairness metrics for visible demographic attributes (e.g., sex or age) considered in isolation. This strategy not only overlooks potentially more informative latent stratifications, which may reveal deeper sources of model error and inequity, but also fails to scale when multiple demographic attributes are considered simultaneously due to the resulting sparsity of training data within each subgroup. We deal with these issues by introducing the label-free hidden-cohort fairness (LHCF) training paradigm that instead of maximizing fairness over visible demographic attributes, it optimizes fairness across latent subpopulations discovered from image appearance. By clustering images into K appearance-based cohorts and applying fairness optimization over them, LHCF uncovers underlying sources of model error and avoids the combinatorial sparsity of multi-demographic attributes, reducing disparities across both single and multiple demographic attributes. We demonstrate on our proposed fairness benchmark, HIDFairBench, that LHCF provides state-of-the-art fairness results on single and multiple demographic attributes, despite never using demographic labels for training. Our results position hidden-cohort fairness as a practical, scalable, and robust alternative to demographic-based fairness optimization for trustworthy medical image analysis.

Existing methods in Multimodal Knowledge Editing (MKE) have advanced the ability to correct outdated or inaccurate knowledge in Multimodal Large Language Models (MLLMs). However, they exhibit a critical limitation: while effectively modifying target factual pairs, they fail to generalize edits to logically related queries and often cause unintended alterations to unrelated but visually or semantically linked information. We identify and formalize two underlying failure modes causing this issue: Causal Misalignment, which confines edits to the specific sample, and Feature Entanglement, which causes unintended alterations to coupled but irrelevant information. To address these issues, we propose Localized and Disentangled Knowledge Editing (LDKE), a new framework that achieves precise and generalized editing by localizing fact-specific model layers and disentangling target-relevant inputs from irrelevant ones. Our approach introduces a Fast Localization module to identify and update critical layers efficiently, along with a Disentanglement Classifier that routes inputs appropriately to preserve unrelated knowledge. Extensive experiments across various benchmarks and MLLMs demonstrate that LDKE achieves superior performance in propagating edits to related contexts while maintaining high locality.

The growing number of submitted papers has motivated the exploration of Large Language Models (LLMs) as a means to support and augment the peer review process, particularly in terms of improving its speed and scalability. Yet, it remains unknown whether LLMs engage with scientific manuscripts in the same manner as human reviewers, or whether they merely produce review-looking text. To address this, we introduce the Peer Review AI Benchmark (PRAIB), a novel framework comprising thoroughly defined metrics that measure review specificity, style, and behavior of engagement. To complement the PRAIB framework, we conduct a large-scale empirical study leveraging a dataset of 11,000 reviews generated by five proprietary and open-source models for 1,000 ICLR and NeurIPS papers. Spanning the 2021--2025 period, these machine-generated reviews are compared against original human feedback across diverse prompting strategies to identify systematic behavioral divergences. Our analysis reveals that the generated reviews diverge significantly from feedback provided by human reviewers: LLM ratings are less variable, positively biased, and overconfident, and their cross-reference patterns are model-dependent and distinct from human norms. Furthermore, when evaluated through PRAIB, we observe that LLMs tend to generate longer, more complex reviews, yet frequently overlook the atomic weaknesses noted by human reviewers. By characterizing where and how LLMs reviewing behavior departs from human norms, PRAIB provides the community with a diagnostic tool for identifying which aspects of the review process LLMs can reliably support today and which require further development before deployment.

Data quality is a critical factor in the effectiveness of machine learning models. Label errors, present even in widely used benchmarks, introduce noise into training data and reduce model generalization. In this work, we conduct a comparative analysis of two automatic label error detection methods - Confident Learning and Dataset Cartography - on three Russian text classification corpora of varying size, number of classes, and domain: ru_emotion_e-culture (49,123 examples, emotion classification), RuCoLA (8,524 examples, linguistic acceptability), and TERRa (2,337 examples, textual entailment recognition). We use the pre-trained rubert-base-cased model fine-tuned on each corpus. To verify the meaningfulness of filtering, we conduct control experiments with random removal of an equivalent number of examples. Results show that the effectiveness of both methods depends strongly on dataset characteristics: on large corpora with low noise levels, filtering does not improve performance, while on small datasets with high noise, Confident Learning achieves a significant F1-macro improvement. Dataset Cartography demonstrates more conservative behavior, removing fewer examples. Across all corpora, targeted removal by both methods outperforms random removal, confirming the meaningfulness of the approaches.

Modern open-world agents such as OpenClaw exhibit powerful cross-environment execution capabilities yet introduce broad new safety risk sources. Meanwhile, advanced frontier AI models drastically lower attack barriers, rendering current agent alignment frameworks inadequate for real-world deployment. To tackle these emerging threats, we propose a lightweight and scalable agent safety alignment framework. Specifically, we update the agent safety taxonomy to accommodate emergent risks from Codex and OpenClaw execution scenarios. We further build a taxonomy-guided data engine with influence-function purification to train lightweight AgentDoG 1.5 variants (0.8B, 2B, 4B, and 8B parameters) using only around 1k samples, achieving comparable performance with leading closed-source models (e.g., GPT-5.4). Based on AgentDoG 1.5, we construct a highly efficient agentic safety SFT and RL training environment, which reduces deployment overhead in Docker-level environments by two orders of magnitude. Finally, we deploy AgentDoG 1.5 as a training-free online guardrail for real-time safety moderation. Extensive experimental results indicate that AgentDoG 1.5 achieves state-of-the-art performance in diverse and complex interactive agentic scenarios. All models and datasets are openly released.

Agentic search enables LLMs to solve complex multi-hop questions through iterative reasoning and external search. Despite the effectiveness, these systems often suffer from a critical limitation in practice: agents fail to recognize their own knowledge boundaries, blindly triggering searches when internal knowledge suffices and failing to terminate search even when adequate evidence has been collected. The lack of self-awareness leads to severe \textbf{over-search}, incurring substantial inference latency and prohibitive computational cost. To this end, we propose SAAS, a novel RL framework designed to cultivate dynamic self-awareness that precisely regulates search behavior without compromising accuracy. SAAS introduces three key components: (i) a search boundary modeling mechanism, which identifies the search boundary under the evolving policy by contrasting search-disabled and search-enabled rollouts; (ii) a boundary-aware reward module, which translates this boundary awareness into trajectory-level penalties, suppressing unnecessary and redundant searches; and (iii) a stage-wise optimization strategy, which leverages a sequential curriculum to prioritize reasoning over search regularization, thereby avoiding reward hacking. Extensive experiments demonstrate that SAAS substantially reduces over-search, while maintaining accuracy. Our code is anonymously released at https://github.com/XMUDeepLIT/SAAS.

While Large Language Models (LLMs) can convincingly simulate personas in explicit self-reports, they often deviate in implicit behavioral decisions, revealing a substantial Knowledge-Decision Gap ($G_{\text{KD}}$). Existing benchmarks struggle to measure this asymmetry due to limited construct validity, multi-dimensional entanglement, and distributional biases in LLM-based evaluation. To address these issues, we propose ActTraitBench, a human-grounded evaluation framework for measuring personality consistency in LLMs. Grounded in empirical human data, ActTraitBench establishes one-to-one mappings between psychometric facets and behavioral paradigms, and applies a Distributional Calibration via Quantile Mapping procedure to align LLM-judge score distributions with human norms. Experiments on 14 mainstream LLMs reveal a pervasive knowledge-decision asymmetry, where larger and more capable models often exhibit stronger behavioral divergence despite highly consistent self-reports. To mitigate this gap, we further introduce the Chain of Cognitive Alignment (CoCA), a plug-and-play inference-time intervention that improves alignment in reasoning-capable frontier models while exposing clear capability limitations in smaller architectures.

Reinforcement learning (RL) refines large language models (LLMs) by directly optimizing model behavior through reward signals. While accurate state value estimation is critical for stable training in classical RL, it remains an underexplored challenge in LLM post-training. In this work, we introduce the State Value Estimation Benchmark (SVEB) to assess state estimation within existing RL frameworks and show that critics in standard approaches like PPO collapse to a coarse group-average baseline. To address this, we propose two techniques: Numca, which leverages numerical spans as gradable milestones for state value estimation, and Hista, a framework that uses LLM's hidden states as representation to weighted average disjoint rollouts and their return. Extensive experiments demonstrate that both methods yield more accurate state value estimates and enhance training performance across different RL algorithms and model sizes without incurring significant computational overhead.

Vision-Language Models (VLMs) such as CLIP demonstrate strong zero-shot generalization, but their performance significantly degrades in cross-domain scenarios with scarce target-domain training data (Cross-Domain Few-Shot Learning, CDFSL). In this paper, we focus on the target-domain few-shot finetuning in the CLIP-based CDFSL task. Prevailing finetuning paradigms uniformly align all image patch tokens with their corresponding textual embeddings. However, we find a counterintuitive phenomenon: actively pushing away certain low-similarity image tokens, termed "tail tokens", from their textual embeddings consistently improves target-domain performance. We delve into this phenomenon and provide a novel interpretation: under great domain shifts and scarce training data, the model can hardly extract semantic information from visual inputs; therefore, the common belief of alignment is valid only for tokens already containing sufficient semantic information; for tail tokens, forcing the alignment would lead to excessive overfitting to the scarce training, while breaking the alignment is more useful. Motivated by this, we propose Adaptive Tail-Head Alignment (ATHA), a novel fine-tuning strategy for CLIP that transforms the conventional uniform alignment paradigm to an adaptive alignment paradigm, with both alignment strengthening and weakening. Extensive experiments on four challenging CDFSL benchmarks validate our state-of-the-art performance. Our code is available at https://github.com/shuaiyi308/ATHA.

Intelligent wearable technology plays an increasingly important role in human-computer interaction, motion, and health monitoring. To ensure comfort and practicality of use, one common form for motion monitoring is to utilize soft wearable sensors. However, many research applications regarding wearable sensors are simplistic and difficult to adapt to different situations. This study proposes a system for estimating the angle of the wrist joint using a customized wristband based on an online incremental learning approach. It is a two-stage estimation method: the first stage updates the model based on the wearer's wrist movement characteristics using online learning, integrating real-time data from an IMU as ground truth. The second stage utilizes the updated model for estimation of wrist joint angle solely with the wristband. In other words, model training is completed during data acquisition, allowing the trained model to be used for subsequent angle estimation. This method offers advantages in adapting to data drift caused by variations in different testing configurations, such as the left and right wrists of the same subject, deviations in the wearing position on the same wrist, and even differences among various subjects. The results indicate that the sensors exhibit good performance under strain variations, and the wrist joint trajectory estimation of the proposed system has an approximate error of 15 degree in different scenarios.

Imitation learning has become a cornerstone for solving complex robotic manipulation tasks. In particular, multimodality, which enables robots to capture diverse yet valid behavioral patterns, has driven the rapid emergence of generative policies as a dominant paradigm in robot learning. However, achieving such multimodality typically relies on stochastic noise initialization and iterative denoising procedures, resulting in substantial training complexity and low inference efficiency. Meanwhile, not all phases of a robotic task inherently require behavioral diversity. Motivated by this insight, we propose the Modality-Adaptive Robot Sampling (MARS) policy, which adaptively invokes tailored stochasticity only when it is truly beneficial, while reverting to an efficient deterministic learning during single-modal phases. In other words, the proper amount of noise is injected only at the proper time. By selectively activating multimodal generation, MARS policy bridges the gap between the multimodal capability of generative policies and the superior training and inference efficiency of deterministic models. Empirical studies across 8 simulated and 4 real-world tasks demonstrate that MARS exhibits robust multimodal expressivity and high efficiency, with a 16.67% success rate improvement and an 83.20% inference latency reduction in real-world tests. Counterintuitively, MARS also outpaces deterministic policies in training efficiency on near-deterministic tasks by more effectively modeling nuanced action diversity.

Video Motion Magnification (VMM) reveals imperceptible dynamics but often suffers from structural inconsistencies under complex geometric transformations. Existing learning-based methods generally face a trade-off between the limited global context of CNNs and the high computational cost of Transformers. In addition, current training protocols, largely dominated by simple linear motion, fail to capture the geometric and imaging complexities encountered in real-world videos. To address these issues, we propose GeoMag, a geometric-aware VMM framework built upon State Space Models to achieve globally consistent motion amplification with linear complexity. We further construct Geo-200K, a large-scale synthetic dataset that introduces rich geometric transformations together with sensor-realistic degradations, improving the diversity and realism of training signals. Extensive experiments on synthetic and real-world benchmarks show that GeoMag consistently outperforms prior methods in visual fidelity and computational efficiency, while producing fewer artifacts and better structural consistency.

The impressive performance of generalist large language models (LLMs) such as GPT and Claude in healthcare raises a critical question: will domain-specific medical specialist models become obsolete? We argue that the future of medical artificial intelligence (AI) lies not in building monolithic medical foundation models, nor in replacing human expertise, but in orchestrating collaboration among generalist LLMs, domain-specific specialist models, and clinicians. We propose HetMedAgent, a heterogeneous medical multi-agent framework that enables conflict-aware evidence fusion, uncertainty-based clinician intervention triggering, and adaptive threshold calibration. Experiments on three real-world clinical decision-making tasks demonstrate that the synergy between generalist LLMs and domain-specific specialist models significantly outperforms using either type of model alone, validating the irreplaceable value of specialist models in modality-specific analysis. HetMedAgent represents a shift from building medical LLMs or foundation models to multi-agent collaboration, achieving a balance between general reasoning capabilities and domain-specific precision.

The dominance of colonial languages in African education and scientific communication limits how hundreds of millions of speakers of African languages access and produce scientific knowledge. A core obstacle is the lack of established scientific terminology in these languages. We introduce AfriScience-MT, a parallel corpus covering six African languages (Amharic, Hausa, Luganda, Northern Sotho, Yorùbá, and isiZulu) across 11 scientific domains. Professional translators, working with expert science communicators, translated plain-language summaries of scientific papers into each target language and created new terms where none existed. We benchmark machine translation systems and large language models in zero-shot, few-shot, and fine-tuned settings. Our results show that closed-source models outperform all open-source models at both the sentence and document levels: GPT-5.4 and Gemini-3.1-Flash-Lite lead with average sentence-level COMET scores of 68.3 and 68.0, respectively, and tie at an average document-level COMET of 48.3. Among open systems, fine-tuned NLLB-1.3B reaches 67.3 at the sentence level, and TranslateGemma-12B reaches 44.0 at the document level with 1-shot in-context learning. We release AfriScience-MT to support benchmarking and document-level scientific MT for African languages.

Legal NLP benchmarks overwhelmingly evaluate a single language or aggregate tasks that differ fundamentally across jurisdictions, making cross-lingual comparison impossible. We introduce Multi-Legal-Bench, the first cross-jurisdictional legal benchmark that evaluates identical tasks across six countries (Ukraine, France, Netherlands, Poland, Czech Republic, Lithuania), four language families, and 134 million court decisions. The benchmark defines five tasks court-type classification, judgment form classification, case-outcome prediction, legal norm extraction, and cause category prediction mapped to structured metadata from national court registries, forming a deliberately sparse 5x6 task-jurisdiction matrix (20 of 30 cells filled). We evaluate 7 frontier LLMs under zero-shot and 3-shot prompting via AWS Bedrock, with 4 additional small/medium models (3-12B) for scaling analysis. Our results reveal that: (1) task-dependent few-shot effects discovered in Ukrainian replicate across all jurisdictions; (2) no single model dominates any language rankings shift with both task and jurisdiction; (3) cross-lingual few-shot transfer does not follow language proximity: UA->FR (Romance, -2.1 pp) transfers better than UA->PL (Slavic, -13.7 pp), with label-set alignment predicting transfer quality better than language family; and (4) tokenizer fertility, despite a 2.3x spread, does not significantly predict cross-lingual accuracy (r=-0.27, p=0.14), suggesting that model architecture and pretraining data dominate tokenizer efficiency. We release all data, prompts, and model predictions.

High-performance GPU kernels are essential for efficient LLM deployment, yet optimizing them remains expertise-intensive. Recent LLM-based code generation makes automatic GPU operator generation promising, but operator optimization remains a hardware-aware search problem. Existing LLM-based methods face a granularity mismatch: coarse hints are reusable but hard to execute, whereas detailed memories are actionable but enlarge the search space and obscure optimization bottlenecks. The key challenge is therefore to organize optimization experience at an appropriate granularity. To address this issue, this paper proposes HTAM (Hierarchical Transition-Attended Memory), a coarse-to-fine framework for LLM-based operator optimization. HTAM builds a two-level Hierarchical Transition Graph (HTG) to organize coarse global directions, detailed local strategies, and transition experience between optimization steps. During each evolution step, HTAM selects a global direction from the current state and recent optimization history, retrieves the corresponding local strategy memory, and uses it to guide concrete CUDA code generation. Experiments on the full KernelBench suite demonstrate that HTAM consistently improves correctness, fast-solution rate, and speedup over LLM-based baselines, while backend and Robust-KBench studies indicate transferable benefits from structured memory.

Emotional support plays an important role in dialogue systems, and its success depends on adapting to a user's evolving and implicit needs across multi-turn interactions while leveraging the strong reasoning capacity of large language models. However, since signals about user needs are often weak, indirect, and can only be disambiguated through multi-turn interaction, existing emotional support methods often struggle to acquire and generalize relevant conversational knowledge efficiently. To bridge this gap, we introduce User-Aware Active Knowledge Acquisition (UKA), a gradient-free active dialogue learning framework that explicitly represents uncertainty about user needs and incorporates active learning into both knowledge acquisition and response selection.We propose a Theory-of-Mind uncertainty estimation mechanism that allows the model to prioritize responses, thereby eliciting more informative user feedback. UKA is capable of efficiently exploring user-aligned conversational knowledge during training while maintaining robustness at test time. Experiments across multiple dialogue benchmarks and model architectures demonstrate that our approach consistently outperforms strong baselines in dialogue quality and user alignment.

Grounded claim factuality checking is important for large language model (LLM) applications such as retrieval-augmented generation, as it helps users assess the correctness of generated outputs. Existing metrics using entailment classifiers require dataset-specific threshold tuning, while LLM-based approaches often use direct prompting, which underutilises the reasoning capabilities of LLMs. We address this by formulating grounded claim factuality checking as a true/false reading comprehension task and prompting LLMs with explicit test-taking strategies for efficient reasoning. Our method reduces token usage by over 80% compared to unguided open-ended reasoning, and achieves competitive performance to more expensive alternatives across two factuality benchmarks, setting a new state of the art on one. To further reduce inference cost, we train small language models (SLMs) to replace LLMs in the checking pipeline. Using supervised fine-tuning (SFT) and a self-revision mechanism, the SLMs learn to improve their factuality judgements. Experimental results show that the resulting SLMs perform on par with strong baselines, combining low inference costs with generating supporting rationales to support interpretability. Code and datasets will be released upon acceptance.

User satisfaction with AI assistants is highly personalized: the same response may satisfy one user but disappoint another depending on what each user expects and what they have asked for before. Existing automatic evaluation methods mostly measure generic response quality, making it difficult to judge whether a response satisfies a user at a specific turn. We study this problem as personalized turn-level user conversation satisfaction evaluation. We build a conversation satisfaction evaluator that combines compact user memories with target-turn context to produce satisfaction scores and dissatisfaction-oriented rationales. Meta-evaluation against human satisfaction annotations shows that personalized memory and post-hoc score calibration improve ordinal agreement and dissatisfied-turn detection over supervised, retrieval-based, and generic LLM-as-a-judge baselines. We further introduce PersTurnBench, a personalized turn-level user conversation satisfaction benchmark that uses the verified evaluator to assess generation models via replay. By holding the replay state fixed, PersTurnBench enables controlled comparison of generic generation models and memory-augmented personalized systems without new human labels for every candidate model. The evaluator and benchmark let researchers compare candidate generation models on personalized satisfaction without collecting new user feedback for every model.

Real-world evaluation of vision-language-action (VLA) policies still rests on binary success rate at a fixed timeout with $N \le 25$ rollouts per condition, almost always without confidence intervals or paired statistical comparison; these cohort sizes struggle to resolve close comparisons reliably. We introduce PhAIL (Physical AI Leaderboard, https://phail.ai), an open real-robot benchmark on a Franka FR3 (dataset, per-rollout artifacts, and end-to-end reference implementation) of a distributional evaluation methodology: the time-to-success cumulative distribution function (CDF) as the evaluation primitive, with two separated jobs. The first is scoring via Human-Relative Throughput (HRT), a dimensionless scalar with bootstrap confidence intervals, anchored to same-fixture human teleoperation. The second is a significance test (Kolmogorov-Smirnov, computed per-object and macro-averaged across objects). On four publicly-available VLAs, the macro-averaged KS test resolves two close comparisons (GR00T vs. ACT, OpenPI vs. ACT) at $N \le 30$ rollouts per (model, object) cell where binary-threshold metrics do not; the closest pair (OpenPI vs. GR00T) remains unresolved within our budget. The best evaluated VLA is $\sim 7\times$ slower per operation (RMST ratio) than the human reference.

Traditional large-scale formation planning either oversimplify the formation representation which leads to poor performance, or they employ complete collaborative relationships, which results in excessive computational load. To achieve high-performance and large-scale formation planning, we transform the Optimal Formation Position Sequence \cite{c1} (OFPS) calculation problem into a spatiotemporal Point Cloud Registration (PCR) problem. Each agent derives its OFPS by distributively computing the matching result between current positions and the desired formation positions of all other agents. Then each agent optimizes the cooperative formation trajectory by using OFPS. We leverage the PCR method with outlier rejection to rapidly perform large-scale formation position registration. This prevents suboptimal trajectories and failed agents from propagating through the cooperative network and affecting more agents. Consequently, we uniformly achieve resilient, efficient, and distributed trajectory planning for large-scale swarms. The effectiveness and the superiority of the proposed method are demonstrated through large-scale simulations of 120-drone formation, and rigorous benchmarking against state-of-the-art (SOTA) methods.

Self-supervised learning (SSL) has produced a diverse landscape of vision transformers (ViTs) whose pretrained representations support a wide range of downstream tasks. Towards a better understanding of these models, a body of work has assessed the mechanics of their self-attention as well as the types of information captured across their representations, revealing, for example, stark differences between models trained with contrastive learning (CL) and masked image modeling (MIM). However, these advances in model understanding have not yet fully permeated the broader community, where insights specific to CL models are sometimes generalized to MIM models. To make model understanding straightforward and intuitive for a broad audience, we propose a simple and easily interpretable visualization protocol. Our protocol is based on visualizing unsupervised semantic segmentation results, yet our goal is not to maximize segmentation performance. Instead, it allows us to convey model behaviors that consistently emerge across images. Benchmarking a diverse set of SSL models across layers and representations, we obtain novel insights into distinct positional biases and scaling behaviors, including strong boundary artifacts in DINOv3-Large model tokens. These insights complement and help communicate a range of previous findings. Our protocol further enables a clear visual distinction between positional effects and the closely related but distinct locality bias, which has been studied much more extensively in the literature. The protocol is publicly available on GitHub and we believe it will catalyze further model understanding for a broad community.

Agent harnesses increasingly determine the performance of language-model systems by deciding how models call tools, receive feedback, verify intermediate states, store memory, and revise solutions. Yet current test-time scaling analyses often parameterize this process by raw expenditure -- tokens, tool calls, operations, wall time, or cost -- which does not distinguish useful feedback from redundant or unstable interaction. We introduce \emph{Effective Feedback Compute} (EFC), a trace-level scaling coordinate that credits feedback only when it is informative, valid, non-redundant, and retained for subsequent decisions, and we normalize it by task demand when comparing tasks with different feedback requirements. Across synthetic controllable tasks, executable code tasks, real benchmark traces, held-out splits, and a prospective validation batch, EFC-based coordinates consistently predict failure rates better than raw-compute baselines and a strong multivariate SAS baseline. In controlled scaling, raw tokens and tool calls explain limited variation ($R^2=0.33$ and $0.42$), SAS reaches $0.88$, while Oracle-EFC and Estimated-EFC reach $0.94$ and Oracle-EFC/$D_{\mathrm{task}}$ reaches $0.99$. Matched-budget interventions show that improving feedback quality raises success from $0.27$ to $0.90$ while raw cost and tool calls are fixed. On mixed real traces, NRS-EFC/$D_{\mathrm{task}}$ reaches $R^2=0.92$ while raw compute has near-zero or negative fit, and it remains the best predictor in a prospective holdout ($R^2=0.85$). These results suggest that harness scaling is governed less by how much computation is spent than by how efficiently raw budget is converted into durable, task-sufficient feedback.

Large language models are highly sensitive to prompts, but this sensitivity is usually studied through task-relevant instructions, demonstrations, or reasoning cues. In this paper, we study a different form of prompt sensitivity: whether prompts that are semantically unrelated to the task can nevertheless steer model behavior. We call them spurious prompts and show their surprising efficacy. We also propose a simple black-box search procedure for discovering them. Across reasoning and question-answering benchmarks, using models ranging from 0.8B to 27B parameters and spanning three model families, we show that spurious prompts can improve performance, often matching or outperforming standard prompting baselines and task-aware prompt optimization. We further show that they can steer models toward unintended behaviors, such as repeatedly selecting the first answer option, producing incorrect answers, returning an even, prime or small number without explicitly instructing the model to do so. These findings reveal a new kind of prompt sensitivity: LLMs can be systematically steered by prompts that are unrelated to the task they are asked to solve. Our code is available at https://github.com/Batorskq/spurious

Large language models in Agentic AI systems consume tool schemas and execution results and emit tool invocations as structured data. The default language for that exchange, JSON, was designed for application-to-application interchange rather than token efficiency, so its structural elements impose substantial token overhead. Recent work proposes token-optimized alternatives such as TOON (Token-Oriented Object Notation) and TRON (Token Reduced Object Notation) as more compact replacements, but these formats have been evaluated only on isolated comprehension or generation tasks. Whether their token reductions hold inside end-to-end agentic loops therefore remains an open question. We evaluate TOON and TRON on four agentic benchmarks (BFCL, MCPToolBenchPP, MCP-Universe, StableToolBench) and five open-weight LLMs, decoupling input compression from output compression to measure comprehension and generation independently. TRON reduces tokens by up to 27% with accuracy within 14pp of the JSON baseline. TOON achieves up to 18% reduction at a similar 9pp accuracy cost, but additionally cascades on multi-turn parsing failures and collapses parallel tool-call output for most models.

Reconstruction-based inference assigns a class by comparing class-wise reconstruction residuals; Sparse Representation Classification (SRC) is a canonical instance whose reliability depends on the geometry of the learned representation. We adopt a strict training-inference separation: SRC is used only as a fixed test-time rule and is never differentiated, unrolled, or optimized during training. In a span-level idealization based on class-conditional spans and their associated projection residuals, we formalize residual-ordering stability through a residual margin and characterize geometric obstructions -- span overlap, dominance, and near-overlap via small principal angles -- that can collapse this margin in worst-case directions. This span-level theory is primary: it specifies when the idealized residual family is well-separated, and it provides a conditional solver-level interpretation for practical residual approximations (e.g., OMP) insofar as they remain close to the span-level residual ordering. Under explicit coverage and separation assumptions, we derive a quantitative lower bound on the (idealized) residual margin. Guided by these targets, we propose geometry-shaping objectives that promote masked within-class self-expressiveness, discourage cross-class reconstruction pathways and inter-class span alignment, and prevent collapse -- without invoking SRC residuals or predictions during training. Experiments on images (COIL-100), text (TREC), and EEG connectivity evaluate all representations under identical fixed SRC/OMP inference and report residual margins and geometric diagnostics; cross-entropy is included only as a reference geometry under the same evaluation protocol.

LLM agents acting in structured environments fail in operational rather than conversational ways, and reliability depends on procedural knowledge of the environment. Prior self-improvement methods accumulate natural-language guidance without checking that each new item preserves previously correct behavior, so a note that fixes one trajectory can silently regress another. We introduce GRASP (Gated Regression-Aware Skill Proposer), which treats agent improvement as a sequence of edits to a bounded skill library, admitting each candidate only if it produces a net improvement on a balanced held-out probe under a hard regression budget. We evaluate GRASP across five base models (gpt-oss-120b, DeepSeek V4 Flash, Gemini 3.1 Flash Lite, GPT-4.1, GPT-5.4) on two FHIR-based clinical benchmarks. On MedAgentBench, GRASP lifts gpt-oss-120b from 40.6% to 88.8%, exceeds the strongest of five self-improvement baselines by 21.0 points, and improves every other base model by 17.2 to 40.3 points. Ablations attribute the gain to comparative proposal generation, the acceptance gate, and the hard regression budget rather than to skill writing itself, which without validation is no better than using no skills. The mechanism generalizes beyond the clinical domain, improving agents on three of four non-clinical environments and remaining flat only where the action space is open-ended. Frozen libraries transfer across models, where skills from a stronger model improve weaker executors beyond what they learn for themselves while the reverse does not, an asymmetry that no ungated baseline reproduces.

When Large Language Models (LLMs) are deployed in Chinese-language settings, a troubling pattern emerges: safety systems that work well in English break down. These systems struggle to cross linguistic and cultural bound-aries, leaving models exposed to adversarial prompts that exploit Chinese-specific evasion techniques, including Pinyin romanization, character decomposition, internet slang, and hedging tone. To address this gap, we introduce ChiSafe-PAS (Chinese Safety Pilot Annotation Set), a human-annotated benchmark of 1,897 adversarial Chinese prompts spanning four high-stakes domains: self-harm and violence, drug and illicit trade, fraud, and satire. Of these, 1,544 entries carry complete gold-standard annotations: a 3-class response label (REFUSE, SAFE-REDIRECT, RESPOND), a nine-category obfuscation taxonomy, a risk-level rating, and annotator rationale. We describe the dataset design, annotation process, and obfuscation taxonomy in detail. Our primary goal is practical: to give the research community a high-quality, culturally grounded resource for benchmarking LLM safety alignment. In doing so, we engage three broader tensions in the field: the blurring boundary between training and evaluation data, the need for domain coverage grounded in real-world risk, and the limits of scale as a substitute for cultural expertise.

Ground robots often carry payloads, implements, or other attachments that turn their effective footprint into complex, non-convex shapes. Navigating safely through clutter then requires reasoning about this true geometry, yet most local planners simplify it with convex or inflated proxies and rasterize sensor data into occupancy grids or distance fields. Both choices eliminate feasible motions when clearance is comparable to the footprint geometry. We present EXACT-MPPI, a training-free local navigation framework that maps local point-cloud observations and sparse guidance directly to motion commands, without any intermediate map representation. The framework embeds an analytic, exact signed-distance evaluator into a Model Predictive Path Integral (MPPI) controller. The footprint is represented as a simple polygon for general convex or concave planar shapes, with a rectangle-cover specialization for faster evaluation of rectilinear footprints, enabling footprint-aware collision costs without convex decomposition, inflation, or learned encoders. During each MPPI rollout, observed obstacle points are transformed into the predicted body frame and evaluated against the footprint. All operations are batched in JAX, leveraging GPU parallelism for real-time receding-horizon control. Experiments show that EXACT-MPPI accelerates batched distance evaluation over a learned point-to-robot baseline, preserves feasible motion where convex-footprint planners fail, and remains robust under dense static and moving obstacles. The same framework deploys on differential-drive, Ackermann, omnidirectional, and hybrid-mode platforms by changing only the footprint description and motion model without per-platform training. Pairing exact footprint geometry with sampling-based predictive control thus offers a practical, training-free path to footprint-aware local navigation across diverse robots.

Real-time inference of vision-language-action (VLA) models is essential for robotic control. While visual token pruning has shown strong potential for accelerating inference, most existing methods mainly base pruning decisions on shallow-layer cues and risk discarding visual information required by deep layers. To address this issue, we propose SAFE-Pruner, a plug-and-play pruning framework that incorporates attention cues of future layers into pruning decisions. Specifically, we identify semantic attention consistency, the tendency that VLA models concentrate their attention probability mass on the same semantic entity across execution steps. Based on this observation, we design a forward-looking strategy to forecast the token saliency in deep layers, which prevents the premature removal of critical tokens and leads to more stable acceleration. We further introduce an adaptive subtask division strategy to detect abrupt attention shifts, thereby improving forecasting accuracy and pruning reliability. Extensive experiments in simulation and real-world settings demonstrate that our method achieves up to 1.89x speedup with a minimal degradation in success rate of less than 1.7%, while outperforming state-of-the-art methods by up to 1.9%.

Monocular 3D shape recovery is fundamental to geometric understanding, yet achieving robust generalization across arbitrary viewpoints and unseen object categories remains a significant challenge. In this paper, we present a generalizable deformation learning framework that reconstructs 3D objects by explicitly deforming a category-level shape template to match the target observation. To address complex shape variations between the template and the target, we introduce a geometry-guided feature modeling mechanism. This process first enriches foundation features with template topology to yield a geometry-aware representation, which is then explicitly correlated with the target observation to guide precise deformation. Furthermore, to bridge the disparity between the fixed template and arbitrary target views, we propose a view-adaptive feature aggregation module. This module leverages multi-view template features and their corresponding camera poses to enrich the canonical template representation, ensuring robust feature alignment regardless of the target's perspective. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods in handling large shape variations and diverse viewpoints, exhibiting strong generalization to novel categories and effectively supporting downstream real-world dexterous robotic manipulation tasks. Project homepage: https://GODeform.github.io/

Real-time safety filtering for large language model (LLM) applications requires classifiers that can detect unsafe prompts, toxic language, jailbreak attempts, and unsafe responses without the cost profile of large guardrail models, and that can distinguish benign sensitive text from genuinely covert harmful content. In this paper, we introduce Opir, a family of encoder-based guardrail models built on the GLiClass architecture. Opir includes multi-task models for binary safe/unsafe classification, multi-label toxicity classification, jailbreak classification, and zero-shot unsafe prompt and response categorization. We also release edge variants with fewer than 100M parameters dedicated to binary safe/unsafe categorization. The models are trained on a three-level taxonomy containing 996 categories across 16 top-level labels, 126 mid-level labels, and 854 leaf labels. Opir's training data combines taxonomy-grounded unsafe prompts, adversarially mined hard negatives, benign safety-preserving examples, generated response examples, multilingual translations, and portions of the Aegis2 and WildGuard training subsets. We also open-sourced an evaluation harness that supports GLiClass and GLiNER2 backends as well as decoder-based models, and covers binary safety classification, multi-label categorization, toxicity, jailbreak detection, prompt safety, response safety, response refusal, and prompt subcategory views across public benchmark families. Across an expanded comparison spanning 12 safety-classification tasks and 17 category tasks against eight contemporary guardrail systems -- including both GLiNER2-based and generative guardrail models -- Opir variants are competitive on or ahead of the strongest open-weight baselines on the majority of benchmark datasets while operating with a substantially smaller deployment footprint.

Applying reinforcement learning to improve factual accuracy in knowledge-intensive question answering faces a reward design dilemma. Response-level rewards provide only coarse supervision and cannot distinguish correct from incorrect statements within a reasoning trace. Sentence-level alternatives offer finer-grained feedback, but typically rely on NLI verifiers, LLM judges, or knowledge-verification pipelines that are expensive to deploy at RL scale and often unreliable for rare-entity facts, where accurate reward signals are especially important. We propose CorVer (Corpus Verify), a lightweight, plug-in-ready process reward that replaces neural verifiers with a corpus-grounded signal derived from Wikipedia co-occurrence statistics. CorVer assigns sentence-level credit and maps it to token-level advantages via a simple alignment, requiring only a 0.5B extractor and a single corpus lookup per sentence. Across 30 (model, benchmark) cells spanning six instruction-tuned models (3B to 14B) and five QA benchmarks, CorVer improves over the raw baseline for every cell, with an average TriviaQA gain of +4.1 pp. It also outperforms four neural-verifier baselines in 18 of 20 cells under their feasible configurations, while training 4.8 to 8.4x faster.

Cross-Video Reasoning (CVR) has emerged as a critical frontier in multimodal intelligence, requiring models to retrieve, align, and aggregate evidence distributed across multiple videos. Current Multimodal Large Language Models (MLLMs) often struggle with CVR, as simple single-pass strategies encode multiple videos into a shared compressed context, potentially obscuring rare but critical evidence. In this paper, we propose AgentCVR, a multi-agent framework that treats CVR as an active evidence-acquisition task. AgentCVR employs a Master Agent to iteratively coordinate specialized Visual and Audio Agents for targeted evidence extraction. To ensure efficient training, we introduce Script-Simulated RL, which optimizes the agent's policy with LLM-generated semantic scripts and a lightweight text-based simulator, bypassing costly multimodal inference during online exploration. Experimental results on a comprehensive CVR benchmark show that AgentCVR outperforms single-pass baselines and achieves comparable performance to state-of-the-art closed-source systems, particularly in complex cross-video alignment and localization. To ensure reproducibility, our code is available at https://github.com/wang-jh24/AgentCVR.

Large language models recall knowledge reliably in English but often fail on the same query posed in a lower-resourced language -- a crosslingual consistency gap that remains underexplored for Indian languages and their code-mixed counterparts. To study this gap, we introduce IndiKLAR, an Indic extension of the KLAR-CLC benchmark covering 18 of the 22 scheduled Indian languages and pairing them with code-mixed variants for 11 widely used language pairs, with native-speaker verification of both monolingual and code-mixed variants for these 11 settings. This three-way alignment offers a unique opportunity to examine how knowledge recall consistency varies across the spectrum of English, code-mixed, and native Indian language inputs. Evaluating across nine open-weight models, we find that the native-language accuracy gap to English can reach $\sim$0.50, while code-mixed inputs close most of it -- bringing performance within $\sim$0.05 of English without any model-level intervention. Motivated by this, we evaluate several prompting strategies that vary in how language conversion is exposed, including a two-stage translate-then-answer setup, a one-stage joint translation-and-answer prompt, and Translate-in-Thought (TinT) -- a single-step strategy in which the model converts the input internally and emits only the final answer. Across the performance trajectory native $\rightarrow$ code-mixed $\rightarrow$ English, we identify a consistent flip point -- the boundary between incorrect and correct prediction -- that lies between the native and code-mixed settings. Interestingly, this holds whether the trajectory is induced by the input surface form or by the model's internal conversion process.

Randomized controlled trials are a cornerstone of medicine and the social sciences as they enable reliable estimates of causal effects. However, they are costly and time-consuming to conduct, motivating interest in predicting causal effects from existing experimental evidence. Recent advances in large language models (LLMs) have demonstrated strong performance on knowledge-intensive tasks, raising the question of whether these models can be used for forecasting causal effect sizes. To investigate this, we introduce Query2Effect, a new large-scale benchmark consisting of more than 72,000 natural language questions aligned with experiment descriptions, created to simulate realistic information-seeking scenarios by varying query specificity along dimensions of implicitness, abstraction, and ambiguity. We then propose a two-step framework that first generates a synthetic structured representation of a query before predicting effect size using a supervised encoder model. Experiments show that finetuning plays a crucial role in improving prediction performance, with absolute error reducing by -27% up to -71% compared to prompted out-of-the-box LLMs, and that our two-step framework is beneficial for out-of-domain generalization, highlighting the benefits of separating semantic interpretation from numerical effect estimation.

End-to-end agent-memory benchmarks report a single hit@k per retriever, confounding lexical leakage (uncontrolled query/gold/distractor entity overlap) with tag-mixing (preferences, services, tools averaged together). We propose entity-collision, a system-agnostic protocol that pins the BM25 floor by construction -- every distractor shares the answer's entity tokens -- and stratifies queries by discriminator tag, so any lift over BM25 is attributable to the embedder. Applied to an open-source agent-memory testbed across 5 tags x 3 embedders x 5 collision degrees with paired-bootstrap 95% CIs, the protocol reveals a two-axis pattern: a 256-d hash trigram helps only on closed-vocabulary lexical tags at deep collision; MiniLM-384 dominates both axes; and a 2.7x-parameter BGE-large does not uniformly improve on MiniLM -- it wins on intent-style queries but loses on lexical ones. Encoder capacity alone is not the binding constraint. The synthetic intent-tag null replicates on LongMemEval (n=500) as a single-session-preference recall cliff. Adaptive vector-weight routing on LoCoMo is a measured null: 11.7pp of oracle headroom exists, but no signal we tested recovers it. All 26 result tables and 37 reproduce scripts are version-controlled and verified by a public registry; the protocol is exercised on a deterministically governed memory testbed (event-sourced decision log, DAG-state-machine schema lifecycle) so every reported CI is reproducible byte-for-byte from the ingest stream.

Contrastive Language-Audio Pretraining (CLAP) models are widely used for audio understanding and support modality-agnostic condition swapping in many zero-shot applications. However, their performance is heavily affected by the modality gap between audio and text embeddings. Existing explanations mainly attribute this gap to the cone effect, treating it as a shift between mean embeddings, yet correcting the mean alone yields only limited improvements. Alternative hypotheses, such as information imbalance and dimensionality collapse, have also been proposed, but they remain insufficiently verified and have not been thoroughly studied in the audio domain. Meanwhile, several works attempt to decompose multimodal contrastive embeddings into interpretable concepts, but none explicitly analyze the modality gap from the perspective of concept decomposition. In this work, we introduce COMET (Concept space Organization and Modality gap Explanation with PLS-SVD Transformation), a novel partial least squares singular value decomposition (PLS-SVD) framework for CLAP that unveils a broader perspective of the modality gap. Our framework reveals that only a small, interpretable subset of axes, which captures shared concepts, contributes substantially to similarity computation, and that the mean component represents only partially the modality gap. Building on this insight, we propose a simple spectral truncation method that mitigates the modality gap in a training-free manner. The method enables zero-shot audio captioning with condition swapping to approach fully supervised performance, without requiring large auxiliary memory banks or expensive computation. At the same time, it achieves substantial embedding dimensionality reduction while preserving strong performance on retrieval and audio captioning tasks.

Vision-language models (VLMs) have made strong progress on high-level image-text alignment, yet their ability to perceive subtle visual differences remains limited. We study this problem in rendered web interfaces, where localized visual changes are both a diagnostic test of fine-grained perception and a practical requirement for GUI agents and design tools. We introduce \textbf{DiffSpot}, a code-driven benchmark for open-ended spot-the-difference on web interfaces. DiffSpot constructs controlled image pairs by mutating a single CSS property of a target element in self-contained HTML, re-rendering the page, and recording the changed property, element, and mutation magnitude. A grounding gate retains only pairs whose rendered pixel difference is confined to the target element. The benchmark contains 4{,}400 pairs, including 3{,}900 has-diff pairs balanced across 13 CSS-property operators and three difficulty tiers, plus 500 no-diff pairs for hallucination control. Evaluating 13 frontier VLMs zero-shot, we find that even the best model identifies only $40.7\%$ of true changes, with Hard-tier Recall below $23\%$ for every model. DiffSpot further shows that difficulty is strongly property-dependent: across CSS operators, neither pixel magnitude nor CLIP distance reliably predicts Recall.

Although large language model (LLM) based multi-agent systems (MAS) show their capability to solve complex tasks and achieve higher performance over single agent systems, they lead to huge computational overheads because of heavy communication between agents. Previous research has made efforts to train a sparse multi-agent graph or fine-tune a planner to orchestrate the workflow better. However, such extra training processes introduce computational costs and limit MAS to specific domains, therefore compromising their generalizability. In this paper, we propose CONCAT, a training-free multi-agent collaboration framework based on CONsensus and Confidence-driven Ad hoc Teaming to efficiently organize agent interactions. Specifically, agents are clustered based on their initial answers, and leaders of each cluster are selected based on the agents' confidence. Then, a heuristic function based on the Theory of Mind is designed to predict the collaboration benefits between every two leaders according to their answers and confidence. Finally, an ad hoc multi-agent network is organized after evicting a percentage of communications based on the predicted benefits. Experiments across three LLMs and three benchmarks show that CONCAT achieves up to 2.02x higher efficiency (accuracy/latency ratio) than LLM-Debate and outperforms training-aware methods such as AgentDropout, while reducing average latency by 50.1% on Qwen2.5-14B-Instruct, without any task-specific training.

In scene graph generation, a central challenge is modeling polysemous predicates whose meanings shift across contexts. Prior approaches address this issue by decomposing predicates into multiple static prototypes or retrieving semantically similar exemplars. However, these strategies keep predicate representations static and cannot reorganize semantics to reflect image-specific evidence, leading to systematic confusions in ambiguous contexts. We propose AlignG, which learns context-conditioned predicate semantics via prototype feedback. AlignG infers context-conditioned predicate semantics from the relation candidates within each image and feeds the adapted semantics back to recalibrate relation representations. The learning objective anchors this adaptation to global semantic centers, preventing semantic drift while still allowing selective reorganization when the scene provides consistent relational cues. Experiments on VG-150 and GQA-200 show consistent improvements over state-of-the-art baselines, with F@100 improvements of +1.4 on VG-150 and +2.7 on GQA-200 under SGDet. We further visualize per-image prototype similarity shifts and observe coherent context-dependent reorganization where prototypes selectively merge or separate predicates according to scene evidence. The code is available at https://github.com/Namgyu97/AlignG-SGG.pytorch.

Confidence estimation for Vision-Language-Action (VLA) models is essential for robots to perform manipulation tasks in the open world, providing crucial signals for risk-sensitive decision-making and failure anticipation. Existing confidence estimation methods typically rely on ensemble-based paradigms or action-token probabilities to predict the likelihood of task success. However, they still encounter challenges in computational efficiency and cross-architecture generalizability. These methods usually require repeated sampling, leading to inference inefficiency, and are restricted to VLA models with discrete action outputs, making them difficult to apply to continuous action spaces. To address this issue, we propose VLAConf, a one-class discriminative confidence framework. By leveraging frozen pretrained VLA internal representations, VLAConf directly estimates step-wise anomaly scores in a single forward pass using a lightweight confidence head, thereby eliminating the overhead of exhaustive resampling. We additionally use step-conditioned modeling to encode rollout-phase information along the manipulation trajectory. Experiments on the LIBERO benchmark demonstrate that VLAConf significantly improves the quality of the confidence signal constructed for post-hoc calibration, outperforming existing baselines by a large margin in inference efficiency. The effectiveness of VLAConf is further validated in real-robot experiments. To access the source code and supplementary videos, visit https://sites.google.com/view/vlaconf.

Multi-modal Retrieval-Augmented Generation (MMRAG) has emerged as a powerful paradigm for enhancing Multimodal Large Language Models in knowledge-intensive question answering by integrating external visual, textual, and structural knowledge. However, existing MMRAG frameworks suffer from critical limitations, including noisy and irrelevant retrieval, cross-modal semantic misalignment, lack of adaptive reasoning, and incoherent generation across local and global contexts. We introduce \textbf{CogniVerse}, a novel MMRAG framework that addresses these challenges through a cognitive-inspired, mathematically rigorous approach. Drawing from human-like reasoning, CogniVerse integrates three synergistic components: (1) a Cognitive Reflection Module that dynamically assesses retrieval necessity and filters relevant multi-modal content, reducing noise and computational overhead; (2) a Multi-modal Retrieval Module that aligns embeddings in a Riemannian manifold using information geometry and refines knowledge graphs via spectral graph theory, ensuring precise and coherent retrieval; and (3) a Hierarchical Generation Module that employs an optimal transport-based loss to balance token-level accuracy and global semantic coherence. Extensive experiments demonstrate that CogniVerse significantly outperforms state-of-the-art systems in both accuracy and coherence, while reducing retrieval latency.

As autonomous agents become more capable of performing real-world tasks, distinguishing scheming behavior from benign task pursuit may become a central AI control problem. Existing monitors often rely on chain-of-thought access or internal activations, or use prompted frontier models, all of which can be unavailable, unreliable or expensive in deployment. In this work, we study action-only deliberative monitors: smaller open-weight models trained to detect scheming and sabotage from agentic trajectories without accessing the monitored agent's reasoning or model internals. Our method, inspired by deliberative alignment, uses a scheming specification to elicit structured rationales from a frontier teacher, filters them with a separate judge, and distills the highest-quality rationales into open-weight monitors with supervised fine-tuning and reinforcement learning. We train on five datasets, and evaluate across six out-of-distribution agentic misalignment benchmarks. We show that applying our method to Qwen3.5-27B yields higher performance than all low-cost frontier models as prompted monitors (Gemini 3.1 Flash-Lite, GPT-5.4 Nano, and Claude Haiku 4.5) and than Gemini 2.5 Pro, while also achieving lower marginal inference cost (token-metered USD per 1,000 evaluations). Stronger prompted frontier monitors (Gemini 3.1 Pro, GPT-5.4, Claude Sonnet 4.6, and Claude Opus 4.6) achieve higher performance but at roughly $16$--$34\times$ higher marginal inference cost. Several of our trained monitors are positioned on the empirical cost--performance Pareto frontier among the monitors we evaluate, providing practical low-cost, low-FPR alternatives to prompted frontier models.

Vision-language models (VLMs) are increasingly used to answer questions about physical scenes, yet most evaluations reduce performance to a final answer. This hides whether the model perceived the right objects, represented the right physical state, predicted a plausible transition, or merely selected the right option for the wrong reasons. We introduce \wmw, an evaluation framework for auditing the \emph{language-expressed physical commitments} of VLMs. Instead of scoring only $I,q\mapsto a$, we ask models to produce a typed trace $I,q\mapsto(s_0,Δs,s_1,a)$: an initial state, a state transition, a resulting state, and an answer. A hybrid verifier then checks schema validity, state grounding, transition consistency, and answer-trace compatibility, yielding typed error labels such as object, relation, force, transition, temporal, unit/scale, and faithfulness errors. We release \tracebank, a controlled trace resource with \nSeed schema- and recomputation-validated synthetic scenarios across \nFamilies physics families, \nPairs minimally perturbed contrastive preference pairs, verifier code, audit guidelines, and model outputs. We evaluate \nModels VLMs on both controlled and external physical-reasoning examples. \wmw reveals failures that answer-only evaluation misses: 35\% of correct answers from mid-tier models are backed by physically invalid traces. Verifier-guided reranking recovers up to 7 percentage points of trace validity without sacrificing answer accuracy, and trace-level preference tuning reduces hidden inconsistency by 41\% relative. The contribution is not another final-answer physics benchmark, but a reusable protocol for measuring whether a VLM's stated physical world can be true at the same time as its answer.

Reinforcement learning (RL) is a natural fit for agentic knowledge base question answering (KBQA), where a model must issue executable actions, observe knowledge-base feedback, and eventually return an answer. However, current RL-based KBQA systems mainly optimize sparse rewards from the final answer, leaving intermediate action errors weakly supervised. This is especially limiting for logical-form annotated KBQA benchmarks: gold logical forms can be converted into executable action sequences, but existing pipelines use them mainly for warm-start data construction rather than for on-policy RL updates. We propose GAPD, a training-time Gold-Action Policy Distillation framework that adds dense token-level guidance to outcome-based RL. To align gold actions with on-policy student rollouts, GAPD uses MID-ANCHOR MATCHING: it treats the intermediate entities reached during student exploration and gold execution as state anchors, and matches student states to gold states through these explored entity sets. The current policy conditioned on this aligned gold action serves as a stop-gradient teacher, whose token distribution is distilled back to the ordinary student policy over generated action-token spans. GAPD consistently surpasses the current state of the art on WebQSP, GrailQA, and GraphQ.

Large Language Models (LLMs) have shown promise as educational tutors, yet effective tutoring requires more than solving problems: it must provide progressive Socratic guidance and balance multiple pedagogical objectives across multi-turn interactions. However, training such tutors remains challenging due to limited-fidelity and weakly controllable student simulation, under-specified pedagogical reward modeling, and unstable multi-objective optimization. To overcome these limitations, we propose PEARL, a pedagogically aligned reinforcement learning framework for training Socratic tutoring agents, consisting of three key components. First, we introduce a controllable student simulator that decouples latent cognitive states from response generation to model diverse abilities and misconceptions. Second, we develop a generative reward model that jointly evaluates pedagogical quality and objective correctness for policy optimization. Finally, we propose a stable multi-objective RL scheme that discretizes rewards within each dimension and aggregates normalized advantages across dimensions, preventing high-variance objectives from dominating updates. Experiments on multiple benchmarks show that PEARL achieves the best performance among open-source models and remains competitive with leading proprietary LLMs, despite using only a 30B policy model.

Human tactile perception of materials relies on complex multisensory touch cues, yet the relationship between low-level tactile signals and perceptual representations remains poorly understood. This knowledge gap hinders the integration of touch in digital environments and the development of robots capable of human-like tactile perception. Here, we present an interpretable computational framework for modeling human material perception and recognition using multisensory touch data. Our framework comprises three interconnected models: Model 1 maps finger-surface interaction features to psychophysical sensory attributes, Model 2 classifies materials based on these perceptual representations, and Model 3 directly classifies materials from tactile features. The results showed that combining information from pressing, static contact, and sliding interactions improves prediction accuracy, and that thermal cues are particularly informative for both perceptual modeling and material classification. These findings highlight the importance of thermal and compliance cues, which remain underrepresented in current robotic fingers and haptic displays. Incorporating such cues may enhance artificial systems' ability to approximate human material perception and guide the design of more perceptually grounded haptic interfaces.

When using reinforcement learning (RL) for contact-rich robotic manipulation, vision can provide task-relevant information that accelerates learning beyond what proprioception alone can achieve. However, vision-enabled policies tend to overfit to the visual conditions seen during training, limiting their robustness and transferability. We present a human-in-the-loop RL framework that employs teacher-student distillation to achieve robust performance across multiple task variants, trained entirely in the real world without requiring domain randomization or data augmentation. A vision-enabled teacher distills its knowledge into a vision-free student that relies solely on pose, twist, and wrench sensing, combining fast training with strong task generalization. On the real-world NIST assembly benchmark board, our approach achieves 95\% overall success after approximately 50 minutes of training on 3 representative tasks, including robust generalization to 8 unseen task variants. Fine-tuning with distillation achieves full success on the most challenging task. We demonstrate that the resulting policies outperform baselines in both robustness and adaptability.

Vision-Language-Action~(VLA) models have shown strong potential for general-purpose robotic manipulation, yet they still struggle to generalize to unseen tasks that necessitate transferring relevant experience across objects, scenes, and action patterns. This paper proposes VLA-Pro, a plug-and-play framework designed to enhance cross-task generalization by storing task-relevant procedural memories at training time and transferring these memories during inference. Specifically, VLA-Pro stores task-specific LoRA adapters as parameterized procedural memories during training. At inference time, VLA-Pro retrieves relevant procedural memories based on the current multi-modal context and dynamically fuses these memories for generating the current action chunk. Experiments on RoboTwin, RLBench, and real-world manipulation tasks show that VLA-Pro consistently improves cross-task generalization across multiple backbones, achieving up to a 207% relative improvement in simulation and increasing real-world success rate from 5.8% to 65.0%. These results suggest that procedural memory retrieval and adaptation provide an effective mechanism for transferring manipulation experience to novel tasks while preserving modularity and execution stability.

Vision-Language-Action (VLA) models are a powerful paradigm for generalist robotic control. However, their high computational cost and limited control frequency hinder real-time robotic manipulation, especially when large vision-language backbones and iterative action heads run at every control step. Existing VLA acceleration methods often optimize individual components or rely on fixed acceleration rules, treating different control steps with largely fixed computation and overlooking the non-uniform reasoning demands of sequential embodied control. Inspired by human motor control, where cognitive and feedback resources concentrate on goal-sensitive stages, we argue that VLA models should learn when to invest full computation and when to reuse prior computation. We propose ElegantVLA, a plug-in phase-adaptive inference framework that accelerates VLA models through intra-model dynamic compute scheduling. ElegantVLA introduces a lightweight scheduler that observes temporal representation similarity, robot-motion cues, and episode progress to jointly allocate computation across the vision encoder, LLM, and action head. For perception-language reasoning, the scheduler selects a five-level Vision-LLM compute mode, from full recomputation to multi-step temporal reuse, based on visual-language representation stability. For action generation, it selects a three-level denoising mode, reusing intermediate denoising states during stable motion while preserving full refinement for goal-sensitive stages. By coordinating these decisions, ElegantVLA offers a general acceleration framework for modern VLA pipelines with explicit action-generation modules, without modifying or retraining the base model. Experiments on GR00T and CogACT achieve up to 2.55x and 3.77x speedup, and on six real-world GR00T tasks ElegantVLA cuts computation by 2.18x while raising control frequency from 13.8 Hz to 26.3 Hz.

Vision-Language-Action models have achieved remarkable progress in robotic manipulation, yet they suffer from a critical limitation: a lack of 3D scene understanding. This deficiency manifests as three intertwined challenges: weak extraction of 3D spatial positions without enforcing multi-view consistency, inadequate 3D instance understanding, and fragile reasoning under occlusion. Although mature 3D perception methods exist, their direct integration into VLA pipelines is hindered by architectural incompatibility and by heavy reliance on costly instance-level annotations. To address the above challenges, we propose 3DVLA, a plug-and-play framework that injects robust 3D reasoning into pretrained VLAs without requiring extra manual labels or discarding VLM priors. Specifically, 3DVLA tackles the three challenges through: (1) pervasive 3D feature encoding with explicit multi-view consistency constraints across all modalities and a Spatially-Conditioned Geometry Aggregation method, (2) an instance estimation module with high-level instance tokens for 3D instance awareness, and (3) a masked self-supervised 3D encoding branch that retains its predictor for visual token completion to handle occlusions. We integrate 3DVLA with multiple VLA baselines and evaluate on LIBERO-Plus and RoboTwin 2.0. Results show consistent and significant gains in manipulation performance, validating both the effectiveness and plug-and-play compatibility of our approach.

Reliable midair docking between small unmanned aerial vehicles (UAVs) is essential for modular aerial cooperation and manipulation, but it requires precise relative-pose control and repeatable platform under tight thrust and payload constraints. We present a dual-drone docking platform where two quadrotors operate in a leader-follower formation and dock using a lightweight modular frame with passive magnetic latching. A progress-aware mission supervisor manages phase transitions: approach, alignment, capture, and settle. This platform integrates a complete hardware-software stack (ROS 2 with Crazyflie/PX4 interfaces) and synchronized logging for benchmark evaluation. We evaluate the platform in simulation and real-world experiments using quantitative metrics such as formation error, baseline and yaw consistency, docking success rate, time-to-dock, and failure-mode statistics. The platform enables statistically grounded comparison of docking supervision and synchronization strategies and provides a practical testbed for modular aerial cooperation and repeatable midair aerial manipulation.

This paper presents a phase-conditioned, force-aware framework for robust deformable object manipulation. Standard imitation learning policies such as Action Chunking with Transformers (ACT) rely on a Markovian assumption at inference, causing state aliasing when visually similar observations require contradictory actions and preventing autonomous recovery from execution failures. We address this with a closed-loop hierarchical architecture. A FiLM-conditioned ACT encoder modulates feature extraction based on the current task phase, enabling a single unified policy to produce phase-specific behaviors while sharing action dynamics across phases. A multi-modal phase predictor fusing visual, force, and pose feedback estimates the phase in real time, detecting contact failures that are invisible to vision alone and autonomously triggering recovery trajectories. The system is completed by a hybrid impedance controller for compliant execution and a haptic teleoperation interface for force-aware data collection. Ablation studies show that FiLM-based modulation significantly outperforms both unconditioned and token-level conditioned baselines, and t-SNE analysis confirms that FiLM induces well-separated, phase-specific feature representations. Validated on hanging and removing a T-shirt with dual arms, the closed-loop system improves the hanging success rate from 56\% to 87\% through autonomous error recovery. Code and videos: https://leledeyuan00.github.io/phaser/

Natural language interfaces can simplify interaction with multi-robot systems, especially when non-expert users need to issue high-level commands. Acoustic manipulation using ultrasonic phased arrays also enables contactless object handling for applications such as healthcare, laboratory automation, and precision transport. However, combining large language models (LLMs) with distributed acoustic mobile robots remains underexplored. This paper presents a decentralized framework for natural language-driven coordination of acoustic robots for contactless object manipulation. The system converts spoken instructions into executable multi-robot task plans using Whisper-based speech recognition, LLM-based semantic parsing, structured JSON task representation, and distributed scheduling. The JSON schema encodes robot assignments, temporal dependencies, spatial constraints, and synchronization requirements for sequential, parallel, and synchronized execution. The system is implemented on two TurtleBot3-based acoustic robots, each equipped with an ultrasonic phased array for contactless object transport. Experiments were conducted in three scenarios: sequential execution, parallel multi-robot transport, and synchronized cooperative manipulation. The system achieved task success rates of 96 percent for sequential tasks, 86 percent for parallel execution, and 70 percent for synchronized collaborative transport. These results show that natural language commands can be transformed into distributed robot actions for contactless manipulation, highlighting the potential of LLM-driven automation for human-robot interaction in distributed robotic systems.

Bimanual coordination is essential for many real-world manipulation tasks, yet learning bimanual robot policies is limited by the scarcity of bimanual robots and datasets. Single-arm robots, however, are widely available in research labs. Can we leverage them to train bimanual robot policies? We present MonoDuo, a framework for learning bimanual manipulation policies using single-arm robot demonstrations paired with human collaboration. MonoDuo collects data by teleoperating a single-arm robot to perform one side of a bimanual task while a human performs the other, then swapping roles to cover both sides. RGB-D observations from a wrist-mounted and fixed camera are augmented into synthetic demonstrations for target bimanual robots using state-of-the-art hand pose estimation, image and point cloud segmentation, and inpainting. These synthetic demonstrations, grounded in real robot kinematics, are used to train bimanual policies. We evaluate MonoDuo on five tasks: box lifting, backpack packing, cloth folding, jacket zipping, and plate handover. Compared to approaches relying solely on human bimanual videos, MonoDuo enables zero-shot deployment on unseen bimanual robot configurations, achieving success rates up to 70%. With only 25 target robot demonstrations, few-shot finetuning further boosts success rates by 65-70% over training from scratch, demonstrating MonoDuo's effectiveness in efficiently transferring knowledge from single-arm robot data to bimanual robot policies.

Non-uniform scaling control of formation enables multi-agent systems to adjust their shape by scaling with different ratios along different coordinate axes, offering enhanced flexibility in complex environments. However, like most existing formation maneuver strategies, it typically assumes a fixed set of agents, limiting its applicability in scenarios requiring dynamic team expansion. This paper introduces a distributed control framework that enables a formation to incorporate new agents during non-uniform scaling maneuvers in arbitrary dimensions while preserving the spectral properties of the graph Laplacian. Simulation examples validate the effectiveness of the theoretical results.

Vision-Language-Action (VLA) models with integrated reasoning have been proposed for end-to-end autonomous driving, assuming a tight coupling between reasoning and trajectory generation. However, the robustness of such systems under realistic input perturbations remains largely unexplored. We show that these models are highly vulnerable to realistic input perturbations, achieving up to 89% attack success rate (ASR) on reasoning and up to 72% on trajectory manipulation in closed-loop simulation, leading to increased collision rates and degraded safety metrics. Using NVIDIA's recent Alpamayo models as representative industry-developed VLAs, we conduct the first systematic black-box study of reasoning-enabled VLA models under realistic textual input corruptions, evaluating their impact on reasoning and driving behavior. We introduce a reasoning-aware evaluation framework capturing both semantic and structural aspects of reasoning, along with safety-centric measures. We also introduce a benchmark for evaluating attacks and defenses on reasoning-trajectory interactions in autonomous driving. Our results highlight the need for rigorous evaluation and improved defenses to ensure the safety of reasoning-enabled VLA systems in autonomous driving.

Swarm and field robotics face significant barriers to real-world validation due to the high cost and development time to deploy hardware. This paper introduces the ``Bionic Swarm,'' a novel system that lowers these barriers by abstracting away many of the tasks that are difficult to implement on robots but which do not contribute to the overall algorithm evaluation, giving these tasks to human users. These human users take directions from a smartphone web-app that takes measurements from Bluetooth-connected sensors and relays them to a centralized server. This server runs the swarm algorithm and directs actions to the human users. We evaluate this system through the experimental validation of a geotechnically-focused search algorithm named Score-Biased-Search, which functions by assigning a ``score'' to each location on a reconstructed map, then biases search patterns through areas of higher expected scores, and which exhibits superlinear map reconstruction relative to the number of search agents. After presenting simulation results for the algorithm, we then apply the algorithm on the Bionic Swarm platform to validate its function in a real-world, outdoor setting. This work demonstrates that this human-in-the-loop approach significantly lowers the barrier to entry for field and swarm robotics research.

Are current Vision Language Models (VLMs) ready to comprehend and reason about complex embodied interactions in 3D environments? We introduce Embodied3DBench, a robot-centric benchmark targeting low-level spatial intelligence in embodied 3D environments. To systematically evaluate these foundational perceptual capabilities, the benchmark includes 6 task categories divided into two core groups: Spatial Structural Understanding (Grounding, Spatial Relation Prediction, and Multi-view Correspondence) and Interaction-Oriented Perception (Affordance Prediction, Grasp Point Prediction, and Trajectory Prediction). The benchmark spans 12 subcategories and contains over 21k high-quality question-answer pairs. We evaluate 13 state-of-the-art models, and the results show that while current models exhibit relatively strong high-level spatial reasoning, such as understanding object-to-object positional relations, they remain fragile in interaction-oriented perception, highlighting a significant lack of robust 3D-aware interaction priors. To actively bridge this capability gap revealed by our benchmark, we further synthesize a large-scale training dataset comprising 1.3M QA pairs. Notably, fine-tuning on this dataset yields significant improvements in low-level spatial intelligence. Ultimately, Embodied3DBench fills a critical gap by providing both a systematic evaluation framework and a scalable data solution, setting a clear target for the development of interaction-aware multimodal systems.

A primary bottleneck in contact-rich manipulation is the difficulty of collecting real-world data. Sim-to-real reinforcement learning offers a scalable alternative, but the simulation-reality gap prevents information-dense modalities like touch from being effectively used. Existing sim-to-real methods often mitigate this gap by simplifying tactile data into coarse low-dimensional features -- sacrificing the richness required for complex manipulation. In this work, we introduce Center-of-Pressure (CoP), an effective tactile representation grounded in physical principles that preserves dense contact information while maintaining robustness for sim-to-real transfer. To support this representation, we propose a sensor calibration scheme based on differentiable dynamics, enabling the estimation of taxel orientations without requiring ground-truth force measurements. We evaluate CoP on two blind, challenging contact-rich manipulation tasks: peg-in-hole insertion and ball balancing. Across both tasks, policies conditioned on CoP achieve zero-shot sim-to-real transfer on a multi-fingered hand, and outperform both coarse binary-contact and raw-taxel baselines. Analysis of learned policy states further suggests that CoP-conditioned policies encode task-relevant physical properties, such as object mass, as an emergent byproduct of control.

This study presents the first evaluation of general-purpose imitation learning for surgeon-robot collaborative assistance in open surgery, targeting suture following: the grab-pull-release motion an assistant performs at every stitch. We collect 160 teleoperated demonstrations (32,374 frames) on an open-source robot arm, benchmark four architecturally diverse imitation learning policies (ACT, Diffusion Policy, SmolVLA, $π_0$) across 28 trained models evaluated in 32 configurations along three clinically motivated dimensions: dataset size, camera viewpoint, and background variation. Our results demonstrate that under ideal conditions, the four policies achieve $50$-$75\%$ task success, with depth error as the dominant failure mode across all architectures. Among all policies, $π_0$ achieves the strongest results with a pretrained vision-language backbone, demonstrating superior data efficiency, greater robustness to background variation, and smoother trajectories compatible with surgical workflow. When deployed in a surgeon-robot suturing trial, $π_0$ yields a $92\%$ stitch completion rate. These findings establish collaborative robotic assistance in open surgery as a feasible target for imitation learning and highlight depth perception and end-effector design as key priorities for clinical translation.

We discover that VLA architectures fail in fundamentally different, predictable ways at the motor-command level. Running VQ-BeT, Diffusion Policy, and ACT on identical evaluation protocols (n=450 episodes across PushT and ALOHA 14-DOF bimanual manipulation), we find: (1) direction reversal rate is a universal failure predictor across all three architectures (AUROC=0.93, 0.79, 0.91; p<0.001); (2) jerk monitoring is predictive only for discrete-token architectures, following a discrete-to-continuous gradient (0.88, 0.69, 0.41); (3) velocity violations alone are non-predictive everywhere (AUROC 0.41-0.69), yet velocity checking is the most common safety mechanism in VLA deployment code; and (4) for continuous-family VLAs, velocity monitoring provides effectively zero predictive signal (AUROC=0.52 on ACT, 0.41 on Diffusion), proving that architecture-matched monitor selection is essential. These results quantify a monitoring consequence of the well-known discrete/continuous VLA distinction: the two families produce qualitatively different failure signatures that require different monitors. No single monitor works universally; architecture-matched selection is required. This finding was enabled by SafeContract, a training-free, black-box action monitoring toolkit with conformal calibration. Code: https://github.com/krishnam94/vla-edge

Vision-Language-Action (VLA) models offer a promising paradigm for generalist robotic policies, yet their adaptation is hindered by data inefficiency and poor generalization. We argue that these bottlenecks stem from the prevailing Direct Instruction-to-Control Mapping, which forces models to memorize monolithic trajectories rather than reusable motion patterns, i.e., primitives. We propose PrimitiveVLA, a framework that shifts this paradigm toward a Primitive-Centric Disassemble & Assemble paradigm. Supported by a shared Multimodal Canonical Representation (MCR), PrimitiveVLA unifies two phases: (1) Fine-tuning-phase Disassembly, which uses an automated pipeline to disassemble demonstrations into reusable primitives; and (2) Inference-phase Assembly, which employs a VLM-based planner and an LLM-generated switch module for robust closed-loop execution. By disassembling tasks into reusable primitives, PrimitiveVLA enables VLA models to learn invariant motion patterns instead of task-specific trajectories. Extensive experiments show that our framework improves data efficiency and achieves superior zero-shot generalization across unseen and long-horizon tasks.

Ensuring both safety and efficiency in decision-making for autonomous driving systems remains a fundamental challenge. Traditional Deep Reinforcement Learning (DRL) suffers from unsafe random exploration and slow convergence, while Large Language Models (LLMs) demonstrate inherent latency in real-time inference operations. To address these limitations, this paper proposes SARAD, a novel safety-aware hybrid framework that synergizes LLMs and DRL for autonomous driving. SARAD substitutes the random exploration of DRL with Retrieval-Augmented Generation (RAG)-enhanced, LLM-guided decisions sourced from a dynamic expert knowledge repository. An attention discriminator is proposed to integrate the prior knowledge of LLMs into DRL policy optimization. A collision predictor module, fine-tuned with historical collision data, is further designed to improve vehicle safety. Extensive experiments show that SARAD achieves significant performance improvements in the Highway-Env simulator, validating the effectiveness of the proposed model in autonomous driving.

The pursuit of humanoid athletic sprints is hindered by a scarcity of humanoid-viable kinematic reference data and the inability of existing frameworks to maintain stability during sprints. To overcome these limitations, we introduce SPRINT, a novel framework driven by efficient, frequency-adaptive spectral priors. By characterizing the fundamental periodicity of human locomotion in the frequency domain using a reference library of five discrete motion sequences, these priors generate kinematically feasible joint trajectories across a broad velocity spectrum, successfully extrapolating to speeds that exceed the reference distribution. Guided by these pretrained priors, the SPRINT policy achieves zero-shot sim-to-real transfer in field experiments on the Unitree G1 platform, reaching a peak sprinting velocity of 6 m/s and demonstrating seamless gait transitions while preserving biomimetic naturalness. Ultimately, this work establishes frequency-adaptive spectral priors as a highly data-efficient foundation for humanoid athletic sprints. The project page is available at https://anonymous.4open.science/w/SPRINT-138A/.

Vision--language--action (VLA) policies are trained to imitate actions; their loss never asks them to estimate reward, progress, or future success. Their frozen representations nevertheless carry such information, and it can be read out and used to guide action choice without retraining the policy. From mixed successful and failed manipulation trajectories on LIBERO-Goal, we recover Monte-Carlo outcome targets using lightweight linear probes on frozen features. The targets are consistently predictable from OpenVLA, Pi0.5, DINOv2, and CLIP features, and substantially less so from baselines built on progress, time-to-go, task identity, or proprioception. To rule out task and temporal shortcuts, we evaluate the probes under same-task, same-timestep matched comparisons: Pi0.5 probes still reach roughly 92% pairwise ordering accuracy, while label-shuffled controls stay at chance. Used as a test-time selector over sampled Pi0.5 action prefixes, the same probe turns this offline finding into behavior: on push-plate, success rises from 26.7% under greedy decoding to 44.3%, with a second positive case on wine-rack. The gains are not universal and require additional inference compute, but the underlying finding is clean: frozen VLAs already encode information about success that their imitation objective never explicitly demands.

Magnetically actuated microrobots have been used as wireless, non-contact manipulation tools at microscales, making them promising for minimally invasive applications. However, their control remains challenging due to indirect actuation, limited sensing, and nonlinear magnetic interactions. In this work, we propose Mag-VLA, a vision-language-action (VLA) model for dexterous magnetic microrobot manipulation using two robotic arms with mounted magnets for dynamic magnetic-field construction. Bimanual coordination enables capabilities such as microrobot reorientation that are difficult or infeasible with a single arm, but it also introduces coupled control challenges, as the policy must generate coordinated trajectories for both actuators within a shared workspace. Our framework adapts a Qwen2.5-VL-7B backbone using Low-Rank Adaptation (LoRA) to process visual observations and language instructions for action prediction. To capture task progression, we introduce a motion-aware phase classifier and a phase-conditioned Action Chunking Transformer (ACT) decoder for temporally coherent multi-step control. We further construct a teleoperated magnetic microrobot manipulation dataset covering three task configurations. Ablation studies show that the ACT-based decoder substantially outperforms alternative generative action heads. In real-robot experiments, Mag-VLA achieves a 90% approach success rate across all tasks and transport success rates of 80%, 70%, and 50% as task difficulty increases. These results demonstrate that hierarchical VLA modeling provides a promising framework for magnetic microrobot manipulation.

We present a hybrid robotic skin that combines electrical impedance tomography (EIT) with pneumatic tactile sensing to improve force reconstruction capability. The developed robotic skin is fabricated entirely by 3D printing and spray coating, making it affordable and easy to build. A Tikhonov-regularized inverse reconstruction, paired with per-pad pneumatic calibration, enables accurate large-area tactile sensing with a simple measurement scheme. For validation, we conducted load-cell indentation experiments; the results showed consistent force reconstruction across locations within a pad. Compared with an EIT-only baseline, sensitivity non-uniformity was also reduced, with the coefficient of variation decreasing from 0.31 to 0.14, indicating that the proposed approach addresses a longstanding limitation of EIT. We further demonstrated chest-mounted integration on a humanoid robot and found that the pneumatic signals remained reliable across diverse contact scenarios, including multiple simultaneous contacts on the same sensing pad. These results indicate a practical path toward accurate, scalable whole-body tactile sensing in real robotic systems.

Optical tweezers (OT) provide piconewton-scale manipulation for delicate biomedical tasks, where visuo-haptic feedback can improve operator awareness by conveying interaction-force cues and trap-stability information. However, visuo-haptic teleoperation frameworks for complex-shaped optical microrobots remain underdeveloped, particularly in multi-trap manipulation scenarios. This paper presents a digital twin framework for virtual visuo-haptic teleoperation of complex-shaped OT-driven microrobots. The framework integrates a digital twin environment, image-based pose and depth estimation, microrobot motion simulation, and model-based haptic rendering within a Robot Operating System (ROS)-connected bimanual teleoperation system. For force modeling, we combine a Multi-Sphere Distributed Manipulation (MSDM) model with optical-force estimation from the Optical Tweezers Toolbox, enabling simulator-driven visuo-haptic feedback. The framework reproduces representative microrobot motion trends and provides haptic force rendering that is numerically consistent with the fitted optical-force model. In simulated cell-delivery tasks, haptic feedback reduced the standard deviations of the contact-force metric and the microrobot-to-trap-center distance metric by 53.2% and 55.2%, respectively, and improved task success from 30% to 80%. These results demonstrate the framework's effectiveness for evaluating visuo-haptic teleoperation strategies for complex-shaped optical microrobots.

Self-supervised online traversability estimation enables robots to continuously learn from unlabeled open-world experiences and adapt their navigation behavior toward safe and efficient trajectories. Existing approaches either rely on handcrafted proprioceptive traversability scores, limiting robot-agnosticism, or cluster prior data, preventing online learning. Moreover, many continual learning methods incur substantial memory and computational costs, hindering onboard deployment. We introduce COTRATE, an online learning framework for continuous traversability estimation from multimodal, unlabeled robot experience. Our method first infers robust traversability scores using a robot-agnostic, learning-based online terrain assessment module operating on proprioceptiveand inertial signals. These scores then supervise a visual traversability network through a novel alignment loss that associates visual embeddings with online terrain assessments.To mitigate forgetting during continual learning with minimal overhead, we propose a diversity-aware feature selection strategythat preserves performance using a compact replay memory. We further show that the learned traversability representation supports knowledge transfer across different robot platforms with different locomotion kinematics. We evaluate COTRATE on a dataset of \approx 50,000 images collected with two robotic platforms across 11 outdoor terrains, and benchmark it on navigation tasks in three representative outdoor environments. We make the dataset, code, and trained models publicly available.

Direct physical guidance is a natural means of teaching and interacting with robots, and robotic skins make a key contribution by enabling sensitive contact sensing and localization. This paper presents a tactile-proprioceptive sensor fusion framework for natural physical human-robot interaction. Tactile cues from pneumatic skin pads serve as contact indicators that bypass the ambiguity between frictional residues and applied external forces, enabling highly sensitive contact detection without explicit friction identification. We fuse these cues with motor-current-based proprioception to reconstruct multi-axis contact forces on the robot surface. To maintain accuracy during motion, we employ a temporal convolutional network (TCN) to mitigate friction hysteresis during stick-slip transitions, reducing uncertainty at contact onset and yielding smooth, responsive guidance. We validate the approach on a skin-integrated robot arm: (i) multi-axis forces are reconstructed in stationary contacts, and (ii) simultaneous force estimation and kinesthetic teaching are demonstrated. Results indicate improved sensitivity and responsiveness across diverse contact conditions compared with tactile-only and proprioceptive-only baselines, supporting tactile-proprioceptive fusion as a reliable pathway to safe, intuitive physical human-robot interaction.

Safe physical interaction is critical for deploying robotic manipulators in human-robot interaction and contact-rich tasks, where uncertainty, external forces, and actuator limitations can compromise both performance and safety. We propose an online adaptive impedance control framework that enforces joint-state safety while achieving compliant interaction under uncertain dynamics. The approach combines a quadratic-program-based safety filter with a novel composed position-velocity non-smooth control barrier function (NCBF), enabling joint position and velocity constraints to be enforced through a unified relative-degree-one barrier. Unknown dynamics are compensated online using an interval type-2 fuzzy logic system, while actuator torque limits are handled through soft constraints with exact penalty recovery of feasible solutions. A disturbance-observer-enhanced safety mechanism improves robustness against modelling errors and external interaction forces. Using composite Lyapunov analysis, we prove forward invariance of the safe set and the uniform ultimately boundedness of the impedance-tracking error. Simulations on a 7-DOF manipulator with severe parametric uncertainty and external interaction wrenches demonstrate safe constraint satisfaction and robust impedance tracking.

This paper presents a magnet-based robotic skin that integrates a multilayer soft lattice with distributed Hall-effect sensor arrays and a tactile super-resolution model. External contact forces are converted to magnetic field changes by embedded permanent magnets, and the lattice spreads these changes across the sensing domain. This gives each sensor a large, overlapping receptive field and enables a large sensing area with minimal blind spots. Lattice parameters are tunable, enabling joint adjustment of mechanical compliance and transduction characteristics. An implicit modeling workflow and selective laser sintering (SLS) 3D printing support rapid fabrication of conformal, high-complexity structures. A convolutional neural network trained on experimental measurements estimates contact location and normal force in real time. Experiments validate localization accuracy and indicate scalability to larger surfaces, suggesting applicability to whole-body robotic skin and safe human-robot interaction.

This paper addresses the problem of identifying parsimonious explicit piece-wise polynomial relationships that might involve a relatively large number of raw features. The algorithm leverages a recently proposed identification algorithm that yields parsimonious implicit relationships enabling to derive normality characterization in the context of anomaly detection and localization. The algorithm proposed in this paper goes a step further by deriving explicit piece-wise representations that are built using the set of polynomials involved in the implicit representations. The framework is illustrated on the problem of identifying parsimonious explicit representations of the inverse model of a 6-axis manipulator robot. Moreover, further experiments on a 4-axis robot are also shown which are designed to investigate the generalization capability of parsimonious models compared to state-of-the-art DNNs structures, when models face unseen contexts of use.

Real-world navigation is fundamentally driven by Points of Interest (POIs), yet reaching a precise POI remains a critical "final-meters" challenge. Existing Vision-Language Navigation (VLN) benchmarks of POI-goal navigation often suffer from coarse granularity or significant sim-to-real gaps due to generated scene. To bridge this gap, we present POINav-Bench, the first benchmark designed for closed-loop evaluation of real-world POI-goal navigation. It comprises 11 commercial areas reconstructed from real-world captures using 3D Gaussian Splatting (3DGS), covering 126,398 $m^{2}$ in total and spanning 163 distinct POIs. With traversability-aware annotations and reference trajectories, POINav-Bench enables high-fidelity evaluation of navigation agents in realistic, POI-rich real-world environments. Building on this, we propose the POINav Brain-Action Framework where a Brain module performs POI-grounded reasoning to guide an Action module in predicting continuous waypoints for real-world execution. We further curate the POINav-Dataset, containing 70K real-world signage-entrance pairs. Experiments show that our framework provides a viable path toward refining real-world POI-goal navigation.

We present ProgVLA, a compact vision-language-action (VLA) model designed for reliable robot manipulation under tight compute and memory budgets. The model specifically focuses on efficiently processing long multi-modal sequences by maintaining an explicit representation of task progress over extended horizons. To this end, ProgVLA integrates two key components. First, a multi-modal encoder with a two-stage Perceiver resampling scheme compresses variable-length visual, language, and proprioceptive streams into a fixed set of control-ready context tokens, substantially reducing sequence length while preserving cross-modal grounding. Second, an auxiliary set of progress heads is trained with offline reinforcement learning (RL) objectives to jointly learn critics over normalized remaining-horizon targets. This provides the policy with an internal estimate of task progress and enables advantage- and success-weighted flow-matching imitation learning. On two well-established multi-task robot manipulation benchmarks, a 0.1B-parameter ProgVLA model reaches success rates that are competitive with, and on long-horizon and harder task tiers exceed, substantially larger pretrained baselines. Ablations indicate that the learned context resampler and task-adaptive visual fine-tuning are the largest single contributors, while progress-aware training provides a consistent additional gain that is concentrated on long-horizon and multi-object tasks. We further validate the approach in real-world toy-kitchen environments.

Generating collision-free and smooth motions remains a central challenge in robotic manipulation, particularly in cluttered environments and narrow passages where feasible regions are highly constrained and fragmented. We propose a trajectory optimization framework that performs geometry-aware updates directly in function space using natural functional gradients. The method optimizes a Gaussian-smoothed surrogate objective that regularizes the optimization landscape through smooth trajectory perturbations while preserving trajectory-level structure. Because the updates are defined intrinsically in function space, trajectory regularity can be controlled independently of a particular time discretization. We derive a practical Monte-Carlo estimator of the natural functional gradient that requires only black-box trajectory evaluations, making the method applicable when analytic gradients are unavailable or unreliable due to collision checking and contact-rich simulation. Experiments on constrained robotic manipulation tasks demonstrate that the proposed method improves trajectory feasibility and produces smoother motions than representative planning and trajectory optimization baselines in environments with narrow geometric clearances. Additional results, videos, and implementation details are available at the project page: https://kisangpark.github.io/natural-functional-gradient/

Deep reinforcement learning (DRL) has been shown to achieve high performance on locomotion control tasks in MuJoCo benchmarks such as HalfCheetah, Ant, and Walker2D. However, visualizing the motion structures internally obtained by a trained policy function implemented as a deep neural network remains challenging. It is known from biomechanics and related fields that locomotion control is realized through the repetition of motion phases such as the stance phase and swing phase. In this study, we propose a framework for uncovering latent motion phase structures from trajectories generated by locomotion control policies through interaction with the environment. The proposed method extends the clustering features from state observations alone to augmented features including actions, next states, and next actions, and introduces a method for determining the number of clusters that suppresses self-transitions. Applying the proposed method to three environments -- Ant-v5, HalfCheetah-v5, and Walker2D-v5 -- we successfully identified phase structures with clearer and more regular transition rules than those obtained by the existing method.

Canary deployment routes a fraction of traffic to a new software version, monitors metrics, and rolls back on regression. Mainstream controllers (Argo Rollouts, Spinnaker, Flagger) change the deployed system's cryptographic identity during the canary window. The drift is harmless for stateless microservices but breaks the claim that "the agent you certified is still the agent you have" for safety-critical embodied agents, forcing re-certification per canary. We present ICAN-Deploy (Identity-stable CANary Deployment), a middleware construction whose state machine holds the identity hash invariant across the canary window by separating capability names (frozen, hashed) from capability versions (mutable runtime state). We implement ICAN-Deploy inside a runtime governance layer for LLM-driven robots and verify invariance by closed-form proof, AST lint, and TLA+ model-checking, then corroborate over N=100 real canary cycles on a Franka Panda arm in MuJoCo (zero drift; entry latency 95% BCa CI [1.52, 2.01] ms). A feature-flagged strawman that folds versions into the manifest falsifies on the same workload. A system certified once at identity-creation time can then ship arbitrary capability evolution under that same certification, within the version-and-name envelope.

Instructing robots from demonstrations can be done through different teaching modalities, each with different usability and performance trade-offs. This paper compares kinesthetic guidance, joystick teleoperation, and hand gestures in a user study with eight participants. We evaluate replay success, modified NASA-TLX workload, and common teaching errors across three manipulation tasks. Kinesthetic guidance produced the shortest demonstrations, lowest workload, and highest success on the more orientation-sensitive and contact-rich tasks. Joystick teleoperation performed best on simple peg picking. Hand-gesture teaching, although less reliable overall, performed better than expected and in some cases achieved results comparable to kinesthetic guidance.

We propose an optimization-based framework for robust contact-rich manipulation. Recent contact-implicit methods enable online hybrid planning across contact modes, allowing closed-loop manipulation for a given target state and contact location sequence of the robot and object. However, most existing approaches lack the ability to autonomously reason and generate diverse contact location sequences and manipulation trajectories, i.e., active contact location selection, which limits their applicability to relatively simple tasks. Active contact location selection is challenging due to complementarity in contact dynamics and the sparse gradients, making the design of a unified framework for contact selection and planning difficult. To address these challenges, we introduce Simultaneous Contact Selection and Planning (SCSP), a cascaded optimization framework comprising Contact Selection Optimization (CSO) and Contact Planning Optimization (CPO). CSO leverages a surrogate contact model and discrete-continuous optimization to efficiently resolve the nonsmoothness and coupling in contact selection, enabling online global searching of optimal contact locations. CPO performs prior-guided contact planning by evaluating the reference contact locations produced by CSO and generating corresponding manipulation trajectories in real time for redundant manipulators. Extensive simulations and real-world experiments demonstrate that SCSP produces diverse manipulation behaviors and robust control under inaccurate dynamics and perceptual noise. We further validate the generalization of the framework on challenging manipulation tasks. Project website: \href{https://sites.google.com/view/scsp-robot}{https://sites.google.com/view/scsp-robot}.

Visual odometry (VO) is a fundamental component in robotics and augmented reality. RGB-D direct VO benefits from metric depth measurements, but it can degrade in challenging environments, where dynamic objects, occlusions, illumination changes, and unreliable depth violate the short-horizon photometric and depth-geometric consistency assumptions used by direct alignment. Existing approaches mitigate these issues through semantic filtering, explicit occlusion reasoning, illumination adaptation, or hand-crafted geometric criteria, but often rely on external modules or fixed assumptions tailored to individual failure modes, limiting their flexibility and ability to handle diverse challenges in a unified manner. In this work, we propose Con-DSO, a consistency-aware RGB-D direct sparse odometry framework that predicts dense photometric and depth-geometric consistency uncertainty from temporally adjacent RGB-D frame pairs. The consistency network is trained using flow-guided photometric errors and projective depth-consistency errors, allowing consistency violations to be represented as pixel-level uncertainty. These pairwise uncertainty predictions are converted into a host-side quality prior for keyframe-based tracking. The prior is then applied to VO through quality-aware support-pixel selection and decoupled photometric-geometric weighting during pose estimation, enabling continuous attenuation of unreliable observations rather than hard rejection or threshold-based gating. Experiments on five public RGB-D benchmarks show substantial gains over direct RGB-D VO baselines, with over 20\% absolute trajectory error reduction on ICL-NUIM and 50\%--80\% reductions on RGB-D Scenes V2, TUM/Bonn Dynamic, and OpenLORIS sequences.

World Action Models (WAMs) improve robot manipulation by using video-based future representations to condition action generation. In pixel-space WAMs, however, the best action condition is not necessarily the fully denoised video. Controlled denoising-depth scans show that video refinement can reduce action error up to a state-dependent point, after which the gain may saturate or even reverse when late predictions become less action-relevant or physically unreliable. This suggests that action generation should use a state-dependent point along the video noise trajectory rather than a fixed terminal denoising depth. We introduce State-Adaptive Noise Trajectory Scheduler (SANTS), a lightweight scheduler for video-to-action diffusion policies. At each video decision point, SANTS reads the current video-state representation and noise level, then jointly predicts a cumulative stopping hazard and a relative noise-progression ratio. SANTS is post-trained with a path-level reward computed after the frozen action branch generates the final action chunk, so the scheduler is optimized for downstream action quality rather than intermediate video fidelity, while redundant video-state updates are explicitly penalized. Experiments show that SANTS reaches \(94.4\%\) overall success on RoboTwin 2.0 and \(73.1\%\) average success across seven real-robot tasks, while reducing latency by \(81.7\%\) and \(79.0\%\) relative to full video denoising, respectively. These results indicate that adaptive selection along the video noise trajectory can preserve the control benefits of WAM-style future reasoning while removing much of its redundant inference cost.

Learning visuomotor policies via behavior cloning typically involves mimicking expert demonstrations collected by human operators. However, natural human demonstrations inherently contain high-frequency noise, such as intermittent jerks, pauses, and action jitter. Training policies to directly imitate these raw trajectories inevitably causes the model to inherit these suboptimal behaviors. This pathology is particularly pronounced in diffusion-based policies, where iterative denoising steps can inadvertently amplify high-frequency artifacts at the expense of meaningful fine-grained details. To address these limitations, we present a novel frequency-based algorithm that enables implicit spectral maneuvering and smooth action generation. Our method, Frequency Guidance Operator (FGO), steers the generation process of diffusion polices by progressively driving the noisy samples through intermediate sub-frequency manifolds with expanding spectral bands. Validated on 15 robotic manipulation tasks from 5 benchmarks, FGO achieves superior performance in enhancing action smoothness and temporal consistency while preserving the details necessary for successful task execution. Project website: https://henrywjl.github.io/frequency-guidance-operator/

The biological spine of quadrupeds enables sagittal flexion/extension, lateral bending, and axial rotation, playing a crucial role in highly agile and dexterous locomotion. While numerous studies have integrated active spinal joints into quadrupedal robots to enhance agility, most designs simplify control complexity by reducing spinal degrees of freedom (DOF), failing to achieve the spatial tri-axial rotation characteristic of biological spines. Consequently, replicating a multi-DOF biomimetic spine and effectively leveraging it to empower the agile locomotion of quadrupedal robots remains a significant research challenge. In this study, we present S-Cheetah, a quadrupedal robot featuring a 3-DOF bio-inspired serial active spine capable of biomimetic spatial tri-axial rotation. To empower the robot to fully utilize this active spine, we developed a specialized reinforcement learning framework to actively promote the engagement of the introduced spine and maximize the robot's locomotive capabilities by integrating an acceleration curriculum learning strategy with tailored reward functions, such as a gallop gait reward, a spine undulation reward, and a spine steering reward. Experimental results demonstrate that S-Cheetah can achieve a peak speed of 6.9 m/s using the rotary G2 gallop gait and an in-place turning rate of 7.2 rad/s. Besides, the system exhibits an emergent, feline-inspired aerial self-righting capability, allowing it to land stably on four feet from arbitrary orientations during free fall. Finally, through extensive evaluations across diverse locomotion tasks, we prove that the introduction of the proposed 3-DOF spine comprehensively enhances the locomotive agility of quadrupedal robots. Project website: himmy-robotics.github.io/scheetah

Tactile sensing is essential for robots to achieve human-like gentle manipulation. However, existing Vision-Language-Action (VLA) models struggle to exploit tactile feedback for gentle manipulation due to scarce aligned vision-tactile-language data and the lack of effective closed-loop force feedback mechanisms. To address these challenges, we introduce Tabero, a benchmark and model suite for gentle, language-conditioned robotic manipulation that demands fine-grained contact force perception. First, the Tabero benchmark addresses the scarcity of tactile data by presenting a data-efficient pipeline that repurposes open-source robot manipulation trajectories to generate diverse vision-tactile-language tasks, and establishes a multidimensional evaluation protocol that measures task success alongside physical interaction quality. Second, we propose Tabero-VTLA, an architecture with a decoupled force-position command interface; the resulting force-position commands are executed by a fixed hybrid controller to enable real-time, force-aware manipulation. Evaluated on Tabero, our model maintains high task success while reducing average grip force by over 70\% under gentle instructions, demonstrating its ability to modulate interaction forces based on multimodal experience. Our code is publicly available at https://github.com/NathanWu7/Tabero.

Video generative models have emerged as a promising robotics backbone, capable of generating videos that depict the completion of complex tasks across embodiments and environments. Recent work proposes robot foundation models that jointly predict future observations and actions by finetuning video models with action-labeled data. In this paper, we test the limits of an alternative approach: leave the video planner as-is while training an embodiment-specific inverse dynamics model (IDM). This decoupling offers several natural benefits: the video planner remains embodiment-agnostic, different video models can be interchanged easily without re-training the IDM, and the IDM can be independently trained with readily available self-play data. We present a closed-loop, video-to-action policy that combines an action-free video world model with a carefully-designed IDM based on the robot embodiment Jacobian. We demonstrate that our IDM design is both data-efficient and scalable to high-dimensional action spaces. Our policy, which we coin the Video-to-Embodied Robot Action Model (VERA), achieves strong performance across simulated and real-world benchmarks, including zero-shot Panda arm manipulation and 16-DoF Allegro-hand dexterous cube re-orientation. The same video planner can be used across multiple embodiments by pairing it with different embodiment-specific IDMs. Our results show that decoupled video planning plus faithful video-to-action translation is a viable alternative route towards zero-shot, cross-embodiment, and generalizable robot control. More results are available on our project website: https://vera.csail.mit.edu.

Vision-Language-Action (VLA) models demonstrate promising generalization in robotic manipulation, driven by advances in large-scale vision and language pre-training. This progress can be misleading. Despite the zero-shot perception and language capabilities of VLAs, their overall task performance often degrades under distribution shifts, revealing gaps in how these systems translate high-level understanding into robust behavior. To systematically study this gap, we introduce Colosseum V2, a large-scale simulation benchmark for evaluating VLA generalization in robot learning across diverse conditions. The benchmark comprises 28 tasks spanning 13 task categories and two robot morphologies, covering a wide range of manipulation primitives and long-horizon behaviors. Built on the ManiSkill simulator, Colosseum V2 enables fast, GPU-parallelized evaluation and supports both in-domain and out-of-domain testing at scale. We evaluate state-of-the-art methods, including Action Chunking Transformers (ACT) and Pi0.5, and reveal limitations in both base performance and generalization. We demonstrate strong correlations between simulation and real-world metrics that support the ecological validity of the benchmark. By standardizing tasks, metrics, and evaluation protocols within a unified benchmark, Colosseum V2 enables reproducible and fair comparisons, reduced evaluation overhead, and accelerated progress toward general-purpose robot policies.

Imitation learning is a promising approach for training humanoid robots to both walk and manipulate, but it requires a large number of demonstrations, which are time-intensive and difficult to collect via teleoperation. Existing data-generation algorithms can automatically synthesize demonstrations for manipulators, but they are ineffective on humanoids because their high-dimensional composite action spaces involve arms, legs, and torsos. We present HumanoidMimicGen, a method for generating humanoid legged loco-manipulation data. Our method adapts contact-rich whole-body skills from a handful of source demonstrations to new states, generalizing across changes in object pose. By interleaving these single- and dual-arm skills with whole-body locomotion and manipulation planning, the method generates stable, collision-free data across diverse scenes and layouts. To evaluate our approach, we introduce a new simulated loco-manipulation benchmark containing nine diverse tasks that test humanoid loco-manipulation capabilities. There, we demonstrate that HumanoidMimicGen automatically generates large datasets for imitation learning and enables a systematic study of how data generation and policy learning decisions impact model performance. We show that whole-body visuomotor policies co-trained with data generated by HumanoidMimicGen outperform those trained only on real-world data by 20%.