University of Michigan（密歇根大学安娜堡分校）

Modern LLMs are alarmingly susceptible to surprisingly simple immaterial changes of input prompts: a casual hint, an incorrectly labeled few-shot example, or a fake prior assistant turn often flips an originally correct answer. We study where this susceptibility, spanning sycophancy and related cue-induced biases, lives inside the model. Across five model families and seven BCT bias types, we extract a per-bias direction from hidden states and triangulate it through three measures: probing, leave-one-dataset-out transfer, and causal intervention. The susceptibility is largely installed by alignment tuning rather than pretraining: pretrained base models barely cave to these biases, and their activations carry no cue-specific signal beyond question content. Within aligned models, each bias becomes a single coherent direction that we can both decode and steer along, recovering the unbiased answer across every family we test. The biases stay representationally distinct, however: cross-bias entanglement is model-specific rather than a property of the bias category, and even behaviorally similar biases occupy different directions. The same intervention also serves as a modest debiasing tool, recovering a meaningful share of bias-induced errors while preserving most correct answers across all instruct families. Cue-induced bias is therefore best understood not as a single flaw in LLMs but as a family of distinct, causally active directions that alignment tuning installs.

Test-time scaling improves foundation-model inference by spending additional computation, but robot control requires deciding whether extra compute is useful before executing an action. World Action Models (WAMs) make this decision natural: each rollout exposes both an action chunk and predicted future observations. We propose \methodgated, a training-free selective test-time scaling framework for WAMs. We first instantiate \method, a fixed-budget Best-of-$N$ selector that ranks sampled rollouts by cross-view depth reprojection consistency of their predicted futures, computed with a frozen geometry foundation model. \methodgated\ adds a lightweight action--future consistency gate that invokes \method\ only when the initial rollout appears internally inconsistent. Across five benchmark--backbone settings on RoboCasa, LIBERO Long, and RoboTwin~2.0, fixed-budget \method\ improves $N{=}8$ task success in every setting, e.g., raising the RoboCasa group average from $66.3\%$ to $68.4\%$ with Cosmos Policy and from $80.8\%$ to $82.5\%$ with X-WAM. With gating enabled, \methodgated\ recovers on average $74.8\%$ of the always-on success gain while triggering additional sampling on only $26.2\%$ of decision points. Offline diagnostics show that cross-view reprojection is a strong task-label-free selector, and we identify false low-score selections as a failure mode that helps explain why performance can saturate or degrade as $N$ increases.

Quality control in smart manufacturing increasingly relies on data-driven methods, particularly deep learning, to automate the inspection of manufactured parts. Recent advances in three-dimensional (3D) metrology have enabled fine-scale assessment of dimensional accuracy, surface quality, and shape conformity. However, deep learning methods for point-cloud-based inspection require large volumes of labeled data covering part designs and defect types, which are costly and time-consuming to obtain. Moreover, defective parts are intrinsically rare in mass production, and the resulting class imbalance can degrade model performance and make rare defect types difficult to detect. Synthetic data generation (SDG) offers a promising approach to address these challenges by producing large, balanced, and fully annotated datasets. Yet, applying SDG to precision components requires representing part geometry and defect morphology parametrically, so that design and quality can be co-varied. This article describes MFGNet-Gear, a publicly available synthetic 3D dataset comprising 24,000 paired polygon meshes and point clouds across 12 gear designs and 4 quality classes, with 500 instances per design-quality combination. Gear geometries are generated with parametric computer-aided design software, with dimensional parameters perturbed by $\pm$0.0254 mm and defect parameters sampled from distributions representing defect morphologies. For each mesh, 100,000 points are uniformly sampled using Open3D and stored as N $\times$ 3 coordinate text files. Metadata labels identify the gear design and quality class, supporting part design classification, geometric defect detection, representation learning, and dataset benchmarking. MFGNet-Gear provides an open-source dataset for deep learning-based 3D metrology, with a reproducible generation pipeline extensible to additional part designs.

We propose Q-Guided Value-Gradient Matching (Q-VGM), an off-policy reinforcement learning method for a central difficulty in fine-tuning flow-matching vision-language-action (VLA) policies: improving an expressive flow-matching action expert with a learned $Q$-function. Effective improvement must exploit the critic's first-order signal $\nabla_A Q$, yet flow policies make this hard: backpropagating values through the multi-step denoising chain is unstable at VLA scale, and the tractable action likelihoods required by policy gradients are unavailable under iterative denoising. Existing value-based methods therefore backpropagate through the full chain, use the critic only for test-time selection or guidance, or distill critic-improved actions as terminal labels that never supervise the velocity field. Q-VGM instead casts policy improvement as optimal control over the denoising dynamics, where the optimal residual velocity is the gradient of a denoising-time value function: clean-action estimates improved by iterative $Q$-gradient ascent with keep-best selection are converted into residual velocity targets that directly supervise the velocity field -- no action likelihoods, no backpropagation through the denoising chain, fully offline on a fixed replay buffer, and no critic at inference time. The critic is an action-sensitive stepwise IQL critic on compact latent states from the frozen VLA backbone. This enables a few-shot-initialization, learn-from-experience paradigm: starting from a few-shot-SFT $π_{0.5}$ policy, Q-VGM improves the policy from its own rollouts without additional expert supervision, raising the average LIBERO success rate from 79.0% to 92.5%, outperforming all same-backbone, same-critic baselines, and attaining high success rates on four real-robot manipulation tasks, including fine-grained plug insertion.

The rise of generative AI has enabled the production of high-fidelity synthetic tabular data across fields such as healthcare, finance, and public policy, raising growing concerns about data provenance and misuse. Watermarking offers a promising solution to address these concerns by ensuring the traceability of synthetic data, but existing methods face many limitations: they are computationally expensive due to reliance on the inverse process of large diffusion models, struggle with mixed discrete-continuous data, or lack robustness to common post-processing attacks. To address these limitations, we propose TAB-DRW, an efficient and robust post-editing watermarking scheme for synthetic tabular data. TAB-DRW embeds watermark signals in the frequency domain: it normalizes heterogeneous features via the Yeo-Johnson transformation and standardization, applies the discrete Fourier transform (DFT), and adjusts the imaginary parts of adaptively selected entries according to precomputed pseudorandom bits. To further enhance robustness and efficiency, we introduce a novel rank-based pseudorandom bit generation method that enables row-wise retrieval without incurring storage overhead. Experiments on five benchmark tabular datasets show that TAB-DRW achieves strong detectability and robustness against post-processing and adaptive attacks, while preserving high data fidelity and fully supporting mixed-type features.

Structured 3D representations such as keypoints and meshes offer compact, expressive descriptions of deformable objects, jointly capturing geometric and topological information useful for downstream tasks such as dynamics modeling and motion planning. However, robustly extracting such representations remains challenging, as current perception methods struggle to handle complex deformations. Moreover, large-scale 3D data collection remains a bottleneck: existing approaches either require prohibitive data collection efforts, such as labor-intensive annotation or expensive motion capture setups, or rely on simplifying assumptions that break down in unstructured environments. As a result, large-scale 3D datasets and benchmarks for deformable objects remain scarce. To address these challenges, this paper presents an affordable and autonomous framework for collecting 3D datasets of deformable objects using only RGB-D cameras. The proposed method identifies 3D keypoints and robustly tracks their trajectories, incorporating motion consistency constraints to produce temporally smooth and geometrically coherent data. TrackDeform3D is evaluated against several state-of-the-art tracking methods across diverse object categories and demonstrates consistent improvements in both geometric and tracking accuracy. Using this framework, this paper presents a high-quality, large-scale dataset consisting of 6 deformable objects, totaling 110 minutes of trajectory data. Project page: https://roahmlab.github.io/trackDeform3D-core-tracking/

Achieving safe, efficient, and kinematically feasible planning in dynamic environments remains a significant challenge, as planners must simultaneously handle moving obstacles, sensor uncertainty, and strict motion constraints. To address this problem, we propose an energy-parameterized diffusion planning framework that learns a conservative energy field to realize safe and stable generalization across diverse scenarios. The energy-parameterized diffusion formulation enables flexible integration of multiple constraints, allowing the planner to generalize to previously unseen environments without retraining. To ensure real-time safety during deployment, we further incorporate a lightweight safety filter that enforces safety and kinematic feasibility constraints in real-time. Additionally, we develop a scene-agnostic, MPC-based data generation pipeline to produce large-scale, dynamically feasible training trajectories. In simulation, the proposed method achieves real-time performance with a mean planning time of 0.21s and a low planning failure rate of 0.57%. Real-world experiments on the F1TENTH platform further validate the effectiveness of the proposed framework. Under sensor uncertainty in previously unseen dynamic environments, the planner consistently generates collision-free trajectories, which remain safe after being tracked by a simple controller, maintaining a mean obstacle clearance of 0.26 m, demonstrating strong robustness and practical applicability. Project page: https://rstp-comp-diffuser.github.io.

The link function, which characterizes the relationship between the preference for two trajectories and their returns, is a crucial component in designing RL algorithms that learn from preference feedback. Most existing methods, both theoretical and empirical, assume that the link function is known (often a logistic function based on the Bradley-Terry model), which is arguably restrictive given the complex nature of preferences, especially those of humans. To avoid mis-specification, this paper studies preference-based RL with an unknown link function and proposes a novel zeroth-order policy optimization algorithm called Sign-SZPO. Unlike typical zeroth-order methods, which rely on the known link function to estimate the value function differences and form a gradient estimator, Sign-SZPO only estimates the sign of the value function difference. It then constructs a parameter update direction that is positively correlated with the true policy gradient, eliminating the need to know the link function exactly. Under mild conditions, Sign-SZPO provably converges to a stationary policy with a polynomial rate in the number of policy iterations and trajectories per iteration. Empirical evaluations further demonstrate the robustness of Sign-SZPO under link function mis-specifications.

This paper presents a scalable, open-source visuotactile sensing system for tensegrity robots that enables six-axis wrench estimation and contact detection. The proposed endcap sensor integrates an elastomeric shell, a 3D-printed thermoplastic polyurethane (TPU) interface, and a rigid base housing an embedded camera and LED illumination ring. A novel gyroid-infill bonding technique is introduced to form a durable elastomer-TPU interface without adhesives, yielding a lightweight and modular design compatible with large-scale tensegrity structures. A tactile-to-wrench neural network maps shear vector fields to six-dimensional force and torque measurements. Experimental results demonstrate accurate and stable wrench estimation with a mean squared error (MSE) of 0.1531 on static validation data and out-of-domain generalization under dynamic motion. Furthermore, full-system integration on a 12 kg tensegrity robot confirms the sensor's ability to reliably identify ground contacts. The system substantially improves the practicality of tactile feedback for tensegrity robots, offering a low-cost, reproducible, and physically interpretable pathway toward contact-aware proprioception and state estimation. Open source files are available at \href{https://github.com/Jonathan-Twz/tensegrity-gelfoot}{github.com/Jonathan-Twz/tensegrity-gelfoot}

3D Gaussian Splatting (3DGS) captures scenes by coupling explicit geometry (position, covariance) with view-dependent photometry (Spherical Harmonics). However, building $\mathrm{SE}(3)$-equivariant architectures on these primitives presents a fundamental representation bottleneck. Color has been treated as a signal rather than a geometric entity, making it nontrivial to unify symmetry across geometry and appearance as the camera frame changes. While translations are handled by relative coordinates, rotations act heterogeneously across attributes: $μ\mapsto Rμ$, $Σ\mapsto RΣR^\top$, and $f_\ell\mapsto D^\ell(R)f_\ell$. This mismatch complicates strict equivariance, leading existing methods to either discard or flatten SH coefficients, thereby breaking symmetry. We propose a unified solution rooted in representation theory: for SH degrees $\ell\le2$, photometry is algebraically isomorphic to a rank-2 geometric tensor. We prove that the Wigner-$D$ action on these SH coefficients can be exactly reformulated as the conjugation action on $3\times3$ matrices. Leveraging this, we introduce the Unified Matrix Embedding, a lifting that maps all Gaussian attributes into a unified carrier space, $\mathfrak{gl}(3)$. Building on the "Color-as-Geometry" formulation, we present E3DGS, a rigid-body ($\mathrm{SE}(3)$) equivariant architecture that processes 3D Gaussians without Clebsch-Gordan tensor products. Evaluations on object vision and action-conditioned Gaussian world modeling demonstrate that our unified approach yields strong robustness under camera-frame changes and improved data efficiency.

Warning: This paper contains several toxic and offensive statements. Modern large language models (LLMs) are typically aligned through large-scale post-training to ensure fair and reliable behavior. In this work, we investigate how easily such guardrails can be broken by Group Relative Policy Optimization (GRPO). We show that one-shot GRPO training on a single biased example is sufficient to induce systematic bias, with stereotype-driven reasoning generalizing across attributes, categories, and benchmarks. We further find that models differ in their susceptibility based on the initial likelihood of producing biased outputs. Our results reveal a critical vulnerability in post-training: alignment can be overridden by a single example.

The proliferation of Large Language Models (LLMs) necessitates valid evaluation methods to provide guidance for both downstream applications and actionable future improvements. The Item Response Theory (IRT) model with Computerized Adaptive Testing has recently emerged as a promising framework for evaluating LLMs via their response accuracy. Beyond simple response accuracy, LLMs' chain of thought (CoT) lengths serve as a vital indicator of their reasoning ability. To leverage the CoT length information to assist LLM evaluation, we propose the \textbf{La}tency-\textbf{R}esponse \textbf{T}heory (LaRT) model, which jointly models both the response accuracy and CoT length by introducing a key correlation parameter between the latent ability and the latent speed. We derive an efficient stochastic approximation Expectation-Maximization algorithm for parameter estimation. We establish rigorous identifiability results for the latent ability and latent speed parameters to ensure the statistical validity of their estimation. Through both theoretical asymptotic analyses and simulation studies, we demonstrate LaRT's advantages over IRT in terms of superior estimation accuracy and shorter confidence intervals for latent trait estimation. To evaluate LaRT in real data, we collect responses from diverse LLMs on popular benchmark datasets. We find that LaRT yields different LLM rankings than IRT and outperforms IRT across multiple key evaluation metrics including predictive power, item efficiency, ranking validity, and LLM evaluation efficiency. Code and data are available at https://github.com/Toby-X/Latency-Response-Theory-Model

The increasing adoption of neural networks in learning-augmented systems highlights the growing need for model safety and robustness, especially in safety-critical domains. While recent advances in neural network verification offer formal guarantees on worst-case behavior, existing approaches require users to manually define model specifications, an error-prone, incomplete, and time-consuming process. In this paper, we present AutoSpec, the first comprehensive framework for automatically generating and evaluating neural network specifications for learning-augmented systems. AutoSpec introduces a tree-based algorithm that adaptively partitions the input space to generate specification sets aligned with model behavior, as well as a statistical certification framework that provides rigorous accuracy guarantees for each specification. We also propose a principled evaluation framework that defines interpretable metrics for specification accuracy and coverage, establishing a benchmark for future research. Experiments across four diverse applications show that AutoSpec outperforms both manually defined specifications and existing baseline algorithms, improving the F1 score by up to 53% over human-defined specifications and 73% over the strongest baseline.

A common bottleneck in two-stage recommendation is embedding staleness: when a user rates a new item, their embedding remains fixed until the next retrain cycle. We propose mutable sketches, which store each user's preferences in a KP-tree (a sparse segment tree with sum aggregation), fit a low-rank projection once, and recompute embeddings on-the-fly as ratings arrive. We prove that each new observation monotonically tightens the prediction error envelope (Theorem 1), a guarantee that FunkSVD and eALS lack. On KuaiRec, the mutable sketch achieves 0.810 RMSE at 1.8% data read vs. ALS 0.822 at 100%, with 8x faster per-batch updates. A new user receives personalized recommendations in <1 ms after their first rating, with no model retraining required. A comparison of sampling strategies across density regimes shows that the KP-tree's norm-proportional sampling provides 40-130% better item coverage on sparse data (<1% density), while uniform sampling suffices on dense matrices.

Robots are increasingly integrated into everyday contexts, including museums, where they can both entertain and educate visitors. To enhance visitor experience and engagement, we present a novel mixed-agent tour guide system that combines a physical robot with a projected virtual agent that actively participates in the tour through conversation and interaction, achieving the interaction richness of two mobile agents from a single platform. We validate the system through a within-subjects study with 30 participants to assess engagement, quality of experience, and learning performance. Participants experienced different conversational styles and agent configurations, and data were collected via surveys, behavioral sensors, and interviews. Results showed that engagement and quality of experience remained consistent across conditions. Learning performance revealed a significant gender-moderated difference: the mixed-agent conditions improved learning performance for female participants. This suggests that the proposed dyadic conversational style in this paper influenced learning performance differently by gender. Nonetheless, in interviews, participants reported a greater preference for mixed-agent teams regardless of gender, citing interaction as a key factor in their experience.

Federated fine-tuning of large pre-trained models increasingly relies on Low-Rank Adaptation (LoRA) to reduce communication and computation, but heterogeneous clients can make adapter aggregation unstable. We identify the data-parameter interference as a geometric source of this instability. This interference is controlled by the alignment between LoRA update subspaces and client activations, suggesting that federated LoRA aggregation should be viewed not only as parameter averaging but also as subspace allocation. We propose Dynamic Subspace Boosting (Dysco), a plug-in method that allocates client-specific LoRA subspaces in a federated and dynamic manner. In each round, clients compute activation-insensitive subspaces from local representations and transmit only the resulting bases; the server then constructs client-specific merged subspaces through a closed-form solution that maximizes compatibility with other clients' insensitive directions. To handle representation drift, Dysco performs multi-round subspace boosting to preserve past update directions while adapting to future representations. We provide a convergence analysis that embeds the data-parameter interference as an aggregation-error term in a standard federated optimization bound, and prove that Dysco's server-fixed merged subspaces yield a tighter upper bound on this error. Experiments on controlled synthetic federated tasks and on MIMIC-IV clinical-note classification with Llama-3.2-1B show that Dysco substantially reduces interference, reduces the final-round synthetic training loss by up to 9 times relative to baselines under the orthogonal-subspace partition the theory identifies, improves all five tested FL algorithms by up to 4.3% on MIMIC, outperforms recent federated LoRA methods, and adds only 0.9% wall-clock overhead. Our code is available at https://github.com/illidanlab/Dysco.

Zero-shot object-goal navigation (ZSON) requires navigating unknown environments to find a target object without task-specific training. Prior hierarchical solutions mainly focus on either scene understanding and representations (belief) or high-level decision-making and planning (policy), yet treat the option, i.e., the subgoal candidate that belief proposes and policy selects, as an interface inherited from adjacent modules rather than a design axis in its own right. In practice, options are predominantly single waypoints scored by destination utility: a lone destination hides the value gathered en route, and a flat list obscures the relationships among candidates. Our insight is that the option space should be a tree of paths. Full paths expose en-route information gain that destination-only scoring systematically neglects; a tree of shared segments enables coarse-to-fine LLM reasoning that dismisses or pursues entire branches before examining individual leaves, compressing the combinatorial path space into an efficient hierarchy. We instantiate this insight in REST (Receding Horizon Explorative Steiner Tree), a training-free framework that (1) builds an explicit open-vocabulary 3D map from online RGB-D streams; (2) grows an agent-centric tree of safe and informative paths as the option space via sampling-based planning; and (3) textualizes each branch into a spatial narrative and selects the next-best path through chain-of-thought LLM reasoning. Across the Gibson, HM3D, and HSSD benchmarks, REST consistently ranks among the top methods in success rate and path efficiency.

Large language models (LLMs) have demonstrated significant potential to accelerate scientific discovery as valuable tools for analyzing data, generating hypotheses, and supporting innovative approaches in various scientific fields. In this work, we investigate how LLMs can handle the transition from conceptual research ideas to well-structured research plans. Effective research planning not only supports scientists in advancing their research but also represents a crucial capability for the development of autonomous research agents. Despite its importance, the field lacks a systematic understanding of LLMs' research planning capability. To rigorously measure this capability, we introduce the Idea2Plan task and Idea2Plan Bench, a set of benchmarks built from ICML 2025 and Nature Mental Health papers released after major LLM training cutoffs. Each benchmark instance includes a research idea and a grading rubric capturing the key components of valid plans. We further propose Idea2Plan JudgeEval, a complementary benchmark to assess the reliability of LLM-based judges against expert annotations. Experimental results show that GPT-5 achieves the strongest performance on the benchmark, though substantial headroom remains for improvement. Our study provides new insights into LLMs' capability for research planning and lays the groundwork for future progress.

Multimodal speaker identification systems typically assume the availability of complete and homogeneous audio-visual modalities during both training and testing, and assume each speaker only speaks a single language. However, in real-world applications, such assumptions often do not hold. Visual or audio information may be missing due to occlusions, camera or microphone failures, or privacy constraints. Multilingual speakers introduce additional complexity due to linguistic variability across languages. These situations constitute substantial challenges for the robustness and generalization capabilities of multimodal speaker identification systems. Aim of the POLY-SIM 2026 challenge is to address these aspects of speaker identification and to provide a standardized setup for the comparison of the proposed solutions.

Multi-robot-arm motion planning is a key challenge in deploying multiple manipulators for industrial tasks such as manufacturing. Existing search-based and sampling-based solvers often require significant computation time to produce collision-free, high-quality motions suitable for safe real-world execution. In this work, we introduce a new suite of multi-robot-arm motion planners capable of near real-time motion generation, combining classical planning algorithms with state-of-the-art vectorized collision-checking techniques. Based on CPU SIMD instructions, our new planners accelerate their primary bottleneck, collision checking, and achieve up to two orders of magnitude speedup in both motion planning and execution postprocessing for multi-arm manipulation tasks. We also release our implementation to lower the barrier for research and development of multi-robot-arm planning and manipulation problems. Code is available at https://vamp-mr.github.io/vamp-mr

Lower-limb powered exoskeletons have demonstrated substantial improvements in mobility and function, but most designs place actuation components lateral to the legs or remotely at the waist or back. These configurations often extend beyond the body's natural envelope, making devices intrusive in everyday use and potentially limiting societal adoption. We posit that rethinking actuation geometry could enable exoskeletons that conform more closely to the body. Here, we explored an actuation layout in which the actuator encircles the limb in a plane orthogonal to the limb axis, potentially reducing its spatial footprint around the body. We developed the Limb-Encircling Actuator (LEA) and characterized its electromechanical properties using a custom-built testbed. The LEA also features a novel radial bearing layout with potential as a lightweight or lower-cost alternative to traditional large-diameter bearings. The actuator achieved a continuous torque density of 7.5 Nm/kg with a mass of 894 g. Despite this high torque density and innovative layout, the system remained difficult to contain close to the body. These results highlight opportunities and challenges in limb-encircling actuation and provide insights into torque-dense exoskeleton designs that could integrate more readily into everyday apparel if challenges in actuator sizing and geometry are overcome.

Traffic forecasting is highly challenging due to complex and nonlinear spatial and temporal dependencies. Self-attention mechanisms have been widely adopted to model dynamic and long-range dependencies, achieving state-of-the-art performance, but suffer from limited scalability due to quadratic computational and memory complexity. To address this, we propose an Efficient Multi-Attention Graph Network (EMAGN) that linearises the spatial attention mechanism itself, inspired by the theory of fast high-dimensional Gaussian filtering. Two learned clustering matrices C_k and C_v adaptively group key and value vectors into M super-clusters, reducing complexity from O(N^2 d) to O(NMd) without sacrificing the flexibility of attention for dynamic dependency modelling. Experimental results on PEMS-BAY and METR-LA show that EMAGN achieves accuracy within 2.7-3.2% MAE of full-attention GMAN while reducing training time by 32%, inference time by 38%, and GPU memory by 58%. Critically, at K=16 attention heads, full-attention GMAN runs out of memory on a standard 11 GB GPU entirely while EMAGN continues to operate, demonstrating a categorical expansion of feasible model configurations. EMAGN also surpasses Linformer and Performer in both accuracy and efficiency within the same backbone, owing to its traffic-network-aware adaptive clustering.

As Large Language Models (LLMs) increasingly utilize massive context windows as working memory for autonomous tasks, their reliability fluctuates significantly depending on how information is distributed in real-world corpora. We investigate how fact placement, corpus-level distributions, and anti-hallucination ("Don't Make It Up") prompts influence model behavior by introducing a model-agnostic extended needle-in-a-haystack benchmark designed for scalability, which we apply to evaluate Gemini-2.5-flash, ChatGPT-5-mini, Claude-4.5-haiku, and Deepseek-v3.2-chat. Unlike prior work, we separately evaluate literal extraction, logical inference, and hallucination risk. We identify two critical failure modes: Distributional Collapse, where performance degrades significantly when evidence is dispersed; and a Safety Tax, where anti-hallucination prompts cause over-conservative refusal of present facts and evidence, sharply reducing accuracy. Our results suggest that many failures stem from ineffective context utilization, as models struggle to prioritize relevant information even when it is present. These findings highlight the need for model-specific robustness and effective context management to ensure reliable deployment in long-horizon agentic workflows.

A humanoid that can walk should not relearn locomotion from scratch to jog or run. Yet current approaches often obtain gait diversity by prescribing gait schedules, imitating motion clips, training experts to switch between or distilling skills into one policy. These strategies can produce impressive behaviors, but offer limited flexibility across continuous speed commands, terrains, and morphologies. We study skill growth with GaitSpan, a framework that expands a pretrained, basic walking policy into faster locomotion. It treats walking as a seed skill: reusable motor structure for balance, support, body coordination, and contact transition that can be regenerated at new rhythms, extended into longer/higher strides, and corrected by residual adaptation. This expansion has three aspects: 1) rhythm generation, which modulates the frozen walking policy with multiple internal clocks and learns command-conditioned combinations of the resulting canonical actions; 2) stride shaping, which rewards dynamic locomotion patterns appropriate for higher commanded speeds using a physically grounded objective inspired by spring-loaded inverted pendulum dynamics; and 3) residual adaptation, which captures motion details not accounted for by rhythm generation or stride shaping. GaitSpan is the first to deliver a single command-conditioned humanoid policy that spans walking, jogging, and running-like regimes covering a continuous speed range, transfers across morphologies, and deploys zero-shot on unseen sim-to-sim, and real-world terrains. Compared with baselines either trained with multi-experts or imitation from humans, it learns faster and achieves stronger gait performance.

Autonomous driving requires long-horizon closedloop decision making in dynamic traffic environments. Latent world models offer an effective framework for this problem by enabling imagination-based decision making in compact latent spaces. However, multi-source observations contain controlirrelevant redundancy, whereas reliable driving decisions rely on risk-relevant relations, future dynamics, and continuous action adjustments. This mismatch makes observation reconstruction and absolute action modeling suboptimal for learning decisionrelevant latent dynamics. We propose LIDAR-AD, a decoderfree Latent-Interaction Dreamer with Action-Residual Chains for autonomous driving. LIDAR-AD replaces observation reconstruction with redundancy-reduced latent alignment, encouraging compact representations of risk-relevant relations in multi-source driving inputs. It further models vehicle control as residual action updates and uses residual-action sequence contrastive learning to align multi-step residual-driven rollouts with future latent states. A deterministic analysis shows that the latent-tanh residual parameterization preserves interior action reachability while representing smooth long-horizon control as compact local updates. Together, these designs improve risk-aware state abstraction, continuous-control modeling, and long-horizon dynamics prediction. Extensive experiments across diverse simulated driving scenarios demonstrate that LIDAR-AD consistently outperforms world-model baselines, achieving the highest reward and the best success rate among learning-based methods. Evaluations on nuPlan-derived log-reconstructed scenarios further demonstrate the transferability of LIDAR-AD under real-world traffic layouts.

Visual perspective taking--inferring how the world appears from another's viewpoint--is foundational to social cognition. We introduce FlipSet, a diagnostic benchmark for Level-2 visual perspective taking (L2 VPT) in vision-language models. The task requires simulating 180-degree rotations of 2D character strings from another agent's perspective, isolating spatial transformation from 3D scene complexity. Evaluating 103 VLMs reveals systematic egocentric bias: the vast majority perform below chance, with roughly three-quarters of errors reproducing the camera viewpoint. Control experiments expose a compositional deficit--models achieve high theory-of-mind accuracy and above-chance mental rotation in isolation, yet fail catastrophically when integration is required. This dissociation indicates that current VLMs lack the mechanisms needed to bind social awareness to spatial operations, suggesting fundamental limitations in model-based spatial reasoning. FlipSet provides a cognitively grounded testbed for diagnosing perspective-taking capabilities in multimodal systems.

We present SheetMind, a modular multi-agent framework powered by large language models (LLMs) for spreadsheet automation via natural language instructions. In this paper, we introduce a hierarchical agentic system consisting of three specialized agents: Manager Agent that decomposes complex user instructions into subtasks; an Action Agent that translates these into structured commands using a Backus-Naur Form (BNF) grammar; and a Reflection Agent that validates alignment between generated actions and the user's original intent. We evaluate SheetMind on the 221-task SheetCopilot Benchmark with GPT-3.5-Turbo. SheetMind achieved 100% execution success and 54.8% functional correctness, exceeding SheetCopilot (44.3%) while maintaining perfect execution reliability. We also conduct ablation study on a separately curated dataset to confirm that the full three-agent configuration consistently outperforms all partial variants. Lastly, we integrate our system into Google Sheets via a Workspace extension.

Existing studies of LLM-as-judge scoring bias work predominantly at the input-output level: they perturb inputs, measure score deltas, and propose prompt-level mitigations. We argue that the same biases admit a representation-level account in the judge's hidden state, complementary to the input-output view and operationally useful in ways it does not afford. We report three findings, across seven judges, seven bias types, and nine benchmarks. Geometry: baseline judging inputs occupy a tight activation manifold while biased inputs are displaced along a low-dimensional, type-specific subspace that sharpens with depth and is recovered consistently by three families of estimators. Causal control: steering hidden states along this subspace drives scoring in both directions, forward shifts reproducing biased scoring on clean inputs and reverse shifts restoring baseline scoring on biased ones, while matched-norm random directions produce shifts an order of magnitude smaller. Operational: a simple linear projection onto the same bias-direction features anticipates judge failures on three entirely unseen benchmarks, substantially outperforming text-based alternatives. Reading bias as activation geometry, rather than as input-output noise, unifies geometric structure, causal control, and operational prediction within a single framework. The project page is available at https://xzx34.github.io/unfair-judge/

Language models are increasingly used for moral decision-making across diverse linguistic and cultural contexts, yet existing work overlooks multilinguality on three aspects: 1) multilingual evaluation benchmarks use direct translation, failing to adapt culture-specific items; 2) inference-time methods for moral reasoning rely on static, English-centric scaffolds and lack grounding in moral theory; 3) training methods for moral decision-making typically require expensive supervision from stronger models or human annotators. We address these gaps with three contributions. First, we introduce MCLASH, a multilingual moral decision-making benchmark to capture culturally situated moral intuitions and social norms across languages. Second, we propose MET (Multilingual Ethics with Theory-grounded reasoning), a two-step prompting method built on expert-curated, theory-based grounds drawn from psychology and philosophy: the model first selects situation- and culture-specific grounds, then reasons over them in the native language of the user. Third, we introduce MET-D (MET-Distillation), which enhances the second step through a self-distillation training stage that requires no external supervision. MET-D improves macro-F1 over the base model on all three models of different sizes and families (Qwen3-4B, Qwen3-8B, Gemma3-4B), by an average of 3.71 points on MCLASH and 4.23 on MMoralExceptQA, with a peak MCLASH gain of 12.94 points for Malay on Qwen3-8B. We further reveal that MET-D increases native-language reasoning by 62.13 points on average, and that beneficial grounds differ systematically across cultures. Together, these contributions open the path for culture-aligned, theory-grounded multilingual moral reasoning.

A trained molecular property model can be refined at test time by correcting each prediction with the measured labels of the most similar training molecules, a retraining-free procedure we call neighbor fusion; evidential neural networks make it principled by using their aleatoric and epistemic uncertainty to parameterize a Bayesian update. Our main contribution, PG-EVIKAL, learns a property-distance metric to re-rank structurally similar neighbors by their property relevance before fusion, building on EVIKAL (scalar Kalman filter) and GP-EVIKAL (Gaussian process variant handling correlated neighbors). Evaluated on 16 molecular datasets, PG-EVIKAL reduces RMSE relative to the evidential model baseline on 14 of them, with a median reduction of 19.4%, and improves calibration; in sequential-assay scenarios it further incorporates newly measured molecules, refining predictions as they arrive without retraining. This work demonstrates that evidential uncertainty decomposition is not merely a calibration objective but an actionable inference resource that enables test-time refinement of molecular property predictions.