Papers

Tool-using agents are continually updated with new interaction data. After each policy update, however, previously estimated action credits may become stale. Recomputing them from scratch can require many additional tool calls and environment interactions, making repeated updates increasingly expensive. We ask a simple question: when does historical action credit actually need to be updated? Our key observation is that a change in action value does not necessarily imply a change in the decision. Historical credit can still be useful as long as policy-induced drift is too small to overturn the existing action ranking. Building on this idea, we introduce pairwise branch sensitivity to capture how strongly a policy update affects the downstream regions that distinguish two candidate actions. We then derive a first-order anchored credit-transport estimator that updates historical credit using old interventional trajectories, and propose a Decision-Sufficient Credit Gate (DSC-Gate) that chooses whether to reuse, transport, or resample credit. Experiments show that branch sensitivity explains credit drift substantially better than global policy distance. With sufficient historical data, credit transport reduces estimation error, while its benefit to decision making is concentrated on updates that affect action-distinguishing branches. On a fully independent test set, DSC-Gate changes mean regret by only +0.00004 relative to a gap-based gate while reducing mean new tool steps from 472 to 286, a 39.4% reduction. We observe the same pattern after a real tool-agent parameter update. Overall, our results show that agents do not need to recompute action credit after every policy update: much of the historical evidence can be reused or cheaply corrected, reducing the additional interaction required to keep action decisions up to date.

Existing video deraining methods typically exploit neighboring frames through either explicit alignment or implicit spatiotemporal aggregation. Explicit alignment relies on accurate motion estimation, which can become unreliable under dense rain, while implicit aggregation avoids alignment but lacks explicit guidance on the directional and temporally coherent structure of rain. This leaves a gap between reliable temporal aggregation and explicit modeling of rain motion. To address these limitations, we propose FluidRain, a lightweight video derainer that uses divergence-free rain flow to guide Loop-in-Loop attention across scales and neighboring frames. Motivated by fluid mechanics, we model rain motion as a divergence-free image-space flow and use it to organize multi-scale and temporal aggregation. Specifically, FluidRain first estimates a rain-flow field for each frame and projects it onto the divergence-free subspace. The resulting flow steers window attention along rain streaks, enabling neighboring frames to be aggregated without explicit alignment. Since rain-flow structure is preserved across scales and nearby frames, Loop-in-Loop reuses the same attention operator across both dimensions, resulting in a three-frame model with only 0.80M parameters. Experiments on four benchmarks show that FluidRain remains competitive with substantially larger restoration models. We further examine how temporal evidence scales with different input views. To evaluate whether the model remains reliable when rain motion changes across frames, we introduce RainSyn-Gust, which injects controlled changes in rain-streak direction into existing benchmarks. We also develop a physics-based no-reference metric that evaluates real-rain removal without requiring clean targets.

Despite strong performance on standard benchmarks, it remains unclear whether large language models (LLMs) evaluate social pragmatics in ways that align with human judgments. We evaluate LLM politeness judgments using two English-language datasets with complementary annotation formats: continuous human ratings and three-way categorical labels. Across the seven evaluated models, we find that inter-model agreement is stronger than model–human agreement. Strategy-level analyses suggest that model–human alignment is associated with explicit linguistic cues, while some rapport-building strategies occur more frequently in misaligned cases. In the categorical task, model predictions exhibit systematic neutral compression, characterized by the overproduction of Neutral labels and the underprediction of Impolite labels. This pattern persists when expert consensus is used as the reference on a diagnostic subset. Our findings highlight the need for pragmatic evaluations that go beyond aggregate agreement metrics by examining directional patterns of model–human disagreement across different human references.

Robot policies deployed in real environments naturally accumulate mixed-quality experience, including successful executions, partial progress, and failures. Although these rollouts provide valuable information for further learning, directly incorporating them into imitation learning may reinforce undesirable behaviors, while offline reinforcement learning often suffers from unreliable value estimation under sparse rewards and limited data coverage. We consider a practical post-deployment setting where learning relies only on naturally accumulated autonomous rollouts, without additional human corrections or exploratory interaction. To effectively exploit such experience, we propose Predictive Action Chunk Learning (PACL). PACL first learns a predictive chunk-level critic that evaluates temporally extended action sequences and augments temporal difference learning with future latent prediction, providing richer supervision for long-horizon value estimation. The learned critic then converts chunk-level Q-values into discrete quality conditions, which guide a diffusion actor to learn jointly from these mixed-quality experiences without treating all behaviors as equivalent supervision. At inference, the actor generates multiple action chunks and the critic selects the highest valued candidate. Experiments across simulated and real-world robot manipulation tasks show that PACL consistently improves the pretrained policy and outperforms strong imitation learning and offline reinforcement learning baselines.

Low-rank adaptation (LoRA) has become a popular parameter-efficient fine-tuning method for large language models. A key challenge in LoRA is how to determine the rank of each adaptation matrix, as rank directly controls its capacity and efficiency. Existing adaptive-rank methods typically allocate ranks according to manually designed importance scores, which are not directly derived from an optimization objective. In this work, we propose $\ell_p$-LoRA, a principled rank-allocation method based on $\ell_p$ regularization with $0<p<1$, which is a classical sparsity-inducing technique in signal processing and statistics. Specifically, we regularize the energy of each rank-one LoRA component, encouraging redundant components to vanish while preserving important ones. We derive the corresponding proximal subproblem and reduce the matrix optimization to a two-dimensional problem, leading to an implicit thresholding criterion for identifying redundant components. Experiments on natural language understanding and question-answering tasks demonstrate that the proposed method achieves competitive performance with existing LoRA baselines.

Most of the memory of a 3D Gaussian Splatting model holds spherical-harmonic colour coefficients, yet each Gaussian is seen only from the narrow cone of directions of the training cameras. We turn this into a distortion metric that other compressors can adopt: a per-Gaussian observation Gram matrix, accumulated from viewing directions and blending weights, is the exact first-order map from coefficient changes to squared image error and needs only the model and the camera poses. Under it, degree reduction becomes a closed-form projection that generalises truncation, degree allocation a Lagrangian rate-distortion problem, and vector quantisation the matrix-weighted Lloyd algorithm, of which Compressed3D's quantiser is the scalar case. Swapped into Compressed3D with everything else unchanged, the metric raises PSNR by +0.49 dB before fine-tuning, with SSIM and LPIPS following, and at matched rate still gains +0.32 dB without a single training image. A training-free stack built on the metric alone is 15% smaller than the image-free GSICO at equal quality on Mip-NeRF 360.

The Node.js ecosystem heavily relies on NPM packages, and software supply chain attacks targeting malicious NPM packages are rampant. Malicious code primarily triggers during package installation, import, and runtime. Traditional static analysis fails to understand code semantics; machine learning-based methods rely on feature extraction, which suffers from concept drift; existing LLM solutions suffer from high invocation costs, high data security risks, and poor performance. To overcome these limitations, we propose DistillGuard, a lightweight malicious NPM package detection framework that combines static graph semantic analysis and LLM knowledge distillation. DistillGuard first acquires multi-granular features through three static analysis modules. Then, it leverages online LLM to distill high-quality security knowledge and structured labels. Finally, it uses LoRa to efficiently fine-tune the open-source Qwen3-8B model to support offline deployment. Experiments show that DistillGuard achieves an accuracy of 95.3%, a precision of 99.4%, and an F1 score of 93.8%, outperforming state-of-the-art tools, improving the F1-score by 11.1 to 30.0 percentage points over the baselines. Our empirical research further reveals the stages of malicious attacks and the distribution of malicious behaviors. We also summarized eight typical API attack chains for malicious activities, providing practical insights for NPM supply chain security defense.

We develop and analyze HDG methods for two central problems in finite element exterior calculus, the Hodge-Dirac problem and the Hodge-Laplace problem, in arbitrary dimension $n$. Our analysis allows for equal-order polynomial spaces with $\mathcal{O}(1)$ penalty parameters, by contrast with previous work on the Hodge-Laplace problem requiring an underlying conforming complex and $\mathcal{O}(h)$ or $\mathcal{O}(h^{-1})$ penalties. For the Hodge-Dirac problem with degree-$r$ polynomials, our error estimates give optimal order-$(r+1)$ convergence in all form degrees under suitable regularity hypotheses. For the Hodge-Laplace problem, we prove optimal order-$(r+1)$ convergence for the $k$-form solution and order-$(r+\frac{1}{2})$ convergence for the $(k \pm 1)$-form auxiliary variables. We obtain improved order-$(r+1)$ auxiliary-variable estimates when $k=1$ and/or $k=n-1$, which in particular sharpens some recent HDG error estimates for the two-dimensional vector Poisson equation. Our analysis also encompasses cases of lower regularity, including solutions with reentrant-corner singularities on non-convex domains, and requires only mild mesh regularity conditions. The results are illustrated by numerical experiments in dimensions two and three.

Video understanding is increasingly performed by multi-stage LLM agents that separate temporal grounding, visual observation, and reasoning. Yet these stages are typically evaluated on different benchmarks and distributions, making it difficult to determine where hallucinations originate. We first organize existing benchmarks around these stages and show that their scores provide inconsistent diagnostic signals: stronger stage-level performance does not reliably imply lower downstream hallucination, and even benchmarks targeting the same capability can disagree. We therefore introduce a causal stage-intervention protocol that overwrites individual stages while holding the downstream task fixed. Across 60,008 runs on three video-agent architectures, we find that grounding is the dominant source of downstream error, with roughly four times the causal impact of corrupting visual observations. Successful grounding depends primarily on locating the correct region rather than precise temporal overlap, explaining why standard mIoU metrics poorly predict downstream reliability. We further find that incorrect evidence is substantially more harmful than missing evidence. Finally, auditing existing benchmarks against these interventions reveals that their scores do not reliably predict causal cascade sensitivity and can fail under distribution shift. These results motivate intervention-based, stage-aware evaluation for trustworthy video agents.

Generative AI chatbots are increasingly embedded in everyday life, yet most large-scale studies describe global patterns. This paper presents an Australia-focused analysis of WildChat, a public dataset of real-world ChatGPT interaction logs. Using descriptive analysis and a multi-layer classification scheme, we analysed 37,845 conversations identified as Australian, examining language diversity, work relevance, interaction intent, topic distribution, turn-taking, temporal change, work activities, and Australia-related domains. Our findings show that the Australian subset is strongly action-oriented and comparatively work-oriented, with most interactions classified as doing and a majority of conversations classified as work-related. The dataset also shows multilingual use and a growing presence of self-expression over time. Australia-related conversations frequently invoke local institutions, laws, regulators, education systems, companies, cultural references, and public services. Finally, we outline implications for future research, including local AI evaluation, multilingual participation, context-aware design, and safeguards for everyday high-stakes domains.

We present a personalized Korean visual speech recognition (VSR) system and quantify, on the nine-camera OLKAVS corpus, the gap between the population-level benchmark score and an individual user's error. A video-only Conformer initialized from English-trained weights attains 9.95 - 12.19% character error rate (CER) under the corpus protocol against the published 26.64, and 19.00 - 21.52 on unseen wording. Per speaker, CER spans 1.0 to 52.2%, with seen wording lowering CER by 7.0 - 9.0 points and professional delivery and spontaneous speech raising it by 8.5 - 10.5 and 12.7 points. A low-rank adapter with 4.6% of the parameters, trained on 4 to 29 minutes of the user's frontal video, lowers the CER of twelve high-error speakers by 2.13 to 3.58 points, transfers to every camera without loss, and keeps 85% of the full fine-tuning gain at 12% of its cost to other speakers. Cameras above the mouth plane add about six CER points as a constant offset that training on all views keeps small.

Cross-embodiment learning has shown that a single model, such as a vision-language-action (VLA) model, can learn state representations and manipulation skills that can be applied across heterogeneous robots to accomplish various tasks. We hypothesize that the same holds for safety enforcement. The reasoning required to satisfy a safety constraint, such as detecting an obstacle, recognizing that it should be avoided, and selecting a safe abstract action, is largely shared across robots. What differs across embodiments is how the abstract safe action is realized: morphology, kinematics, and dynamics determine which actions are safe and feasible. Consequently, the same action can be safe for one robot and unsafe for another. This is especially important for generalist manipulation policies that operate in a common end-effector action space without explicitly capturing how safety depends on the robot's morphology and kinematics. We propose embodiment-conditioned safety filtering, in which a Hamilton-Jacobi reachability-based value function and its corresponding safety-maximizing policy are shared across robots. Using a morphology-aware latent representation of the robot and its environment, we perform Hamilton-Jacobi reachability analysis directly in latent space so that the learned safety concepts can generalize across embodiments while remaining explicitly conditioned on each robot's morphology and kinematics. We evaluate our approach across five bimanual robot embodiments and five manipulation tasks with whole-body collision-avoidance constraints. Our results show that a single policy, jointly trained across five manipulation tasks and four embodiments, exhibits zero-shot generalization to a held-out embodiment, reducing the nominal policy's collision rate. They also show that training using more embodiments improves generalization.

LLM-based retrievers and rerankers have advanced passage ranking, yet both paradigms interact with the corpus in a single pass and commit to the resulting candidate set, leaving relevant documents permanently unrecoverable once missed. We introduce Seek, Self-Evaluative Exploration for Knowledge Retrieval, a training-free framework that addresses this limitation through iterative corpus interaction at test time. At each round, an LLM generates pseudo-passages conditioned on accumulated relevance feedback, a retriever surfaces fresh candidates, and a dedicated assessor assigns graded relevance judgments that guide subsequent rounds. On TREC Deep Learning, Seek matches trained rerankers in ranking quality while consistently improving Recall@100 over single-pass BM25. On the reasoning-intensive BRIGHT benchmark, Seek with Qwen2.5-7B achieves an 82% relative gain over BM25, surpassing all trained baselines, and Seek with GPT-4.1 reaches 37.4 average nDCG@10, exceeding the strongest baseline by 37%.

We study spectral graph neural networks built from Hermite polynomials and propose HermNet, a simple model that combines a nodewise predictor with normalized Hermite propagation. Its sparse recurrence requires neither eigendecomposition nor a learned basis. We distinguish the basic model from optional coordinate calibration, response normalization and Gaussian derivative regularization. Hermite and other complete polynomial bases span the same degree-bounded filter space, but their coordinates can produce different optimization behavior under limited training budgets. We analyze this behavior through spectral signal energy, label sampling, changes in learned features and the bias–variance trade-off of regularization. Controlled synthetic experiments identify a regime in which plain HermNet outperforms matched polynomial-basis alternatives, including with a jointly trained nonlinear predictor. Curvature regularization further improves HermNet when the same functional penalty is available to every comparator. Fixed-predictor controls support the advantage under short training budgets, but longer training removes the plain-model lead. Matched real-data comparisons show accuracy deficits, and architectural and numerical studies identify further limits. Together, the analysis and experiments clarify when Hermite propagation is useful and how calibration and regularization affect its performance.

Identifying the minimal set of nodes whose removal breaks a complex network apart, also referred as the network dismantling problem, is a highly non-trivial task with applications in multiple domains. Whereas network dismantling has been extensively studied over the past decade, research has primarily focused on the formulations of the optimization problem for single-layer networks, neglecting that many, if not all, real networks display multiple layers of interdependent interactions. In such networks, the optimization problem is fundamentally different as the effect of removing nodes propagates within and across layers in a way that can not be predicted using a single-layer perspective. Here, we propose a dismantling algorithm named MultiDismantler, which leverages multiplex network representation and deep reinforcement learning to optimally dismantle multi-layer interdependent networks. MultiDismantler is trained on small synthetic multiplex graphs; when applied to large, real and synthetic networks, it displays exceptional dismantling performance, clearly outperforming all existing methods that rely on a single-layer approach to network dismantling. We show that MultiDismantler is effective in guiding strategies for the containment of diseases in social networks characterized by multiple layers of social interactions. Also, we show that MultiDismantler is useful in the design of protocols aimed at delaying the onset of cascading failures in interdependent critical infrastructures.

Visual Navigation Models (VNMs) enable robots to navigate from egocentric visual observations without geometric localization and planning, but long-range navigation still requires pre-built maps. This paper presents the Remote Visual Navigation Model (ReVNM), which uses a single remote surveillance camera to serve as both an observation source and an implicit environmental map for visual navigation. While the use of remote cameras could eliminate the need for pre-built maps as well as onboard vision processing, their limited field of view instead of egocentric observations makes it hard to achieve collision-free navigation. The lack of existing data with diverse remote viewpoints, which are crucial for training robust VNMs, further complicates the challenge. In this work, we propose a learning-by-synthesis approach to address this two-fold challenge. Our ReVNM extends a state-of-the-art VNM architecture with an exocentric-to-egocentric (exo2ego) module that predicts an egocentric depth observation from remote-camera observations. This helps the VNM to plan a path while considering obstacles in front of the robot. Trained only on randomly generated worlds with diverse obstacle layouts and camera viewpoints, ReVNM can generalize well to real robot navigation without additional fine-tuning. Experiments in both simulation and real-world environments confirmed the effectiveness of the proposed approach.

Post-training quantization (PTQ) reduces the cost of on-device text-to-speech (TTS), but published evaluations cover one system or method. We evaluate PTQ across TTS architectures under one protocol with three core models, weight and activation ablations of eight more, and two held-out models quantized blind. Four-bit per-channel weights reduce UTMOS, a predicted mean opinion score, by 2.8 on Supertonic and 0.07 on Kokoro, and per-tensor scaling can cause severe degradation even at 8 bits. The same bit width yields different outcomes, because the sensitive component is model-specific and not reliably predicted from the model class. A staged ablation procedure identifies it, and per-layer GPTQ can restore it to within 0.1 UTMOS. Real int8 and int4 kernels reproduce the simulated ordering at hardware-dependent cost. On a Mac mini, a 4-bit weight kernel runs Supertonic at 0.60x the fp32 latency while int8 is slower, so each configuration requires validation on the target runtime.

Physical changes or sensing errors can invalidate embodied agents' task-relevant beliefs. AquaMend compares re-probing, rollback, and supported continuation on a probe-belief-action graph under an expected-loss objective covering sensing, physical recovery, and uncorrected failures. A joint posterior guides a one-step policy with conditional detection-power screening. The per-belief three-way optimum requires independence, separability, and fully resolving probes; the general policy has no global optimality guarantee. Across 32 paired scenarios in a self-constructed simulation benchmark, AquaMend recovers in 28/32 cases and reduces mean complete loss by 21.6% versus restart. Its paired loss difference from decision-theoretic troubleshooting (DTT) is not statistically significant after Holm correction. Against the all-candidate ablation, online decision time decreases by 12.3% overall but increases by 3.4% in the uncovered late stage.

Cross-country recommendation on modern e-commerce platforms is typically deployed with disjoint user and item ID spaces across markets, removing the shared anchors that conventional cross-domain methods rely on. Generative recommendation (GR) mitigates this by mapping items into a shared token space and training a unified model, but existing approaches keep behavior sequences strictly country-specific, so knowledge transfer occurs only at the parameter level and remains absent at the data level. Inspired by code-switching corpora in multilingual natural language processing, we propose CMRec, a cross-country GR framework that injects cross-country supervision at the data level via dual-constrained, context-aware code-mixing. CMRec first learns a shared semantic codebook from multi-modal content and behavioral co-occurrence across countries. It then uses this codebook to synthesize mixed-country sequences via token-level substitutions that satisfy both static (content) and dynamic (e.g., price, audience, popularity) constraints. Finally, it introduces a context-aware loss that reweights mixed samples according to their plausibility in the current sequence. Experiments on two real-world multi-country datasets and an online A/B test show that CMRec substantially improves recommendation quality in data-sparse countries while preserving performance in data-rich countries, achieving +1.77% advertising revenue and +2.64% orders on a large-scale e-commerce platform.

This paper is devoted to the non-diffusive computation of solutions to $N$-dimensional hyperbolic systems of conservation laws (HCL) and more specifically the accurate approximation of simple waves. We develop the LeafNet algorithm, which relies on: i) a reformulation of HCL as a coupled system involving smooth solutions, and ii) an accurate physics informed algorithms approximating smooth functions. An error estimate analysis and two-dimensional numerical experiments are proposed to illustrate the presented strategy.

Quasi-brittle fracture is prevalent in concrete, rock, ceramics, composites, and masonry, and its simulation faces a trade-off among accuracy, efficiency, and simplicity. The classical Lattice Spring Model (LSM) captures cracking via bond breakage without remeshing, but its elements are empirical and limited to a few tessellable shapes with fixed Poisson's ratios. We propose a tessellated Isotropic Elastic Lattice Spring Model (IELSM) that discretizes continua into polygonal elements with axial springs and a volumetric constraint, achieving isotropic elasticity on arbitrary polygonal tessellations. Macroscopic isotropy reduces to governing equations whose solvability gives a theoretical criterion for element admissibility, proving conventional tessellable elements and extending to arbitrary regular N-gons and concave elements. Exploiting boundary interpolation compatibility with finite elements, IELSM is assembled by direct node sharing, without interface elements or kinematic constraints. Coupled with an isotropic damage model, a pure bending test and four fracture benchmarks show that the coupling preserves displacement accuracy, yields crack paths and load-displacement curves agreeing with experiments and outperforming standard FEM, and is insensitive to mesh refinement. IELSM can also be restricted to damage-prone regions, with the remainder modeled by finite elements via node sharing. For the benchmarks, this reduces nodes by 41.2-80.9% and CPU time by 34.2-85.2% versus full-domain IELSM. The framework advances IELSM element construction from empirical trial and error to theoretical determination and simplifies coupling to node sharing, offering a balanced route for complex quasi-brittle fracture analysis.

While end-to-end autonomous driving systems show promise, their application to micromobility vehicles is hindered by simulators failing to capture specific kinematics, such as differential drives and omni-wheels. This paper pro- poses a sim-to-real-aware, vehicle-specific end-to-end learning environment for the WHILL Model CR on AWSIM and ROS 2. To minimize the sim-to-real gap, physical parameters are optimized via Bayesian optimization using real-world data, reducing trajectory errors across various driving scenarios. Additionally, this study introduces a synchronized architecture tailored for the stable training of world model-based agents. An end-to-end policy trained with DreamerV3 exhibited learning progress and achieved task completion in a simulated obstacle avoidance setting. Furthermore, this policy demonstrated direct sim-to-real transfer to the physical vehicle, enabling the vehicle to navigate around a cardboard box in a real-world corridor replica without fine-tuning. This paper provides a practical foundation for sim-to-real micromobility policy studies.

Passive long-wave infrared (LWIR) hyperspectral ranging enables distance estimation in low-light and nighttime scenes by exploiting atmospheric absorption features in thermal radiance received through the atmosphere.Joint estimation of temperature, emissivity, and distance is computationally expensive. Reference-range joint inversion also uses a distance-invariant effective attenuation coefficient, which can bias range estimates.We introduce transmittance extraction and distance alignment (TEDA), which decouples range estimation from temperature–emissivity inversion. In the first stage, a baseline estimator with a data-fidelity term invariant to the known absorption direction yields two closed-form smoothing branches for the slowly varying thermal continuum. An observation-derived gate combines the branches, and subtracting the blended baseline in the log domain recovers atmospheric transmittance. The second stage estimates range by matching the recovered transmittance to sensor-domain transmittance models recomputed for each candidate distance. Monte Carlo simulations show that TEDA effectively reduces the ranging bias caused by the distance-invariant attenuation coefficient approximation. In a measured scene, TEDA's mean range estimates are closer to the LiDAR medians than those of reference-range joint inversion in both evaluated patches. TEDA processes a complete $256\times256$ region of interest in 8.19~s versus 159.47~s for reference-range joint inversion, an approximately 20-fold speedup.

Prediction markets combine trading with public discussion of the same events. We examine whether comment-derived signals predict subsequent activity, buying direction, and changes in the leading outcome. A correlation sweep across 79 non-political Polymarket markets guides six classification experiments comparing comment features, trading features, and their combinations. On live blocks containing comments, attention nearly matches trading history in predicting heavy trading within 18 hours (PR-AUC 0.786 versus 0.790, against prevalence 0.606), with its relative advantage concentrated in short markets. Comment content carries directional information: toxicity ranks future buying direction above chance in 43 of 53 scored markets, while adding attention, sentiment, and stance to flow history increases ROC-AUC from 0.780 to 0.788. Leadership changes are predicted primarily by market state. Stance shifts against the leader before reversals in 23 of 27 evaluable markets, but adds no clear improvement in individual-block forecasting. These results distinguish attention from directional support and price uncertainty. They establish predictive associations consistent with discussion and trading responding to shared information, without identifying a causal effect of comments on markets.

Group-based reinforcement learning (RL) methods, such as GRPO and its variants, have become a leading paradigm for training reasoning and agentic large language models (LLMs). While their group-normalized advantage estimation is reliable at the response level, it becomes systematically biased at the step level, since coarse-grained trajectory-level advantages are hard to accurately reflect the contribution of individual steps (i.e, failed trajectories may contain valuable steps). Revisiting the foundational RL definition, we notice that GRPO's success on single-turn tasks stems from its advantage estimation strategy, which adheres to the basic definition: the mean reward of multiple actions sampled from the same state constitutes a credible state-value estimate. Extending the faithful estimation to step-level would in principle demand sampling multiple actions from each intermediate state, which is too costly on a per-state basis. To mitigate this issue, we propose a Graph-based Faithful sTep-level credit-assignment framework (GRAFT) that grafts all rollout trajectories into a trajectory graph, recovering node state-values via Bellman iteration on the graph, and assigning credit to each edge by the node value difference. Theoretically, the estimated step-level advantage faithfully adheres to the basic advantage definition in RL. To further ensure the reliability of step-level advantage estimation, we further propose Graph GAE, which extends GAE to the trajectory graph for reducing the impact of state-value estimation bias. Experiments across a range of multi-turn agentic benchmarks show consistent gains over GRPO and superior performance compared to recent agentic RL algorithms. Code will be available at https://github.com/xcyao00/GRAFT.

A novel second-order consistent splitting scheme is proposed for the natural convection equations. The scheme is constructed using Taylor expansions about the time level $t^{n+k}$, where $k$ is a parameter to be determined. It is proved to be stable for all $k>4$, with the present study focusing primarily on the case $k=5$. Compared with the conventional splitting scheme based on Taylor expansions about $t^{n+1}$, the proposed scheme exhibits improved stability. By employing the Sobolev inequality and Gronwall's lemma, we rigorously establish stability of the proposed scheme and derive error estimates in both two and three spatial dimensions. Finally, several numerical experiments are presented to verify the stability and accuracy of the proposed scheme and to demonstrate its effectiveness.

While recent advances in perceptive locomotion have enabled humanoid robots to traverse structured terrains, agile parkour in highly discontinuous environments remains an open challenge. In particular, crossing sparse footholds and narrow support regions requires precise foothold selection, effective use of visual observations, and consistent alternating foot placement during fast transitions. In this paper, we present a perceptive humanoid parkour framework that enables stable traversal across terrains with limited foothold availability using only onboard depth observations. The framework features a saliency-guided temporal perception module that combines a saliency prior with gated memory. It retains informative depth features across frames, enabling reliable foot placement from partial observations. By introducing an alternation loss, our symmetry regularization encourages alternating gait patterns and improves traversal robustness. Extensive experiments show that our method significantly improves success rate and foothold accuracy on challenging terrains in both simulation and the real world.

Humanoid parkour policies can traverse various terrains, but task completion may mask challenges of harsh landings, edge contacts, and unstable stance contacts. Humans naturally regulate foot-terrain interaction through tactile feedback, modulating contact compliance according to terrain stiffness. This highlights a key domain gap between humans and humanoid robots: the absence of rich tactile sensing in most humanoid systems. We address this problem with TactileStep, a deployable tactile learning framework that brings sole pressure sensing into humanoid locomotion control for softer touchdowns and more stable support. TactileStep aligns tactile simulation with the real pressure insole, allowing the policy to learn from the same contact features available on hardware. During training, we use tactile and motion cues to recognize different foot-contact phases and apply phase-aware rewards that encourage safer landing and more stable stance. Evaluated in simulation and on a Unitree G1 humanoid across diverse terrains, TactileStep reduces peak touchdown force by up to 48.8% and peak A-weighted impact noise by up to 30.1 dB over a strong perceptive baseline, while increasing stance contact area by up to 23.8%.

Recent advances in video object segmentation with Multimodal Large Language Model (MLLM) reasoning have demonstrated the effectiveness of using a single textual token, such as SEG, to predict segmentation masks across images and videos. However, we observe that this single-token strategy lacks the granularity required to precisely localize multiple objects across time in video segmentation tasks. To address this limitation, we develop Multi-Token Reasoning for Video Object Segmentation, or MoVISA. MoVISA uses multiple segmentation tokens, such as SEG0 and SEG1, to represent an object across different frames. This design enables more fine-grained alignment between language prompts and spatio-temporal mask predictions, improving both performance and interpretability. On the challenging MeViS, DAVIS17, ReVOS, and Ref-Youtube-VOS benchmarks, our model achieves a 13.2 percent J and F improvement on MeViS and an 8.4 percent J and F improvement on ReVOS. Code and models will be released.

Current Vision-Language-Action (VLA) models often struggle with high-precision robotic manipulation. We attribute this limitation primarily to their visual attention being dispersed across task-irrelevant regions. To address this issue, we propose ActGaze, a training approach that guides VLA policies to gaze on task-relevant regions, much like humans gaze on critical visual cues while executing precise movements. Unlike prior methods that rely on external labels for gaze supervision, ActGaze derives spatial supervision directly from the VLA's own action objective by using counterfactual visual interventions to identify regions that are critical for action prediction. Extensive real-robot experiments on four high-precision robotic manipulation tasks demonstrate that ActGaze induces more focused visual attention on task-relevant regions and consistently outperforms the base VLA policy and other visual-grounding approaches.