Z Z Wu

Zichen Ding, Jiaye Ge, Shufan Jiang et al.

As Large Language Models (LLMs) evolve into autonomous agents, the need for unified evaluation infrastructure becomes critical. However, current evaluation pipelines remain highly fragmented and tightly coupled, hindering reproducibility and causing redundant engineering. To address this, we introduce AgentCompass, an open-source, lightweight, and extensible infrastructure for evaluating LLM-based agents. AgentCompass organizes the evaluation process around three independent components, namely Benchmark, Harness, and Environment, thereby enabling flexible configurations without requiring the reimplementation of complex execution logic. Furthermore, it features a fault-tolerant asynchronous runtime and comprehensive trajectory analysis tools to transparently diagnose nuanced failure modes like reward-hacking. Natively supporting over 20 benchmarks across five capability dimensions, AgentCompass provides the community with a scalable and reproducible infrastructure for advancing agent research.

Zishuo Li, Bowen Yang, Changtao Miao et al.

Egocentric videos of human manipulation provide scalable supervision for embodied intelligence, yet existing resources rarely combine low-cost continuous capture, manipulation-level structured annotations, and reusable tools for robot learning. We present Open-AoE, an open, community-oriented egocentric manipulation dataset and toolchain spanning the full pipeline from smartphone capture to model training. Its first release contains approximately 2,000 hours of manipulation video collected in natural environments by 500+ contributors using 400+ smartphones. The dataset provides text annotations, MANO-based hand poses, camera trajectories, and temporally localized atomic actions. Open-AoE further includes a data processing pipeline that transforms raw recordings into structured samples through temporal action segmentation, semantic annotation, hand reconstruction, and camera trajectory reconstruction. Meanwhile, we provide a separate downstream toolchain supports visualization, cross-embodiment retargeting, model-specific data conversion, and training recipes for VLA policies, WAMs, and World Models. By integrating scalable capture, structured processing, and downstream adaptation, Open-AoE reduces the barriers to both data contribution and reuse, providing practical open infrastructure for embodied model training, human-to-robot transfer, and world modeling.

Hao Yang, Yanyan Zhao, Kewei Zhao et al.

Understanding driver emotion and state is critical for the next generation of intelligent in-cabin systems that ensure safety and enhance human-vehicle interaction. However, existing public datasets for in-cabin affective computing are largely limited to visual modalities and rarely include conversational information, making it difficult to capture the linguistic and interactive cues underlying driver emotion. To address these gaps, we introduce InCarEmo, a multimodal dataset for in-cabin emotion recognition and driver state monitoring. InCarEmo integrates RGB and infrared video, in-cabin audio, and dialogue text collected from scripted in-cabin scenarios designed to simulate realistic driver behaviors, covering diverse lighting conditions and driving contexts. The dataset supports three primary tasks: 1) multimodal emotion recognition, 2) fatigue detection, and 3) distraction monitoring. In addition to the original Chinese data, we construct an auxiliary English benchmark to support preliminary cross-lingual evaluation. We provide a unified benchmark with extensive baseline results across unimodal and multimodal methods, including analyses under modality-missing and noise conditions. Experimental results demonstrate the benefits of multimodal fusion and reveal remaining challenges under real-world noise and low-light conditions. By releasing InCarEmo, we aim to establish a comprehensive foundation for robust, interpretable, and human-centric in-cabin affective understanding, promoting safer and more empathetic driver-vehicle interaction.

Xingze Gao, Chuanrui Hu, Hongda Chen et al.

Agent self-evolution in long-horizon LLM systems is largely procedural: useful experience is not merely stored information, but reusable procedures for searching, debugging, and verification. Yet current evaluations do not isolate this form of transfer. Agent benchmarks test single-episode task solving; memory benchmarks target information retention rather than procedural reuse. We introduce EvoAgentBench, a benchmark for agent self-evolution via Ability-guided transfer across four agentic domains: web research, algorithmic reasoning, software engineering, and knowledge work. EvoAgentBench extracts trace-grounded Abilities from agent executions, canonicalizes them into operational units, and builds domain-specific Ability Graphs linking tasks that share procedural overlap. By design, every test task is backed by verified training-side Ability support. Across a 528/267 train/test split, two scaffolds, and three backbones, curated Ability content transfers reliably across model families, but no current automatic method sustains positive gain in all settings. EvoAgentBench shifts self-evolution evaluation from aggregate accuracy comparison to fine-grained diagnosis of experience encoding, routing, and uptake. The benchmark is publicly available at https://huggingface.co/datasets/EverMind-AI/EvoAgentBench.

Yishan Lv, Jing Luo, Xinyu Yang et al.

Singing Quality Assessment (SQA) has become increasingly important for practical multimedia applications and Music AI systems, yet existing studies predominantly focus on short singing clips and remain insufficient for full-length songs. Unlike clip-level assessment, full-length song SQA requires modeling how singing quality varies across different audio segments and how these local variations influence the overall evaluation of vocal performance. Moreover, the scarcity of segment-level annotations makes effective supervision challenging, as directly assigning a single overall score label to every segment tends to treat different segment qualities as equivalent. To address these challenges, we propose SongSQA, a two-stage framework for full-length song SQA. In the first stage, a Segment Score Predictor is trained with pseudo labels generated by a pre-trained teacher model, enabling segment-level singing quality prediction without requiring manual segment annotations. In the second stage, a Song Quality Aggregator integrates segment features and predicted segment scores into unified segment embeddings, and employs a learnable song embedding together with self-attention to capture the connection between segment-level vocal performance and overall song quality. In this way, SongSQA dynamically aggregates critical quality cues across the song to produce a holistic quality prediction, while also generating a temporal segment-level quality curve. Experimental results demonstrate the effectiveness of SongSQA for full-length song SQA, achieving up to a 13.95% relative improvement in KTAU over the strongest baseline, while consistently improving other evaluation metrics across all datasets.

Haoxuan Li, Tianci Gao, Jianhe Li et al.

Understanding the brain increasingly depends on integrating evidence across scales, modalities, and disciplines. Addressing a single research question therefore requires a coordinated sequence of operations, from surveying prior work to executing analyses and interpreting results in light of domain knowledge. AI agents promise to accelerate this process, but current agents lack domain expertise in brain science, may fabricate claims, drift during multi-step reasoning, and offer few defined points for expert intervention. These failures are especially costly in brain science, where conclusions feed into downstream scientific claims and depend on laboratory-specific expertise and careful human judgment. We present \textbf{BrainPilot} a \textbf{fully open-source} multi-agent system that accelerates brain science research with traceable logs and agent-verified results. A principal investigator (PI) agent coordinates specialist agents grounded in curated domain knowledge: a unified brain science knowledge base containing 7{,}233 indexed items and a skill library of 72 reusable methodology units across seven research domains. Every major step is recorded in the Graph of Trace, an auditable record that links subgoals, tool use, evidence, and claims and allows researchers to follow and inspect the workflow. An Auditor agent further integrates fabrication checking into the workflow. For evaluation, we run three brain science tasks from Agents' Last Exam, introduce our own benchmark, \textbf{BrainPilotBench-v0}, and present additional end-to-end case studies. Across these evaluations, BrainPilot with an open-source backbone model attains performance comparable to state-of-the-art agent framework with less costs.

Shen Zhou, Jinghui Zhang, Wenbo Huang et al.

All-in-one image restoration aims to recover clean images degraded by multiple corruption types using a single unified model. Existing methods typically rely on image-level prompts or shared guidance to handle diverse degradations. However, such a paradigm becomes inadequate when degradations are spatially heterogeneous or even coexist in mixed forms within a single image. Yet spatially adaptive guidance alone is not sufficient, since accurate restoration also requires each spatial query to reliably aggregate complementary information from local neighborhoods and global contexts. To this end, we propose QuReC, a unified framework for all-in-one image restoration. QuReC consists of a Degradation-Guided Query Reconstruction Module (DQRM) and a Local-Global Response Calibration Module (LGRCM). Specifically, DQRM matches each spatial query against a degradation prototype space to reconstruct a query-specific degradation-aware representation, thereby providing fine-grained spatially adaptive restoration guidance. To further stabilize this query-wise matching process, we introduce a weakly supervised prototype matching learning strategy to improve optimization stability and degradation semantic consistency. Meanwhile, LGRCM performs local-global dual-branch aggregation and calibrates the aggregated responses with learnable priors, improving the reliability of feature aggregation and the coordination between local detail modeling and global context modeling. Extensive experiments demonstrate that QuReC achieves superior performance on multiple all-in-one image restoration benchmarks. The code is released at https://github.com/zhoushen1/QuReC.

Xunkai Li, Guohao Fu, Yuming Ai et al.

Multimodal-attributed graphs (MAGs), whose nodes carry modalities such as images and text alongside topological structure, now pervade applications including social platforms, e-commerce, and biomedical networks, offering richer semantic signals than single-modality graphs. In practice, such graphs are fragmented across privacy-restricted silos owned by different platforms and institutions, so learning a broadly transferable model over them demands collaborative training that never exposes raw data. This places the task at the intersection of multimodal graph learning and federated learning, yet existing methods cover only one side of it. To address the challenges from these two perspectives, we propose FedGAMMA, casting federated multimodal graph foundation learning as a two-stage semantic-structural alignment problem of federated pre-training and prompt-based fine-tuning. During pre-training, a shared-private semantic enhancer disentangles cross-modal commonality from modality-specific information, aligning it through optimal transport, a topology-aware graph fusion module decouples semantic and structural views via semantic residual graphs and dual positional encodings, and a dual-channel affinity-aware aggregation mechanism estimates client similarity from feature and graph centroids without exposing raw data. During fine-tuning, FedGAMMA adapts the pretrained encoder through lightweight graph-aware prompts, a shared prompt pool with controlled exploration, and channel-wise prompt synchronization. Experiments on twelve multimodal graph datasets show FedGAMMA consistently surpassing a broad range of baselines across downstream tasks, with gains of up to 12.96%. FedGAMMA further outperforms competitive baselines accross multi-domain datasets on multiple tasks with up to 5.71% under few-shot learning scenario.

Zihao Zhang, Haoyu Zhao, Siqian Yang et al.

Despite the success of diffusion models in Video Frame Interpolation (VFI), existing methods still suffer from two critical limitations. First, latent diffusion inevitably loses fine-grained details when reconstructing images from latent representations back to the pixel space. Second, multi-step sampling incurs prohibitive memory consumption and inference latency. To address these issues, we propose SPEED, a one-step pixel diffusion framework for high-quality VFI. Specifically, SPEED employs a progressive multi-stage architecture with dynamic patch scaling to effectively learn multi-scale motion, structural, and appearance representations. Furthermore, we propose a novel Noise-Update-Only Attention mechanism to prevent semantic degradation of the clean condition frames while reducing the computational overhead by nearly 50%. Besides, we introduce a Drift-aware Timestep Sampling strategy coupled with a tailored training objective to directly predict images in the pixel space, enabling one-step inference without compromising the quality of the generated frames. Extensive experiments show that SPEED achieves state-of-the-art performance. On SNU-FILM, SPEED reduces LPIPS by 8.8% while delivering 63.3% faster inference and 10.6% lower memory usage. On challenging 4K benchmarks, it further surpasses prior methods by up to 51.5% in LPIPS.

Jiachi Zhang, Zhuoyu Wu, Quanjun Wang et al.

While lightweight polyp segmentation is highly desirable for low-cost deployment, reported performance gains often stem from upgraded backbone encoders, complex decoders, or heavy refinement branches. Consequently, it remains difficult to isolate whether a lightweight correction mechanism is inherently effective on its own. We address this limitation by formulating refinement as a prediction-space recursive correction task, introducing a recursive controller that operates directly on backbone logits. Under a fixed recursion budget, this controller aggregates discrepancy and uncertainty evidence, updates a compact state tracking recent correction utility, and applies additive residual logit corrections. By design, this correction path remains small, host-portable, and deployment-explicit. Utilizing a unified Kvasir-trained protocol, we evaluate our approach across seven lightweight backbones on Kvasir-SEG and three transfer datasets, measuring segmentation accuracy (Dice/IoU) alongside deployment efficiency (parameters, GMACs, and peak memory). The controller yields consistent improvements in the source domain, achieves competitive performance against both training-side baselines and heavier structural refiners on representative hosts, and delivers selective transfer gains with minimal static overhead. Code is available at https://github.com/tyui99/Gain-Aware-Prediction-Space-Recursive-Controller.

Jiachi Zhang, Zhuoyu Wu, Wenqi Fang

Post-refinement can improve colonoscopy segmentation after host inference, but many designs still rely on extra correction heads or multi-stage pipelines with non-negligible parameter or computational cost. For polyp segmentation, host predictions are often already reasonable globally, with remaining errors clustered around ambiguous boundaries and difficult local structures. These residual errors matter in colonoscopy images because useful masks need correct lesion coverage and clean contour delineation across subtle mucosal transitions. This setting favors selective local repair in prediction space over reprocessing the entire mask. We therefore propose RIGS-Refiner, a lightweight post-refinement plugin for risk-guided recursive refinement in prediction space. Starting from a frozen host anchor prediction, RIGS-Refiner extracts lightweight image priors and prediction cues, applies risk-guided update, and writes back residual corrections through a shared recursive cell. The module adds only +519 parameters and +0.631 GFLOPs, keeping the refinement path compact for deployment. Experiments use Kvasir-SEG for training and Kvasir, ClinicDB, ColonDB, and ETIS for evaluation under two frozen hosts, namely PraNet and SegFormer-B0. Results show consistent gains on both hosts and a favorable efficiency-accuracy trade-off against representative post-refinement methods. Code is available at https://github.com/tyui99/RIGS-Refiner.

Weize Quan, Zhengwei Wu, Kai Wang et al.

View-based point cloud completion aims to recover a complete 3D shape from a partial point cloud, guided by a single-view image. However, existing approaches often suffer from limited performance due to weak modality alignment and limited self-geometry enhancement. To overcome these challenges, we propose a unified geometry-aware framework that integrates efficient modality alignment and adaptive geometry enhancement, mainly to address cross-modal geometric inconsistency of view-guided point cloud completion. Specifically, we propose a geometry-aware modality alignment by integrating a shared self-attention Transformer and cross-modality reconstruction supervision, which aims to bring features of the image and point cloud close to each other in a shared latent space describing the 3D object. To enhance the perception of global shape and local geometric details, we propose an adaptive geometry-aware self-attention module, which simultaneously considers local geometry-aware attention computation and the spatially-variant feature fusion. In addition, we apply a geometry-perceptive anchor refinement module to reorganize the anchor points (representing a local region of the shape) under appropriate supervision, further boosting the completion performance of our method. Extensive experiments on both synthetic and real-world datasets demonstrate that our method achieves superior performance over existing approaches. Our code will be available at https://github.com/weizequan/MAGE.

Yang Shen, Kai Lu, Min Xie et al.

Large-buffer reads increasingly connect data-intensive applications to high-speed storage. They amortize system-call overhead and create a larger in-kernel window for organizing page-cache work and submitting I/O. However, Linux buffered read primarily exploits only the former benefit. Within a large read, its conventional interleaved path repeatedly switches among fine-grained page-cache operations, amplifying metadata and serial orchestration overheads and failing to consistently expose enough in-flight requests to modern parallel SSDs. We present MARS, a multi-stage accelerated read stack for synchronous large-buffer buffered reads. MARS treats each large-range read as one unit of work and stages page-cache operations by data structure and dependency. During I/O waits, it handles user-buffer page faults and performs reorderable data copies early. Opportunistic kernel workers then copy remaining data in parallel and, when the backend provides sufficient parallelism, optionally submit I/O in parallel. We implement MARS in Linux 6.6.58. For MiB-scale fio reads, MARS improves bandwidth by up to 6.56 times over Linux. On five NVMe SSDs in RAID0, it reaches 36.87 GiB/s for 128 MiB random reads, 4.44 times Linux. MARS also accelerates DuckDB/Parquet queries by 1.80--2.15 times and ExecuTorch model loading by 3.17--3.61 times.

GigaWorld Team, Angen Ye, Angyuan Ma et al.

World Action Models (WAMs) improve robot policy learning by jointly modeling actions and future visual observations, using future scene evolution as dense supervision for physically grounded action generation. However, a common design in existing WAMs is to explicitly generate future videos at inference time, incurring substantial computational overhead and hindering real-time closed-loop deployment. GigaWorld-Policy addresses this issue with an action-centered formulation, where future visual dynamics are used during training while action-only decoding is used at inference time. Building upon this framework, we present GigaWorld-Policy-0.5, an enhanced action-centered WAM designed for more efficient robot control. During pretraining, GigaWorld-Policy-0.5 adopts a mixed Action-Conditioned World Modeling (AC-WM) and WAM training strategy. This strengthens the coupling between visual dynamics and robot actions and improves the transferability of action representations for downstream policy learning. For efficient inference, GigaWorld-Policy-0.5 introduces a Mixture-of-Transformers architecture that separates visual dynamics modeling and action generation into specialized experts, reducing active computation during action-only inference and achieving 85 ms inference latency on a local RTX 4090 setup. In addition, we employ an agent-based AutoResearch pipeline to systematically search training configurations, enabling more efficient identification of optimal experimental setups while reducing the time and manual intervention required for hyperparameter tuning. Experiments and ablations show that GigaWorld-Policy-0.5 preserves the training benefits of future visual dynamics while improving inference efficiency for robot control.

GigaWorld Team, Angyuan Ma, Boyuan Wang et al.

Evaluating embodied robot foundation models remains a critical bottleneck; unlike large language models efficiently assessed via digital benchmarks, robotic policies require slow, costly real-world rollouts limited by hardware and human supervision, which has driven interest in world models as surrogate policy evaluators, yet the key properties that make a world model reliable for policy assessment remain poorly understood. This work presents a systematic study of world models for robotic policy evaluation and introduces WMBench, a benchmark constructed from real-robot teleoperation data and matched policy rollouts covering diverse manipulation tasks to enable controlled comparisons across model families, action encodings, rollout horizons, and evaluation metrics. Using WMBench, we analyze 7 video world models, 4 action representation schemes, and over 324,000 simulated policy rollouts paired with real robot executions, further enriching our analysis with large-scale community submissions from the CVPR 2026 GigaBrain Challenge, curated synthetic trajectories, and a training videos spanning more than 12,000 hours. Our experiments deliver three core insights: evaluator quality is dominated by long-horizon, action-faithful rollout consistency rather than short-term visual realism; pretraining gains stem not only from data scale but from balancing general world knowledge with robot-specific controllability; and architectural choices including action encoding, memory design, and evaluator-focused post-training strongly determine alignment with real-world robot behavior. Drawing on these results, we derive a practical design roadmap and realize it in \textit{GigaWorld-1}, a world model specially optimized for policy evaluation, and we fully release our code, models, datasets, and toolkits to advance scalable evaluation research for embodied foundation models.

Xianfu Cheng, Shiwei Zhang, Jiyu Zhao et al.

Agentic Reasoning has become a transformative force in financial analysis due to its ability to integrate large-scale information and generate reliable and accurate content. However, when handling complex real-world problems, different agents still show significant performance variation. In this work, we design Finance-LaTeX SKILL, a skill for synthesizing financial documents with complex layouts based on expert knowledge. Using an agent workflow built on this skill, we generate 2,000 professional financial documents along with 6,000 high-quality question-answer pairs. To evaluate the overall capability of agents, we introduce FinanceComplexQA, a comprehensive open-ended generation benchmark for financial documents that closely resembles real-world scenarios. It contains 2,026 deep research tasks targeting 1009 financial documents. FinanceComplexQA has 8 key features: bilingual support; coverage of six mainstream scenarios and seven tasks; expert-level document reasoning questions; deep research of complex layouts; relatively stable and permanent reference answers; and precise evaluation through an Agent-as-a-Judge with multiple evaluation metrics. Using FinanceComplexQA, we conduct a comprehensive evaluation of leading RAG systems and agentic reasoning tools for financial document QA. Through identifying and analyzing failure cases, we provide an in-depth study of their capabilities in numerical computation, multi-hop reasoning, content summarization, and industry analysis.

Lianghua Huang, Zhi-Fan Wu, Yupeng Shi et al.

We present Wan-Streamer v0.3, which reframes our native-streaming interaction model under a single organizing view: a video is a world plus an event stream. The world is the persistent context in which a video unfolds, including the environment, scene, subjects, ambient acoustic conditions, voice characteristics, and other relatively stable conditions. The event stream is everything that changes over time within that world, including scene or environmental changes, subject behavior, speech, and other sounds. This yields a general-purpose pretraining task over large amounts of real video: given a world and incoming input, predict how the world moves, changes, and responds in real time. The resulting competence can be specialized to a broad family of real-time downstream tasks. We instantiate it on real-time full-duplex audio-visual interaction, where the event stream is the agent's speech together with free-form behavior. Functionally, the model's multimodal understanding process is vision-language-action-like: it maps multimodal user input to language-form speech and behavior actions. Wan-Streamer v0.3 preserves the v0.2 operating point: 640x368 video at 25 FPS, a 160 ms streaming unit, approximately 200 ms model-side response latency, and approximately 550 ms total interaction latency under a 350 ms bidirectional network budget.

Zicong Wu, Panagiotis Kalozoumis, S. M. Hadi Sadati et al.

Safe navigation within the spinal subarachnoid space is constrained by its narrow, compliant, and delicate anatomy. Conventional catheters and continuum robots rely on proximal pushing, generating friction and shear along the tissue device interface that limit distal controllability and increase the risk of neural injury. Here, we present a 2 mm diameter eversion-growing robotic platform that enables friction minimised extension and steering within the human spinal subarachnoid space, validated through computational modelling, phantom experiments, and intact human cadaver studies. The robot integrates a miniature endoscope for real time intrathecal visualisation and advances by pressure driven tip eversion, localising motion to the distal tip while minimising translational sliding of the deployed body. Phantom experiments demonstrated reductions of 65.2% in mean interaction force and 48.0% in peak interaction force compared with matched push-based insertion. Physics based modelling showed that eversion based growth redistributed tissue loading, reducing local stress concentrations and interfacial shear relative to conventional insertion. In an intact human cadaver, the system achieved 150 mm of controlled intrathecal extension with concurrent fluoroscopic and endoscopic visualisation, providing access across multiple vertebral levels from a standard lumbar entry point. Postprocedural laminectomy and durotomy revealed no observable macroscopic disruption of the dura mater or surrounding neural structures. These results provide the first mechanically characterised and multimodally validated demonstration of eversion-based robotic navigation in intact human spinal anatomy, establishing a quantitative and procedural foundation for future intrathecal interventions. Further validation in larger anatomical cohorts and under physiological conditions will be required before clinical translation.

Lianghua Huang, Zhi-Fan Wu, Yupeng Shi et al.

We present Wan-Streamer v0.2, a latency-preserving upgrade of the native-streaming, end-to-end audio-visual interaction model. v0.2 keeps the v0.1 modeling formulation, but raises the interactive output stream from 192x336 to 640x368 while preserving approximately 200 ms model-side signal-to-signal latency at 25 FPS. The higher-resolution stream supports scene-grounded mid-shot agents whose posture, gaze, hands, nearby objects, and local scene layout remain legible during real-time conversation. To support the larger visual stream without adding user-visible delay, v0.2 keeps the thinker as a single-GPU low-latency path for streaming perception, the short language/state Transformer pass that builds the generation cache, and final decoding. The performer becomes a multi-GPU Ulysses-style context-parallel group for the expensive next-unit latent generation. Each performer rank writes incoming K/V into a pre-sharded local cache. The long high-resolution latent video sequence is split across ranks for denoising and gathered through Ulysses communication, while the much shorter audio latent sequence is generated without sequence sharding. In this split, the thinker's language/state computation reaches the performer only as K/V conditioning, so no separate language sequence has to be communicated inside the performer group. This concentrates additional hardware on visual generation while preserving the compact thinker-performer boundary, keeping total remote interaction latency at approximately 550 ms when a 350 ms bidirectional network budget is included.

Zhiyuan Wu, Zhuo Chen, Shan Luo

Vision and touch are complementary modalities essential for robotic perception and manipulation. While vision provides global object context, touch offers precise local information at contact points. Integrating these modalities for contact localization, i.e., predicting the location of touch on an object's surface, poses significant challenges due to the need for accurate spatial alignment between tactile data and visual geometry. To address this challenge, we propose VTLoc, a novel visual-tactile framework that localizes contact points from tactile readings using a 3D point cloud as visual input. VTLoc introduces two key components: a geometric multi-modal alignment module, which reconstructs a pseudo-point cloud from fused visual-tactile features and aligns it with the visual point cloud to enforce spatial consistencies across modalities; and an iterative localizing updater, which iteratively refines the predicted contact location using fused visual-tactile features. Evaluated on a new benchmark of 100 real-world objects, VTLoc improves single-touch contact localization by reducing local-to-global correspondence ambiguity.

Zezhong Qian, Xiaowei Chi, Chak-Wing Mak et al.

Video models are evolving into vision foundation models, yet they still lack human-like multi-step reasoning. Streaming autoregressive diffusion models are efficient but limited in reasoning, while bidirectional diffusion enables global revision with high inference costs due to dense frame-level denoising. Both paradigms struggle to achieve logical consistency and low-latency streaming for complex reasoning tasks. We propose HDR (Hierarchical Denoising for Visual Reasoning), a unified framework that integrates hierarchical latents into causal video generation for multi-step reasoning. HDR organizes video latents into a tree-structured hierarchy, enabling coarse-to-fine reasoning before streaming output. Coarse denoising layers preserve uncertain hypotheses for global planning, while finer layers progressively refine them into concrete visual states. A sparse hierarchical attention pattern (SHAP) further reduces temporal attention costs. We introduce a level-stratified multi-step video reasoning benchmark with out-of-distribution cases, covering six tasks: maze navigation, Tower of Hanoi, one-line drawing, sliding puzzle, Sokoban, and water pouring. Compared with streaming autoregressive diffusion baselines, HDR improves success from 34.22 to 60.29 (76.2% relative gain) and increases average progress from 76.00 to 89.56, demonstrating more consistent reasoning trajectories. HDR maintains low-latency streaming at 0.70 seconds per latent, achieving 54.2 times faster inference than bidirectional diffusion. It also retains 82.9% of full-data performance with only 2% training data, compared with 52.0% for bidirectional diffusion. Real-world robot experiments further demonstrate HDR's potential for physical interaction and world modeling. Project demo: https://hierarchical-diffusion-reasoning.github.io/.

Yuyao Zhang, Junjie Gao, Zhengxian Wu et al.

Recent advances in Tool-Integrated Large Language Models have made web search a core capability of information-seeking agents. However, as interaction histories grow, agents increasingly struggle to track task progress. When search attempts fail to yield useful evidence, current single- and multi-agent systems can become trapped in repetitive loops, wasting search budgets and ultimately compromising the quality and completeness of the final output. We introduce SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state. First, we formulate open-domain information seeking as relational schema completion with grounded citations, where agents discover entities, populate attributes across linked tables, and anchor each value to source evidence. Then we design Search-Oriented Context Management (SOCM), which externalizes the evolving state into Frontier Task, an Evidence Graph, a Coverage Map, and Failure Memory. Built on SOCM, SearchOS applies a pipeline-parallel scheduling mechanism that overlaps the execution of sub-agents and continuously refills freed slots with tasks targeting unresolved coverage gaps to improve utilization and throughput. To schedule and control the execution of search agents, SearchOS introduces a Search Tool Middleware Harness that intercepts model and tool interactions to record grounded evidence and react to stalls or budget exhaustion, and provides a reusable hierarchical skill system comprising strategy and access skills to augment the agents' search process and avoid repeating failed search patterns across runs. On WideSearch and GISA, SearchOS leads all metrics among the evaluated single- and multi-agent baselines, paving the way toward robust information-seeking collaboration.

Rui Shu, Chun Yong Chong, Xin Zhou et al.

Coding agents are ranked almost entirely by resolve rate: whether their final patch passes the target tests. Yet two agents can reach the same outcome through very different processes, and a single pass/fail label says nothing about why a run failed or why an accepted run spent extra steps, time, or tokens. This process evidence lives in the trajectory, which records a run's searches, reads, edits, tool calls, validation, and reversions. However, raw traces are heterogeneous and hard to compare across runs. We present TraceProbe, a trajectory-diagnostic framework that recovers what resolve rate hides. TraceProbe normalizes each raw run into a canonical nine-type action taxonomy with deterministic effect labels, then applies two rule-based modules: Insight names single-trajectory anti-patterns adapted from established debugging practice (e.g., search loops, verification skips), while Converge aligns pairs of runs and classifies where their behavior diverges under controlled references. Applying TraceProbe to 2,500 trajectories from five production settings on SWE-Bench Verified, we find that (i) file choice is too coarse to separate success from failure, whereas function selection and completion behavior localize it; (ii) Insight anti-patterns act mainly as corpus-level difficulty clues, with search loops the most stable; and (iii) even resolved runs differ in how quickly they reach relevant code and how much failed work they incur. Trajectory structure thus adds auditable diagnostic context to outcomes by localizing inspection targets, suggesting failure hypotheses, and prioritizing runs for review.

Yuquan Xue, Le Xu, Zeyi Liu et al.

Reinforcement learning (RL) can overcome the demonstration-coverage limitation of imitation learning (IL) by allowing robots to improve through trial-and-error interaction beyond the states observed in demonstrations. However, deploying RL on real robots remains constrained by high interaction costs, since each physical rollout is costly and reflects only one realized action-outcome path. To address this challenge, we propose WorldSample, a physically grounded data augmentation framework for real-robot RL that closes a real-synthetic loop between physical rollouts, world-model generation, and policy improvement. Grounded on real rollouts, WorldSample generates high-fidelity synthetic transitions through a post-trained world model, which greatly lowers the visual hallucination. Specifically, rather than simply using these transitions as real-world experience, WorldSample introduces Policy-Paced Learning (PPL) to regulate the training process through sample selection and scheduling, balancing useful augmentation against value overestimation and mitigating the hallucination-induced noise. Experiments on robot manipulation tasks involving contact-rich and precise tasks show that WorldSample improves policy success rate by 28% while reducing training steps by 59% compared with baselines. Furthermore, WorldSample improves world model visual fidelity by 19.4dB in PSNR and 0.47 in SSIM over demonstration-only post-training, validating the effectiveness of the real-synthetic loop for both policy and world model performance.