C Shijia Yu

Yixian Zhang, Huanming Zhang, Feng Gao et al. · tsinghua

Language-conditioned manipulation requires both precise contact-rich control and robust reasoning over language, scenes, and long horizons. End-to-end Vision-Language-Action (VLA) models provide strong local visuomotor skills, but they are trained on in-distribution task trajectories and often fail under deployment perturbations such as semantic retargeting, goal re-binding, spatial-layout shifts, and unstable local contacts. LLM coding agents provide complementary semantic and compositional reasoning, but purely analytic primitives struggle with irregular grasping, constrained placement, and articulated-object interaction. We present Harness VLA, a memory-augmented agentic framework that exposes a frozen VLA as a retryable contact-rich primitive and composes it with a small fixed library of analytic primitives for grounding, staging, transport, navigation, and release. Rather than expanding the skill library, the harness learns the operating range of these fixed primitives from task-specific execution traces, global success rules, and failure models. By lifting semantic re-grounding, non-contact execution, and VLA re-staging to the planner while reserving the frozen VLA for local contact-rich phases, Harness VLA extends pretrained VLAs beyond their original trajectory distribution without finetuning. Across perturbed tabletop, household kitchen, and clean-to-randomized bimanual manipulation, Harness VLA improves over the strongest relevant baselines by 38.6 and 25.4 percentage points on LIBERO-Pro and RoboCasa365, respectively, and reaches 58.4% on RoboTwin C2R.

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.

Xinjie Wang, Liu Liu, Taojun Ding et al.

We present EmbodiedGen V2, a generative 3D world engine for building executable sim-ready environments for embodied intelligence. Sim-ready 3D asset generation has advanced rapidly, yet assembling such assets into policy-ready task environments remains largely manual, limiting scalable closed-loop learning. EmbodiedGen V2 addresses this gap through a unified sim-ready representation that connects cross-simulator assets, interaction affordances, task-driven worlds, large-scale multi-room scenes, and stateful Vibe Coding into a generative, editable, and reusable simulation pipeline. The generated environments support manipulation, navigation, mobile manipulation, cross-simulator deployment, and embodied policy training. In evaluation, the asset pipeline achieves 96.5% human acceptance and 98.6% collision success, and 83.3% of task-driven worlds are directly usable for downstream simulation without manual modification. Online reinforcement learning with generated environments further improves simulation success from 9.7% to 79.8%, and transfers to real robots with task success increasing from 21.7% to 75.0%. These results establish EmbodiedGen V2 as scalable simulation infrastructure for training, evaluating, and deploying embodied policies.

Xinye Yang, Zhiyuan Ma, Hongze Yu et al.

Real-world learning for dexterous hands remains brittle because high-dimensional hand actions amplify imitation errors and make reinforcement-learning exploration prone to contact-breaking motion. While combining imitation learning (IL) with online reinforcement learning (RL) can reduce manual supervision, unconstrained exploration in raw hand-action spaces is sample-inefficient and risky for physical hardware. We introduce a latent motion prior module (\prior{}) that maps recent hand-action histories to a compact, history-conditioned latent prior and decodes continuous latent commands into executable high-dimensional hand targets. Built on this prior, \method{} is a three-stage real-world dexterous learning framework: it pretrains \prior{} from demonstrations, trains a visuomotor policy that predicts native arm commands and latent hand-action offsets, and improves the policy with online residual RL in the same latent hand-action space. This shared, decodable interface lets residual exploration make local corrections near demonstrated, contact-consistent hand motions rather than perturbing every finger joint independently. We evaluate \method{} on four real-robot dexterous manipulation tasks against raw, linear, and discrete hand-action interfaces. Starting from small task-specific demonstration sets, \method{} achieves a 56.25\% average IL success rate and raises it to 98.75\% after online RL, reaching 100\% final success on three tasks and 95\% on the remaining task.

Chenhao Yu, Hongwu Wang, Weitao Zhang et al.

Humanoid robots are increasingly expected to perform contact-rich tasks that require not only accurate whole-body motion but also robust physical interaction with surrounding objects and humans. Although recent advances in humanoid motion imitation and whole-body control have achieved remarkable tracking performance, existing datasets and benchmarks primarily focus on kinematic motion while largely overlooking synchronized interaction forces. As a result, current evaluations fail to capture how external interaction forces affect tracking accuracy, stability, and control robustness. In this paper, we present ThorArena, a benchmark for evaluating force-aware humanoid interaction based on human demonstrations with synchronized motion and force measurements. We collect a real-world interaction dataset that simultaneously captures whole-body human motion and forces exerted by both hands across six representative physical interaction tasks. Based on these demonstrations, we propose force-aware evaluation metrics that jointly assess whole-body tracking accuracy, robustness under different force levels, control effort, and episode survival through the Force-Aware Tracking Score (FATS) and complementary diagnostic metrics. We further establish a unified benchmark protocol that replays recorded interaction forces in simulation and provides a standardized evaluation interface for different humanoid control policies. Experiments on representative whole-body control policies demonstrate that force-aware evaluation reveals substantial performance differences that remain largely hidden under conventional no-force evaluation. ThorArena provides a practical and reproducible framework for studying force-aware humanoid interaction and offers a new benchmark for evaluating contact-rich humanoid behaviors.

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.

Zhoutong Ye, Chengwen Zhang, Zhaibin Cui et al.

Gaze target estimation, the task of predicting where a person is looking in a scene, is crucial to understanding human attention and intent. It is a challenging task that combines high-level understanding of global scene semantics and precise spatial reasoning using human appearance (e.g. pose, eye orientation). As a result, human-level performance remains elusive for existing models, limiting their practical application. To this end, we propose PaGE (Practical Gaze Estimator), a gaze estimation model that explicitly models the complex interaction between scene and head features. Using a PaGE model with a large ViT-H+ backbone as the teacher, we further distill student models with lighter backbones on a much larger and more diverse unlabeled dataset. The architectural improvements and novel training recipe allow PaGE to achieve state-of-the-art performance on several gaze estimation tasks, outperforming humans in 7 out of 9 metrics while reducing the human-AI gap by at least 60% in the remaining 2. The distilled student models retain most of the teacher's performance while being lightweight enough for practical deployment on robots and consumer devices. The code and model checkpoints are available at our project page.