Motus2: A Self-Evolving General World Model for Dexterous Manipulation
2026-08-31 • Robotics
RoboticsArtificial IntelligenceComputer Vision and Pattern RecognitionMachine Learning
AI summaryⓘ
The authors introduce Motus2, a world model designed to help robots learn to manipulate objects skillfully by combining seeing, planning, and evaluating in one system. Motus2 has three parts: a policy that suggests actions, a simulator that predicts what will happen, and an evaluator that checks how good outcomes are. These parts work together repeatedly to improve the robot’s actions using both expert examples and lessons from mistakes. The model also improves by training on a lot of different kinds of vision and robot data, including touch sensors, and runs on a robot with human-like hands and eyes. This approach aims to help robots learn more complex manipulation tasks by continuously refining themselves.
World modelsDexterous manipulationPolicy improvementClosed-loop learningEgocentric dataSimulatorValue modelTactile feedbackRobot-domain adaptationBiomimetic platform
Authors
Hongzhe Bi, Zihao Zhou, Yihang Tang, Jingrui Pang, Shuhe Huang, Haitian Liu, Runqing Wang, Shuai Huang, Yichen Wang, Yiming Cheng, Ruowen Zhao, Zhenghua Li, Hengkai Tan, Xiaolong Liu, Jinhui Wan, Jiabao Liu, Min Zhao, Fan Bao, Jun Zhu
Abstract
General embodied agents should perceive, predict, act, evaluate, and improve within a unified system. World models have shown great promise in building such agents, yet existing models typically append an action output head to a world simulator, without coupling them into a closed decision-and-learning loop for policy improvement. We present Motus2, a self-evolving general world model for dexterous manipulation. Motus2 advances world modeling through model scaling and data scaling. For model scaling, a single model with shared weights exposes three control interfaces: a policy (world-action model), a simulator (action-conditioned world model), and an evaluator (value model). The policy proposes candidate action chunks, the simulator predicts their visual consequences, and the evaluator assesses the predicted outcomes. Their coupling forms a closed decision-and-learning loop for policy improvement. This formulation uses curated expert demonstrations for action learning, while failed and suboptimal interactions provide valuable evidence for dynamics modeling and value learning. For data scaling, Motus2 progresses from large-scale monocular egocentric data to synchronized stereo egocentric data, followed by robot-domain adaptation with robot trajectories and supplementary human-robot alignment data. Motus2 further studies global-autoregressive and hybrid-memory extensions of its sliding-window context, adds tactile feedback for contact-aware control, and is instantiated on a fully biomimetic platform with stereo vision, dual arms, dual dexterous hands, and tactile sensing. Together, egocentric data scaling and closed-loop general world model scaling provide a general path toward self-evolving dexterous manipulation.