Motus2 world model hits 84% success on dexterous tasks with self-evolving closed loop
burny_tech · x · 2026-09-08
Motus2 is a self-evolving general world model for dexterous manipulation that closes the loop between prediction, evaluation, and policy improvement.
- A single shared-weight model serves as policy, simulator, and value model: it proposes action chunks, imagines their visual consequences, and scores them, improving via model-based RL
- Failed and suboptimal interactions feed dynamics and value learning
- Data scaling: 130K hours of egocentric pretraining, moving to stereo egocentric data plus robot-domain adaptation
- Adds tactile feedback and global-autoregressive/hybrid-memory context extensions, deployed on a biomimetic robot platform
It reaches 84% average success across five dexterous tasks, with MBRL and test-time planning further boosting performance.
More from Embodied
- RealSense D585 Pro AI-Powered 3D Stereo Depth Camera Unboxed for Physical AI — chrismatthieu · 2026-10-26
- Physical world agent swarm demo shared on X — pduan · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26
- 'Lamp Swarm': a desk of lamp-shaped robots shows what physical AI work looks like — dee_hw · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26
- SJTU's VaM lets robots learn online from mistakes, 95%+ success over 4-hour autonomous runs — chris_j_paxton · 2026-10-26