DexForesight Boosts Dexterous Manipulation Success From 49.4% to 60.5% With Implicit Spatial Foresight
Sharpa-Robotics · hf · 2026-11-11
Sharpa Robotics, with SJTU, CUHK and FDU, introduced DexForesight, a framework that injects dynamics-aware world representations into a dexterous policy's latent space instead of relying on explicit future prediction. During training it learns an action-conditioned spatial world model from offline scene depth, hand kinematics and demonstration trajectories, predicting 3D hand-point motion and visual feature changes — including occluded finger motion. The frozen world model's predictive queries are then aligned with policy queries to condition action generation. At deployment only the policy remains, running on RGB, language and robot state with no depth input. On 11 DexJoCo simulation tasks it lifts base-policy success from 49.4% to 60.5%; on three real-robot tasks it hits 58.3% vs 50.0% for the strongest baseline.
More from Embodied
- Humanoid Factory Deploy Stalls: Teleop Intervention Every 6 Minutes After 15 Weeks — ZeeshanZiaML · 2026-11-11
- CoRL 2026's wildest moments: 'Why do you care about delay?' and 'haptic doesn't matter' — breadli428 · 2026-11-11
- JKU Linz's new Institute for Machine Intelligence hires 2 PhD students in robot learning, deadline in a week — HildeKuehne · 2026-11-11
- Darpinian: Tesla should wait for the AI5 chip before scaling Cybercab production — Darpinian · 2026-11-11
- Maestro paper: cheap Gemini 2.5 beats π₀.5 baselines at zero-shot robot control — dineshjayaraman · 2026-11-11
- OpenDDE: open-source all-atom drug discovery engine hits 51% on hard antibody-antigen benchmark — teortaxesTex · 2026-11-11