Peking University's DexPolicy lifts dexterous manipulation success to 85% via annealed exploration
PekingUniversity · hf · 2026-10-02
DexPolicy from Peking University makes RL exploration scale an explicit function of training steps for trajectory-guided dexterous manipulation, annealing from broad to narrow exploration while keeping loss, architecture, reward, and optimizer fixed — addressing the conflict between noise needed for contact discovery and noise that hurts precise control.
- Compared PPO, critic-free GRPO continuation, and flow-parameterized PPO (FPO): on five YCB objects, mean deterministic Target success rose 49.4%→68.1% (FPO), 14.1%→45.4% (GRPO), 32.0%→35.7% (PPO);
- On a real RealMan RM75 arm with an Inspire/RH56 hand (360 trials), FPO lifted mean success from 25.0% to 85.0%, GRPO to 63.3%, PPO to 43.3%;
- Training return, deterministic success, and robustness to execution noise dissociate — schedules should be judged by terminal task success under intended execution conditions. Code and website are open-sourced.
More from Embodied
- PSYONIC's prosthetic Ability Hand now used by 90+ robotics orgs, raising Series A — DhowCIO · 2026-11-19
- Simulator misspecification breeds overconfidence: sim-to-real study on SPEED+ — Nikolay Dolgov · 2026-11-19
- Doppel world model renders the robot body and beats action-only baselines across 8 embodiments — mangahomanga · 2026-11-19
- FloMo: a world action model using 3D motion as the bridge from video pre-training to robot control — CSProfKGD · 2026-11-19
- MIT's paperclip-weight flying robot is first microrobot with onboard sensing and compute — MIT News AI · 2026-11-19
- Ken Goldberg's 30-year arc: from Telegarden crowdsourced farming to Dex-Net and Ambi Robotics — jfiance · 2026-11-19