ExRL Boosts Sample-Efficient RL Adaptation of Long-Chunk Robot Policies via Execution Length Control
Xu Luo · hf · 2026-11-11
Researchers from UESTC, Tongji, and HK PolyU present ExRL, a method for sample-efficient online adaptation of long-chunk robot policies.
- A lightweight adapter trained with off-policy RL predicts how long each action chunk should execute, addressing the sample-efficiency collapse existing RL methods suffer as chunks grow longer.
- On contact-rich and long-horizon real-world tasks (Tool Hang, Stack Cube, Power Insertion, Clothing Folding, Parts Organization), ExRL lifts a π0.5 base policy to high success rates with only tens of minutes of physical interaction or a few hundred episodes.
- ExRL-SAC maintains high learning efficiency at large chunk sizes (25-50) where ResFIT, DSRL, and DICE-RL degrade.
Project page: exrl-project.github.io
More from Embodied
- 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
- Tesla demos Optimus walking inside a world model at CoRL, real policy in control — Scobleizer · 2026-11-11
- NUS Proposes WM-PPO: World-Model-Predicted Successor States Improve PPO Policy Learning — NationalUniversityofSingapore · 2026-11-11
- DexForesight Boosts Dexterous Manipulation Success From 49.4% to 60.5% With Implicit Spatial Foresight — Sharpa-Robotics · 2026-11-11
- NTU Study Finds Time-Indexed Memory Makes VLA Robots Mistake Pauses for Progress — NanyangTechnologicalUniversity · 2026-11-11