Reproduce “Grow Your Limits: Continuous Improvement with Real-World RL for Robotic Locomotion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Laura Smith, Yunhao Cao, Sergey Levine
APRL enables a quadruped robot to learn walking with an action constraint that adapts to dynamics-model prediction error. The constraint expands as behavior becomes predictable and contracts when prediction error rises. On flat ground, after 80,000 steps, APRL reached 0.62 m/s versus 0.44 m/s for a fixed-range baseline, though the baseline learned faster initially. These results compare performance and do not establish safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.