Reproduce “Learning to Walk in the Real World with Minimal Human Effort” 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
Sehoon Ha, Peng Xu, Zhenyu Tan, Sergey Levine, Jie Tan
The authors present a real-robot learning system that trains direction-specific walking policies and schedules tasks to keep a quadruped near its workspace center. It also uses safety-constrained Soft Actor-Critic learning to penalize excessive torso tilt. On flat ground, two of three runs required no intervention; the third required two resets. Challenging surfaces still required 20–30 manual resets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.