Reproduce “Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience” 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
Kai Ye, Yifan Jin, Qianhang Luo, Hongru Zhang, Jiedong Zhang, Shuyuan Hu, Song-Yuan Wu, Jihai Xu and 2 more
Evo-RL selectively reuses robot experience to improve deployed policies. It retains human-controlled corrections from successful interventions and uses estimated advantages to select demonstration segments and segments from successful autonomous rollouts for fine-tuning. On PiperX screw–sleeve insertion, it records 88.24% ASR versus RECAP’s 57.78%, with ASR counting stage attempts across five units, not whole-task success, under a five-minute trial limit.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.