Reproduce “ExRL: Sample-Efficient Online Adaptation of Long-Chunk Policies via Execution Length Control” 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
Xu Luo, Jiaying Yang, Hao Wu, Junlin Xie, Zijin Hong, Ji Zhang, Lianli Gao, Jingkuan Song
ExRL adapts robot policies by learning how many actions from a predicted chunk to execute before replanning, while leaving the base policy’s actions unchanged. Its decision is conditioned on the robot’s state and proposed chunk and trained with off-policy reinforcement learning. On a LIBERO-90 task, π₀.₅ exceeds 90% success within 200,000 environment steps; this is a simulation result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.