Reproduce “Sample-Efficient Reinforcement Learning for Policy Improvement via Flow-Reversed Action Anchors” 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
Carl Qi, Weijun Dong, Xiaojie Wang, Peiyi Chen, Rahul Sajnani, Mingyuan Zhang, Amy Zhang, Jose Barreiros and 3 more
AnchorQ improves frozen robot policies by mapping reference motions into noise-space anchors, sampling local particles around them, and using a learned critic to select refinements. On the held-out LIBERO-90 15-task split, it reports 86.0% success after 250,000 environment steps, versus 37.2% for DSRL. This is simulation evidence and does not establish gains beyond the tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.