Reproduce “Reinforcement Learning via Advantage Guidance” 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
Zhiyuan Zhou, Sergey Levine
Advantage Guidance trains an offline policy to respond to estimated action quality while retaining supervised flow-matching training. At sampling, it contrasts high- and low-quality conditions to favor higher-advantage actions rather than merely matching a requested value. The method uses IQL-based advantage estimates and is evaluated on 26 simulated sparse-reward tasks. It reports higher aggregate success than prior methods, but does not lead every task family.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.