Reproduce “Outcome-Balanced PPO and Policy-Induced Flow Rectification for Flow-Based VLA Policies” 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

Yuhan Wang, Ruiyang Wang, Jiaguan Zhu

The authors improve online training of flow-based vision-language-action policies for robot manipulation by reweighting advantages across tasks and outcomes, then periodically refitting the action flow when stochastic and deterministic sampling diverge. In four simulated benchmarks, their method averages 88.9% success versus 86.6% for π_RL. This is simulated evidence; the reported physical trial tests a different method combination.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.