Reproduce “Data-efficient Hindsight Off-policy Option Learning” 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

Markus Wulfmeier, Dushyant Rao, Roland Hafner, Thomas Lampe, Abbas Abdolmaleki, Tim Hertweck, Michael Neunert, Dhruva Tirumala and 3 more

HO2 trains hierarchical reinforcement-learning policies from replayed trajectories by inferring latent options recursively from states and actions, then differentiating through that inference to update the controller, skills, and termination rules. In simulated pixel-based robot tasks, option policies improve performance and sample efficiency over mixture policies, especially on complex stacking; results use proprioception too and do not establish real-robot performance.

Summary by alphaXiv.

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