Reproduce “Deploying Physically and Epistemically Compliant Agents by Distilling Belief-Space Task and Motion Planners” 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
Phillip J.K. Christoffersen, Rachel Ma, Patrick Barragán, Dylan Hadfield-Menell
The authors train a policy from offline belief-space planner trajectories to act on raw observations and history, using Dual Averaging and ρ-Optimal Reinforcement. In Scavenger, the policy exceeds 90% success on visual-hint and language-hint tasks, while flat and hierarchical PPO remain at or below 5%. These single-seed simulated results assume a tractable planning model and do not certify safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.