Reproduce “Long-Horizon Multi-Object Rearrangement via Distilled Tree Search over Continuous Actions” 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
Kallol Saha, Yining Hong, Amber Li, Yejin Choi, Maxim Likhachev, David Held
DISCO turns short-horizon pick-and-place demonstrations into long-horizon rearrangement behavior by using tree search in simulation to generate training examples and distilling them into a policy and value function. At deployment its learned model acts from point-cloud input without simulator-based search, though robot motion planning remains. On held-out simulated shelf-packing tasks, it achieved 80.0% success versus 20.0% for RL-PD, averaged over 32 seeds.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.