Reproduce “Tidiness Score-Guided Monte Carlo Tree Search for Visual Tabletop Rearrangement” 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
Hogun Kee, Wooseok Oh, Minjae Kang, Hyemin Ahn, Songhwai Oh
TSMCTS (Tidiness Score-Guided Monte Carlo Tree Search) enables robots to autonomously tidy tabletops using an RGB-D camera, guided by a learned vision-based tidiness score. The framework achieves an 88.5% success rate in diverse simulated environments and 85% in real-world scenarios, demonstrating generalization to unseen objects and zero-shot sim-to-real transfer.
Key insights
- A learned, continuous vision-based tidiness score can effectively generalize across diverse environments and unseen objects, providing robust guidance for autonomous rearrangement.
- Monte Carlo Tree Search, when guided by such a tidiness score, can autonomously plan optimal pick-and-place sequences for complex tidying tasks without explicit goal configurations.
- Simplified visual patch movement for next-state prediction within MCTS provides sufficient fidelity for effective planning, enabling efficient real-time operation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.