Distilled tree search improves long task robot object rearrangement
Long-Horizon Multi-Object Rearrangement via Distilled Tree Search over Continuous Actions
Robotics
Summary
Moving many objects in a row using robots is hard because small mistakes can add up over time. The authors developed a new way called DISCO that combines looking ahead with careful simulations and learning from these plans. This approach helps the robot figure out better moves in continuous spaces and gets better over time by learning from its own searches. After training, the robot can act quickly without needing to simulate each time. DISCO works better than previous methods in both computer simulations and real robot tests.
What this means in practice
- •For robotics engineers: Design robot controllers that handle long sequences of object moves with fewer errors using distilled tree search techniques.
- •For industrial automation teams: Improve automation of complex object rearrangement tasks requiring precise multi-step handling in warehouses or factories.
Authors
Kallol Saha, Yining Hong, Amber Li, Yejin Choi, Maxim Likhachev, David Held
Abstract
Despite recent advances in policy learning, long-horizon manipulation remains difficult because learned policies must avoid compounding errors while preserving future feasibility. While search-based planning can explicitly reason over such future consequences, it quickly becomes expensive in high-dimensional continuous action spaces. This raises a central question: how can long-horizon reasoning from search be transferred into policies in continuous action spaces? To this end, we propose DISCO: Distilled Tree Search over Continuous Actions. DISCO uses progressive-widening MCTS in physics simulation to generate long-horizon plans. During search, a learned policy proposes continuous actions, while a learned value function scores actions by their expected progress toward task completion. Using GPU-parallelized simulation, DISCO periodically distills MCTS-generated plans into the policy-value model, improving the action proposals and value estimates used in subsequent searches. Finally, the learned policy-value model can be directly deployed without the need for simulator-based search at test time. Across simulated and real-world rearrangement tasks, DISCO outperforms behavior cloning, reinforcement learning, and search-based baselines.