Contact Trajectory Optimization via MCTS

This repo hosts the source code for the paper Efficient Object Manipulation Planning with Monte Carlo Tree Search

Dependencies

• Pinocchio for rigid body dynamics
• Bullet Utils for interfacing PyBullet with Pinocchio
• Robot Properties NYU Finger for NYU finger URDFs and configuration files
• Other standard dependencies can be found in requirements.txt.

Notebooks

We prepared a few notebooks

• /demos mcts[miqp]_playground.ipynb shows step-by-step how to generate a contact plan for a given motion with force plots and plan visualization. mcts[miqp]_bullet_simulation.ipynb simulates the contact plan with an impedance controller in PyBullet.
• /experiments contains notebooks to reproduce the respective experiments in the paper.
• /train contains notebooks to generate the training data and to train the neural networks.

Citing

@article{zhu2022efficient,
  title={Efficient Object Manipulation Planning with Monte Carlo Tree Search},
  author={Zhu, Huaijiang and Meduri, Avadesh and Righetti, Ludovic},
  journal={arXiv preprint arXiv:2206.09023},
  year={2022}
}

Maintainer

• Huaijiang Zhu

Copyrights

Copyright(c) 2023 New York University

License

BSD 3-Clause License