Body Transformer: Leveraging Robot Embodiment for Policy Learning

Paper Website

Body Transformer (BoT) is an architecture tailored for robot learning, which exhibits better performance in both imitation and reinforcement learning scenarios.

Summary

Please refer to the corresponding readme files in the subdirectories for more detailed instructions:

• Imitation Learning
• Reinforcement Learning
• Unitree A1 Sim-to-Real

Citation

If you find BoT useful for your research, please cite this work:

@article{sferrazza2024body,
  title={Body Transformer: Leveraging Robot Embodiment for Policy Learning},
  author={Sferrazza, Carmelo and Huang, Dun-Ming and Liu, Fangchen and Lee, Jongmin and Abbeel, Pieter},
  journal={arXiv preprint arXiv:2408.06316},
  year={2024}
}

References

This codebase contains some files adapted from other sources:

• Robot Parkour Learning: https://github.com/ZiwenZhuang/parkour
• IsaacGymEnvs: https://github.com/isaac-sim/IsaacGymEnvs
• rl_games: https://github.com/Denys88/rl_games
• PyTorch: https://github.com/pytorch/pytorch