Awesome

<h1 align="center"> Resisting Stochastic Risks in Diffusion Planners with the Trajectory Aggregation Tree 🌳 </h1> <div align="center"> ICML 2024 (Spotlight) </div> | <a href="https://github.com/langfengQ/tree-diffusion-planner">Website</a> | <a href="https://arxiv.org/abs/2405.17879">Arxiv</a> | <a href="https://proceedings.mlr.press/v235/feng24b.html">Paper</a> | <a href="https://icml.cc/media/PosterPDFs/ICML%202024/34197.png?t=1719634719.7916923">Poster</a> | | <a href="https://github.com/langfengQ/tree-diffusion-planner/tree/maze2d">Code in Maze2D</a> | <a href="https://github.com/langfengQ/tree-diffusion-planner/tree/kuka">Code in Kuka</a> | <a href="https://github.com/langfengQ/tree-diffusion-planner/tree/locomotion">Code in Locomotion</a> |

This repository contains the complete code of TAT built upon the pre-trained Diffuser. The codes are organized into three separate sub-branches for easy access:

Maze2D environment: access the code here (maze2d branch).
Kuka Block Stacking environment: access the code here (kuka branch).
MuJoCo Locomotion environment: access the code here (locomotion branch).

⚠️ Note: the master branch does not contain any codes. Please refer to the maze2d, kuka, and locomotion branches for the full implementations.

🚀 Quick Start

To get started with this repository, follow these steps:

Clone the repository and verify the available branches:

git clone https://github.com/langfengQ/tree-diffusion-planner.git
cd tree-diffusion-planner/
git branch -a
# * master
#   remotes/origin/HEAD -> origin/master
#   remotes/origin/kuka
#   remotes/origin/locomotion
#   remotes/origin/master
#   remotes/origin/maze2d

The command git branch -a lists all the branches, and you should see the following branches: master, maze2d, kuka, and locomotion.

Switch to a specific branch to access the corresponding environment's code:

For the Maze2D environment:

git checkout maze2d
# Branch 'maze2d' set up to track remote branch 'maze2d' from 'origin'.
# Switched to a new branch 'maze2d'

For the Kuka Block Stacking environment:

git checkout kuka
# Branch 'kuka' set up to track remote branch 'kuka' from 'origin'.
# Switched to a new branch 'kuka'

For the MuJoCo Locomotion environment:

git checkout locomotion
# Branch 'locomotion' set up to track remote branch 'locomotion' from 'origin'.
# Switched to a new branch 'locomotion'

Follow the branch-specific README: Now you are ready to explore and experiment with TAT in your chosen environment! 🥳 Then, you can refer to that sub-branch's README file for instructions on setting up the environment, running experiments, and additional configurations.

❓ Issue

If you have any questions about the code, please feel free to open an issue!

📝 Citation

@inproceedings{feng2024resisting,
  title={Resisting Stochastic Risks in Diffusion Planners with the Trajectory Aggregation Tree},
  author={Feng, Lang and Gu, Pengjie and An, Bo and Pan, Gang},
  booktitle={International Conference on Machine Learning},
  pages={13175--13198},
  volume={235},
  year={2024},
  organization={PMLR},
}

🙏 Acknowledgements

This implementation is based on the Diffuser repo.