DRP Pre-release

This repository contains the pre-release version of DRP: Deep Reactive Policy, with the IMPACT model and checkpoints.

Note: We will be adding a lot more features and examples for the full release, aiming to be released sometime in June 2026. Stay tuned!

Installation

# Create the conda environment
conda create -n drp python=3.8
conda activate drp

# Clone and install DRP
git clone -b v0.0.0 git@github.com:deep-reactive-policy/drp.git
cd drp
pip install -e .
pip install -r requirements.txt

# Install robofin
git clone -b v0.0.1 git@github.com:Jim-Young6709/robofin.git
pip install -e robofin/

# Install pointnet2_ops. This step may take a few minutes
git clone -b drp git@github.com:Jim-Young6709/pointnet2_ops.git
pip install -e pointnet2_ops/ --no-build-isolation

Test the Installation

Run the example inference script, which launches an interactive DRP GUI:

python drp/drp_inference.py

drp_example.mp4

Citation

If you find this codebase useful in your research, please cite:

@article{yang2025deep,
  title={Deep Reactive Policy: Learning Reactive Manipulator Motion Planning for Dynamic Environments},
  author={Jiahui Yang and Jason Jingzhou Liu and Yulong Li and Youssef Khaky and Deepak Pathak},
  journal={9th Annual Conference on Robot Learning},
  year={2025},
}

@inproceedings{dalal2025neural,
  title={Neural MP: A Neural Motion Planner},
  author={Dalal, Murtaza and Yang, Jiahui and Mendonca, Russell and Khaky, Youssef and Salakhutdinov, Ruslan and Pathak, Deepak},
  booktitle={2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={1--8},
  year={2025},
  organization={IEEE}
}