This repository contains the code for the paper "Safe Offline Reinforcement Learning using Trajectory-Level Diffusion Models", which was presented at the ICRA2024 Workshop "Back to the Future: Robot Learning Going Probabilistic". The paper can be found here.

Install on Ubuntu

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt

Run

python scripts/generate_dataset.py
python scripts/train.dy
python scripts/eval.py

Reference

@inproceedings{romer2024safe,
  title={Safe Offline Reinforcement Learning using Trajectory-Level Diffusion Models},
  author={R{\"o}mer, Ralf and Brunke, Lukas and Schuck, Martin and Schoellig, Angela P},
  booktitle={ICRA 2024 Workshop Back to the Future: Robot Learning Going Probabilistic}
}

Acknowledgements

We use some code from Diffuser.