stevenyangyj / p3

Official code for the paper: Pareto Policy Pool for Model-based Offline Reinforcement Learning

P3

Code for "Pareto Policy Pool for Model-based Offline RL", presented in ICLR 2022.

Key Dependencies

python==3.6.13
- d4rl==1.1
- ray==1.0.0
- gym==0.18.3
- torch==1.7.1
- tensorflow==2.3.1
- mujoco-py==2.0.2.13

Quick Start

python p3.py

Notes

Pretrained environment models and behaviour cloning policies can be downloaded via Google Drive.

Citing P3

If you use the code in P3, please kindly cite our paper using following BibTeX entry.

@inproceedings{
yang2022pareto,
title={Pareto Policy Pool for Model-based Offline Reinforcement Learning},
author={Yijun Yang and Jing Jiang and Tianyi Zhou and Jie Ma and Yuhui Shi},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=OqcZu8JIIzS}
}

Acknowledgement

We appreciate the open source of the following projects:

MOPO, MOReL, and D4RL