SCPO is a safe reinforcement learning algorithm. This repo is a fork of Stable Baselines3.
Note: Stable-Baselines3 supports PyTorch >= 1.11
SCPO requires Python 3.7+.
Install the Stable Baselines3 package:
pip install -r requirements.txt
We use environments from Bullet-Safety-Gym. Please follow the installation steps from https://github.com/SvenGronauer/Bullet-Safety-Gym.
If you want to run pytorch in gpu mode, please install cuda and pytorch separately https://pytorch.org/
Example code for training can be found at train.py. To train models with the best hyperparameters, please check train_best_hyper.py.
Check play.py.
 |
 |
|---|---|
 |
 |
If you find the repository useful, please cite the study
@article{mhamed2023scpo,
title={SCPO: Safe Reinforcement Learning with Safety Critic Policy Optimization},
author={Mhamed, Jaafar and Gu, Shangding},
journal={arXiv preprint arXiv:2311.00880},
year={2023}
}