hari-sikchi / pytorch_CPO

Simple implementation for Constrained Policy Optimization in Pytorch

Constrained Policy Optimization (CPO) implementation in PyTorch

This is PyTorch implementation of Constrained Policy Optimization (CPO) [ArXiv].

If you use this code in your research project please cite us as:

@software{Sikchi_pytorch-CPO,
author = {Sikchi, Harshit},
doi = {10.5281/zenodo.5121027},
license = {MIT},
title = {{pytorch-CPO}},
url = {https://github.com/hari-sikchi/pytorch_CPO}
}

Requirements

• pytorch
• safety_gym
• mpi4py

Instructions

To train an CPO agent on the pointgoal1 task run:

python cpo.py --env=Safexp-PointGoal1-v0 --cost_lim=<cost threshold> --exp_name=<exp path>

This will produce the exp_path folder, where all the outputs are going to be stored.