Code release for the paper: "Exterior Penalty Policy Optimization with Penalty Metric Network under Constraints", IJCAI 2024.
This repository is based on the omnisafe.
Create conda env: conda create -n epopmn python=3.8
Activate conda env: conda activate epopmn
Install: conda env create -f epopmn.yaml
Register the EPO algorithm into the omnisafe library.
Activate conda env: source activate epopmn
Run the code by:
cd experiments
python train_exp.py --algo EPO_PMN --env-id SafetyPointGoal1-v0 --parallel 1 --total-steps 10000000 --device cpu --vector-env-nums 1 --torch-threads 1
EPOPMN is released under Apache License 2.0.