This repository is the official implementation of Feasible Actor-Critic: Constrained Reinforcement Learning for Ensuring Statewise Safety. The code base of this implementation is the Parallel Asynchronous Buffer-Actor-Learner (PABAL) architecture, which includes implementations of most common RL algorithms with the state-of-the-art training efficiency. If you are interested in or want to contribute to PABAL, you can contact me or the original creator.
Important information for installing the requirements:
To install requirements:
pip install -r requirements.txt
To train the model(s) in the paper, run this command:
python train_scripts4fsac.py --env_id Safexp-PointButton1-v0 --seed 0
To test and evaluate trained policies, run:
python train_scripts4fsac.py --mode testing --test_dir <your_log_dir> --test_iter_list [3000000]
When contributing to this repository, please first discuss the change you wish to make via issue, email, or any other method with me before making a change.