Model-Free Safe Reinforcement Learning through Neural Barrier Certificate

This repository contains the official implementation of Model-Free Safe Reinforcement Learning through Neural Barrier Certificate published in IEEE Robotics and Automation Letters.

Installation

Clone the repository:

git clone https://github.com/jjyyxx/srlnbc.git
cd srlnbc

Create conda environment:

conda env create -f environment.yml
conda activate srlnbc

Install Safety Gym. See https://github.com/openai/safety-gym.

Install MetaDrive. See https://github.com/metadriverse/metadrive.

Install srlnbc:

pip install -e .