Safe and Stable RL (S2RL) Driving Policies Using Control Barrier and Control Lyapunov Functions

This repository contains code for the paper Safe and Stable RL (S2RL) Driving Policies Using Control Barrier and Control Lyapunov Functions Publication Link : https://ieeexplore.ieee.org/document/9737420

System Requirements

The code has been tested in systems with the following OS

• Ubuntu 20.04.2 LTS

Installation

1. Setup conda environment

$ conda create -n env_name python=3.8.5
$ conda activate env_name

2. Clone the repository to an appropriate folder
3. Install requirements

$ pip install -r requirements.txt
$ pip install -e .

Usage

All the trained policies are avialable in the policies folder

The pre-trained policies are available in the Policies folder. Following environement files are available for training

1) KinematicBicycleGymCutACC - Straight Drive with limited steering
2) KinematicBicycleGymTurn - Training Left/Right Turning agents
3) KinematicBicycleGymLane - Training Left/Right Lane change
4) KinematicBicycleGymCutIn - Training Left/Right Cut In

The main program takes the following command line arguments

1) --env : environment name (default is KinematicBicycleGymCutACC)
2) --actor : filepath to the actor network (default is Policies/ACCS2RLPPO/ppo_actorKinematicBicycleGymACC.pth)
3) --critic : filepath to the critic network (default is Policies/ACCS2RLPPO/ppo_criticKinematicBicycleGymACC.pth)

The hyperparameters can be changed in the hyperparameters.yml file


Note : Change the default arguments inside the main.py file otherwise the command line may become too long


### Testing

To test a trained model run:

$ python main.py --test

### Training the policy for a env

$ python main.py --train

Specify the appropriate environment for training

The GIF demonstrations are stored inside the demonstrations folder