Skip to content

Implementation of Reinforcement Learning algorithms on the OpenAI Gym Highway-Environment

Notifications You must be signed in to change notification settings

aniketmpatil/RL-Highway-Env-Project

Repository files navigation

Reinforcement Learning algorithms for Lane Keeping and Obstacle Avoidance for Autonomous Vehicles

CS525: Reinforcement Learning - Worcester Polytechnic Institute, Fall 2022

Members: Rutwik Bonde, Prathamesh Bhamare, Aniket Patil - Master of Science in Robotics Engineering


Motivation:

Our goal is to train RL agents to navigate ego vehicle safely within racetrack-v0 environment, third party environment in the Open-AI gym and benchmark the results for lane keeping and obstacle avoidance tasks.


Training Outputs


Procedure:

Training:

  1. Clone this repository.

  2. Create an empty python environment:

python3 -m venv rl_venv
  1. Activate the environment and install libraries
source rl_venv/bin/activate
pip3 install -r requirements.txt
  1. Configure the paramters in the config/params.yaml file.

IMPORTANT: For each run, ensure that you modify the exp_id so that the log and checkpoint files are not overwritten. Use the agent tag to select between "PPO", "DDPG" and "A3C" agents.

  1. Run the main.py file
python3 main.py

Testing:

  1. In the same environment as above, change the config/params.yaml file by setting train tag to False. This will set the mode to Testing. Set the save_video tag to True to generate the video output for testing.

  2. Add path to the model in load_model. Use the models in the models folder to load pretrained weights.

About

Implementation of Reinforcement Learning algorithms on the OpenAI Gym Highway-Environment

Topics

Resources

Stars

Watchers

Forks