PyRace Reinforcement Learning

This repository contains implementations of reinforcement learning algorithms for the PyRace environment, a 2D racing game.

Implementations

1. Q-Table Approach: Traditional RL with discretized state space

   • File: Pyrace_RL_QTable.py

2. Deep Q-Network (DQN) Approach: Modern deep RL implementation

   • File: Pyrace_RL_DQN.py
   • Details: See DQN README

Environment

The custom PyRace environment is a 2D racing game where an agent learns to navigate a track. It's implemented as a custom Gymnasium environment in the gym_race package.

Requirements

Install all dependencies using:

pip install -r requirements.txt

Usage

Running the Q-Table Implementation

python Pyrace_RL_QTable.py

Running the DQN Implementation

python Pyrace_RL_DQN.py

You can modify these files to:

• Train a new model
• Continue training from a saved model
• Play using a trained model without further learning

Model Selection

Both implementations allow for:

1. Training from scratch
2. Loading a saved model and continuing training
3. Loading a saved model to play without learning

See the respective files for instructions on how to switch between these modes.

Acknowledgements

This project is part of the Reinforcement Learning course.