Pong RL

A reinforcement learning environment implementing the classic Pong game using Pygame and Gymnasium. The environment includes both a playable version and training capabilities using Stable Baselines 3.

▶️ Watch full video

Features

• Custom Gymnasium environment for Pong
• Human playable mode with keyboard controls
• AI training using PPO (Proximal Policy Optimization)
• Real-time visualization of training progress
• Configurable game parameters (speed, size, colors)
• Visual effects (shadows, smooth animations)

Usage

Playing the Game

Run the play script to play against the hardcoded AI:

python scripts/play.py

Controls:

• Up Arrow: Move paddle up
• Down Arrow: Move paddle down
• ESC: Quit game

Training an Agent

To train a new agent:

python scripts/train.py

The training script will:

• Create a new run directory in logs/
• Save the best model during training
• Log metrics to TensorBoard
• Display periodic evaluation episodes

Evaluating a Trained Agent

To evaluate a trained model:

python scripts/eval.py

Development

For development, install additional dependencies:

pip install -r requirements-dev.txt

Acknowledgments

• Built with Pygame
• RL implementation using Stable Baselines3
• Environment interface with Gymnasium