RL GridWorld Game

A reinforcement learning project for grid world navigation using Stable Baselines3.

Overview

Train AI agents to navigate through grid-based mazes using PPO and DQN algorithms. The agent learns to find the shortest path from start (@) to goal ($) while avoiding obstacles (#).

Features

• Multiple Algorithms: Supports PPO and DQN reinforcement learning algorithms
• Custom Maps: Various difficulty levels (easy, medium, hard) with different maze layouts
• Real-time Training: Visualize training progress and agent behavior
• Model Persistence: Save and load trained models for evaluation

Quick Start

1. Install Dependencies:

pip install stable-baselines3 gymnasium matplotlib numpy

2. Train a Model:

python train.py --map-dir maps/ --algo ppo --timesteps 50000

3. Test the Model:

python test.py --map map.txt --model model.zip

Project Structure

├── gridworld_env.py # Custom grid world environment
├── train.py # Training script ├── test.py # Testing and evaluation
├── maps/ # Maze map files
├── models/ # Saved trained models
└── logs/ # Training logs

Map Format

Maps use simple text-based format:

• @ - Start position
• $ - Goal position
• # - Obstacle
• * - Free space