Maze Game with Machine Learning

An interactive maze game that uses reinforcement learning (Q-learning) to find optimal paths through procedurally generated mazes.

Features

• 🎮 Interactive maze generation
• 🧠 Machine learning pathfinding
• 📊 Real-time statistics display
• 🎯 Goal-oriented learning
• 🔄 Continuous training capability
• 🐍 Built with Python and PyTorch

Installation

1. Clone the repository:

git clone https://github.com/rajaminyak/maze-game-ml.git
cd maze-game-ml

2. Install required packages:

pip install pygame torch numpy

Usage

Run the game:

python main.py

Controls

• SPACE: Let ML agent make a move
• R: Reset/generate new maze
• T: Train ML agent (500 episodes)
• D: Toggle debug mode

How It Works

Machine Learning Components

1. State Space (What the AI "sees"):

   • Distances to walls in 4 directions
   • Distance to goal
   • Angle to goal

2. Action Space (What the AI can do):

   • Move Up
   • Move Down
   • Move Left
   • Move Right

3. Reward System:

   • +100 for reaching goal
   • -1 for hitting walls
   • -0.1 step cost
   • Distance-based penalties

Neural Network Architecture

QNetwork Structure:
Input Layer (6 neurons) 
    → Hidden Layer (64 neurons) 
    → Hidden Layer (64 neurons) 
    → Output Layer (4 neurons)

Learning Process

• Uses epsilon-greedy strategy for exploration
• Experience replay for better learning
• Adaptive learning rate
• Continuous improvement through training

Visual Elements

• 🟢 Green Circle: Player
• 🔴 Red Square: Goal
• ⬛ Dark Gray: Walls
• ⬜ Light Gray: Paths
• 🔷 Blue: Visited cells
• 🟡 Orange: Training indicator

Performance Metrics

The game displays real-time statistics:

• Steps taken
• Goals reached
• Current epsilon value (exploration rate)
• Training episode (during training)
• Current reward

Debug Mode

Debug mode (D key) provides additional information:

• Episode details
• Reward calculations
• Path analysis
• Training progress

Project Structure

maze-game-ml/
│
├── main.py           # Main game file
├── requirements.txt  # Dependencies
└── README.md        # Documentation

Technical Details

Key Classes

1. QNetwork: Neural network for decision making
2. AdvancedMazeGame: Main game logic and ML integration

Machine Learning Parameters

learning_rate = 0.001
gamma = 0.99         # Discount factor
epsilon_start = 1.0  # Initial exploration rate
epsilon_min = 0.01   # Minimum exploration rate
memory_size = 10000  # Experience replay buffer size
batch_size = 32      # Training batch size

Development

To contribute:

1. Fork the repository
2. Create your feature branch
3. Make your changes
4. Submit a pull request

Troubleshooting

Common issues and solutions:

1. Game freezes during training:

   • Reduce maze size
   • Lower episode count
   • Check system resources

2. Agent not learning:

   • Increase training episodes
   • Adjust reward values
   • Check maze connectivity

Future Improvements

• Add different maze generation algorithms
• Implement A* comparison
• Add visualization of neural network decisions
• Support for larger mazes
• Save/load trained models

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• PyTorch documentation
• Pygame community
• Reinforcement learning resources

Authors

• Your Name
• Contributors welcome!