An AI-powered Snake game built using Python, Pygame, and PyTorch. The AI agent learns to play the Snake game using Q-Learning with a Deep Neural Network.

Watch a quick demo of the AI Snake Game in action:
🔗 Click here to watch the video

• Python
• Pygame
• PyTorch
• Conda (for environment management)

• AI agent learns to play the Snake game using Reinforcement Learning.
• Real-time gameplay with dynamic UI updates.
• The agent improves its performance over time using Q-Learning.
• Graphical Plot to visualize the agent's progress (score and mean score).
• Fully customizable codebase with detailed comments for learning purposes.

Ensure you have the following installed on your machine:

• Python 3.7+
• Conda

git clone https://github.com/bebopkenny/Snake-AI-Game.git
cd Snake-AI-Game

conda create -n pygame_env python=3.7
conda activate pygame_env

pip install -r requirements.txt

python game.py

python agent.py

The AI agent is trained using Q-Learning, a type of reinforcement learning where the agent learns to maximize rewards by taking specific actions.

• State Representation: The agent observes the game state, including the position of the snake's head, food location, and possible collisions.
• Action Space: The agent can move in three directions: straight, left, or right.
• Reward System:
  • Positive reward (+10) for eating food.
  • Negative reward (-10) for hitting a wall or itself.

The agent uses a Linear Neural Network built with PyTorch to predict the best possible action based on the current state.

A real-time graph is displayed during training to visualize the agent's progress:

• Number of Games vs. Score
• Mean Score over Time

The graph is generated using Matplotlib.

• Kenny Garcia
• Email: kennygarcia15@yahoo.com
• Portfolio: kennygarcia.net

• Add more complex reward structures to improve the agent's decision-making.
• Optimize the neural network architecture for faster training.
• Implement additional gameplay features such as obstacles and different game modes.

This project is open-source and available under the MIT License.