/Snake-Q-Learning

An experimental Snake game where a Deep Q-Network (DQN) agent learns to navigate the environment, avoid obstacles, and find food using reinforcement learning techniques.

Primary LanguagePythonMIT LicenseMIT

Snake Game Reinforcement Learning

This repository is a deep dive into the evolution of the Snake Game using Reinforcement Learning (RL). It features three progressively complex models—Classic, Advanced, and Expert—each exploring unique gameplay mechanics and RL strategies.

Models

1. Classic Snake

  • A standard Snake game with basic mechanics.
  • Introduces obstacles that the snake must avoid to survive.
  • Reinforcement learning agent implemented to learn optimal movements.

2. Advanced Snake

  • Adds dynamic obstacles that move around the game environment.
  • Increased difficulty with adaptive gameplay mechanics.
  • Special food is introduced, granting temporary immunity upon consumption.

3. Expert Snake

  • Features multi-agent gameplay with competitive reinforcement learning.
  • Two snakes compete to maximize their scores:
    • Eating food.
    • Avoiding collisions with obstacles, walls, and each other.
  • Dynamic environment with both static and moving obstacles.

Blog Series

Learn more about the ideas and experiments behind these models in the blog series: Building a Deep Q-Network Agent for Snake A Reinforcement Learning Journey

Setup Instructions

Prerequisites

  • Python 3.8 or higher
  • Install required dependencies from requirements.txt: 
    pip install -r requirements.txt

###Acknowledgments Special thanks to Patrik Lober for his valuable contributions and insights that greatly enhanced the project.