Glitch – Reinforcement Learning Agent For Playing Atari Games

Status: Archived – No Longer Maintained

Glitch is a reinforcement learning project that implements a Deep Q-Network (DQN) agent designed to play classic Atari games. The agent learns to make optimal decisions in a discrete action space using deep neural networks to approximate Q-values.

Overview

This project demonstrates how reinforcement learning can be applied to Atari game environments using deep learning techniques. It serves as a reference or starting point for those interested in building game-playing AI agents.

Supported environments include:

• Pong
• Breakout
• Space Invaders
  (and other Atari games via Gymnasium)

Key Features

• Deep Q-Network (DQN): Learns Q-value approximations through deep neural nets.
• Atari Environment Integration: Utilizes environments from the Arcade Learning Environment (via OpenAI Gym).
• Discrete Action Space: Handles finite, defined sets of game actions.

Requirements

• Python 3.x
• Virtualenv (recommended)
• Gymnasium (with Atari support)
• TensorFlow

Install dependencies with:

pip install -r requirements.txt

Disclaimer

This project is no longer maintained and may not function as intended. It is archived for reference purposes only. Users are welcome to fork and modify the codebase as needed.

License

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

---

Author: Kaan Bertuğ Ersoy
GitHub Profile