breakout

A simplified version of Atari Breakout, played by you or an AI.

how to use

1. Install pygame and numpy

   pip3 install pygame
   

   pip3 install numpy
   

2. Run main.py (using Python 3) from the appropriate directory

   python3 main.py
   

3. Play Breakout, train an AI to do it, or watch a pre-trained AI try and play the game!

how it works

The AI is a Deep Q-Learning Network that responds to rewards and punishments from the moves it makes in the game environment. Whenever it hits the ball it gets a reward (gets a positive number), but whenever it misses the ball it is punished (gets a negative number).

In this current version, the AI can only see the x and y distance between itself and the ball, yet it still shows improvement over time! In a future version it would be great if the AI could see the whole board and trade its fully-connected neural network for a convolutional neural network. It is for this reason that I have left the game window quite small, so that hopefully one day the AI will be able to take in that full image (hopefully without much lag) and make a move.

More information about the project can be seen in this crude video: https://www.youtube.com/watch?v=nAjENKu2_6c