FlappyAI

Flappy bird automation using Neuroevolution of Augmenting Topologies (NEAT) in Python

Everything Used

• Genetic Algorithm especially NEAT concept
• Unsupervised Learning
• Neural Network

• NEAT-Python used in developing the Genetic Algorithm (NEAT) and also the Neural Network (Forward Propagation)
• Matplotlib and Pillow used in the visualization of the neural network
• Pygame used for creating the game (Environment)

Files Documentation

• Bin (All of the python scripts are here)
  • environment.py > Helper class that control the game itself (Rendered, Pipe, Bird, Gravity, and also game speed)
  • evolve.py > Genetic Algorithm for generating the best individual
  • main.py > Using the generated best individual from evolve.py and then put the individual to the game alone
  • visualize.py > Helper class that visualize the neural network in another window
• Img (Assets that is used by the game)
• Model (Where the best individuals are stored)

Resource

• NEAT - Introduction
• NEAT-Python lib

Installation

In case you want to try it on your local machine

1. Clone
2. Enter the virtual env
   • in windows powershell you can

   cd Scripts
   ./activate
   

3. And now you can run the scripts inside /bin

• You don't need to install the requirements inside requirements.txt when you use the virtual env

Notes

• In the main.py, default best bird is still hard coded (I think I just deleted the .pickle files but still manage to stored those value, you can customize and make your own bird farm)
• Using the above hard coded sample, I've never seen the bird fail
• Game speed, visualization of the neural network can be customized in main.py hyperparam
• Feel free to reach me in discord