FlappyAI

Flappy Bird played by a NEAT (NeuroEvolution of Augmenting Topologies) neural network. Birds learn to navigate pipes through evolutionary reinforcement over generations.

How It Works

A population of 100 birds is spawned each generation. Each bird is controlled by a feed-forward neural network with 3 inputs and 1 output:

Input	Description
y	Bird's current vertical position
abs(y - pipe_y)	Vertical distance to the next bottom pipe
abs(y - upPipe_y)	Vertical distance to the next top pipe

If the output is > 0, the bird jumps.

Fitness scoring:

• +0.1 per frame survived
• +5 for passing a pipe
• -1 for dying

The NEAT algorithm evolves the network topology and weights over 50 generations, keeping the top 2 genomes each generation (elitism) and discarding the bottom 80%.

Requirements

• Python 3.8+
• Pygame
• NEAT-Python

Setup

pip install -r requirements.txt

Running

python flappy.py

Note: The game requires graphics and audio assets (graphics/ and audio/ directories with sprite sheets and sound effects) that are not included in this repository. Without them, the game will crash on launch. You'll need to provide your own Flappy Bird-style assets or download them from a free sprite resource.

Configuration

NEAT parameters are in config-feedforward. Key settings:

Parameter	Value	Description
pop_size	100	Birds per generation
num_inputs	3	Neural network inputs
num_outputs	1	Neural network output
fitness_threshold	1,000,000	Stops early if reached
elitism	2	Top genomes preserved each generation
survival_threshold	0.2	Bottom 80% are killed

Testing

This project has no test suite. Run the game and observe the NEAT training output in the console to verify it works.