Flappy Bird AI - Neural Network-Based Evolutionary Agent

📌 Introduction

This project implements an AI agent to play a Flappy Bird-like game using Neural Networks and Evolutionary Algorithms. The agents learn to navigate through pipes by evolving over generations, improving their performance with genetic algorithms.

🎯 Objectives

• Develop an AI agent that learns to play Flappy Bird using Neural Networks.
• Implement evolutionary learning for continuous improvement.
• Simulate a genetic algorithm for natural selection among agents.
• Experiment with mutation for better performance.

🌟 Highlights

• AI-powered gameplay: No human intervention needed.
• Neural Network-based decision-making for flapping.
• Genetic Algorithm for Evolution: Best agents survive and reproduce.
• Vision-based AI: Agents "see" pipes and decide actions.

🔧 Recent Updates / Changes

• Randomized Initial Y: Each bird starts at a random vertical position to avoid converging at the same point.
• Top Collision: Birds now die if they collide with the top boundary, preventing them from “sticking” there.
• Reduced Flap Strength: The flap velocity was lowered from -7 to -6 to give more precise control.
• Increased Pipe Gap (Optional): Pipe openings can be widened to make early learning easier.
• Lower Mutation Rate: Reduced from 80% to 20% for more stable evolution.
• Improved Fitness: Birds gain extra points for each pipe they pass, in addition to lifespan-based scoring.

🔬 Methodology

1. Initialize Population
   Random agents with different neural network weights.
2. Vision Processing
   Agents detect pipes, ground, and boundaries.
3. Decision Making
   The neural network predicts whether to flap or not.
4. Fitness Calculation
   Longer survival + passing pipes → higher fitness.
5. Natural Selection
   The best-performing agents reproduce for the next generation.
6. Mutation & Evolution
   New agents inherit traits with slight mutations to explore new strategies.

🛠️ Software and Libraries Used

• Python (Main language)
• Pygame (For game rendering)
• NumPy (For potential matrix operations)
• Random (For mutations and randomness in evolution)

🚀 Usage

1️⃣ Setup the Environment

Make sure you have Python installed, then install dependencies:

pip install pygame numpy

2️⃣ Run the Game

Execute the main application:

python app.py

3️⃣ Observe the AI

Watch how the agents evolve and improve over time. You’ll see statistics like:

• Generation Count
• Number of Agents Alive
• Average Fitness
• Number of Pipes Passed

🔮 Future Scope

• Implement Deep Reinforcement Learning for more advanced training.
• Introduce multi-layered (hidden-layer) Neural Networks for richer decision-making.
• Add GUI-based visualization of each agent’s network and fitness progress.
• Extend to more complex environments for broader generalization.