🚗 Driving AI with NEAT (NeuroEvolution)

This project showcases an AI that learns to drive a car in a 2D environment using the NEAT algorithm (NeuroEvolution of Augmenting Topologies). No hardcoded pathfinding — the agent evolves behaviors through generations, mutation, and selection. 🧬

🧠 What It Does

🎮 The AI controls a car in a simple 2D driving simulation with checkpoints built with Pygame.

🧬 It learns using NEAT: networks are evaluated based on how far and how safely they drive.

👁️ A visual interface displays the car in action as it learns and improves over generations.

🚀 Features

🔄 No supervised learning – only evolution by fitness

🧠 Networks evolve topologies and weights

📊 Real-time simulation with visualization

🏆 Tracks best fitness, average scores, and generation progress

Here is an image of what it looks like :

There is sevral cars for a generation and we select the best. (Grays had hit a wall, reds are normal and the green is the best of this generation)

📦 Dependencies

• Python 3.x 🐍
• any lib : this time, I don't use neat-python for neuroevolution
• Pygame for visualization 🎮

📝 Notes & Observations

⏳ Like any evolutionary approach, early generations perform terribly — driving in circles or crashing instantly — but over time, the network learns basic control and navigation.

Here, we can see that over 13 generations, the best path have been found (in less than 5 min). So, it is really quick !