SelfCar

SelfCar is a neural-network training simulation where cars learn to drive around a track as fast as possible.

Each generation creates multiple cars with slightly different neural networks. The best-performing car is selected, mutated, and used to create the next generation. Over time, the cars learn to navigate the track better.

Features

• 3 sensors: front, left, right
• Neural-network-based learning
• Multiple cars trained simultaneously
• Adjustable population and mutation strength
• Normal and difficult track modes
• Save and load trained models

Goal

Train the cars and see how high a score you can achieve. My best result so far is ~397 points.

Windows build available for download above.