Neural Racer: cars that learn a track in the browser

Neural Racer is a personal project. Cars learn a track with neural networks and a genetic algorithm, implemented from scratch in the browser. It is not client work. It is a study in training a small network without a backend.

What it is

Each car is a small network. A genetic algorithm selects the ones that stay on the track longest, then the next generation starts from those weights.

The track, the physics, and the training loop all run in the page. There is no server-side model.

Why it is on the site

It sits with the other studies: public work you can open, not a case study about a client.

Open the page if you want to watch a generation learn a line, or to see how the controls are laid out.