What am I looking at?

A population of cars, each driven by its own tiny neural network, trying to navigate traffic — no ML libraries, just linear algebra from scratch.

On the left canvas, a fleet of semi-transparent cars starts from the same position. Each one is equipped with a fan of sensor rays that detect the road edges and other vehicles ahead; those sensor readings are fed into a small feed-forward network whose outputs decide forward, reverse, left and right.

At each frame the simulation highlights the best car — the one that has made it furthest up the road — and the view follows it. Faded cars behind are the rest of the population still sharing the road.

The right canvas visualises the best car's brain: each column is a layer, each node is a neuron, and the lines between them are weighted connections. The dashed “marching ants” along the connections make positive and negative weights easy to spot.

Every car is born with a slightly randomised brain, so each run plays out differently. Click the road canvas (or hit Reset) to spin up a fresh batch.