Skip to content

Show HN: Train racecar neural networks in the browser

zbendefy.github.io
3 pointszbendefy1 comment
On HN

This is small site where racecars learn to navigate a racetrack.

Each racecar has a randomly initialized neural network, receiving input data from distance sensors. In each round the best performing racecars are selected to generate the next round. Cars hitting the wall are instantly disqualified.

It takes a while to train them (I think they could do with fewer neurons), I included a pre-trained network for you to try.

Comments

The learning speed also shows why evolution based learning is not so great compared to backpropagation.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.