Very interesting concept.
I would like to know how the specific evolution algorithm works because, observing those bikes' behaviour, it doesn't look like there's an actual neural network behind but something more like a sequence of actions they learn to do.