Love the article, I'm jealous of the author's ability to visualize this. I played with this a while ago using small neural networks and state preference to do something similar, the NNs became self tuning PID controllers and the bots could fly around the simulation like hummingbirds. Was pretty neat.