Also, IIRC early neural net research covered this approach with Hannibal & Attila decades ago. Edit: I guess there are some differences -- I think this is what

Also, IIRC early neural net research covered this approach with Hannibal & Attila decades ago. Edit: I guess there are some differences -- I think this is what I was remembering: http://people.csail.mit.edu/brooks/papers/AIM-1091.pdf http://people.csail.mit.edu/brooks/papers/AIM-1091.pdf ...but maybe there was further work that is more closely related.