On Mon, Nov 12, 2007 at 08:15:20PM +0000, Andy Farnell wrote:
Good lord Chris, those are very impressive. My first thought wasn't how cool they would look skinned in a game, but how powerful this is to the field of robotics. Build your real robots as models in this gravity/friction environment and use GAs to converge on behaviour (that you could never have time to train IRL) then pop the results back into your real bot.
There was a Robot Soccer exhibition not long ago at the local university here, and one of the exciting things I picked up from that was that the little robot dogs they used in the competition were often trained to walk in exactly the way you describe.
For a long time I've been meaning to code up a nice little 3d/physics/GA environment with PyODE and PyOpenGL to try those kind of experiments and replicate Karl Sims' work. Maybe one day I'll find the time!
If we wanted to connect this back to Pd, we could train 3d GAs to dance in time to Pd made music. :)
Best,
Chris.