Robotic helicopter teaches itself how to fly aerobatics
programming these machines to perform complex aerobatics is a formidable challenge – unless of course they teach themselves. This example developed by Stanford computer scientists does just that, learning to fly by watching other RC helicopters in the air. Not only does this artificial intelligence system produce a spectacular flying exhibition, it’s seen as an important demonstration of robotic learning through observation.
As maneuvers are repeated several times, the trajectory of the helicopter inevitably varied slightly with each flight. But the learning algorithms created by Ng’s team were able to discern the ideal trajectory the pilot was seeking. Thus the autonomous helicopter learned to fly the routine better and more consistently than the experts themselves.
Well that would be one straight out of the “Holy Crap” department then – although I’m not entirely sure that I wasn’t just watching helicopter that wasn’t just completely insane and which was incapable of flying like a sensible grown up helicopter.
I think this approach will be how robots learn to move etc though – once the mechanics get past a certain level of complexity.