Stanfords robotic helicopters teach each other tricks, fly better than a human


A team of Stanford University computer whizzes have found out how to get a troupe of autonomous helicopters to fly better than the expert RC stunt pilot who "trained" them. At the heart of the system is acomputer that quickly analyzes data fed to it from instruments both on the 'copters and on the ground, and sends 20 instructions every second to the units in flight. Previous methods saw autonomous helicopters trying to mimic the instructions an RC pilot would send, but, because of the fickle nature of the wind, this resulted in stiff flying at best, and a crash at worst. By analyzing the way a human operator reacts to his helicopter to keep it aloft during precise tricks, Stanford's 'copters can "watch and learn," and respond to changes in the air more fluidly.

Crazily enough, the helicopters used aren't fancy at all. They're just store-bought RC helicopters, with the complex innards added by the Stanford students. The team includes Professor Andrew Ng, graduate students Pieter Abbeel, Adam Coates, Timothy Hunter, Morgan Quigley, and expert remote controller Garett Oku.

Via Stanford News Service