U.S. scientists say they have created robotic helicopters that can teach themselves to fly by "watching" other helicopters.

Stanford University Professor Andrew Ng, who led the project, said the achievement is an important demonstration of "apprenticeship learning," in which robots use an artificial intelligence system to learn by observing an expert, rather than by having software engineers write instructions.

Stanford's artificial intelligence system learned how to fly by "watching" a four-foot-long helicopter flown by expert radio control pilot Garett Oku.

While it might seem an autonomous helicopter could fly stunts by simply replaying the exact finger movements of an expert pilot using joy sticks on a remote controller, Ng said that's not possible due to many uncontrollable variables such as gusting winds.

So the researchers had Oku and other pilots fly entire air show routines while every movement of the helicopter was recorded. As each maneuver was repeated, the helicopter's trajectory 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.

Ng said there is interest in using autonomous helicopters for such duties as searching for land mines or mapping wildfire hot spots in real time.