Palo Alto (CA) – The best pilot for a helicopter and potentially operate other systems such as cars and even airplanes may not be humans, but software. Stanford researchers said they have developed a system that can teach software how to fly a model helicopter through difficult stunts.

At least once on a typical week day, we come across a story that describes and idea or a technology that is simply fascinating. You know, an idea that prompts you to lean back in your desk chair for a moment and start imagining what this technology would be capable of to accomplish. If there was a daily visionary award I’d have to give this prize to a group of graduate students at Stanford University who came up with a (relatively) simple way to teach software how to do RC helicopter tricks during flights – eventually in a much more sophisticated way than its human teachers.

Broken down, Andrew Ng, the professor directing the research of graduate students Pieter Abbeel, Adam Coates, Timothy Hunter and Morgan Quigley enabled a computer to “watch” an expert fly an RC helicopter and run the model aircraft through and entire air show. The result was a textbook example of “apprenticeship learning” - an artificial intelligence system that replicated and eventually improved the actions of the pilot and the tricks performed. The stunts were "by far the most difficult aerobatic maneuvers flown by any computer controlled helicopter," said Andrew Ng.     

What makes this particular example of apprenticeship learning especially impressive example is the difficulty to fly helicopters and their nature to always tend to an unstable state. "The helicopter doesn't want to fly. It always wants to just tip over and crash," said Garrett Oku, the pilot. This simple fact would make it nearly impossible to program an autonomously flying helicopter by itself and take it through challenging flight routines.  
 
However, even an approach of simply replaying finger movements cannot really work, as the aircraft is continuously exposed to changing environment variables, such as wind gusts. When the Stanford researchers decided their autonomous helicopter should be capable of flying air show stunts, they realized that even defining their goal was difficult. What's the formal specification for "flying well?" The answer, it turned out, was that "flying well" is whatever an expert radio control pilot does at an air show.
 
The researchers found that they needed to come up with algorithms that not only replicated movements of a pilot, but detect the ideal trajectory the pilot was seeking. As a result, the autonomous helicopter learned to fly the routine better—and more consistently—than Oku, the researchers said. A demonstration of the flight is posted on Stanford’s website.

As any autonomous device, the robots were equipped with a boat load of technology, including accelerometers, gyroscopes and magnetometers, the latter of which use the Earth's magnetic field to figure out which way the helicopter is pointed. The exact location of the craft is tracked either by a GPS receiver on the helicopter or by cameras on the ground. The group noted that the entire navigation package could be packed into a larger helicopter.

So, what could you do with software that was taught by the best pilots and learn over time to fly better than their teachers and master challenges in a much more secure way? The researchers pointed to autonomous helicopters to search for land mines in war-torn areas or to map out the hot spots of California wildfires in real time, allowing firefighters to quickly move toward or away from them. "In order for us to trust helicopters in these sort of mission-critical applications, it's important that we have very robust, very reliable helicopter controllers that can fly maybe as well as the best human pilots in the world can," Ng said. Stanford's autonomous helicopters have taken a large step in that direction, he said.

There no mention of larger aircraft - potentially carrying humans – that could use such an approach. Will we be flying one day in air planes that have no pilots at all? I will leave that up to you decide and I will let you answer the question if that is a scneario we should feel comfortable with. But I there is no doubt that teh Stnaford reserachers have found an interesting way to teach computers difficult tasks. Tasks they can't do on their own - yet.