Want to know how to teach robots to pilot themselves without help through the physical world without destroying themselves or anything else in the process?
You could launch a massive, decade-long effort to develop the artificial intelligence required to avoid obstacles and recognize the difference between between-the-lines and upside-down-in-a-ditch.
You could also build the world's largest swarm of miniature robotic submarines, teach them to boat slowly in formation and try to keep from bumping into each other, then send them out to explore underwater like a litter of Disneyfied underwater Roombas put-putting slowly off into adventure.
Or you could put a few massively overpowered quadcopters in a padded room and send them screaming around an 11-foot circle at, put them in a padded room and send them into what looks like the most insanely suicidal game of tetherball ever played by quadcopters screaming around an 11-foot circle at 30 miles per hour while tied to a post.
If you did, you might post the video, as evidence of what you and your engineering buddies would get up to in the robotics lab at school after a few beers on a Friday night.
You probably wouldn't submit it as a piece of serious and legitimate research into multi-unit cooperation, special awareness and obstacle avoidance by unmanned aerial vehicles to the respectable IEEE/RSJ International conference on Intelligent Robots and Systems, which will start Sept. 38 in Hamburg.
That is exactly what a team of robotics researchers at engineering powerhouse of Swiss universities – ETH Zurich – did with its video, which they could only video using high-speed motion-capture cameras set up to catch the motion of insanely overpowered flying bots in the ETH Zurich Flying Machine Arena.
The Arena was built with the help of a student who took a five-year sababatical to found and become rich from warehouse-automation-robotics company Kiva Systems.
The two-minute video shows a test of the learning algorithms built into the quadrocopters, which are trying to learn to fly in formation under challenging conditions, according to the explanation on a YouTube video posted by team member Maximilian Shulz. The drones are tethered to the pole because they spin so fast around it that they generate up to 13 g's of centripetal force, which should be enough to smash them to bits even on padded walls.
Once they learn a bit more, Shulz wrote, they should be able to coordinate so well with each other that they keep so steady a pull on the line connecting them that they can zip in circles at the same speed and in the same spot even if they were no longer tied to the post.
The degree of sensitivity and speed of adjustment needed to fly that suicide circle in place without a tether really would be more impressive than pokey little submarine-bots. It would mean serious control and a respectable level of autonomy. (It would also mean, it turns out, that they could dance; interactively with each other; to music.)
It would also mean the drones the FAA may soon approve for general commercial use rather than just in special cases, could be doing a lot more than just package delivery or casual surveillance or rural crop surveys or other services performed far away from concentrations of people.
Patent documents from Amazon show it plans to have drones that have real routes and are able to recognize and respond to other drones, the car or house of the person to whom they're delivering and will likely come in a variety of sizes, payload capacities and sensors. Some won't be allowed to descend to head height if people are around. Others might be able to navigate inside public buildings, with some restrictions.
The FAA is even considering allowing some drones to operate far enough away from their operators that the operators can no longer see them – a major no-no until now for a fliers whose onboard cameras give only a limited view of what is around them.
A little of the Flying Machine Arena magic would go a long way with the FAA, if it were possible to show that drones could consistently recognize and avoid crashing into anything on their own, but especially humans.
They might also be doing delivery duty inside the office buildings where we work, delivering lunch from drone-powered takeout places, running little errands for us, flying over to see who's at the door, or whether the commuter train is getting close. Within just a few months they could begin filling the air, in numbers that will eventually become high enough that, trouble avoidance or no, the wearable tech accessory of the future may be a helmet designed not to augment our vision of the real world or paint over it with a virtual one.
It may be a nice solid helmet designed to protect us from all the high-speed, autonomously piloted convenience.