Races on drones, small unmanned aerial vehicles, are a relatively new type of fast, dynamic and spectacular sport. In all competitions already held on such races, including the last year’s World Championships in Dubai, people acted as drone pilots, but this situation will not last long. Soon people will have a serious competitor in the form of specialized artificial intelligence systems.
For several years, specialists from the NASA Jet Propulsion Laboratory (JPL) have been working on the creation of the above-mentioned artificial intelligence system. This work is partly funded by Google, and its ultimate goal is to create a visual navigation system that uses data from cameras and other sensors that can effectively control both the spacecraft and the robot car.
The culmination of this work was a kind of race on drones, held on October 12 this year. On the “machines” side were three unmanned aerial vehicles (Batman, Joker and Nightwing) operating under the control of an artificial intelligence system developed by JPL specialists. A people’s honor was defended by a young man named Ken Loo (Ken Loo), who is a pilot of world-class drones.
Custom-built drones can easily speed up to 130 kilometers per hour, but on a challenging obstacle course built specifically for racing, they flew at speeds of 50 to 65 kilometers per hour.
“We compelled our algorithms to compete, which operate according to strict calculations, with a person who acts on an intuitive level and refined reflexes,” says Rob Reid, head of the JPL research group. “As you can see, artificial intelligence provides a smooth and uniform control of the flight, while it is common for a person to accelerate, slow down and make other sharp movements and maneuvers. ”
Compared to the drones operated by Ken Lu, drones under the control of artificial intelligence moved somewhat more slowly, but more steadily and purposefully. In addition to the higher flight speed, Ken Lu demonstrated a number of sophisticated maneuvers bordering aerobatics. But, in the end, the man overcame fatigue, and in time he could not control the drone with the same efficiency. “This track is the most difficult route I’ve ever flown,” says Ken Lu. “Because of the constant strain, I’m tired and lost my concentration in ten laps.”
Nevertheless, at the beginning of the race, artificial intelligence and man were packed around the same time. But after a few dozen laps, Ken Lu studied the track and got used to it, after which he began to manage the drone more deftly, showing creativity where necessary. The official time of passing one lap by Ken Lu was 11.1 seconds, while the artificial intelligence coped with this task in 13.9 seconds.
Nevertheless, the artificial intelligence system demonstrated high stability, while the time of one lap passed by Ken Lu could be quite different from the time of passing the previous circle.
“The race showed that our artificial intelligence systems do not lag far behind the man, but they have great potential for further improvement,” says Rob Reed. “And in a short time, I hope, artificial intelligence will be able to catch up and outrun the most the best of professional people. ”