Well, another win for the machine. After successfully beating humans in chess, Go, and a variety of video games, AI has now surpassed humanity’s best drone racers. Three world-class drone pilots recently lost in a competition for an autonomous, artificial intelligence-powered drone called Swift. The AI-equipped drone, developed by researchers at the University of Zurich, topped 15 of 25 races and recorded the fastest lap time, according to a paper published Wednesday in Nature.
“This work represents a milestone for mobile robotics and machine intelligence, which may inspire the deployment of hybrid learning-based solutions in other physical systems,” the researchers wrote.
Swift Beats Humans in First-Person View Drone Racing Human competitors use a headset connected to a camera on their drone to navigate and pilot a quadcopter through complex obstacle courses at high speed, with the goal of completing the race in the fastest time and avoiding too much damage in the process. Drones in these races can go up to 50 miles per hour while really buzzing. The video below shows the Swift taking on human-controlled drones.
Swift won 15 of the 25 races against human pilots and clocked the fastest overall lap time of 17.47 seconds. That fastest lap time was half a second better than the human best. Three human competitors, Alex Vanover, Thomas Bitmata and Marvin Scheper, have each previously won the drone racing championship.
In this case, the human competitors had one week to learn the new course and train for the race. During the same period, Swift was also undergoing training but in a digitally simulated environment as per the course. According to the paper, Swift used deep reinforcement learning while in the simulation with additional data collected from the outside world. If deep reinforcement learning sounds familiar, that’s because it’s the same training method that previous AI models used to outperform humans in chess, Go, and even StarCraft.
Drone racing differs from previous examples, as it involves a physical object moving through the real world, unconstrained by the set limitations of a board game or digital world. During an actual race, Swift will take video collected by its camera and send it to a neural network capable of identifying gates. A combination of onboard sensors are then used to assist the drone with positioning, speed and orientation. All this happened autonomously, very quickly.
Swift noticed some interesting differences in the way the researchers approached the course as opposed to their human counterparts. The autonomous system, he noted, was more consistent across laps and appeared to take tighter turns. Adding those tight turns can give drones an edge in the race and often shave fractions of a second off lap times.
Swift’s performance shocked the human champions.
“He was crazy,” Bitmata is heard shouting during one of his races against Swift. “The possibilities are endless, this is just the beginning of something that could change the whole world,” Bitmata added in the research paper.
Others, however, like Schaepper, seem left with a sense of disappointment.
“Racing against a machine feels different, because you know the machine doesn’t get tired,” he said.