An artificial intelligence (AI) software has defeated three world champions in drone racing, a “new milestone,” according to its developers. The software named Swift, built by researchers at Zurich University in Switzerland and Intel, won many races against human competitors, during tests last year. waterfordbanquet.com
According to researchers, while IBM’s Deep Blue defeated Gary Kasparov in a chess game in 1996 and Google’s AlphaGo defeated Go champion Lee Sedol in 2016, this is the first time an AI system has defeated human adversaries in physical activity such as drone racing.
Swift defeated 2019 Drone Racing League champion Alex Vanover, 2019 MultiGP Drone Racing champion Thomas Bitmatta, and three-time Swiss champion Marvin Schaepper in races held on a purpose-built track near Zurich in June 2022.
“This is a “new milestone” for AI
Swift, according to the Nature report, won numerous races against each of the human champions and had the fastest recorded race time, beating the best human performer by half a second.
Until recently, autonomous drones took twice as long to fly through a racetrack as those piloted by humans.
According to the researchers, this is a “new milestone” for AI because physical sports are more difficult for autonomous systems than board or computer games because they are less predictable.
“We don’t have a perfect knowledge of the drone and environment models, so the AI needs to learn them by interacting with the physical world,” said Davide Scaramuzza, head of the Robotics and Perception Group at the University of Zurich.
To win, the AI system had to handle quadcopter drones remotely and fly them at speeds surpassing 100 km/h, just like its human competitors.
Swift raced using real-time data obtained by an onboard camera
The lap was considered complete when racers crossed seven square gates in the correct order. They also had to perform difficult movements and acrobatic features.
Swift raced using real-time data obtained by an onboard camera similar to the one used by human racers.
According to the researchers, the system learned to fly by trial and error after a month of simulated flying time, which is equivalent to less than an hour on a desktop computer.
They claimed that training in a simulated environment averted the destruction of drones during the early stages of learning when the system frequently crashed.
“To make sure that the consequences of actions in the simulator were as close as possible to the ones in the real world, we designed a method to optimize the simulator with real data,” said first author of the paper, Elia Kaufmann.
Human pilots, according to studies, are still superior at responding to shifting conditions. AI failed when the conditions were different from what it had been taught for, even if it was just too much light in the room.
It would be less important in terms of the technology’s possible real-world applications, where fast-flying and hence battery-life-preserving drones might be advantageous in areas such as environmental monitoring or disaster response, according to the researchers.