When a computer defeated the then world chess champion Gary Kasparov in 1996, that was already impressive, the researchers announced. Even more so when in 2016 a computer beat the world champion Lee Sedol in the much more complex game of Go. However, these successes all took place in a virtual environment. In a board game, everything is known exactly in advance, they said. "Our autonomous drone has managed to beat humans in the real world," said researcher Leonard Bauersfeld.
The drone competed against several drone pilots: Drone Racing League Champion 2019 Alex Vanover, MultiGP Drone Racing Champion 2019 Thomas Bitmatta and three-time Swiss champion Marvin Schaepper.
The human pilots were able to practice on the track for a week. Afterwards, they competed in head-to-head races against "Swift". The system won several races against each of the human champions (15 of the 25 races in total) and also achieved the fastest recorded race time on the track, half a second ahead of the best time recorded by a human pilot.
Rescue of buried victims
However, the races only served as a test environment for the drone. In the future, such AI-controlled drones could search for missing persons in collapsed buildings, for instance.
According to Bauersfeld, very similar skills are needed for these applications as for the drone races. In these, pilots steer through an obstacle course. The drones have to fly through various gates, for example. This ability to fly precisely through gates could help to fly through narrow passages in collapsed buildings, Bauersfeld said. Speed is also of great importance for such real-world applications because drones only have a very limited battery capacity. So the faster it flies, the further it gets.
Written by: sda
Photos: Leonard Bauersfeld, UZH