Artificial intelligence beats world champions in drones
AI developed that can beat world leaders in drone racing
Scientists at the University of Zurich have developed Swift AI, an algorithm based on deep reinforcement learning designed for drones. In an impressive feat, their creation managed to win 15 out of 25 drone races against world champions.
The learning approach of Swift AI involves a trial-and-error process that led to the researchers crashing hundreds of drones during training. Yet, their dedication paid off, as the AI drone outperformed human-piloted drones on a 3D racetrack.
In these races, the drone navigates through a course filled with obstacles without making contact, or it risks crashing. Pilots control the drones’ flight using onboard cameras.
Swift AI proved to be not only more precise but also faster than human pilots, completing its swiftest lap in 17.47 seconds – 0.5 seconds quicker than its human counterparts. Nonetheless, in 40% of the races, the AI drone fell short due to its sensitivity to lighting conditions and other dynamic variables.
The developers of Swift AI herald this achievement as the first time an AI-powered robot has triumphed over a human champion in a real-world sport created by humans for humans.
