AI Drone Outmaneuvers World Champion Pilots in Physical Sport First
A University of Zurich AI system named Swift defeats world-champion drone racers using only onboard camera vision and sensor data.
Researchers from the University of Zurich and Intel create an AI system called Swift that dominates world-champion drone racers in head-to-head matches. Unlike previous AI victories in chess or Go that rely on strategic calculation, this achievement marks the first time artificial intelligence beats top humans in a real-world, physical sport. Swift competes directly against two world champions and a three-time Swiss national champion on a specialized course set up inside a Zurich aircraft hangar.
The autonomous drone uses the same single-camera vision setup as the human pilots to navigate the tight, technical 3D racetrack. It also benefits from real-time acceleration, speed, and orientation data provided by an onboard inertial measurement unit. This combination of visual and physical data allows the AI to execute incredibly precise and dynamic flights at high speeds that match the intense demands of professional first-person-view racing.
Swift not only defeats the human champions repeatedly but also sets the fastest overall race time on the 25 by 25-meter course. This breakthrough demonstrates that AI systems now possess the reaction times and physical control necessary to outperform humans in fast-paced, unpredictable real-world environments rather than just on a screen or a board.