AI-Powered Drone Takes on Champions in High-Speed Racing Battle
In a groundbreaking development, researchers at the University of Zurich in Switzerland have created an AI-powered drone that defeated professional human drone racers. The quadcopter, named Swift, outperformed some of the best drone pilots in the world, winning 15 out of 25 races.
Equipped with first-person-view headsets connected to a camera on the drone, the champion racers struggled to keep up with Swift’s speed and agility. The AI-powered machine reached speeds of up to 62 mph (100 kph) while flawlessly navigating through obstacles, securing victory after victory.
What sets Swift apart is its ability to rely solely on an onboard computer, a single camera, and an inertial sensor. It doesn’t rely on an external position-tracking system, making it a unique contender in the world of autonomous drones.
Davide Scaramuzza, the head of the Robotics and Perception Group at the University of Zurich, explains that physical sports pose a greater challenge for AI because they are less predictable than board or video games. The drone and environment models are imperfectly known, requiring the AI to learn and adapt through interaction with the physical world.
Traditionally, autonomous drones took twice as long as human-piloted drones to navigate a racetrack unless they were connected to an external position-tracking system. However, Swift’s real-time reaction capabilities, driven by deep neural networks and an artificial neural network, based on data collected by an onboard camera, enabled it to perform at the highest level.
To train Swift, the researchers initially used a simulated environment to avoid damage to numerous drones. Once the AI software was sufficiently developed, it underwent refinement through real flights using an actual drone.
Swift’s impressive performance came during races against notable champions, including Alex Vanover, the 2019 Drone Racing League champion, Thomas Bitmatta, the 2019 MultiGP Drone Racing champion, and three-times Swiss champion Marvin Schaepper.
The racetrack, measuring 25 by 25 meters, featured seven square gates that had to be crossed in the correct order to complete a lap. The competition involved challenging maneuvers like the Split-S, a move that demands the drone to execute a half-loop at full speed while half-rolling.
While Swift achieved the fastest lap time, it faced challenges when conditions deviated from the ones it was trained on. Brighter lights on the track affected its performance, highlighting the adaptability of human pilots compared to the autonomous machine.
The research conducted by the University of Zurich team not only aims to showcase the technological advancements in drone racing but also has several practical applications. With limited flight time due to battery constraints, autonomous drones are more efficient when flying at higher speeds, making them valuable for search and rescue missions, forest monitoring, and even space exploration.
Swift’s triumph in drone racing marks a significant milestone in artificial intelligence. While AI has proven its capabilities in games like chess and Go, this is the first time it has dominated a physical sport. The relentless pursuit of innovation by researchers like those at the University of Zurich is propelling AI into new and uncharted territories.
As the world witnesses the extraordinary feats accomplished by Swift, it is clear that artificial intelligence has the potential to revolutionize various industries, opening up new possibilities and posing both exciting opportunities and challenges. The fusion of human skill and machine intelligence is redefining the limits of what is achievable, promising a future where humans and AI work together seamlessly for progress and innovation.