Named Swift, the autonomous system surpassed world-champion human pilots in 15 out of 25 races held on a challenging track with intricate turns and manoeuvres devised by a skilled drone-racing expert. The system, appropriately named, merges AI-based learning algorithms with a solitary camera and onboard sensors that identify the drone’s environment and actions. Crafted by Elia Kaufmann, a robotics engineer at the University of Zurich, in collaboration with Intel Labs researchers, the goal was to create a solution that doesn’t necessitate inputs from external motion cameras, diverging from the approach taken by previous self-racing drones.
Reaching the level of professional pilots with an autonomous drone is challenging because the robot needs to fly at its physical limits while estimating its speed and location in the circuit exclusively from onboard sensors.
Elia Kaufmann and colleagues write in their paper
Drone-racing pilots use headsets that provide a “first person” perspective via a drone-mounted camera, capable of speeds up to 100 kilometres per hour. Similarly, Swift employs an onboard camera and an inertial sensor to gauge acceleration and rotation. These measurements are fed into two AI algorithms, which analyze the drone’s position relative to square gates on the obstacle course. Control commands are then generated accordingly.
Although it didn’t win 40 percent of the races, Swift outperformed each human pilot on multiple occasions and achieved the fastest recorded race time—half a second quicker than the top human time.
