UZH researchers have successfully designed an AI system, called ‘Swift’, that managed to beat human world champions in drone racing. More specifically, it won multiple races against three world-class champions in first-person view (FPV) drone racing. This sport typically has human pilots flying quadcopters at speeds exceeding 100 km/h, which are controlled remotely while wearing a headset linked to an onboard camera.
To train Swift, the researchers used a simulated environment where it learned to fly on its own by trial and error, using reinforcement learning. This helped the team by not destroying multiple drones in the early stages of learning when the system often makes errors. However, they did devise a method to optimize the simulator with real data to make it as realistic as possible. This allowed the drone to fly autonomously, thanks to very precise positions provided by an external position-tracking system, while also recording data from its camera. After only a month of training, the AI-powered drone was ready to take on human pilots.
- ๐๐ถ๐ด๐ต๐๐๐ฒ๐ถ๐ด๐ต๐ ๐ฎ๐ป๐ฑ ๐ฃ๐ผ๐ฟ๐๐ฎ๐ฏ๐น๐ฒ - The drone is lightweight and compact, weighing...
- ๐ก๐ผ ๐ฅ๐ฒ๐ด๐ถ๐๐๐ฟ๐ฎ๐๐ถ๐ผ๐ป ๐ก๐ฒ๐ฒ๐ฑ๐ฒ๐ฑ - Under 249 g, FAA Registration and Remote ID are not required if...
- ๐๐ ๐๐ฒ๐ป๐ฑ๐ฒ๐ฑ ๐๐น๐ถ๐ด๐ต๐ ๐ง๐ถ๐บ๐ฒ - Enjoy longer flights with DJI Mini 2 SE, which offers a 31-min max flight...
Physical sports are more challenging for AI because they are less predictable than board or video games. We donโt have a perfect knowledge of the drone and environment models, so the AI needs to learn them by interacting with the physical world,โ said Davide Scaramuzza, head of the Robotics and Perception Group at the University of Zurich.
