One camera, no GPS, 100 km/h: TU Delft explains its winning racing drone
Delft's MonoRace drone beat three former drone racing world champions in Abu Dhabi in 2025. Now the team describes in Science Robotics how it flies on its own.
Photo: Choinowski / Wikimedia Commons, CC BY-SA 4.0 (cropped and resized)
Video: YouTube · MAVLab TU Delft · youtube.com
Researchers at TU Delft have described in the journal Science Robotics (DOI 10.1126/scirobotics.aee9250, published 14 October) how their autonomous racing drone MonoRace works. The drone from the Micro Air Vehicle Lab (MAVLab) in the Faculty of Aerospace Engineering beat all other autonomous teams and three former human world champions at the 2025 Abu Dhabi Autonomous Drone Racing competition. A video on the MAVLab's YouTube channel shows the race.
For perception, navigation and control, MonoRace uses only a single forward-facing camera and a motion sensor (IMU) – without GPS or external position tracking. A neural network detects the racing gates in the camera image, while motion data and a model of the drone estimate position and speed. Instead of a conventional multi-layer control system, a small neural network generates motor commands directly, 500 times per second, computed on the onboard flight controller chip. The drone reached up to 100 km/h.
The lead author is Stavrow Bahnam, and the team includes Professor Guido de Croon. According to TU Delft, de Croon said autonomous racing is a way of pushing robotic AI to its limits. The researchers say the same principles could also apply to vehicles and mobile robots.