An FPV drone capable of flying at nearly 100 km/h has beaten three former world champions, without a pilot at the controls.
The event took place in 2025 in Abu Dhabi, during a competition for autonomous racing FPV drones. On a course filled with gates to pass through, the machines had to accelerate, turn and correct their trajectory in a fraction of a second. Unlike conventional drones, no one guided them remotely.
(A) Close-up of the supplied competition drone.
(B) Robot performing an S-shaped maneuver.
(C) Overview of the course. The races consisted of two laps and 11 starting gates.
The winner came from Delft University of Technology in the Netherlands. Named MonoRace, its flight-control system uses artificial intelligence to make decisions on its own. The team first defeated the other autonomous drones, then three former human world champions in knockout races.
What makes the performance remarkable is the simplicity of the onboard hardware. The drone has only a camera, a sensor that measures its movements and a small computer. It uses neither GPS for navigation nor cameras installed around the track to track its position.
Its camera observes the course gates. The computer uses their images and the sensor measurements to estimate where the aircraft is and how it is moving. It can then determine the necessary correction before passing through the next opening.
Instead of calculating a trajectory and then sending commands to several conventional control systems, a small neural network directly adjusts the motors' power. It does so 500 times per second, fast enough to react at high speed.
This artificial intelligence must nevertheless deal with imperfect images and sensors that are sometimes disrupted by acceleration. The researchers therefore designed a method to correct the drone's position estimate using data recorded during its flights. This helps it remain reliable without external equipment.
Although the victory dates back to 2025, the team published details of how the system works in Science Robotics in October 2026. The tests show that a relatively lightweight AI can pilot a very fast aircraft in a competition environment it does not control.
The goal is not simply to win races. The same principles could help other machines make rapid decisions with few sensors and limited computing power, such as mobile robots or devices operating in places where GPS performs poorly.