The journal says the onboard framework enables aerial robots to race autonomously at speeds of up to 100 km/h.
Science Robotics announced its October 2026 issue with a cover depicting a drone deployed with MonoRace. The journal describes MonoRace as an onboard framework that enables aerial robots to race autonomously at speeds of up to 100 km/h.
8.4K
3 Sources, first seen ago
A paper published in Science Robotics details how Delft University of Technology researchers built an autonomous racing drone that reached 100 km/h using just one forward-facing camera, an inertial measurement unit and a small onboard processor. The journal highlighted the system, called MonoRace, on its October 2026 cover.
High-speed flight leaves little room for sensing or control delays. MonoRace uses a neural network to identify racing gates in images from a monocular rolling-shutter camera. It combines those detections with IMU readings and a model of the drone to estimate position and motion without GPS or an external motion-tracking system.
The researchers' preprint says they also developed an offline calibration step that uses known gate geometry and data recorded aboard the drone. This helped refine the camera parameters that are especially important when images are blurred or distorted at speed.
A compact neural controller then sends commands directly to the motors at 500 Hz, bypassing several intermediate control loops. According to the research team, the controller was trained with reinforcement learning so it could respond quickly while accounting for the drone's physical limits.
MonoRace was tested at the 2025 Abu Dhabi Autonomous Drone Racing Competition, where organizers controlled the hardware and track rather than the Delft team. In a direct knockout tournament, it beat every competing autonomous team and three former world champion human pilots.
The system also kept flying through problems including camera interference and IMU saturation. Those results matter because the drone could not rely on the external position measurements often available in laboratory demonstrations.
The team argues that the same approach could eventually help other robots make fast decisions with limited onboard resources. That is a prospective application, not a result demonstrated by the racing trial: the published work tested MonoRace on a drone course, with its controller optimized for speed.
@SciRoboticsThe October 2026 issue of Science #Robotics is out! This month’s cover depicts a drone deployed with MonoRace, an onboard framework that enables aerial robots to race autonomously at speeds of up to 100 km/h. Learn about this research and more: https://scim.ag/4i2T4iM@SciRoboticsThe October 2026 issue of Science #Robotics is out! This month’s cover depicts a drone deployed with MonoRace, an onboard framework that enables aerial robots to race autonomously at speeds of up to 100 km/h. Learn about this research and more: https://scim.ag/4i2T4iM@SciRoboticsThe October 2026 issue of Science #Robotics is out! This month’s cover depicts a drone deployed with MonoRace, an onboard framework that enables aerial robots to race autonomously at speeds of up to 100 km/h. Learn about this research and more: https://scim.ag/4i2T4iM@SciRoboticsThe October 2026 issue of Science #Robotics is out! This month’s cover depicts a drone deployed with MonoRace, an onboard framework that enables aerial robots to race autonomously at speeds of up to 100 km/h. Learn about this research and more: https://scim.ag/4i2T4iM—
Not ranked yet
—
Not ranked yet
