AI-powered drone beats human champions in high-speed race

Developed by researchers at the Delft University of Technology's Micro Air Vehicles Lab in the Netherlands, MonoRace won the 2025 Abu Dhabi Autonomous Racing League competition. It posted faster times than three former world-champion human drone pilots and beat all rival autonomous teams.

The drone relies on a monocular camera and an inertial measurement unit to gauge acceleration and rotation. A neural network identifies racing gates and estimates the drone's position, sending motor commands to the four propellers 500 times per second. The system weighs 966 grams and measures roughly 30 x 28 x 10 centimeters, allowing the entire AI to run onboard.

Study co-author Guido de Croon told Live Science that the model was trained with reinforcement learning entirely in simulation. "It actually learns to fly from scratch," he said. Early training focused on avoiding crashes and reaching the first gate; as training progressed, the system learned to navigate longer course sections at higher speeds. The researchers varied drone parameters such as weight, voltage and propeller condition so the AI could handle real-world variations.

On the Abu Dhabi course, MonoRace completed two laps around 11 gates in 16.56 seconds, hitting a top speed of approximately 63 mph (102 km/h). Some video footage of the system's development includes computer-generated sequences, which de Croon said were intentional visual effects to illustrate the training process.

De Croon suggested MonoRace could become a training tool for human drone pilots, "a bit like powerful chess programs have also inspired humans playing the game." More broadly, he said the system demonstrates progress in physical AI — how a robot can perceive, decide and act at high speeds with limited sensing and computing power.

Analysis

Why This Matters

The achievement shows that AI can match or exceed human performance in complex, high-speed physical tasks where split-second decisions are critical.
The system's reliance on minimal hardware (single camera, small onboard computer) suggests that sophisticated autonomous flight could become cheaper and more accessible.
If the same approach is adapted for energy efficiency rather than speed, it could reduce power consumption in everyday robots.

Background

Drone racing is a competitive sport where pilots fly quadcopters through obstacle courses at high speed, often using first-person-view goggles. Autonomous drone racing has been an active research field for years, with teams attempting to close the gap between human and machine performance. Previous autonomous systems often relied on GPS or external motion-capture systems for precise positioning, which MonoRace avoids by using only onboard sensors.

Key Perspectives

Researchers (Delft University of Technology): The team views MonoRace as a proof-of-concept for robust, lightweight AI that can operate under real-world constraints. They emphasize that reinforcement learning in simulation, with varied drone parameters, allowed the system to generalize to different hardware. Human pilots and racing community: While the system beat former world champions, human pilots remain competitive in agility and adaptability across diverse environments. The result may push the sport toward hybrid human-AI formats. Critics/Skeptics: The training was done entirely in simulation, which could limit performance in unpredictable real-world conditions not captured in the simulator. The system also relies on clearly marked gates, raising questions about its ability to handle unstructured environments.

What to Watch

Future autonomous racing competitions and whether MonoRace's approach is adopted by other teams.
Efforts to transfer the neural control system to commercial drones for tasks like inspection or delivery.
Research into adapting the same reinforcement learning framework for energy-optimal control in other robotics domains.

Checked

3 claims in this story checked against 1 source: 3 supported.

Show each claim

SupportedAI-powered drone beats human champions in high-speed raceLatest from Live Science: “A drone powered by artificial intelligence (AI) has completed a championship race course faster than some of the world's best human drone pilots. ... posting faster times than three former human world-champion drone pilots.”
SupportedMonoRace system uses monocular vision and reinforcement learning to navigate courses at over 60 mphLatest from Live Science: “The system, called MonoRace, ... using little more than a single forward-facing camera ... created an AI system capable of pushing a quadcopter through a demanding course at over 60 mph (100 km/h) ... trained the MonoRace model with reinforcement learning”
SupportedA drone controlled by AI has outperformed some of the world's best human pilots in a championship race, according to study published Oct. 14 in Science RoboticsLatest from Live Science: “In a study published Oct. 14 in the journal Science Robotics, scientist created an AI system capable of pushing a quadcopter through a demanding course at over 60 mph ... faster than some of the world's best human drone pilots”