Autonomous quadrotor beats two human pilots in a drone race

A new type of algorithm was faster and more consistent than humans.

Robotics and Perception Group, UZH

Unlike self-driving cars, autonomous drones can generally get from "A" to "B" safely, but could they beat a human pilot in a drone race? So far the answer has been "no way," but now, researchers from the University of Zurich (UZH) have created an algorithm that allowed an AI-powered drone to beat two human pilots on an experimental race track. The work could lead to more efficient drones for rescues, deliveries and other chores.

In the past, researchers have built simplified models of quadrotor systems or flight paths to calculate the optimum trajectory. This time, however, they fully accounted for the drone's limitations. "The key idea is, rather than assigning sections of the flight path to specific waypoints, that our algorithm just tells the drone to pass through all waypoints, but not how or when to do that," said PhD student and author Philipp Foehn.

For the AI versus human race, researchers let the human pilots train on the circuit so the comparison would be fair. They set up external cameras to send the drone's exact position to the algorithm in real time. Once both humans and AI were trained, the algorithm beat the humans on every lap and had more consistent performance to boot.

The research could lead to quicker drones for real-world applications, even in complex environments with multiple waypoints. The next step is to make the system less computationally demanding and allow it work with onboard, rather than external cameras. "This algorithm can have huge applications in package delivery with drones, inspection, search and rescue, and more," said UZH's head of Robotics and Perception Group, Davide Scaramuzza.