University of Zürich researchers have created a new algorithm for handling autonomous flying for a quadrotor drone. The new algorithm was tested against a pair of human drone racing pilots and outperformed them in a race. The new algorithm calculates time-optimal trajectories that fully considers the drone’s limitations.
Drones have many potential uses in the real world, including searching for people in natural disasters and other situations. Drones have to be fast because they have limited battery life giving limited time to complete whatever task they have been assigned. Needing to be fast means they have to complete the task as quickly as possible, and many do traverse several waypoints, including windows, rooms, or specific locations for inspection. The new algorithm allows the drone to adopt the best trajectory and the right acceleration or deceleration to reach each segment.
Typically, the best human drone pilots can outperform autonomous systems and drone racing, but that’s not the case anymore. The algorithm was able to guide the drone through a series of waypoints on the circuit. The algorithm beat the fastest lap of two world-class human pilots on an experimental racetrack. The algorithm is the first to generate time-optimal trajectories that fully consider the limitations of the drone. The idea is rather than assigning sections of the flight path to specific waypoints, the algorithm tells the drone to pass through all waypoints but not how or when to do that.
For the research, the team had the algorithm and two human pilots fly the same quad rotor around the race circuit. External cameras were used to capture the motion of the drones and for the autonomous drones to provide real-time information to the algorithm on where the drone was at any moment. Human pilots were allowed to train on the circuit before the race.
Despite the preparation, the algorithm is faster, with all of its laps recording faster times than the humans were capable of, and its performance was more consistent. Once the algorithm finds the best trajectory, it was able to reproduce that trajectory many times while humans have a harder time being consistent. The team of researchers says their algorithm could have applications for package delivery, inspection, searching rescue, and more.