A new algorithm flies drones faster than world-class human racing pilots.

Drones must be fast to be helpful. Due to their short battery life, drones must be able to complete any task in a reasonable time. This includes inspecting buildings, searching for survivors at disaster sites, and delivering cargo. They may need to go through several waypoints, such as windows, rooms, or specific locations, to ensure they follow the correct trajectory and accelerate or decelerate at each segment.

The algorithm is superior to professional pilots.

Human drone pilots are highly skilled at this task and have consistently outperformed autonomous drone racing systems. A research group at UZH has developed an algorithm to determine the fastest path for a quadrotor, which is a quadrotor with four propellers. It guides the quadrotor through a series of waypoints along a circuit. Davide Scaramuzza is the head of UZH’s Robotics and Perception Group and the Rescue Robotics Grand challenge of the NCCR Robotics. He said the drone beat two top-level human pilots in the fastest lap on an experimental race track.

Scaramuzza says that the algorithm is unique in that it generates time-optimal trajectories that fully consider the drones’ limitations. Previous works relied upon simplifications of either quadrotor systems or flight path descriptions, which could have been more suboptimal. Philipp Foehn (Ph.D. student, first author) says that the algorithm tells the drone where to go rather than assigning specific flight route sections to waypoints.

External cameras provide position information in real-time.

Researchers had two human pilots, and the algorithm flew the quadrotor through a race circuit. External cameras were used to capture the precise motions of the drones and, in the case of the autonomous drone, to provide real-time information to the algorithm about where it was at any given moment. The human pilots had the chance to practice on the circuit before the race for a fair comparison. The algorithm won. All its laps were faster and more consistent than human pilots. It is unsurprising that once it has determined the best trajectory, it can repeat it many times without fail, unlike human pilots.

The algorithm must be less computationally intensive before commercial applications. It currently takes the computer up to an hour to calculate the optimal time-optimal trajectory of the drone. External cameras are used to determine where the drone is at any given moment. Scientists hope to make use of onboard cameras in future research. However, demonstrating that an autonomous drone can fly faster than human pilots is encouraging. Scaramuzza says this algorithm could be used in many areas, including package delivery with drones and inspection, search, rescue, etc.