Artificial intelligence beats world champions in FPV drone piloting for the first time

Swift is an artificial intelligence algorithm that could beat the champions in FPV drone piloting. It was developed by engineers at the University of Zurich.

Here's What We Know

The algorithm is designed to control drones. In 15 out of 25 races, the artificial intelligence was able to beat the world champions in FPV drone racing and set the best lap time.

Swift competed against three champions in this discipline, Alex Vanover, Thomas Bitmatta and Marvin Schäpper. The algorithm was trained in a simulation in which a race track was recreated.

A method based on trial and error was used to train the artificial intelligence. This is called deep reinforcement learning. During the race, the neural network processed data from the camera to detect the gate through which the UAV was to fly.

In parallel, a second neural network collected information from an inertia sensor about the UAV's speed and orientation in space. Based on this data, the artificial intelligence gave commands to the FPV UAV.

As we noted above, Swift AI won 15 out of 25 victories, i.e. live humans turned out to be in 40% of races stronger than the algorithm. At the same time, the artificial intelligence was always faster at the start and took the steeper turns.

Engineers at the University of Zurich believe that their algorithm can be applied in practice. For example, in the inspection of buildings and ships, as well as in search and rescue operations. In addition, the technology should interest the military.

Source: Ars Technica