University of Zurich scientists created a navigation algorithm that allows autonomous drones to perform various tricks using onboard sensors. The neural network can train drones on acrobatic maneuvers via flight simulation software, New Atlas reports.
- After training on the simulator, a drone equipped with the algorithm can develop its own commands for barrel rolls, power loops, and other maneuvers.
- Davide Scaramuzza, a robotics professor at the University of Zurich, said such maneuvers "are challenging even for the best human pilots." Drones that can do these tricks could be more efficient, which translates to better battery life and farther distances, researchers said.
- Their paper, Deep Drone Acrobatics, was recently published on arXiv.