Bold flight maneuvers such as looping or barrel roll are something for experts: An aerobatic pilot must have a lot of experience to keep his aircraft safely in the air in these extreme situations. Drones can do this, even without a human pilot. Researchers at the University of Zurich have developed a navigation algorithm with which they can train drones for daredevil flight maneuvers. Using only on-board sensors and calculations, the drone flies the flight maneuvers independently, pushing itself to its physical limits.
A few hours of simulator training are sufficient for the drone
An artificial neural network converts inputs into control commands delivered by the on-board cameras and inertial sensors. After just a few hours of simulator training, the drone is ready for operation; test runs are not necessary, according to the research team. There would also be no risk to the drone.
Nevertheless, humans are often the better drone pilot
In certain situations, however, the researchers concede that humans are the better pilot. They could react more quickly to external influences. Nevertheless, the algorithm is an important step for the further integration of drones into everyday life. If drones can automatically make full use of their agility, they can become more efficient. This would benefit drone operations – for example in rescue or search missions or in logistics.
Find out more:
Deep Drone Acrobatics – University of Zurich