See how the algorithm controls a swarm of drones
(ORDO NEWS) — Most autonomous drones are capable of taking off, landing and flying in a straight line, but scientists at the University of Zurich have developed a navigation algorithm that allows drones to perform spectacular aerial stunts.
YouTube is replete with videos in which operators force drones to make spectacular flights and perform aerobatics.
They require not only a certain level of skill in controlling the copter, but also fine-tuning both the software and the device itself. However, such experiments are excellent evidence of how great the potential of civilian drones really is.
Taking the same approach to an autonomous control system, a team at the University of Zurich has developed an artificial neural network that can be trained in acrobatic maneuvers using flight simulation software.
This approach allowed the researchers to easily simulate various flight paths and acrobatics, without the need for physical demonstrations that could damage the aircraft.
In just a few hours of training on the simulator, a quadcopter equipped with a navigation algorithm learns to use the built-in camera and sensors to come up with its own control commands for various types of aerial “acrobatics”.
These include roll, dead loop and other maneuvers that require high thrust and angular acceleration of the vehicle.
“Our algorithm can perform acrobatic maneuvers that challenge even the best human pilots,” says David Scaramuzza, professor of robotics at the University of Zurich.
While the spectacle is indeed impressive, the researchers hope the algorithm has more to offer than just a new type of autonomous airshow.
If you increase the battery life in unmanned aerial vehicles, they will be able to fly much longer distances, which means the algorithm can also carry out applied tasks, such as search and rescue operations, as well as the delivery of various cargoes by air.
You can learn more about the algorithm and see the drones in action in the video below. The material describing the study was published in the journal Robotics: Science and Systems :
Contact us: [email protected]