A navigation algorithm enables drones to learn challenging acrobatic maneuvers

Performing acrobatic maneuvers with quadrotors is extremely challenging. Acrobatic flight requires high thrust and extreme angular accelerations that push the platform to its physical limits. Human drone pilots require many years of practice to safely master maneuvers such as power loops and barrel rolls.

Now, the robotics and Perception Group laboratory – part of a joint institute of the University of Zurich and the famous ETH Zurich – aims to offer drones capable of performing challenging acrobatic maneuvers independently, without the need for a pilot. A team of researchers from the University of Zurich and the ETH Zurich, together with the microprocessor company Intel, has developed a navigation algorithm that enables drones to autonomously perform various maneuvers – using nothing more than onboard sensor measurements.

The drones capable of performing such maneuvers are likely to be much more efficient. It can be pushed to its physical limits, make full use of its agility and speed, and cover more distance within its battery life, which benefits conventional search and rescue operations.

In order to prove the efficiency of their algorithm, the researchers flew maneuvers such as a power loop, a barrel roll, or a matty flip, in which the aircraft is subjected to very high thrust and extreme angular acceleration.

Only a few hours of simulation training is enough, and the quadrocopter is ready for use, without requiring additional fine-tuning using real data. Deep Drone Acrobatics rely only on the edge sensors and the front camera – just like a human pilot. The algorithm uses the abstraction of the sensory input from the simulations and transfers it to the physical world.

The core of the new algorithm is an artificial neural network that coverts the inputs supplied by the onboard camera and the inertial sensors directly into control commands. This neural network is trained exclusively by simulating acrobatic maneuvers. The training can be easily scaled and is completely risk-free for the quadrocopter.

This navigation is another step towards integrating autonomous drones in our daily lives,” says Davide Scaramuzza, robotics professor and head of the robotics and perception group at the University of Zurich. “Our algorithm learns how to perform acrobatic maneuvers that are challenging even for the best human pilots.”

However, the researchers acknowledge that human pilots are still better than autonomous drones. “Human pilots can quickly process unexpected situations and changes in the surroundings, and are faster to adjust,” says Scaramuzza. Nevertheless, he is convinced that drones in search and rescue missions or in delivery services benefit from being able to cover long distances quickly and efficiently.