An autonomous daredevil pushes the limits of flight

Why researchers think teaching a drone to fly on the edge of recklessness is a good idea.

drone-acro.jpg

How do you test your mettle if you're a fighter pilot? You pull some gnarly acrobatics that push your hardware (and body) to the limit, such as the Matty flip or the power loop.

But it's not all about ego. Aerospace engineers have long relied on test pilots to push the limits of manned aircraft in order to learn crucial lessons about aerodynamics, thrust, and the complicated materials science behind modern planes.

Drones, especially those designed for difficult environments or severe weather conditions, could benefit from the same kind of acrobatic proofing. To that end, researchers from the University of Zurich (part of the NCCR Robotics consortium) and Intel have developed a quadcopter that can perform incredible aerial acrobatics autonomously. The lessons learned from the autonomous drone may help pave the way for drones that can more fully exploit their agility and speed while maximizing flight efficiency and battery life.

The drone is powered by an algorithm designed to use on-board sensing to determine the physical limits of flight for the drone -- and then to push the drone right up to that line. The resulting hardware can perform stunts reserved for only the most skillful human pilots, like the barrel roll or Matty flip, which pushes the quadcopter to 3Gs of acceleration.

"Several applications of drones, such as search-and-rescue or delivery, will strongly benefit from faster drones, which can cover large distances in limited time. With this algorithm we have taken a step forward towards integrating autonomously navigating drones into our everyday life", says Davide Scaramuzza, Professor and Director of the Robotics and Perception Group at the University of Zurich, and head of the Rescue Robotics Grand Challenge for NCCR Robotics.

The innovation here is autonomy. The algorithm learns in just a few hours of simulation what the drone's limits are, then directs the hardware to fly acrobatic maneuvers accounting for various conditions. An artificial neural network directly converts observations from the on-board camera and inertial sensors to control commands. 

The system comes close to approximating the flying skill of the best human drone pilots, although a gap still remains.

"The best human pilots still have an edge over autonomous drones given their ability to quickly interpret and adapt to unexpected situations and changes in the environment," says Prof. Scaramuzza.