AI and the future of drones

Roboticists from Carnegie Mellon University published a paper earlier this year, entitled “Learn to Fly by Crashing.” The research explores the roboticists tests of an AR Drone 2.0 that they watched teach itself to navigate 20 different indoor environments through trial and errors. In just 40 hours of flying time, the drone mastered its aerial environment through 11,500 collisions and corrections.
“We build a drone whose sole purpose is to crash into objects,” the researchers wrote. “We use all this negative flying data in conjunction with positive data sampled from the same trajectories to learn a simple yet powerful policy for UAV navigation.”