Obstacle detection and avoidance is one area in which drones fail utterly, which is too bad because human pilots just aren’t that good at flying them. The obvious solution? Teach a drone to think like fencer.
Ross Allen and Marco Pavone, researchers at Stanford University’s Department of Aeronautics and Astronautics, have been trying to perfect a drone obstacle avoidance system by training it with the notoriously quick point of a fencer’s sword. Besides simply using a system that spots the incoming foil, the researchers are also incorporating some elements of machine learning, so that the drones of the future will be even more idiot-proof.
Currently the only data they’re feeding this drone comes from an external camera, which feels a bit like cheating, but similar systems being developed by big manufacturers like DJI are actually built into the flying crafts themselves. May the drone in your yard one day be as nimble as the drone in this video.
We only ask Allen and Pavone not to actually give the drone a sword. It ends badly.
[PopSci]