Researchers at Stanford University want to develop a more agile drone—one that’s able to avoid crashes and gracefully dodge moving hurdles. In order to do so, they developed an obstacle avoidance system that uses a drone-mounted camera to feed information to a computer, which processes information in real time. And to teach their drone agility and gauge its reaction time, they trained it against a fencer.
Popular Science explains that the drone, created by researchers Ross Allen and Marco Pavone of Stanford University’s Department of Aeronautics and Astronautics, uses machine learning to improve its abilities. Researchers began fencing with the drone to measure its agility and fine-tune its obstacle-avoidance system. They published their research in a paper [PDF] called “A Real-Time Framework for Kinodynamic Planning with Application to Quadrotor Obstacle Avoidance.”
In the video above, watch as the drone gracefully evades the point of a fencer’s sword, swooping and swerving as the fencer advances. While its reaction time is impressively fast, researchers said they're already gearing up for improvements, including “a range of visual, laser, and ultrasonic sensors.” They made no mention, however, of equipping the drone with a sword so it can fence back.
[h/t Popular Science]
Banner Image Credit: Ross Allen, YouTube