An MIT researcher says he’s developed a system for flying drones safely around areas thick with trees, and to do so at speeds of up to 30 miles an hour.
The development is key, given that most commercial drones are incapable of safely navigating most obstacles. Stories of drones crashing into trees were everywhere last holiday season, as many inexperienced pilots quickly discovered their news toys stuck high in branches.
“Everyone is building drones these days, but nobody knows how to get them to stop running into things,” MIT Computer Science and Artificial Intelligence Lab PhD student Andrew Barry, who developed the system, said in a release. “Sensors like lidar are too heavy to put on small aircraft, and creating maps of the environment in advance isn’t practical. If we want drones that can fly quickly and navigate in the real world, we need better, faster algorithms.”
That’s where Barry’s drone, developed as part of his thesis with MIT professor Russ Tedrake, comes into play. It utilizes stereo-vision algorithms that run 20 times faster than existing software, MIT says, and lets the device detect objects like trees and, in real-time, build a map of its flying area.
The software operates at 120 frames a second, and was designed to extract depth information at up to 8.3 milliseconds per frame.
Barry’s drone has a 34-inch wingspan and weighs just over a pound. He built it using commodity components costing about $1,700, MIT said in its release. It has a camera on each wing and has two onboard processors similar to that found on most cellphones.
There are currently a number of efforts to develop obstacle avoidance systems for drones, and Qualcomm has built a reference design for drones using standard components.