A researcher at the Massachusetts Institute of Technology has created a self-piloting drone that can dodge obstacles at 30 miles per hour.
With a 34-inch wingspan, the drone would cost about $1,700 to build, and uses off-the-shelf parts including two cameras and a processor, according to a statement from MIT CSAIL.
The drone is software that's different from what's usually found in drones. Typically, software in autonomous drones examines images taken by an on-board camera for objects at various distances — 3 feet ahead, then 6 feet ahead, etc — to build a picture of the drone's environment. That gives an accurate picture, but compiling information from measurements taken at all those different distances takes a lot of computing power, thereby limiting the speed the drone can travel to only a few miles per hour, MIT CSAIL said.
Andrew Barry, a student at MIT's Computer Science and Artificial Intelligence Lab, simplified the process by making the software examine images of objects at a constant distance, about 30 feet in front of the drone.
"You don't have to know about anything that's closer or further than that," Barry said in the statement. "As you fly, you push that 10-meter horizon forward, and, as long as your first 10 meters are clear, you can build a full map of the world around you."
As a result, the software, which is open source and available for free online, runs 20 times faster than existing navigational software.
The system is not perfect. Occasionally the software gets estimates incorrect, so it is not yet ready for a cluttered environment such as a dense forest, for instance. But Barry believes the algorithms can be improved.
A copy of the paper can be read here.