While most of the drone industry is working to build drones, one researcher from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) is working to ensure those drones don’t hurt anyone once they are built. Andrew Barry, a Ph.D. student at CSAIL, has devised an obstacle-detection system that allows drones to automatically dip, dive and dart away from objects.
“Everyone is building drones these days, but nobody knows how to get them to stop running into things,” he said. “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.”
Barry has developed a stereo-vision algorithm that detects objects and builds a map of its surrounding in real time. According to Barry, the software, which is open source, is 20x faster than existing software, and extracts information at 8.3 milliseconds per frame.
Traditional algorithms focus on depth fields at multiple distances, limiting the drone’s speed to about 5 to 6 mph. Barry’s algorithm measures distances of 10 meters, allowing the drone to fly at 30 mph.
There are still some limitations to the algorithm, but Barry hopes it can be improved to work in more environments such as thick forests.
“Our current approach results in occasional incorrect estimates known as ‘drift,’ ” he said. “As hardware advances allow for more complex computation, we will be able to search at multiple depths and therefore check and correct our estimates. This lets us make our algorithms more aggressive, even in environments with larger numbers of obstacles.”