Autonomous Drone Dodges Trees at 30 MPH

Zipping above a mown field, the autonomous drone flies towards groupings of trees. Behind, a human-controlled drone follows. From the second drone’s camera view, the observer sees the drone headed for two thin tree trunks. But before hitting, the autonomous drone banks to the left, dodging the obstacle. Andrew Barry, a PhD student at Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Lab (CSAIL), developed an obstacle-detection system, enabling autonomous drones to navigate tree-laden areas at speeds up to 30 mph. “Everyone is building drones these days, but nobody knows how to get them to stop running into things,” said Barry. “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 the real world, we need better, faster algorithms.” Built from off-the-shelf components costing around $1,700, the drone weighs a little over a pound and boasts a 34-in wingspan. A camera and processor are mounted on each wing. According to The Washington Post, the chip powering the drone is the same as the one found in the Samsung Galaxy S3. Barry developed a stereo-vision algorithm that allows the drone to detect objects and build a map in real-time. “Operating at 120 (fps), the software—which is open source and available online—extracts depth information at a speed of 8.3 msec per frame,” according to CSAIL. Due to the drone’s speeds, Barry was able to tailor its algorithm to compute distances up to 10 m away. “You don’t have to know about anything that’s closer or further than that,” he said.
“As you fly, you can push that 10-m zone forward, and, as long as your first 10 m are clear, you can build a full map of the world around you.” However, some kinks still need to be worked out. “Our current approach results in occasional incorrect estimates known as ‘drift,’” Barry 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.” The flight code for the drone is available on GitHub. • CONFERENCE AGENDA ANNOUNCED: The highly-anticipated educational tracks for the 2015 R&D 100 Awards & Technology Conference feature 28 sessions, plus keynote speakers Dean Kamen and Oak Ridge National Laboratory Director Thom Mason. Learn more.