A PhD student in MIT’s Computer Science and Artificial Intelligence lab has designed an autonomous drone that can maneuver its way around obstacles while flying at 30 miles per hour. Creator Andrew Barry says it’s hoped the drone will help create systems that can “fly quickly and navigate in the real world.”
According to a release from MIT, Barry has developed a system that allows a drone to detect objects up to 10 meters (about 32 feet) away and build a map based on what it sees. The system is able to do this about twenty times faster than any other existing software. Barry said that when the drone is travelling at high speeds it doesn't need to build up a detailed picture of what is close by as it's soon moved on.
“Everyone is building drones these days, but nobody knows how to get them to stop running into things,” Barry 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.”
The drone, which weighs just over a pound and has a 34-inch wingspan, was put together using off-the-shelf parts at a cost of only around $1700. A camera on each wing allows the drone to "see" in stereo, and it includes two on-board processors “no fancier than the ones you’d find on a cellphone.”
Previous systems for autonomous drones have meant that they were not able to fly faster than 5 or 6 miles per hour due to the processing power required to analyze the world around them and react to it. Barry and his collaborator, Professor Russ Tedrake, realized that when traveling at faster speeds, things don’t change too frequently. The drone’s camera sees its surroundings at 120 frames per second, managing to perceive depth per frame in about 8 milliseconds, always looking about 10 meters ahead.
“You don’t have to know about anything that’s closer or further than that,” Barry told MIT. “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.”
The drone also uses on-board odometry systems, and uses previous maps it created to help it navigate. Barry says that he hopes to further improve the algorithms so that they can work at more than one depth, and in environments as dense as a thick forest. He’s also open-sourced the algorithm online, allowing anyone to try it out.
Check out the video below to see the automated drone in action.