Delivering packages by drone sounds like a great idea—until they crash. And having a machine navigate the crowded environment of a city, or even a forest, requires a lot of computer power and navigation capability.
A team led by MIT’s Computer Science and Artificial Intelligence Laboratory hopes to solve that problem using a system called NanoMap, which allows drones to fly at 20 miles per hour in densely populated environments. That may sound slow compared to a drone on a flat racetrack, but MIT computer scientist Pete Florence says his survey of the literature shows that NanoMap allows drones to fly faster than other navigation systems do in crowded settings.
Drones commonly navigate using computer vision—more properly known as simultaneous localization and mapping (SLAM)—to generate a map of their surroundings, including the obstacles. That method is imperfect, however; if an object is just slightly out of place, the drone will crash into it. NanoMap foregoes the map and instead navigates around obstacles as it sees them. Although this approach is faster than mapping everything first, the trade-off is that the drone has a reduced understanding of its overall environment.
Most commercial drones rely on GPS signals combined with inertial measurement units to navigate, says Jamey Jacob, director of Oklahoma State University’s Unmanned Systems Research Institute. Sensors such as cameras or lidar are used to detect obstacles, and how fast a drone goes depends on the size of the vehicle, the number of obstacles and the accuracy of the sensors.
“One of the more difficult problems is GPS-denied navigation,” he says. “Without a GPS signal, the drone has no way to know where it is.” Drones can also get confused when GPS signals bounce off of obstacles, which can result in false positional information.
Typically, a drone’s forward speed suffers in a crowded environment, because there are so many obstacles to navigate around, says Kurt Barnhart, executive director of Kansas State Polytechnic’s Applied Aviation Research Center. “There are filters which can greatly assist with this confusion, but there isn’t a ‘one-size-fits-all’ solution,” he says.
Florence says that in the future, he plans to integrate NanoMap with SLAM mapping to get the best from both approaches. NanoMap has been tested in a forest and a warehouse, as well as in simulated environments. The ultimate goal is to improve the accuracy of the flight—not necessarily the speed, Florence says. “We want to bring it from 99 percent reliability to 99.9 percent, and as many more nines as we can get.” His work is supported in part by DARPA’s Fast Lightweight Autonomy program, which aims to improve machine navigation in cluttered environments.