Amazon and other companies have big plans for delivery drones. In order to get to the point where delivery drones are truly safe and ready to become mainstream technologies, though, one of the things that needs work is making them more agile and better able to deal with complex obstacles while flying. That’s something that researchers from MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) have been working to improve.
Their new NanoMap technology promises to give drones the ability to consistently navigate at 20 miles per hour speeds through obstacle-packed locations such as forests or warehouses — in which even the slightest of miscalculations can result in a crash.
“NanoMap is a mapping system that enables drones to fly at high speeds through dense environments like forests and warehouses,” MIT CSAIL graduate student Pete Florence told Digital Trends. “The system’s key insight is that it actively models and accounts for the uncertainty of not being 100 percent confident about where the drone is located in space. This makes it a more flexible approach for flying in real-world environments that you can’t predict in advance, and creates a deeper integration between perception and control.”
NanoMap consists of a depth-sensing system, which seamlessly stitches together a series of measurements concerning a drone’s surroundings. This allows it to anticipate what motion plans to make concerning both what it is currently looking at and also what it might see in the future. That’s different from existing drone piloting technologies that are routinely reliant on intricate maps that tell the drone exactly what is around it at any point.
The idea of a high-speed drone that doesn’t sweat the small details about its exact location sounds, at best, counterintuitive and, at worst, a bit scary, but MIT’s smart tech means it’s surprisingly effective. Without the NanoMap system being employed, MIT’s test drone crashed 28 percent of the time if it went off-course by more than 5 percent. With NanoMap, these crashes were reduced to only 2 percent of flights that veered 5 percent off course.
According to Pete Florence, the technology could theoretically also be used in any piece of hardware involved in navigation, including self-driving cars. There’s still a lot more work to be carried out, though.
“There’s much more that can be done in terms of improving our systems for planning, control, perception and local obstacle avoidance,” he said. “As an example, we intend to work on the system so that it can one day incorporate other pieces of information related to uncertainty, like being able to account for the uncertainty of the drone’s depth-sensing measurements.”