Using drones to deliver packages requires the aerial delivery vehicles to navigate a complex logistics obstacle course at high speeds. MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed a system called NanoMap, that allows drones to consistently fly at 20 miles per hour through dense environments such as warehouses.
Flying around multiple obstacles quickly is computationally complex. Small drones with limited payloads can’t carry enough real-time processing power to perform this navigational feat reliably.
Many existing approaches rely on intricate maps that tell drones where they are relative to obstacles. But this is problematical in a real-world setting where even a small margin of error when estimating the location of objects can easily result in a crash.
The NanoMap system assumes the drone’s position in the world over time to be uncertain, and models for that uncertainty. Building uncertainty into the calculations is a more reliable way to guide the vehicle around obstacles, according to CSAIL.
More specifically, NanoMap uses depth-sensing capabilities to stitch together a series of measurements about the drone’s immediate surroundings. This allows it to construct motion plans for the current field of view and anticipate how the drone should move around hidden fields of view that the system has already logged.
Tests of the system suggest that the new approach is effective. If NanoMap is not modeling uncertainty and the drone drifted just five percent away from where it was expected to be, it would crash more than once every four flights. When the uncertainty is accounted for, the crash rate was reduced to just two percent. The tests also showed that NanoMap is particularly effective for smaller drones moving through relatively small spaces and works well in tandem with a second system that is focused on more long-horizon planning.