Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have determined that for autonomous drones to navigate more precisely, they must rely upon a navigation system predicated upon uncertainty.
The NanoMap system, which allows drones to top speeds of 20 miles per hour while consistently avoiding obstacles presented by dense environments such as forests or skyscraper-lined cities, has been shown to reduce crash rates as low as 2 percent. Using near-constantly scanning sensors instead of traditional pre-determined maps of a given terrain, the NanoMap operating system is based on the assumption that environments rife with uncertainty can’t be effectively mapped pre-flight.
Overly confident maps won’t help you if you want drones that can operate at higher speeds in human environments, graduate student Pete Florence told MIT News. An approach that is better aware of uncertainty gets us a much higher level of reliability in terms of being able to fly in close quarters and avoid obstacles.
The use of this data-collecting technology to plan the motions of a mobile device such as a drone is fairly new ground. While systems commonly referred to as simultaneous localization and mapping (SLAM) have been used in other technology, the speed with which an in-motion operating system must simultaneously collect and utilize data in order to plan movements have made it tricky to incorporate in drones.
Yet, the CSAIL team has proven capable of deploying their version of the SLAM mapping system in their drone, all while it zips around at 20 miles per hour or even faster. Florence, professor Russ Tedrake and research software engineers John Carter and Jake Ware designed the system by allowing the drone’s sensors to rely upon predetermined scenarios and terrains to figure out how to react in a given situation, even if it hasn’t encountered a truly similar obstacle in the past. The system also relies on the underlying assumption that the drone is never 100% certain of its precise geospatial position. This accounting for fallibility is what the researchers believe is the greatest differentiation between NanoMap and other drone navigation systems.
The key difference to previous work is that the researchers created a map consisting of a set of images with their position uncertainty rather than just a set of images and their positions and orientation, says Sebastian Scherer, a systems scientist at Carnegie Mellon University. Keeping track of the uncertainty has the advantage of allowing the use of previous images even if the robot doesn’t know exactly where it is and allows in improved planning.