A lot of research has already been done to create maps robots can use to navigate a given area, such as estimating the distance between themselves and nearby walls, and planning routes around obstacles, said Fallon. But these maps are developed mostly for a single, one-time use, and can't be adjusted to changing surroundings over time. "If you see objects that were not there previously, it is difficult for a robot to incorporate that into its map," he said.
The team also includes John J. Leonard, professor of mechanical and ocean engineering, and graduate student Hordur Johannsson.
The team previously tested the approach on robots that were equipped with expensive laser scanners, but have since implemented it with a Kinect-type camera in a robotic wheelchair, a portable sensor suit, and a PR2 robot developed by Willow Garage. On these devices, the system can continuously locate the robotic hardware within a 3D map of its surroundings while traveling at speeds of up to 1.5 meters per second.
The Kinect sensor's visible-light video camera and infrared depth sensor scan the robot's surroundings as it moves through a new, unexplored area, while the robot builds up a 3D model of the walls of a room and the objects within it. Map details can include location information about the edges of walls and objects within the walls.
When the robot visits the same area again, the system compares the previous images it has taken with the features of the new image it creates until it detects a match. Once the system has decided on its location, any new features it encounters since it took the previous picture of that location are incorporated into the map by combining new and old images.
While the system is making and updating maps, it is also continuously estimating the robot’s motion by measuring the distance its wheels have rotated, with onboard sensors. The system can determine the robot's position within a building by combining the motion data with visual information from the camera and depth sensor, which also serves as a form of error correction, said Fallon.