Large-Scale Mapping with Loop Closure
(EECS 467 Final Project)
We present a robotic mapping system which can produce accurate and consistent maps of large environments. Basic systems for simultaneous localization and mapping (SLAM) can estimate the motion of a mobile robot to a high degree of accuracy, allowing sensor readings from different positions to be fused into a map. But even small amounts of error can accumulate over time, especially when the robot can only observe its immediate surroundings. Our approach enables the robot to recognize that it has returned to a location it has seen before, even if the accumulated error means the map is no longer consistent. The map is then rectified with a pose graph optimization technique, producing a consistent map.