ATLAS: a framework for large scale automated mapping and localization

Abstract

This thesis describes a scalable robotic navigation system that builds a map of the robot's environment on the fly. This problem is also known as Simultaneous Localization and Mapping (SLAM). The SLAM problem has as inputs the control of the robot's motion and sensor measurements to features in the environment. The desired output is the path traversed by the robot (localization) and a representation of the sensed environment (mapping). The principal contribution of this thesis is the introduction of a framework, termed Atlas, that alleviates the computational restrictions of previous approaches to SLAM when mapping extended environments. The Atlas framework partitions the SLAM problem into a graph of submaps, each with its own coordinate system. Furthermore, the framework facilitates the modularity of sensors, map representations, and local navigation algorithms by encapsulating the implementation specific algorithms into an abstracted module. The challenge of loop closing is handled with a module that matches submaps and a verification procedure that trades latency in loop closing with a lower chance of incorrect loop detections inherent with symmetric environments. The framework is demonstrated with several datasets that map large indoor and urban outdoor environments using a variety of sensors: a laser scanner, sonar rangers, and omni-directional video.