Scan Matching based Mobile Robot Localization

Abstract

Mobile robot perceives its position using sensory information acquired by robot motion. This perceiving process is called localization, which is an essential part for autonomous navigation of mobile robot. There are various localization methods to calculate the position of mobile robot according to a perceiving way. This thesis presents localization methods for mobile robots equipped with range sensors, such as laser range finders and ultrasonic sensors. Of the various perceiving ways, scan, a set successive raw sensor values, based approaches are used for localization because they can be applied to non-structured environments without predefined geometric features.First, for laser range finders, the direction augmented probabilistic iterative closest points (DApICP) is proposed. A local feature of sensor data is incorporated to improve the performance of scan matching. From scan data, directions of data points are calculated, then they are used in the error minimization step to find relative translation and rotation between a reference scan and a current scan. This local feature, the direction of data point, allows us to improve the convergence of the DApICP, and obtain the robust matching performance against a large rotation angle between both scans.Second, for ultrasonic sensors, the correlation based scan matching in the Hough domain is introduced. Although ultrasonic sensor have various benefits, they have inevitable problems to be settled, such as the sparseness, angular uncertainty, and undesirable reflections. Constructing local grid map using accumulated sensor data resolves theses problems. From the constructed local grid map, reliable grid cells are extracted, and they constitute a scan. Data points in the scan are transformed into the Hough domain to reflect the local feature of them directly. Then, the correlation based scan matching approach is applied to the Hough transform. Though these process, the position of mobile robot is calculated, and it is consistently fused with the odometry of mobile robot based on extended Kalman filter. This framework allow us to localize mobile robot consistently only using low-cost ultrasonic sensors and the odometry information.As a result, the mobile robot is successfully localized using both laser range finders and ultrasonic sensors. By incorporating the local feature of data points, the performance of scan matching can be improved. Through various experiments, the proposed localization algorithms are evaluated.