Mobile robot localization based on eigenspace analysis

Abstract

A single observation is not enough to obtain accurate results for localization of a mobile robot. Therefore, the authors propose an approach using the eigenspace method to infer the robot position from several observations. Integration of multiple observation results allows flexible reaction to situations with similar images in several positions and directions, or with a partially changed environment. We discuss the method of position inference based on orienting the camera in multiple directions from one position, and the method of travel route inference by observation performed during movement. Effectiveness of the proposed techniques was confirmed by tests in a real-room environment. © 1998 Scripta Technica. Syst Comp Jpn, 28(12): 11–21, 1997