Data association approach to mobile robot localization using non-unique landmarks

Abstract

In global localization problem, the robot should localize itself without the knowledge of its initial pose, i.e., its initial position and attitude. Under GPS-denied environment, non-unique landmarks such as traffic signs can be used as alternative absolute position information. However, such landmarks lead to ambiguous correspondence between landmarks and their features. For a robot to successfully localize itself, this ambiguity should be tackled. To solve the problem, a multiple hypothesis testing method is adopted. As robot detects other landmarks, the number of the hypotheses can be increased exponentially. The nearest neighbor filter technique is applied to prevent the number increase. The capability of the suggested algorithm is examined based on a computer simulation.