New approach in solving the kidnapped robot problem

Abstract

We present a solution for the global localization of a mobile robot which disposes of a prior map and is waked up in an unknown position. This case is known in robotics as the kidnapped robot problem. The algorithm is based on the adaptation of the 2D SURF (Speeded Up Robust Features) image features to 3D landmarks of the environment. In this way the prior map is built, containing only 3D Surf landmarks. When the robot wakes up in an unknown location the robot starts to search after these landmarks, which are compared to the priory known ones based on the SURF descriptors. To estimate the displacement between the online and offline landmarks the RANSAC random sampling method is used. This method also has the advantage that the falsely detected and associated landmarks are eliminated. The experiments were carried out in an indoor office environment.