Global localization of a mobile robot using lidar and visual features

Abstract

Global localization is a well-known problem of estimating the pose of a robot in a learned map without any prior knowledge of its initial pose. It is difficult to achieve global localization with lidar when only a limited number of lidar scans are available, especially in environments with simple and repetitive local geometric features. Camera has the advantage of rich information but suffers from lack of reliable and efficient localization algorithms and relative high computation burden. In this paper, we propose a fast and reliable global localization approach with the capability of addressing the kidnapped robot problem, where both lidar and camera sensors are integrated. We introduce a series of visual keyframes to the environment map with lidar in the SLAM process. A global descriptor is applied to reduce pose search space and a local descriptor is used to further improve localization accuracy and the re-localization trigger mechanism is applied to solve the kidnapped robot problem effectively. Several experiments in indoor environments under different conditions were performed to verify the effectiveness of the proposed approach.