Robust self-localization system based on multi-sensor information fusion in city environments

Abstract

Recently, autonomous vehicle of self-localization based on the high definition (HD) map become more popular due to the availability of HD map and the dropping prices of LIDAR. Many types of studies have improved the local and global accuracy of HD map for accurate localization. However, the global accuracy of a map does not guarantee the accurate self-localization within the map. In this paper, by investigating the scene errors in the map and comparing their characteristics, we introduced four factors that affect the self-localization ability of the map. These factors are highly related to the environment, namely feature sufficiency, layout, local similarity, and representation quality of the map. Then, in order to overcome the limitations brought by the map itself, we proposed a multi-sensor information fusion self-localization system. The system uses GNSS, LIDAR, IMU, and local map to fuse multi-source information. Global mapping is not required for autonomous vehicle's accurate self-localization. To achieve a high precision of global positioning, the system uses Kalman filter to integrate GNSS positioning, slam and inertial navigation solution to improve the robustness of the system, and the method of local map matching is used to eliminate the accumulated error and modify the positioning system. By conducting the experiments in a typical testing environment, we have evaluated the performance of this system by comparing the ground truth with the self-localization error.