Fighting Unreliable GNSS with LiDAR: Autonomous Driving. In The Project KoMo:Dnext

Abstract

Self-driving cars demand high precision regarding perception, especially in the dense traffic of urban canyons. Moreover, navigating and controlling the car on a map-based approach requires precise global ego-localization, e. g. in a geographic coordinate system. While GNSS approaches lack precision and robustness in urban areas, visual odometry and simultaneous localization and mapping (SLAM) approaches only deliver information regarding the relative movements of the car but no geo-referenced information. Within the project KoMo:DNext, the Institute of Transportation Systems at the German Aerospace Center (DLR) studies the fusion of GNSS and SLAM-based methods to leverage their advantages, i. e., to gain a robust and precise global ego-localization solution even in urban areas within dense and dynamic traffic. This work discusses a fusion of GNSS and LiDAR localization solutions developed with Outsight SA, which enables our car to safely navigate through Düsseldorf, even through tight construction sites, which would not have been possible with a pure GNSS-based localization.