A ubiquitous positioning solution of integrating GNSS with LiDAR odometry and 3D map for autonomous driving in urban environments

Continuous and reliable positioning anywhere anytime is one of the critical technologies for autonomous driving, whereas it is challenging yet, especially in urban environments. Due to the presence of buildings, Global Navigation Satellite System (GNSS) is not reliable for degraded positioning accuracy and availability. With the simultaneous localization and mapping (SLAM) approach, Light detection and ranging (LiDAR) has been widely used by autonomous driving vehicles for localization and environmental perception. However, LiDAR odometry (LO) or SLAM is not accurate enough for autonomous driving as it suffers from accumulative errors. This study proposes a GNSS/LiDAR/Map integrated positioning solution to provide ubiquitous positioning for autonomous driving, especially in urban areas, including urban canyons and underground, where GNSS is blocked partly or even fully. Matching LiDAR scans with a 3D map, named 3D map-based global localization (MGL), can provide an absolute positioning solution with respect to the map. A drift error model of LiDAR odometry is proposed to improve the accuracy of LiDAR odometry, and its model parameters are estimated online using either GNSS or MGL absolute locations. Using the graph optimization approach, GNSS and/or MGL precise absolute positioning are integrated with LO relative positioning to provide continuous and reliable positioning even when GNSS is degraded. The proposed solution is validated using the NCLT public dataset, which contains LiDAR data for building 3D maps and 3D map-based global localization, respectively. Based on the NCLT dataset, we simulate three typical urban scenarios to verify the performance of ubiquitous positioning. The experiments show that the proposed drift error correction method decreases accumulated error of LiDAR odometry by 35.5%, and the proposed GNSS/LiDAR/Map solution can provide continuous positioning with the availability of 100%, and improves the positioning accuracy by 49.8%, compared to the state-of-the-art approach of GNSS/LO fusion.