GMRC-Aided LiDAR/GNSS/INS: Ground Map Registration Constrained LiDAR-GNSS/INS Navigation Solution in Urban Canyons

Precise positioning and navigation in urban canyons are crucial for autonomous vehicles, Advanced Driver Assistance Systems (ADAS), and navigation-related technologies. Significant progress has been made by integrating LiDAR with Global Navigation Satellite Systems (GNSS) and Inertial Navigation Systems (INS) to improve positioning accuracy. Pure-SLAM solutions, which fuse LiDAR odometry, reliable GNSS data, and pre-integrated INS through filter- and optimization-based sensor fusion, deliver accurate positioning in typical scenarios. However, these SLAM-based methods are prone to drift, particularly in dynamic environments with varying altitudes, resulting in feature association errors and unreliable GNSS measurements. Map-aided sensor fusion has emerged as a promising approach to mitigate such errors, reducing drift by constraining the error distance in map matching. Recent studies demonstrate that map-aided navigation can significantly limit accumulative errors. However, current solutions have notable limitations: The effectiveness of map-aided navigation heavily depends on the accuracy and timeliness of pre-built maps. Dynamic objects (e.g., pedestrians, vehicles) introduce inconsistencies and gaps in point cloud data, causing map matching errors. Existing approaches often rely on single-stage sensor fusion, where a failure in one subsystem can halt the entire navigation process. To address these challenges, this study proposes a Ground Map Registration Constraint (GMRC)-aided LiDAR/GNSS/INS solution implemented in a two-stage sensor fusion framework. The key contributions of this work are: The introduction of the Ground Map Registration Constraint (GMRC) to enhance the LiDAR-GNSS/INS system, improving positioning accuracy in complex urban environments. The development of a two-stage sensor fusion framework utilizing GMRC, which includes an Extended Kalman Filter (EKF) for GNSS/INS fusion and a factor graph for map constraint integration. Experimental results demonstrate that the proposed GMRC significantly enhances the positioning solution, achieving a standard deviation (STD) of 1.366 m in the open-source dataset.