Tightly coupled GNSS RTK-LiDAR-inertial system for consistent urban navigation

LiDAR-Inertial Odometry (LIO) has emerged as a viable solution for local navigation, especially when GNSS signals are affected by interference and outages. The recent availability of solid state LiDAR with non-repetitive scanning patterns has further enhanced the appeal of solid state LIO (sLIO). However, most state-of-the-art LIO methods rely on absolute constraints by associating newly scanned features to a globally maintained map, making it difficult to effectively integrate GNSS information into a consistent tightly coupled fusion system. In this paper, we introduce a coarse-to-fine LiDAR registration strategy that achieves a consistent estimator by combining both absolute scan-to-map and relative keyframe map constraints, thus transforming the system into a partially dead-reckoning framework. Then, a tightly coupled GNSS RTK-solid state LiDAR-Inertial Navigation System (GsLINS) is proposed through keyframe-based factor graph optimization at the measurement level, with global attitude initialization and pose estimation. The proposed coarse-to-fine strategy proves to be consistent in state estimation and achieves superior accuracy in comparison to other methods. The tightly coupled GsLINS system is validated through various field experiments in diverse urban environments and large-scale scenarios, demonstrating precise and robust navigation performance.