The tight-coupled SLAM system based on LiDAR and improved VGICP method for waterfront environments

Abstract

Simultaneous Localization and Mapping (SLAM) is the core for high-precision positioning and navigation of Unmanned Surface Vehicles (USVs). Existing LiDAR-based SLAM methods are primarily designed for terrestrial environments and often struggle to address the challenges posed by waterfront environments, such as unstructured settings, sparse features, and water surface fluctuations, which degrade localization accuracy. This paper proposes a tight-coupled SLAM approach based on a multi-factor optimization graph to discuss them. It employs an improved Voxel Generalized Iterative Closest Point (VGICP) algorithm to obtain initial odometry information. The state nodes in the multi-factor optimization graph can be optimized by dynamically adjusting the noise covariance of odometry, IMU, and loop-closure factors. It also effectively mitigates environmental noise and sensor errors, enhancing the precision and stability of pose estimation. We achieve high-precision localization and globally consistent map construction in waterfront environments by continuously updating the multi-factor optimization graph. Field tests are conducted on campus lakes and the results show that it is better than the mainstream approaches in accuracy and robustness.