NLOS Identification- and Correction-Focused Fusion of UWB and LiDAR-SLAM Based on Factor Graph Optimization for High-Precision Positioning with Reduced Drift

Abstract

In this study, we propose a tightly coupled integrated method of ultrawideband (UWB) and light detection and ranging (LiDAR)-based simultaneous localization and mapping (SLAM) for global navigation satellite system (GNSS)-denied environments to achieve high-precision positioning with reduced drift. Specifically, we focus on non-line-of-sight (NLOS) identification and correction. In previous work, we utilized laser point cloud maps to identify and exclude NLOS measurements in real time to attenuate their severe effects on the integrated system. However, the complete exclusion of NLOS measurements will likely lead to deterioration in the dilution of precision (DOP) for the remaining line-of-sight (LOS) anchors, counterproductively introducing large positioning errors into the integrated system. Therefore, this study considers the ranging accuracy and geometric distribution of UWB anchors and innovatively proposes an NLOS correction method using a grey prediction model. For a poor line-of-sight (LOS) anchor geometric distribution, the grey prediction model is used to fill in the gaps by predicting the NLOS measurements based on historical measurements. Including the corrected measurements effectively improves the original poor geometric configuration, improving the system positioning accuracy. Since conventional filtering-based fusion methods are exceedingly sensitive to measurement outliers, we use state-of-the-art factor graph optimization (FGO) to tightly integrate the UWB measurements (LOS and corrected measurements) with LiDAR-SLAM. The temporal correlation between measurements and the redundant system measurements effectively enhance the robustness of the integrated system. Experimental results show that the tightly coupled integrated method combining NLOS correction and FGO improves the positioning accuracy under a poor geometric distribution, increases the system availability, and achieves better positioning than filtering-based fusion methods with a root-mean-square error of 0.086 m in the plane direction, achieving subdecimeter indoor high-precision positioning.