Least squares variance component estimation for robust GNSS/INS/LiDAR integrated with large-scale navigation

Abstract Light detection and ranging (LiDAR-simultaneous localization and mapping (SLAM)) systems suffer from cumulative positioning errors in large-scale or dynamic environments. To solve this, we integrate Global Navigation Satellite System (GNSS) for absolute positioning and use an extended Kalman filter (EKF) with least squares variance component estimation (LS-VCE) to optimize multi-sensor fusion. The EKF efficiently incorporates GNSS updates into LiDAR-SLAM, while LS-VCE dynamically adjusts observation variances, replacing empirical noise models with data-driven stochastic models. This integration helps develop a more accurate stochastic model, reducing error accumulation in large-scale navigation. Eperiments were conducted in three outdoor environments: a long-distance path, a short-distance path, and a closed-loop trajectory. Compared to standalone LiDAR-SLAM, the integrated system showed decimetre-level precision in long-distance navigation. After optimizing the stochastic model with LS-VCE, significant improvements were noted. Specifically, the average absolute position errors in the east, north, and up directions were reduced by 44%, 32%, and 31%, respectively. Root mean square errors decreased by 56%, 10%, and 18%, respectively. These results demonstrate that the LS-VCE-enhanced stochastic model more accurately captures the stochastic nature of various observations, improving the overall accuracy and reliability of the GNSS/inertial navigation system/LiDAR fusion system and significantly reducing error accumulation in large-scale environments.