De-Skewing Point Clouds Based on Gaussian Process Regression in a Tightly Coupled LiDAR-IMU SLAM System

By processing inputs from sensors such as LiDAR and Inertial Measurement Units (IMU), Simultaneous Localization and Mapping (SLAM) can provide maps and positioning for autonomous vehicles in unknown environments, enabling autonomous navigation. However, the skew of LiDAR point clouds during the scanning process has not been addressed well, which can affect the localization accuracy. To address this issue, this paper proposes a tightly coupled LiDAR-IMU SLAM (LIS) system. First, the LIS employs Gaussian process regression to de-skew point clouds. It processes IMU data to derive motion vectors that predict the rotational and translational motion of the LiDAR during a scanning period, effectively compensating for motion and producing distortion-free point clouds. Then, LIS integrates LiDAR and IMU measurements through sliding window optimization, producing refined localization and mapping results. To validate the effectiveness of the LIS, evaluations were performed on the sensor data collected in the real world and the public datasets. The LIS achieves high end-to-end localization accuracy, outperforming baseline algorithms such as LOAM LIVOX and FAST-LIO.