A High-Precision SLAM System Based on Lidar, IMU and Wheel Encoders

Abstract

In autonomous driving map generation, vehicle localization plays a crucial role. To address the limitations of single sensor positioning in terms of accuracy and robustness, this paper proposed a multi-sensor fusion SLAM algorithm based on lidar. The algorithm integrated information from lidar, IMU, and wheel encoders. Building upon the FAST-LIO2 framework, this paper first integrated IMU data to obtain an initial pose estimation. Then, the system's velocity was observed and corrected by using the velocity values from the wheel encoders. Subsequently, the distorted lidar point cloud was undistorted using this pose estimation. Finally, the lidar point cloud was matched with the map, and a second pose correction was performed based on point-to-plane residual observations. Using this refined pose estimation, the lidar point cloud was projected onto the world coordinate system, completed the mapping of the surrounding environment. Experimental evaluations were conducted on both dataset and real-world environments, and the results demonstrated that incorporating the wheel encoder velocity as a direct observation of the system's velocity benefited the SLAM algorithm in improving localization accuracy.