Lidar-Inertial SLAM Method for Accurate and Robust Mapping

Yuhang Wang,Liwei Zhang
2023-01-01
Abstract:In recent studies on LiDAR SLAM, the achievement of robust optimized LiDAR odometry is the primary objective. For the mapping part, some studies focus on improving the processing of point cloud, while others aim to the optimization of the result deviation caused by errors. Meanwhile, in the fields of robotics and autonomous driving, multi-sensor fusion solutions based on IMUs are becoming the norm. This paper contributes to the optimization of mapping by leveraging a lightweight LiDAR-inertial state estimator. The proposed method combines information from a 6-axis IMU and a 3D LiDAR to form a tightly-coupled scheme that incorporates iterative error state Kalman filter (IESKF). Due to the continuous error accumulation, trajectory deviations can be significant. To mitigate this, an adaptive distance threshold loop closure detection mechanism is employed. Furthermore, since the algorithm primarily addresses outdoor scenes, ground features collected by sensors account for a significant portion of the computation. Improvements in the ground segmentation method lead to less disturbance during mapping on uneven terrain, enabling the method to effectively ad-dress a wider range of real-world environments. As a result, the proposed method demonstrates excellent stability and accuracy, as verified in experiments conducted on urban dataset and campus environment.
What problem does this paper attempt to address?