A Laser Inertial SLAM System with High Robustness and Low Drift in Complex Environments

The paper proposes an improved version of LeGO-LOAM that utilizes Patchwork++ for ground segmentation, CVC (Curved-Voxel Clustering) for nonground point cloud clustering to remove noise, and incorporates IMU pre-integration factor. In the environment similar to underground conditions with irregular terrains such as potholes and uneven surfaces, as well as conditions containing a significant amount of noise, irrelevant point clouds and obstacles, the accuracy of SLAM decreases while computational complexity increases. Patchwork++ ground segmentation quickly and accurately identifies rugged surfaces, providing a stable reference plane to enhance the quality and accuracy of point cloud data. CVC precisely clusters nonground point clouds to remove noise, reducing computational complexity. IMU pre-integration improves pose estimation accuracy and system stability. The paper conducts a comparison between the proposed method and cutting-edge SLAM algorithms by utilizing KITTI dataset sequences that replicate the mentioned scenarios. The results clearly indicate that proposed method attains superior accuracy.