Abstract
Simultaneous localization and mapping (SLAM) is a key technology for robot localization. Currently, most robot application scenarios are limited to either indoor or outdoor environments. When robots operate across both indoor and outdoor environments, they often experience significant positioning errors or even failures. To address this problem, we propose a LiDAR SLAM approach based on clustering features and constraint separation. First, the ground is removed from the point cloud, followed by rough segmentation of the remaining point cloud. Feature points are then extracted from each rough cluster using normalized principal component analysis (PCA) and reflection intensity information, enabling fine segmentation. Clusters are separated based on their eigenvectors, and pose variables are estimated independently according to the different constraints of the separated clusters in various directions. These estimates are then weighted and fused to complete the front-end odometry. In the backend optimization, scan and local map matching, physical world constraints, and loop closure are employed to jointly optimize pose variables, resulting in a globally consistent point cloud map. We evaluated our method on public and custom datasets, and the results show that, compared to other state-of-the-art methods, our approach achieves better localization accuracy and more robust performance in both indoor and outdoor environments, as well as in combined indoor-outdoor scenarios.
Funding Information
  • National Basic Research Project (JCKY2022110C040)

This publication has 57 references indexed in Scilit: