BEV-LSLAM: A Novel and Compact BEV LiDAR SLAM for Outdoor Environment

Abstract

LiDAR-based SLAM is an essential technology for autonomous robots, benefited from its high accuracy and scale invariance. Interestingly, researchers have been increasingly focusing on establishing simple, efficient, but effective LiDAR SLAM systems recently. In this paper, we propose a novel and compact LiDAR-only SLAM system BEV-LSLAM, leveraging visual features in BEV view for all the steps of pipeline including pose estimation, mapping, loop closing and back-end graph optimization. The proposed BEV features are more stable than traditional geometrical features, which can be adapted to various LiDARs sensors without changing the hyperparameters. In addition, benefited from filtering vulnerable features based on tracking process in consecutive frames, only high-quality feature points are used for lightweight pointcloud map construction. Extensive experiments on UrbanLoco, KITTI, and our 16-channel LiDAR datasets prove the superiority of our approach, compared with state-of-the-art LiDAR SLAM methods.