Cartographer Local SLAM Optimization Using Multistage Distance Scan Scheduler

Abstract

This paper presents the utilization of Google’s simultaneous localization and mapping (SLAM) called Cartographer, and improvement of the existing processing speed using multistage distance scheduler. The presented approach optimizes the Local SLAM part in Cartographer to correct local pose based from Ceres scan matcher by integrating scheduling software, which controls the distance of light detection and ranging (LiDAR) sensor and scan matcher’s search window size. In preceding work, the multistage distance scheduler was successfully tested in the actual vehicle to map the road in real-time. Multistage distance scheduler means that local pose correction is done by limiting the distance scan of LiDAR and search window with the help of scheduling algorithm. The scheduling algorithm manages the SLAM to swap between small scan size (25 m) and large scan size (60 m) LiDAR at a fixed time during map data collection; thus it can improve performance speed efficiently better than full-sized LiDAR while maintaining the accuracy of full distance LiDAR. By swapping the scan distance of sensor between small and long-range scan, and adaptively limit search size of scan matcher to handle difference scan size, it can improve pose generation performance time around 15% as opposed against fixed scan distance 60 m while maintaining similar pose accuracy and large map size.