SLAM Sharing Among Heterogeneous Sensors

Abstract

The advancement of Simultaneously Localization and Mapping (SLAM) has enabled robots to accurately locate themselves in unknown environments with sensors such as LiDARs and Cameras while building a corresponding map. Re-using this map later can ensure accurate and robust localization if the environment does not change significantly. Current SLAM studies mainly focus on improving the performance of SLAM algorithm to gain better localization accuracy. However, the discrepancies between localization sensor and mapping sensor such as accuracy and resolution, may impact the localization in the shared map. The impact factors of map sharing performance has not been widely investigated. Understanding the impact factors can facilitate the implementation of map sharing system to extend the usage of SLAM map. In this paper, we utilize two representative SLAM systems, NDT SLAM and ORB SLAM, to study the potential impact factors of using a shared map for heterogeneous sensors. Specifically, we evaluate the impact of three key factors (map, localization algorithm and sensor) on map sharing performance. With three LiDARs and three cameras, we record a dataset and build two groups of maps of the same environment. By applying these maps for localization, we derive four insights into the relation between localization performance and variability of the critical factors. Specifically, we find that Lidar-based SLAM performs stable to the discrepancy of Lidar sensors. In contrast, visual-based SLAM is sensitive to the shared map’s quality and the camera’s focal length.