Centralized Multi-Robot Collaborative LiDAR SLAM Utilizing Loop Closure Selection

Abstract

Multi-robot systems have attracted much attention due to their high efficiency and robustness. Multi-robot Simultaneous Localization and Mapping (SLAM) is the cornerstone of collaboration, such that each robot can localize and build a map of the workspace. One of the key challenges of this problem lies in utilizing the shared information between robots adequately so that the system can retain the correct information and reject the outliers. Here we propose a centralized multi-robot collaborative LiDAR SLAM framework for robots, each only equipped with a 2D LiDAR without the need for additional sensors. With each robot able to run simple LiDAR odometry onboard, a central server with greater computational capacity collects the information from robots, optimizes their poses globally, and updates their poses through communication. For a large number of loop closures in multi-robot SLAM systems, it is catastrophic to build wrong data associations between pose nodes in the global factor graph. We resort to a loop closure selection method to ensure the correctness and consistency of global optimization. Throughout the verification process on the real Automated Mobile Robot (AMR) platform, the framework exhibits better performance on localization and mapping.