A Lightweight Centralized Collaborative Visual-Inertial SLAM

An individual unmanned platform driving autonomously in complex environments usually encounters problems, such as limited perception range, low mapping efficiency, and poor robustness, making it difficult to consistently and accurately complete SLAM tasks. Collaborative SLAM, where multiple unmanned platforms work together to simultaneously localize and map, can realize high efficiency, robustness, and accuracy SLAM in large environments. This paper proposes a lightweight centralized collaborative visual-inertial SLAM strategy. Each sub-unmanned platform independently operates Visual Inertial Odometry and shares keyframes and map information with the central platform. The central platform utilizes the data contributed by multiple sub-unmanned platforms to establish accurate collaborative pose estimation and a consistent global environment map, optimizing the collaborative estimation through position identification, data association, global optimization. In addition, we lightweight the SLAM scale by removing redundant data to achieve accurate and efficient visual-inertial SLAM. Extensive evaluations on multiple public open-source datasets demonstrate that the proposed algorithm exhibits excellent localization accuracy, robustness and scalability.