GNSS-RTK adaptively integrated with multi-robot LiDAR SLAM for efficient mapping in large areas

Abstract

Multi-Robot Cooperative Simultaneous Localization and Mapping (SLAM) extends traditional single-robot SLAM by enabling multiple robots to collaboratively map environments and localize themselves. Recent advancements have integrated The Light Detection and Ranging (LiDAR) and Inertial Measurement Unit (IMU) data through factor graph optimization (FGO) named 3D LiDAR-based frameworks for multi-robot SLAM. While LiDAR and IMU provide accurate short-term motion estimates, they suffer from drift over time, especially in large areas. The Global Navigation Satellite System Real-time Kinematic (GNSS-RTK) offers precise absolute positioning but is less effective in urban canyons. This thesis addresses these challenges by proposing an adaptive integration of GNSS-RTK with LiDAR and Inertial Odometry (LIO) to enhance mapping efficiency and continuous positioning in urban environments. The proposed method assesses GNSS-RTK solution quality using the incrementally produced point cloud map from LIO, with the mean elevation angle (MEA) mask indicating the openness of the surrounding area. A smaller angle suggests a more open area and a more reliable GNSS-RTK. Global FGO merges reliable GNSS-RTK data with LiDAR and inertial odometry (LIO), incorporating global constraints to counteract pose drift. Testing in Hong Kong's urban canyons demonstrated significant improvements, reducing absolute pose error (APE) by over 75% compared to conventional methods without GNSS-RTK. In the two-stage global and local graph optimization, inter-robot constraints are used to determine transformations between robot coordinate systems and are converted to virtual intra-robot constraints. The effectiveness of these virtual constraints in multi-robot SLAM integrated with GNSS-RTK is evaluated, showing varying contributions across datasets. To optimize local pose graphs, correcting inter-robot constraints is necessary to ensure their positive contribution. This work offers a promising approach for efficient and accurate mapping in complex urban environments, with significant implications for future large-scale mapping projects.