A Distributed SLAM with UWB-VIO Fusion Framework for Enhanced Relative Localization of Multi-UAVs System

The most critical challenge of multi-UAV (unmanned aerial vehicles) systems accomplishing a common mission is to maintain relative localization among UAVs. However, it remains an open and challenging research topic in infrastructure-less environments. In this paper, we propose a novel distributed visual-inertial ultra-wideband (UWB) fusion framework for global localization that overcomes the problem of state estimation in GPS-denied areas, which reduces computational issues confronted by centralized Simultaneous Localization and Mapping (SLAM). Firstly, our solution integrates a UWB ranging front-end on each UAV for ranging and uses Visual-Inertial Odometry (VIO) to estimate the self-motion of each UAV. Using a front-end that fuses UWB measurements with VIO, we correct potential drift issues caused by VIO and perform distributed loop closure detection among multiple UAV s to achieve relative pose estimation. Additionally, we propose a consistency measurement filtering method with the adaptive threshold in three-dimensional space to improve the accuracy of state estimation under continuous non-line-of-sight (NLOS) conditions. Lastly, by using distributed outlier rejection to remove erroneous estimates and fusing measurements from both front-end and back-end pose graph optimization modules, we employ a sliding window method to recover the complete trajectory and guarantee global consistency of pose estimates. We believe our method provides a comprehensive solution for the global localization in multi-UAV systems, eliminating the environmental constraints of previous positioning methods and providing opportunities for other researchers to apply UAV cluster motion planning techniques in complex environments.