Abstract: The navigation and SLAM technologies based on multi-sensor fusion are currently the mainstream direction of development, and their research and applications in complex scenes have increasingly attracted widespread attention. Nevertheless, there exists a relative dearth of multi-sensor datasets specifically designed for complex scenes, particularly those that incorporate ultra-wide band (UWB) sensors. To facilitate users to test and verify multi-source sensor fusion algorithms in complex scenes and explore the shortcomings and potential development directions of multi-source sensor fusion and SLAM technology,both unmanned vehicles and drones are utilized to gather visual, inertial, and UWB data in seven diverse scenes, including dynamic scenes, non-line-of-sight scenes, and large-scale scenes. Furthermore, a high-precision optical motion capture system is employed to provide real-time six-degree-of-freedom true positions and orientations for the dataset. Finally, four state-of-the-art open-source algorithms, namely VINS-MONO, VINS-FUSION, VIR-SLAM and ORB-SLAM3, are utilized to conduct experimental verification and analysis on all scene sequences. The experimental results demonstrate that the data from all scene sequences are effective and usable.

Key words: multi-sensor fusion, dataset, complex scenes, UWB, unmanned platform