VIR-Fusion: A Robust Visual-Inertial-UWB Fusion SLAM System for Corner Cases

Accurate and robust state estimation through multi-sensor fusion is a critical research area with wide-ranging applications. In this work, we introduce VIR-Fusion, an optimizationbased Visual-Inertial-UWB (Ultra-wideband) Simultaneous Localization and Mapping (SLAM) system. The framework incorporates robust techniques for detecting and handling sensor anomalies. These methods effectively manage corner cases for each sensor, ensuring overall system accuracy. To mitigate scale drift under zero-velocity conditions, a dedicated zero-velocity detector is employed to reliably identify the system’s motion state, enabling zero-velocity updates and further enhancing robustness. By integrating visual, inertial, and UWB measurements within a unified factor graph, the system provides accurate and reliable indoor localization. Experiments on both publicly available datasets and real-world environments demonstrate that VIR-Fusion outperforms existing SLAM approaches, particularly in challenging scenarios.