A multi-sensor LiDAR-visual-inertial SLAM method based on adaptive geometric observer

In recent years, the field of low-altitude technology and engineering (such as UAV autonomous flight, low-altitude inspection, aerial robots, etc.) has placed higher demands on high-precision and highly robust autonomous localization and environmental perception, especially in typical low-altitude application scenarios such as narrow spaces, weak GPS signals, and complex environmental structures, which pose new challenges for robotic autonomous localization and mapping. Simultaneous localization and mapping (SLAM), as the core technology to achieve this goal, mainly integrates multiple sensors such as LiDAR, cameras, and IMUs to achieve high-precision and robust localization in complex environments. Among them, LiDAR-visual-inertial SLAM further enhances the system’s localization performance and environmental adaptability through deep multi-sensor fusion. However, existing LiDAR-visual-inertial SLAM systems are sensitive to parameter settings in different environments (such as narrow or open areas), and fixed parameters are difficult to accommodate diverse scenarios. Moreover, LiDAR and cameras may not always provide high-quality localization information simultaneously in certain environments. To address these issues, this paper proposes a multi-sensor LiDAR-visual-inertial SLAM method based on an adaptive geometric observer. The proposed method designs a multi-strategy adaptive parameter adjustment mechanism, which can dynamically optimize system parameters according to point cloud distribution characteristics and the degeneration status of LiDAR-inertial odometry (LIO) and visual-inertial odometry (VIO) subsystems, significantly improving system robustness in diverse low-altitude environments. Meanwhile, to address the varying effectiveness of LiDAR and visual information in different scenarios, a hierarchical adaptive geometric observer weighting method is proposed, which adaptively fuses the outputs of LIO and VIO subsystems based on their degeneration status, further improving pose estimation accuracy. Experimental results show that the proposed method achieves excellent localization accuracy and robustness on multiple public datasets, reaching the current international advanced level. It provides strong technical support for autonomous localization and navigation in complex environments in the field of low-altitude technology and engineering, and has significant engineering application value and prospects for promotion.