RA-LIO: A Robust Adaptive Tightly-Coupled Lidar-Inertial Odometry

This paper proposes a robust adaptive lidar-inertial odometry (LIO), which achieves fast robot localization and mapping by fusing high-frequency IMU and low-frequency Lidar based on a tightly-coupled iterated error state Kalman filter (IESKF) framework. Specifically, to solve the problem of excessive data difference between IMU frames and poor distortion correction caused by the mobile robot's aggressive motion, we propose a fast adaptive interpolation motion compensation method, which accurately restores the true structure of the environment through reasonable interpolation. Furthermore, to address the problem of degeneration in robot localization in complex environments, we propose a degeneration analysis method based on constraint normal vectors to enhance the system's perception capability in degraded scenes, and mitigate the adverse effects of degeneration through an adaptive voxel filter method. Experimental results on public datasets and our own datasets demonstrate that our system outperforms state-of-the-art systems in terms of localization accuracy and the ability to handle degraded scenes. Additionally, our system also provides the method for relocalization.