LiDAR-Inertial Odometry (LIO) is typically implemented using an optimization-based approach, with the factor graph often being employed due to its capability to seamlessly integrate residuals from both LiDAR and IMU measurements. Conversely, a recent study has demonstrated that accurate LIO can also be achieved using a loosely-coupled method. Inspired by this advancements, we present a novel LIO method that leverages the recursive Bayes filter, solved via the Extended Kalman Filter (EKF) - herein referred to as LIO-EKF. Within LIO-EKF, prior and likelihood distributions are computed using IMU preintegration and scan matching between LiDAR and local map point clouds, and the pose, velocity, and IMU biases are updated through the EKF process. Through experiments with the Newer College dataset, we demonstrate that LIO-EKF achieves precise trajectory tracking and mapping. Its accuracy is comparable to that of the state-of-the-art methods in both tightly- and loosely-coupled methods.

翻译：激光雷达-惯性里程计通常采用基于优化的方法实现，其中因子图因其能够无缝融合激光雷达与惯性测量单元残差而被广泛使用。相反，近期一项研究表明，通过松耦合方法同样可以实现精确的激光雷达-惯性里程计。受此进展启发，我们提出一种新型激光雷达-惯性里程计方法，该方法利用递归贝叶斯滤波器并通过扩展卡尔曼滤波器求解——本文将其称为LIO-EKF。在LIO-EKF框架中，先验分布与似然分布通过惯性测量单元预积分以及激光雷达点云与局部地图点云的扫描匹配计算，并通过扩展卡尔曼滤波过程更新位姿、速度及惯性测量单元偏差。通过使用Newer College数据集进行实验，我们证明LIO-EKF能够实现精确的轨迹跟踪与地图构建。其精度在紧耦合与松耦合方法中均与当前最先进方法相当。