Accuracy and Speed Improvement of Lidar Inertial Odometry and Mapping by using Gravity Estimation and Incremental Voxels

LiDAR-inertial fusion-based Simultaneous Localization and Mapping (SLAM) is becoming increasingly important for enabling autonomous vehicles and mobile robots to perceive and localize themselves within unknown environments. Currently, two predominant tightly-coupled LiDAR-inertial Odometry (LIO) methods are widely used: filter-based and graph-based optimization. While the former is faster, the latter is more accurate. LIO-SAM, known as one of the state-of-the-art graph-based optimization methods, still assumes that the gravity vector remains constant, affecting the accuracy of pose estimation. Additionally, it requires high computational resources, especially for large-scale environments, which causes slower runtime per scan. In this paper, based on conventional LIO-SAM algorithm, we propose an enhanced LiDAR-inertial SLAM framework to solve the above issues. We integrate gravity estimation for better IMU motion prediction, resulting in improved pose estimation. Furthermore, we adopt the incremental voxels (iVox) data structure to replace traditional K-Dimensional tree, resulting in better runtime performance. Extensive experiments demonstrate that our proposed framework provides reasonable improvements about 30% in terms of both pose estimation accuracy and average runtime per scan.