Research on High-Frequency Odometry in Laser Slam Using Point-Submap Matching

Abstract

The stability of front-end odometry output is the cornerstone of real-time mapping and high-precision localization. However, in autonomous driving, factors such as sensor types, driving speed, and signal occlusion can impact system performance, particularly causing trajectory drift, reduced accuracy, and decreased reliability during high-speed travel. This study employs a point-by-point update approach for odometry, achieving high-frequency odometry output and mapping updates nearly matching the point sampling rate, effectively eliminating point cloud distortions and providing solid support for tracking high-speed motion and navigation localization. This method emphasizes maintaining registration accuracy and robustness under high-frequency output conditions. Moreover, by utilizing backward propagation to optimize the estimation of posture and position, and ignoring inter-frame processing, this research implements an iterative extended Kalman filter, achieving tight coupling with IMU data. Additionally, the use of the iVox data structure accelerates point cloud searching, enhancing system processing efficiency. Testing in various scenarios has demonstrated that our method exhibits superior performance and robustness across different scene data, showcasing its potential application in the field of autonomous driving.