Research On Tightly Coupled 3D Lidar Inertial Odometry Algorithm In Unmanned Vehicles

As society’s demand for travel increases and research on autonomous driving technology continues to advance,autonomous unmanned vehicles have become a current trend of development,of which,localization technology is one of the cores to realize autonomous driving of unmanned vehicles.Simultaneous Localization and Mapping（SLAM）technology enables an unmanned vehicle to locate itself and build a map of the environment based on the data provided by its own mounted sensors while moving in an unknown environment without a priori information.3D LiDAR has the advantages of high measurement accuracy,independent of sunlight,etc.,and is widely used in unmanned vehicles for simultaneous localization.It is widely used in the technology of simultaneous positioning and map building for unmanned vehicles,but it has poor robustness.Lidar-based localization is challenging due to the unstructured scenes in park with unstable features,bumpy and curvy roads,and dynamic environmental objects.To address the above challenges,a novel hybrid localization method that combines filter-based tightly coupled lidar-IMU odometry with optimization-based SLAM is proposed.Therefore,fusion of 3D LiDAR information with other sensor information to improve the robustness of localization is a hot research topic in recent years.In order to improve the robustness,accuracy and efficiency of unmanned vehicle localization,a tightly coupled localization algorithm based on IMU and LiDAR is proposed and investigated in this paper.The specific work is as follows.The theoretical foundation of the IMU and LiDAR tightly coupled positioning system framework and laser SLAM system proposed in this paper is introduced.The working principle of the new solid-state LiDAR is briefly introduced,as well as the detailed study of the measurement model of LiDAR and IMU,and on this basis,the IMU pre-integration method and the updated zero-bias IMU pre-integration formula are derived in depth.Aiming at the problem of motion distortion in 3D LiDAR during unmanned vehicle motion,we improve the quality of the original LiDAR point clouds by removing motion distortion in the point clouds and removing dynamic obstacle points in the point clouds with the assistance of IMU for 3D LiDAR.A novel iteration-based point cloud storage data structure is innovatively used,and based on which a fast 3D LiDAR front-end direct matching process is proposed,a fast ground point cloud extraction and segmentation method is proposed as the ground constraint term for global optimization,and the IMU pre-integration is used to give the track projection between key frames as the IMU constraint factor.For the cumulative error generated by 3D LiDAR front-end odometry and the error generated by IMU pre-integration due to zero-bias update,an iterative error-state Kalman filterbased local optimization method for tightly coupled 3D LiDAR and IMU is proposed to correct the IMU pre-integration while suppressing the cumulative error of 3D LiDAR.For the large amount of 3D LiDAR point cloud data,a key frame strategy is proposed,which effectively reduces the number of global optimization and map construction of the back-end factor map and reduces the complexity of the algorithm.To address the limitation that the local optimization only depends on the previous frame,the 3D LiDAR and IMU global optimization method based on factor map optimization is used to obtain the combined system positional and global maps by using 3D LiDAR observations,IMU pre-integration,ground constraints and loopback detection as constraints.For the loopback detection,a pupil heuristic loopback detection descriptor is proposed to balance the computational power and accuracy of the algorithm,and a combination of coarse and fine detection is proposed to effectively reduce the computational power of the closed-loop detection module and improve the accuracy of the closed-loop detection,which improves the stability of the tightly coupled IMU and 3D LiDAR positioning system.An unmanned vehicle application platform equipped with 3D LiDAR and IMU is built,and the proposed tightly coupled localization method based on IMU and LiDAR is verified by comparing the results of simulation experiments based on the campus scene dataset of Guangdong University of Technology with RTK-GPS data as the true value and with open source algorithms Livox-mapping and LINS-Livox.Finally,the proposed scheme in this paper is applied to an unmanned vehicle platform and tested in real operation.The results show that the IMU-LiDAR tightly coupled localization method can obtain high accuracy localization results and global maps in small,large and non-closed-loop scenes,improve the stability and robustness of the LiDAR-IMU fusion SLAM system,and achieve the decimeter-level localization effect comparable to the civil GNSS localization system in the non-closed-loop outdoor large scenes and the decimeter-level map building effect in the closed-loop scenes.The map building effect can be achieved in closed-loop scenes.A series of experimental results show that our proposed novel hybrid localization method is effective.