Research On Mapping And Localization Algorithm Of Unmanned Vehicle Based On Lidar And Integrated Inertial Navigation

Multi sensor fusion mapping and positioning technology can improve the accuracy and robustness of autonomous positioning and navigation for unmanned platforms,and has a broad application prospect in the fields of autonomous driving,robotics,unmanned aerial vehicles,and other fields.With the continuous development of autonomous driving technology,multi sensor fusion mapping and positioning technology has become one of the hottest research topics.Based on the national key research and development plan project “Research and Practice on Group Intelligent Consignment Vehicles for Airport Luggage Transfer”(2021YFE0203600),this thesis studies a mapping and positioning algorithm based on the fusion of lidar and integrated inertial navigation for outdoor unmanned vehicles.The specific research work is as follows:(1)A front-end odometry algorithm based on point cloud direct matching was designed to solve the problem of adapting to different manufacturers’ lidar,directly used the down-sampled current frame point cloud and the distance from the point to the surface of the local map as residuals,and uses IMU integration values as initial values for optimization using the LM method.This not only improves the accuracy and robustness of lidar odometry but also makes it easier to adapt to different types of lidar from different manufacturers,which is conducive to practical engineering applications.(2)A solution to the z-axis drift problem common in most lidar and lidar-inertial odometry was presented by introducing GNSS elevation information as a constraint.Compared with LIOSAM,which has good performance for lidar-inertial odometry,the z-axis drift problem is greatly improved.(3)A mapping system based on lidar and combined inertial navigation was constructed.Constraint factors such as lidar odometry,IMU pre-integration,loop closure,and GNSS odometry were used to established a factor graph model to reduce cumulative error during backend optimization and output smooth trajectory.(4)In the case of a priori point cloud map,the Normal Distributions Transform algorithm is commonly used for point cloud matching and localization in engineering.To address the issues of lidar odometry drift and low pose output frequency during registration,the ESKF was used to fuse point cloud and IMU data,using the pose obtained from IMU inertial calculation as the prediction value and the pose obtained from matching lidar point cloud and prior map as the observation value,to output real-time and high-robustness,high-precision,and high-frequency localization information.(5)Experimental verification was carried out on the adaptability,accuracy,and robustness of the algorithm designed in this thesis.Six sets of different radar data were used to verify the adaptability of the mapping algorithm.During the real vehicle testing phase,calibration was performed on the inertial measurement unit and lidar for timing,as well as the intrinsic and extrinsic parameters and data parsing.The algorithm system framework is based on the Ubuntu18.04 and ROS Melodic software architecture,and is implemented in C++.Mapping and fusion positioning algorithms were fully tested on dataset and data collected from a real vehicle,and the experimental results showed that the mapping and positioning algorithms designed in this thesis have high accuracy and robustness,and can be quickly deployed on a real vehicle and easily applied in engineering applications.