Research On Simultaneous Location And Mapping Algorithms Of Planetary Vehicles Based On Lidar Loop Detection

The structure of planetary surface is complex and the environment is harsh.To operate in such an unknown environment,the localization and navigation of the rover is the guarantee of all actions.As an agent robot,the rover carries many sensors,such as vision sensor,laser sensor,inertial sensor,odometry and so on.Among them,lidar is used in many robots because of its high accuracy and small environmental impact.The position and attitude information can be estimated incrementally by means of laser radar scanning matching,but there exists accumulation of errors.However,with the addition of loop detection,errors can be effectively limited.Therefore,on the basis of previous studies,to improve the localization accuracy of planetary rover,this paper deeply studied the laser SLAM based on loop detection.Firstly,the general kinematics model of mobile robot is established based on the differential wheeled robot.Aiming at the inaccuracy of odometer information,the direct linear calibration method and model-based calibration method are studied.Considering the data distortion of lidar caused by motion,a motion distortion processing module is added and the corresponding modules are discussed theoretically and experimentally.In this method,the structure of graph optimization is adopted,the concept of submap is introduced in the front-end,local map is built by probability raster,scan-to-map matching is applied to find the optimal position of laser data in the map and odometry information is taken as an initial estimation,then an incremental laser odometry is constructed.Each sub-map contains a certain number of laser frame point clouds,and takes them as the optimized structure node of graph structure.The back-end uses frame matching to construct corresponding constraints.To solve the problem of large matching errors caused by discontinuous point clouds and scattered distribution,an implicit moving least square method is proposed.The implicit surface in the local continuous point clouds is calculated by sliding window,and the implicit matching points are calculated by normal vector projection of the surface.The results are compared with the traditional ICP method to verify it superiority.Finally,the related methods are tested in simulated and real environment.Using the existing Jackal mobile robot in the laboratory,the robot model is added to the ROS + Gazebo simulation environment with planetary surface scene to complete the algorithm test.The real environment test is carried out under the motion capture system.At the same time,a comparative experiment without loops is added as well.The estimated trajectories in both environments are compared with the corresponding ground truth.The test results show that the trajectory obtained by this method is close to the real trajectory tightly and has a high accuracy of localization estimation.