Research On Environment Map Construction And Navigation Of Orchard Robot Based On LiDAR

Orchard operation robots are applied to unstructured orchard environment for intelligent and refined management of orchards,which can improve fruit production efficiency and reduce fruit farmers’ labor intensity.orchard environment map construction and robot autonomous navigation is one of the key technologies in the research related to orchard robots.In order to optimize the robot’s point cloud map construction and autonomous navigation method in orchard environment,the orchard robot is taken as the research objectin this paper and research on point cloud map construction,fruit tree segmentation and canopy information extraction is carried out.In addition,research on Robot repositioning and autonomous navigation in orchard environment is discussed.The main research contents are shown as follows:1)According to the requirement for high-precision point cloud map construction in semi-structured orchard scenes,a mobile scanning system based on LiDAR＿IMU was designed,which can be deployed on large crawler orchard robot platforms.A lightweight iterative error Kalman filter estimator is used to estimate the LiDAR attitude.The point cloud map is refined and constructed based on the rotation constraint algorithm.The experiment results show that the robustness and accuracy of the point cloud map construction method is effective.The error of the map in each of the coordinate directions is less than 14.8 cm,17.4 cm,24.9 cm,which can meet the needs of fruit tree morphology measurement or robot navigation and positioning.2)A canopy information extraction method based on point cloud map is designed for the demand of fruit tree morphology measurement in orchard environment.Firstly,we use various ground filtering algorithms to ground filter the orchard map,and found that the orchard map processed by CSF(Cloth Simulation Filter)ground filtering algorithm can filter out the ground point cloud well,and the ground filtering rate is 92.54%,which is 10.88% and 26.49% higher than PTDF(Progressive TIN Densification Filtering)and PMF(Progressive Morphological Filter)algorithms respectively;Secondly,to accurately segment fruit trees within an orchard map without ground point clouds,a Euclidean clustering point cloud segmentation method using SOR(Statistical Outlier Removal)filtering fused with KNN(K-Nearest Neighbor)is designed,which is able to accomplish the task of segmenting fruit trees connected by canopy with a segmentation success rate of over 80%.Finally,to accurately extract the point cloud model of fruit trees from the segmentation results,a point cloud classification system using FV(Fisher Vector)-Incpetion network was used to classify the point clouds in the orchard and calculate the canopy volume and tree height of the fruit trees.3)To ensure that stable and accurate attitude information can be obtained by the robot in the orchard,a robot repositioning method based on an a priori map of 3D feature points was proposed.In this method,GNSS(Global Navigation Satellite System)information in the ENU(East-North-Up)coordinate system to obtain the initial attitude of the robot based on the a priori map.The point cloud with attitude information output by the iterative error Kalman filter state estimator was used to establish a distance-based loss function with the a priori point cloud of a specific region.The distance-based loss function of the map was established,and an optimization equation based on the Gaussian Newton algorithm was constructed using the Ceres-Solve nonlinear optimization library to minimize the loss function in order to obtain the robot’s posture information based on the a priori map.Experimental results show that this method can ensure the robot obtaining relocalization information in the orchard map,and the positioning system can still work when the orchard environment changes to a certain extent.Compared with GNSS as the accuracy verification of the robot positioning,the maximum positioning deviation is 19 cm and the average positioning deviation is 7.5 cm.4)In order to achieve the reliable navigation of the robot in the orchard,the orchard navigation research using costly maps was carried out.Firstly,the prior map was preprocessed by direct filtering and PCA(Principal Component Analysis)to obtain part map of the orchard for path planning.The division of tree rows in the orchard is completed according to the map direction,and the least squares method merged the A*algorithm is adopted to complete the global operation path planning in the local map.The robot is realized by the TEB(Time-Eletic-Band)local path planning algorithm path planning to ensure that the robot can complete navigation according to the work path.In the results,this method can ensure successful full-coverage navigation operation of the robot in the orchard with complex road conditions.Under the velocity of 0.4m/s～1.0m/s,the robot can complete the orchard operations.At this time,the average value of longitudinal deviation obtained by the system was 26.7 cm,and the maximum value is below 46.2 cm;the average value of heading angle deviation was 4.09°,and the maximum value was below 8.65°.