Key Technologies Of Robotic Apple Harvester In Constructed Orchard

Apple harvesting is characterized by large labor demand,high labor intensity,and obvious seasonal characteristics.The research and development of the apple harvesting robot are of positive significance to the automation and intelligent upgrading of the fruit industry and to cope with the increasingly emerging trend of aging.At present,the research of the main parts of the apple harvest robot is relatively independent,and the problems of the apple identification and positioning and the attitude control of the apple harvest robot restrict the operation effect of the apple harvest robot.Therefore,this paper takes the apple harvesting robot as the research object and carries out research on the picking posture optimization of the apple harvesting robot,involving the interaction between end-effector and fruit,deep learning and fruit recognition,as well as the function realization and experimental verification of the manipulator's control.The main research contents and conclusions of this paper include:(1)The finite element model of branch-stem-fruit was established to provide a theoretical basis for grasping and damage evaluation of end-effector and optimization of picking action.The branch and stem models were established based on the transverse anisotropic material constitutive model.The axial elastic modulus(E_bz)of the branch,radial elastic modulus(Ebr)of the branch,axial elastic modulus(E_sz)of the stem,and radial elastic modulus(E_sr)of the stem were 181.69±17.88MPa,29.81±4.02MPa,106.42±13.30MPa and 10.34±3.59MPa,respectively.The flexural shear modulus of stem and stem were 337.05±66.52MPa(G_rb)and46.90±11.62MPa(G_rs),respectively.Based on the constitutive relation,the isotropic Poisson's ratio(?_XY)of branch and stem was 0.4,and the estimated axial,torsional shear modulus of the branch(G_{b XY})and stem(G_{s XY})were 10.65 MPa and 3.70 MPa,respectively.The estimated Poisson's ratio of anisotropic plane of the branch(?_br)and stem(?_sr)was 0.20and 0.11,respectively.The separation layer between branches and stems was modeled based on the cohesion model.The maximum traction force(T_i,max),maximum traction displacement(?_m)and fracture energy release rate(G_{i C})of the separation layer are 5.01±0.68 MPa,0.29±0.04 mm and 0.71±0.04 k J/m²,respectively.The comparison of the finite element simulation results of the branch-stem model with the experimental results by Abaqus software showed that the established finite element model could directly predict the fracture process of the branch-stem joint.(2)The effects of different picking methods on fruit separation were analyzed.The effects of fruit separation load and the shape characteristics of fruit-stem-fruit on picking effect were discussed through picking experiments of four basic picking movements:horizontal pulling,vertical pulling,bending and twisting.During the picking process,both horizontal and vertical tension is the main factor that causes the separation of fruit.The tangential force required to separate the fruit is significantly less than the normal force required.When the stem is perpendicular to the branch,the vertical pull may cause the stem to pull out.Fruits with longer stems require greater displacement and angle to break the branch-stem connection,increasing the risk of failure in harvesting.The branch-stem-fruit finite element model was used to simulate the experiment to analyze the effects of four factors(horizontal velocity,vertical velocity,angular velocity,and torsional angular velocity)on the required detach force.The optimized results indicate that horizontal pulling with bending and twisting motion is the potential optimum combination,which can provide a reference for the path planning of the robotic apple harvester.(3)Based on the finite element method,the grasping ability of the flexible three-finger FRE end-effector was analyzed,and the possibility of fruit damage was evaluated.By means of variance analysis and response surface analysis,the significant effects of soft finger material hardness,fruit centroid to palm distance,and fruit size on the tensile force were discussed,and the parameters were optimized.The results showed that TPU hardness,fruit centroid to fruit distance,and fruit size were significant factors affecting the grip-ability of the end-effector.The TPU hardness was determined to be 90HA,and the distance between the fruit center of mass and the palm was 65 mm,using the maximum tensile force as the optimization index.The test results were consistent with the predicted results,and the correlation coefficient R²=0.8221.The damage plasticity model of apple cortex was established by tensile and compression tests.The maximum Mises stress was 0.159 MPa,the maximum flesh stress was0.082 MPa,and the maximum contact pressure was 4.178 N in the simulation test of damage assessment during grasping.In the verification test,the thin-film pressure sensor was installed at the position of maximum contact force in the simulation test,and the maximum contact force was 4.572 N.The error between the simulation value and the actual maximum contact force is 8.62%.Simulation and verification tests show that the flexible three-finger end-effector can achieve lossless grasping of the apple.(4)Based on hand motion capture data and picking motion optimization simulation experiment,two kinds of picking motion were proposed.Move It was used in ROS to carry out path planning,and the feasibility was verified by simulation and experiment.The NOKOV motion capture system is used to collect the upper arm movements of people when picking apples,and the robot can"anthropomorphize"picking through path planning.Complete modeling and planning group configuration in Move It for XARM 5Lite.The RRTConnect algorithm in the open-source motion planning library was used to simulate and test the motion capture and the"horizontal pull and bend"motion path planning.The results show that the path planning of the manipulator based on Move It can achieve the expected goals.(5)The developed apple harvesting robot integrates modules such as vision system,control system,end-effector,and manipulator's arm to realize automatic harvesting function.The laboratory and orchard tests show that the apple harvesting robot has high reliability.The fruit orchard experiment results showed that the total success rate of the apple harvesting robot was 81.60%,among which the success rate of the picking method with anthropomorphic action was 80.17%,and the success rate of the picking method with"horizontal pull and bend"action was 82.93%.The reasons for the failure of picking were deep misalignment,separation failure,and blocked grasp.The average maximum separation power of the fruits in"horizontal pulling and bending"and"anthropomorphic"picking actions was 10.12 N and 8.63 N,respectively.The dynamic load at the end of the anthropomorphic picking action is small,which reduces the overall load of the robot arm,but the success rate of picking is reduced due to the sliding of the fruit.The"horizontal pulling and bending"picking action time is shorter,the success rate of harvest is higher.Neither of the picking operations resulted in stem pullout or damage.Both picking methods show potential for use in harvesting robots.