Research On Human-computer Cooperative Navel Orange Picking Robot Based On Somatosensory Interaction

In the agricultural production of fruit picking is an important step,at present the navel orange fruit picking operation is mainly by hand picking,but manual picking navel orange efficiency is low,especially in large-scale planting of orchards,at the same time,in recent years,China’s economic and social development is very rapid,the surrounding cities for the rural areas of the population siphon effect is obvious,coupled with the impact of population aging,so that people engaged in agricultural production In recent years,the economic and social development of China has been very rapid,and the siphoning effect of the surrounding cities to the rural areas is obvious.In this context,the development and design of an intelligent agricultural picking robot can effectively alleviate the problem of insufficient agricultural labor,improve efficiency,reduce production costs and improve the agricultural production environment.Facing the above-mentioned actual situation,this paper designs a flexible end-effector based on the bionic fin effect for the physical parameters of navel orange fruits,and a pneumatic system for nondestructive picking according to the demand of nondestructive picking;improves the recognition of navel orange fruits by Convolutional Neural Networks(CNN)for the complex characteristics of the actual picking operation environment.A ZED binocular camera was used for spatial localization of the fruits;a somatosensory control mode was designed for fruits that could not be recognized by the mechanical vision system;the IRB120(ABB Engineering Shanghai Ltd)six-degree-of-freedom robot arm was used for the actual needs of fruit picking,and kinematic analysis and trajectory planning were conducted for the robot arm.The specific research contents of this paper are as follows.(1)A flexible end-effect actuator based on the bionic fin effect is designed for the physical parameters of navel orange fruits and a pneumatic system for nondestructive picking is designed according to the demand of nondestructive picking.In order to determine the physical parameters of navel orange fruit,this paper conducted 100 sets of experiments on navel orange fruit based on Guanglu vernier caliper,Xiangshan scale electronic scale and TA.TOUCH texture meter respectively,and the damage limit pressure of navel orange fruit was finally determined to be 21.48 N through the experiments.based on the physical parameters of navel orange fruit,this paper carried out Solidworks modeling of the bionic structure of the end-effector and Ansys pneumatic control system of the end-effector is designed in hardware and software.(2)For the complex characteristics of the actual picking operation environment,the convolutional neural network is improved for the recognition of navel orange fruits,and the spatial localization of fruits based on ZED binocular camera is investigated.In order to improve the correct rate and efficiency of fruit recognition,based on the convolutional neural network Xception framework,the compressed channels further use the average pool to aggregate the feature maps pooled by channels,and change the hybrid attention connection to parallel connection;in order to localize the spatial coordinates of the fruit,first experiments are conducted based on the improved algorithm on the fruit images in the actual environment,and then the ZED binocular camera is used to reconstruct the the 3D coordinate system where the fruit is located.(3)For the fruit that cannot be recognized by the mechanical vision system,we designed a body sensing control system,using the MPU6050 sensor to collect hand information,and wrote a body sensing control program in STM32F103RCT6 to control the pneumatic system to realize the function of human-machine collaboration to control the picking.(4)For the practical needs of fruit picking,the IRB 120(ABB Engineering Shanghai Ltd)six-degree-of-freedom robotic arm is used in this paper.In order to verify the working performance of this robot arm,firstly,the standard D-H method is used to mathematically model this robot arm,then the forward and inverse kinematics of this robot arm is solved using the analytical method,and the kinematic simulation analysis is carried out by Matlab.Finally,the workspace and path planning of the robotic arm are simulated based on the mathematical model.(5)In order to test the actual working performance of the somatosensory interactive picking robot designed in this paper,a simulated picking experiment was conducted on the robot.40 simulated picking experiments were conducted using the fruit model,32 times of automatic picking,27 times of successful picking,with a picking success rate of 84.4%,8times of human-machine collaborative picking,6 times of successful picking,with a picking success rate of 75%,and an overall picking success rate of 82.5%.The test results show that the mechanisms of the picking robot can coordinate and meet the actual picking requirements.