Design Of Intelligent Strawberry Picking Robot Based On ROS

Strawberry is one of the popular fruits in China,and is demanded increasingly from year by year.With the continuous development of intelligent agriculture technology,the elevated planting used in intelligent greenhouses is gradually being promoted.Smart greenhouses in intelligent agriculture mostly use multi-span greenhouses,and the use of the elevated cultivation mode can improve the utilization of space in the greenhouse.Therefore,the design and realization of a fully autonomous agricultural robot to complete the intelligent inspection and picking of strawberries grown on high shelves in intelligent greenhouses is one of the key equipment for future unmanned intelligent agriculture.In this thesis,the robot overall functions are planned and designed.The robot software is developed based on ROS(Robot Operating System)and running on the NVIDIA edge computing platform Jetson Nano.We completed the mechanical design,hardware selection,function realization,control of the whole system and its task planning,and then realized a strawberry inspection and picking robot prototype.Finally,strawberry picking experiments were carried out in a analogy greenhouse elevated cultivation environment to test the performance and verify the effectiveness of the algorithm.The robot uses the Mecanum wheels chassis to realize the movement function,and is based on the Move＿base of ROS to navigate and avoid obstacles.In order to realize the autonomous exploration of the unknown map,we study the existing SLAM algorithm and path planning algorithm and integrate RRT(Rapidly-exploring Randomized Trees)to detect the map boundary points,which guides the robot to navigate to the map boundary points and build a map cyclically.The robot uses a visual servo manipulator to perform the picking function,which is calibrated by applying the Eye-In-Hand method and implements the motion planning based on Move It of ROS.A bionic gripper as end effector was designed to realize strawberry picking.Considering the limitation of resources on the edge computing platform,after the test and comparison,the YOLOv4-Tiny target detection algorithm with less hardware resource consumption is selected to realize the identification and positioning of ripe strawberries.In training process,the sample data set is made and the number of samples is expanded by pictures rotation to avoid over-fitting.Based on the identification and positioning of strawberries,the spatial position of the strawberry stalk is calculated,and the bionic gripper is used to further realize the clamping and shearing of the strawberry stalk,so as to achieve the goal of non-destructive strawberry picking.The feasibility and reliability of the robot prototype have been verified in the analogy greenhouse elevated environment.The system shows some valuable application and can be further marketized and commercialized.