| Tomato picking is an important part of the tomato production process.With the year-byyear expansion of tomato planting area,aging population and reduction of agricultural labor force,it is of great significance to develop tomato harvesting robots and to study tomato harvesting methods.In the process of tomato picking,the robot is required to quickly identify the picking point of tomato stalks,and have good environmental perception and obstacle avoidance movement planning capabilities to complete effective picking.This paper takes tomato bunches as the research object,constructs a tomato bunch harvesting robot system,studies the tomato bunch harvesting methods in actual scenarios,and proposes a motion planning algorithm based on CTB-RRT*,which is verified by Gazebo simulation experiments and real robotic arm picking experiments.The effectiveness and feasibility of the tomato cluster harvesting method.The main research work of this paper is:(1)Building a tomato cluster harvesting robot system.Using the integrated end effector,depth camera,mobile platform and Aubo-i3 manipulator in ROS combined with a variety of algorithms to achieve the platform to build a tomato string collection robot system.According to the characteristics and requirements of bunch tomato harvesting in the greenhouse environment,the optimal installation position of the robotic arm,the picking method of the robotic arm and the movement method of the mobile platform are determined.(2)The location of the picking point of tomato bunches and the establishment of the point cloud model of the picking environment.First,the coordinate conversion relationship between the camera and the robot is obtained through hand-eye calibration,and then the picking point coordinates are obtained using the YOLOv4 string tomato recognition and positioning method,which is used as the target pose of the robot arm motion planning.Finally,the point cloud data of the picking scene is obtained through the visual sensor,and the point cloud data is rasterized to obtain the discretized model of the picking scene,which is used as the obstacle space for the motion planning of the robotic arm.(3)Propose CTB-RRT* motion planning algorithm.Aiming at the problems of low efficiency,long time and high path cost of traditional sampling planning algorithms,three improvements are made on the basis of the RRT*-connect algorithm,and the CTB-RRT*algorithm is proposed.The improvement is: through the method of Cauchy distribution Carry out heuristic sampling to reduce the blindness of sampling;introduce target gravity,and dynamically adjust the step length of random growth direction and target direction to improve local search speed;introduce node rejection strategy to eliminate unnecessary sampling nodes and improve computational efficiency.(4)Experimental research on tomato cluster harvesting.The tomato cluster collection experiment was carried out on the built robot platform,and experimental results show:the cluster tomato recognition and positioning algorithm based on YOLOv4 was used for target recognition.The average recognition and positioning time was 35.58 ms,and the recognition rate was 94%;the CTB-RRT* algorithm was used to plan a collision-free path The average planning time is 0.34 s,and the planning success rate reaches 98%;the average picking time is21.6s,and the overall picking rate reaches 92%,which meets the requirements of real-time picking and verifies the feasibility of the proposed picking method.Compared with the RRT*-connect algorithm,the designed CTB-RRT* algorithm path cost is reduced by 12.6%,the running time is reduced by 69.2%,and the number of expansion nodes is reduced by 76.3%,which greatly improves the real-time performance of robotic arm picking planning. |
PDF Full Text Request