Design and Experiment of Visual Feedback Control in Tomato Picking Bionic Manipulator

Abstract

Tomato picking has always been a challenge problem. The development of modern agriculture urgently requires the development of intelligent and efficient tomato picking robots. However, existing tomato picking robots lack intelligent feedback capabilities. In this paper, the picking control system of the tomato picking robot is proposed, and the visual feedback control module is designed in the self-developed bionic manipulator. In order to avoid empty picking, an image sensor is integrated in the bionic manipulator and the visual feedback control module is realized based on the YOLOv5s detection algorithm. The precision of the detection model trained on the dataset reaches 99.92%, and the recall reaches 99.89%. The results of picking experiments show that the picking success rate under the visual feedback control module reaches 96%, and the average picking time of a single tomato is 14.54s. Compared with the condition without visual feedback control module, the picking success rate increased by 6%, and the average picking time of a single tomato is reduced by 1.52s. Experimental results show that the proposed visual feedback module can effectively improve the success rate and efficiency of bionic manipulator picking.