The work aims to design a picking robot that can replace manual picking, in order to reduce the labor intensity of warehouse picking operations and improve the accuracy of picking. Based on PaddlePaddle's PP-ShiTu image recognition system, accurate identification of goods on shelves and rapid updating of categories were achieved. A method of "uncalibrated 3D vision+AGV motion control" was proposed for guiding shelf product grabbing to solve the visual grasping problem of low-cost robotic arms. The use of two-dimensional code autonomous navigation and intelligent lifting system realized the handling and three-dimensional grasping of goods on shelves. The experiment proved that the designed picking robot achieved accurate picking and handling of goods on shelves, with a test accuracy of 92.25%. The intelligent picking robot constructed based on this method can complete the picking and handling of goods on the shelves of the warehouse.