Abstract: For citrus picking, the hardware platform of picking robot composed of the upper computer, RealSense R200 depth Camera, VS-6556 vertical multi-joint industrial manipulator and three-finger flexible paw was constructed. Windows10 as the development environment, librealsense camera software development kit, OpenCV computer vision library, TensorFlow-GPU and Keras deep learning framework, ORIN2 manipulator control software development kit, Arduino IDE function library and SerialPort serial communication software development kit were adopted. The program design of picking control system based on depth camera and secondary development of mechanical arm included the module program design of visual identification and positioning, hand claw motion control, mechanical arm motion control and picking control. The pick control system citrus positioning and citrus picking experiment was conducted under the condition of random arrangement of citrus in the laboratory environment. The test result showed that the average positioning accuracy error of visual identification module was 1.22 cm, in the process of picking oranges to identify the success rate of 100%, the average recognition time was about 47 ms, citrus picking robot the success rate of 80%, the average picking time was about 15.2 s. The test result verified the feasibility of the platform control system program of the picking robot, and showed that the control system could correctly and efficiently complete the whole citrus picking process.