For citrus trees cultivated by dwarf dense planting, the fruits are randomly distributed in space, which poses difficulties for mechanized picking. To improve picking efficiency, a citrus picking system based on dual robot collaboration is designed and a global scanning picking scheme that scans the entire fruit tree to achieve orderly fruit picking is proposed in this research. The dual-robot calibration with the iterative method and closed-form method is completed in the research. The picking problem is attributed to the single traveling salesman problem, and the trajectory planning and the picking task are completed with genetic algorithm in the research. The picking experiments are designed in the research. As a result, in the picking experiments, the average time for planning the picking sequence of citrus was 0.1184 seconds. In each group of citrus picking experiments, the total picking time averaged 158.9 seconds, the picking success rate is 82%. The picking results show that the built dual-machine system can effectively complete the picking task. The proposed dual-robot picking system can provide a reference for the establishment of other picking robot system.