Abstract: Lycium barbarum L. (L. barbarum), known as goji or wolfberries, is one of the berry species in natural healthcare products. Manual harvesting cannot fully meet the highly seasonal requirement of multiple harvesting cycles per season, due to the labor intensity. Mechanical harvesting machines that achieve high net picking rates tend to have low efficiency, whereas those designed for high efficiency often cause excessive fruit damage. It is often required to adopt mechanical harvesting in commercial production. In this study, a precise multi-point vibration harvester was developed using machine vision. Specifically, the harvester was also designed to be suitable for a double-layer hedgerow cultivation system, which was widely adopted in commercial orchards. Ripe fruit regions were precisely detected to cluster into suitable harvesting zones. The optimal vibration picking point was then determined for each zone. The low contact with ripe fruits reduced the fruit damage due to the high harvesting efficiency. Harvesting precision and efficiency were coordinately optimized for the higher performance than before. The key components included the multi-point vibration picking device, the three-axis harvesting platform, and the cross-row crawler chassis. The structural parameters were determined, including the layout, length, material, vibration frequency, and torsion angle of multiple vibration ends in the picking device, the working space of the three-axis harvesting platform and the architecture of the control system, together with the main structural and driving parameters of the cross-row crawler chassis. Collectively, the effective transmission of vibration energy was realized for the stable operation of the platform, as well as the precise interaction between the harvesting unit and the plant canopy. Additionally, a control system was developed for precise vibration harvesting, providing a feasible technical solution to intelligent mechanical harvesting. A three-factor and three-level orthogonal experiment was conducted to optimize harvesting performance. A systematic investigation was also conducted to explore the effects of excitation position, excitation time, and lateral offset on the harvesting rate of ripe fruits, the harvesting rate of unripe fruits, and the damage rate of ripe fruits. Mathematical models were established to quantify the relationships among these factors and the evaluation indicators. An optimal combination of parameters was determined as an excitation position of 55.14 mm, an excitation time of 3 s, and a lateral offset of 30 mm. Performance tests demonstrated that a single precise multi-point vibration harvesting unit achieved a ripe fruit harvesting rate of 90.85%, an unripe fruit mis-picking rate of 2.86%, and a ripe fruit damage rate of 5.34%, with an operational efficiency of 19.6 kg/h. In a two-sided, double-layer, four-station layout, the synchronous harvesting efficiency significantly increased to 78.4 kg/h, approximately 10 to 16 times that of manual harvesting, which was 5-8 kg/h. The precision multi-point vibration-based harvester shared excellent performance in selectively harvesting mature fruits, while leaving unripe fruits intact, indicating the high efficiency, low fruit damage, and precise mechanical harvesting. Overall, a practical and scalable technical solution can be expected for the intelligent harvesting. A combination of vision-based detection and precision vibration harvesting can also offer valuable insights to improve the quality and efficiency of commercial L. barbarum production, particularly with the low labor intensity. The findings can also provide dependable technical and equipment support for intelligent L. barbarum harvesting.