Abstract: Fruit and vegetable picking robots, as a core branch of agricultural robots, hold significant strategic importance in addressing labor shortages and enhancing the automation level of the fruit and vegetable industry. They have become a cutting-edge research hotspot both domestically and internationally. This paper systematically reviews the current global development trends of fruit and vegetable picking robot technology, deeply analyzes the technological breakthroughs and engineering practices in the evolution of typical complete machine systems from laboratory prototypes to commercialization, and focuses on the latest progress in top international journals and the industry in the past five years.The research systematically summarizes from two major dimensions: key mechanisms and core technologies. In terms of key mechanisms, it elaborates on the technical path of the evolution of the visual perception system from two-dimensional imaging to three-dimensional multi-modal perception, compares the differences in ranging range, accuracy, and environmental adaptability between passive stereo vision and active structured light and time-of-flight methods, and reveals the matching failure problem caused by texture loss in the unstructured environment of farmland. In the walking mechanism section, it deeply analyzes the terrain adaptation mechanisms of wheeled, tracked, and hybrid drive modes in different scenarios such as flat land, hills, and greenhouses, quantitatively compares their obstacle crossing ability, turning radius, energy consumption efficiency, and the impact of vibration on end precision. As the core working component, the end effector is scientifically classified into three mainstream schemes: negative pressure adsorption type, tool shearing type, and flexible grasping type. It systematically expounds their working principles, grasping-picking coordination strategies, and adaptability in different fruit and vegetable scenarios such as apples, strawberries, and tomatoes, and focuses on comparing the damage rate, single fruit operation time, success rate, and crop characteristic constraints of each scheme.In the core technology aspect, it comprehensively reviews the evolution process of navigation and positioning technology from single GNSS to multi-sensor fusion SLAM, and analyzes the robustness improvement strategies for satellite signal occlusion in orchard dense planting environments, such as the tight coupling method of vision-inertial-wheel speed. In the target recognition algorithm section, it summarizes the technological revolution from traditional image processing to deep learning, focuses on analyzing the trade-off optimization mechanism of Faster R-CNN, Mask R-CNN, and YOLO series models in detection accuracy and real-time performance, and discusses the performance degradation problems under occlusion, drastic changes in lighting, and interference from similar color tones. In the aspect of mechanical arm path planning, it deeply evaluates the bottlenecks of swarm intelligence algorithms, rapidly expanding random trees, graph search, and artificial potential field methods in static obstacle avoidance and dynamic response, and reveals the motion redundancy and trajectory smoothing problems in multi-arm collaboration and mobile grasping.The research points out that the current technical bottlenecks mainly lie in five aspects: insufficient flexibility of the end effector leading to mechanical damage rate; poor dynamic adaptability of planning algorithms, with lagging responses to dynamic obstacles such as swaying branches and leaves; severe degradation of perception accuracy in complex environments; lack of multi-machine collaboration resulting in repeated or missed operations; weak all-weather operation capability, with dust-proof and water-proof grades and battery life unable to meet the continuous operation requirements of large fields.Finally, it emphasizes the urgent need to build an integrated perception-decision-control technology system dedicated to agriculture, and to break through existing technical barriers through paths such as collaborative design of software and hardware, multi-modal data fusion, training of agricultural large models, and swarm intelligence scheduling, promoting the development of picking robots towards high efficiency, low damage, and generalization, providing core equipment support for future smart agriculture, and achieving a leap from single-machine autonomy to swarm intelligence.