Abstract: In large-scale planting bases, flower crops such as Lonicerae japonicae and Tagetes erecta are characterized by high bloom density and concentrated single flowering periods．Traditional agricultural mechanized harvesting methods suffer from high damage rates and high omissions when separating high-density flower clusters．Therefore, a harvesting machine suitable for multiple flower crops has been designed, employing YOLOv8s-seg model to achieve efficient and precise navigation path segmentation and flower recognition．A vision-guided robotic arm with interchangeable end-effector simulated manual harvesting motions to achieve low-damage harvesting of multi-species floral crops．Path planning strategy combined A* algorithm for global navigation with an improved DWA algorithm for local obstacle avoidance．A dynamic PID trajectory tracking method was proposed, which improved harvester's walking stability and environmental adaptability compared to traditional PID algorithms．Field experiments confirmed that this harvester achieved high picking accuracy and efficiency with minimal plant damage, and meeting large-scale and continuous harvesting operations requirements．