This paper presents a robotic mushroom harvesting solution, consisting of an actuated scanning vision system integrated into a gantry robot. The system is capable of performing segmentation and pose estimation of the mushrooms on Dutch shelves commonly used in growing farms worldwide. The vision system employs an active stereo RGB-D camera able to capture a 360° scene of the mushroom bed, providing a high quality reconstruction of the mushroom caps. The YOLOv5 algorithm is used for the detection and size classification of the mushrooms, while a two-step model-fitting method is developed for the pose estimation task. The actuated carriage is compact, designed for operation in real mushroom-growing farms and intended to be used together with a soft gripper. The robot has five actuated degrees of freedom (DoFs), three for the linear motion on the shelves, and two DoFs for achieving the desired orientation for the gripper. In a real harvesting scenario, the robot sequentially scans the selected areas and accurately places the gripper in the appropriate angle of attack utilising our pose estimation method together with our visual servoing module for minor adjustments. The results were promising on all trials using 3D printed white button mushrooms on real soil.