Initial experiments in robotic mushroom harvesting

Abstract The locating and picking performance of a robotic mushroom harvesting rig is assessed. The robot system has three main elements: a black and white vision system incorporating a mushroom locating image analysis algorithm, a computer-controlled Cartesian robot and a specialised mushroom picking end-effector. In 18 experiments, on 815 mushroom targets, 689 (84%) were located by the image analysis algorithm and 465 (57%) were picked successfully. Overlapped, small mushrooms and closely packed, touching mushrooms proved most difficult to pick using twist as the primary detachment method. The paper concludes that considerable improvements to picking performance could be achieved by using bend as the primary detachment method and by developing a suitable picking strategy to predict bend direction and picking order.