Development, integration, and field evaluation of an autonomous Agaricus bisporus picking robot

Abstract

The disorganized and densely growing Agaricus bisporus remains challenging for robotic harvesting. Aiming at the lower efficiency, lower success rate, and higher breakage rate in the robotic harvesting process, we propose a fully integrated, autonomous, and innovative harvesting robot to overcome the challenges due to the dense growth characteristics of Agaricus bisporus. An overlapping target detection vision system based on the yolov5s deep learning network is developed to improve the detection accuracy of dense and hidden mushrooms. An intelligent picking planning strategy by fusing multiple algorithms is formulated to guide the robot to effectively pick mushroom clusters with different growth characteristics. A multi-stage telescopic flexible end-effector is designed to pick mushrooms in a narrow space and reduce the picking breakage rate. Finally, we successfully developed a fully autonomous picking robotic system and evaluated it in a mushroom cultivation factory. Experimental results show that the robot can pick Agaricus bisporus continuously with an overall success rate of 94.1 % and an average picking time of 4.23 s per mushroom. These efforts provide a solid foundation for future applications of Agaricus bisporus picking robots.