主催: The Japan Society of Mechanical Engineers
会議名: The 15th International Conference on Motion and Vibration(MoViC 2020)
開催日: 2020/09/09 - 2020/09/11
In recent years, the decline in the number of agricultural workers and their aging have become serious issues in Japan. To solve these issues, a new agricultural approach known as “smart agriculture” that utilizes advanced technologies such as robotics and information and communications technology (ICT) has been developed. However, in the growth management of tomato cultivation, only a few tasks of the planting and harvesting activities are automated. If other tasks could be automated, heavy burden on farmers would be significantly reduced.
This study proposes an automatic tomato harvesting system that combines object detection using deep learning with RGB-D camera (Intel RealSense D415), a robot arm (UFACTORY xArm 5 Lite), and a universal vacuum gripper. Moreover, this system was developed using ROS which is an open source software.
To evaluate the efficiency of the system, harvesting experiments were carried out using two sizes of tomatoes. The experimental results showed that the system achieved a success rate of over 90% with an average total harvesting time of under less than 20 s per single fruit. However, in the case of tufted tomatoes, harvesting could not be performed using the current gripper shape and robot arm motion planning method. The results validated the feasibility of the automatic tomato harvesting system.