Research Paper EAEF 3(1) : 20-24, 2010 Development of an End-Effector for a Tomato Cluster Harvesting Robot* Naoshi KONDO *1, Koki YATA *2, Michihisa IIDA *3, Tomoo SHIIGI *1, Mitsuji MONTA *4, Mitsutaka KURITA *5, Hiromi OMORI *6 Abstract An end-effector was developed for a tomato cluster harvesting robot. This end-effector can harvest not individual fruits but a whole fruit cluster to improve the robot s harvest efficiency. Experiments for harvesting tomato clusters were conducted in a high-density plant training system. According to a harvesting algorithm, the end-effector was able to perform well, even when target peduncle orientations were not given. Although the success rate of harvesting tomato clusters was 50 %, it is considered that this rate would improve if an end-effector is used for the high-wire tomato plant training systems in Dutch systems where the node lengths of plants are long enough to loosely hold the main stems. [Keywords] tomato, harvesting robot, end-effector, fruit cluster I Introduction There have been many applications of robotic technologies to agricultural operations such as seeding, grafting, transplanting, spraying, harvesting, grading and other operations (Kondo and Ting, 1998). Although research on harvesting robots started more than 20 years ago, no fruit harvesting robot has yet been commercialized. One of the reasons for this is the slower operation speed and lower performance of robots compared to humans. That is, the operation speed and success rate of robots harvesting individual fruits one by one are lower than those of human workers. Several tomato harvesting robots were reported so far (Kawamura et al., 1984; Kondo et al., 1996; Monta et al.,1998). Their end-effectors were able to harvest individual tomato fruits by fingers or sucking force. This harvesting method is suitable from the perspective of fruit quality because only properly matured fruits are harvested on the required day. However, this requires repeated operations for each cluster. In case of cluster harvesting, one operation is sufficient for a whole cluster. Some tomato fruit varieties are frequently shipped as whole clusters in European countries and the USA. In addition, cluster harvesting is increasingly common in large-scale greenhouses in Japan. In this study, an end-effector that can harvest a whole tomato cluster with 4-6 fruits was manufactured and tested in a greenhouse using selective compliance assembly robot arm (SCARA). II Materials 1. Plant training system Figure 1 shows a large-scale tomato production system, called the high-wire plant training system (high-wire system) in a Dutch style greenhouse. greenhouse is being increasingly used in Japan. This type of However, a low node-order pinching and high-density plant training system (high-density system) has been a much higher productivity and can be operated intensively as shown in Fig. 2. In this training system, main stems are pinched after the third flower cluster blooms to allow high-density production and easy handling. Figure 3 shows the difference in fruit clusters between both the training systems. It was observed that the high-wire system creates larger angles between peduncles and main stems while some of the plant parts are hidden by leaves, fruits, and peduncles in the high-density system. It appears that it is not easy for a robot to harvest hidden fruits with hidden peduncles in the latter system. * Partly presented at the 121 st Kansai Branch Meeting of the Japanese Society of Agricultural Machinery in March 2009 *1 JSAM Member, Graduate School of Agriculture, Kyoto University, Sakyo-ku, Kyoto, 606-8502, Japan *2 JSAM Member, Graduate School of Engineering, Ehime University, Matsuyama, Ehime, 790-8577, Japan *3 JSAM Member, Corresponding author, Graduate School of Agriculture, Kyoto University, Sakyo-ku, Kyoto, 606-8502, Japan; iida@elam.kais.kyoto-u.ac.jp *4 JSAM Member, Okayama University, Graduate School of Environmental Science, 1-1-1 Tsushima-Naka, Okayama 700-8530, Japan *5 Technology Development Department, S. I. Seiko Co. Ltd., 66 Takaoka-cho, Matsuyama, Ehime 791-8036, Japan *6 JSAM Member, Advanced Greenhouse Production Research Team, National Institute of Vegetable and Tea Science, 40-1 Minaminakane, Taketoyo-cho, Chita-gun, Aichi 470-2351, Japan
KONDO, YATA, IIDA, SHIIGI, MONTA, KURITA, OMORI: Development of an End-Effector for Tomato Cluster Harvesting Robot 21 Fig. 1 High-wire tomato plant training systems in a Dutch style greenhouse. was 400 g, a frictional resistance of at least 5N was necessary to collect the fruit cluster. It was, however, determined that 40 N was necessary as peduncle grasping force of fingers for robot s quick movement with a harvested fruit cluster. Tomato plant phyllotaxis is basically regular so that every cluster comes out in the same direction, but the peduncle orientation cannot be predicted because the main stem twists often. If a machine vision system (Kondo et al., 2009) can detect the orientation, peduncle grasping may be possible using the usual fingers, but the peduncle is sometimes occluded by the main stem or leaves. For these reasons, it was considered that the fruit cluster harvesting end-effector needs multi-directional access to grasp and cut the peduncle from any direction. Peduncle diameter Peduncle Main stem diameter Main stem Cluster Main stem angle Fig. 2 Tomato plants in a low node-order pinching and high-density plant training system. Fruit diameter Fig. 4 Name of plant parts. Table 1 Fundamental physical properties of tomato. Max Min Average Main stem diameter [mm] 20.5 11.8 16.1 High wire High density Fig. 3 Comparison of tomato fruit clusters between the two plant training systems. 2. Physical Properties of Tomato Plants Before designing an end-effector to harvest a whole tomato cluster, the physical and dynamic properties of tomato plants were measured. Figure 4 shows the dimensions of the tomato plant parts; fruit diameter, main stem diameter, main stem angle, peduncle diameter, entire peduncle length and peduncle length between main stem and the first fruit. Table 1 shows the important sizes of tomato plant parts for designing the end-effector mechanism. These properties of tomato plants were measured in the high-wire system because the high-density system has not yet been studied. The cutting resistance of peduncles was around 60 N for the largest peduncle size. Since the larger fruit cluster mass Fruit diameter [mm] 63.2 43.0 52.5 Peduncle length [mm] 131.6 33.0 79.0 Entire peduncle length [mm] 226.9 97.5 148.4 Peduncle diameter [mm] 11.8 4.8 7.0 Main stem angle [ ] 153.0 101.0 122.5 Peduncle angle [ ] 139.0 5.0 69.5 III 1. End-effector Experimental Devices and Methods Figure 5 shows the mechanism of a harvesting end-effector made for a trial based on the physical properties of high-wire tomato plants. This end-effector mainly consists of upper fingers (1), lower fingers (2), a DC motor (3), a ball screw (4), an air cylinder (8), a photosensor (12) limit switches (9-11), a
22 Engineering in Agriculture, Environment and Food Vol. 3, No. 1 (2010) servomotor (13), a roller bearing (14), a coupling joint (15), and a fixer collar (16). Fig. 7 Grasping and cutting functions of the upper finger. Lower fingers were set at the bottom of the end-effector. They also have a peduncle detection sensor that consists of two strain gauges. The strain gauges are attached to the lower fingers to sense a peduncle when the manipulator moves the end-effector up along the main stem. The end-effector can grasp the peduncle between the peduncle grasper and lower finger with a force of 20 N for a peduncle diameter of 8 mm, because its spring constant is 2.33 N/mm. Both fingers are opened and closed through the air cylinder (8) by a linkage mechanism. Fig. 5 Mechanism of end-effector. Upper finger Fig. 8 A stem surrounding posture when a photosensor detects the main stem. Spring Cutter Peduncle grasper Fig. 6 Mechanism of upper finger. Figure 6 shows the mechanism of the upper finger that performs the peduncle grasping and cutting functions. In addition, Figure 7 illustrates the peduncle grasping and cutting functions of the upper finger. Upper fingers only surround the main stem. They are moved up and down by the ball screw (4) and the DC motor (3), while the peduncle grasper slides to grasped the peduncle using spring forces, when the upper finger moves down. Upper finger Main Stem Cutter Peduncle Lower finger Fig. 9 Peduncle grasping and cutting posture (upper fingers moved to the bottom). The motion of the end-effector is as follows: 1. Manipulator approaches a fruit cluster with upper and lower fingers open until the photosensor (12) detects the main stem. 2. Both fingers close and surround the main stem as shown in Fig.8. 3. Manipulator moves the end-effector upwards until one of the strain gauges senses the peduncle. 4. The upper finger moves down. 5. The peduncle is grasped and cut, see Fig. 9. 6. Both upper and lower fingers open. 7. Manipulator moves back to release the harvested fruit cluster.
KONDO, YATA, IIDA, SHIIGI, MONTA, KURITA, OMORI: Development of an End-Effector for Tomato Cluster Harvesting Robot 23 Fig. 10 SCARA with a harvesting end-effector. Fig. 11 Dimensions of SCARA (Mitsubishi Electric Corporation). 2. Manipulator Figure 10 shows SCARA (Mitsubishi Electric Corporation, RH-6SH5520) with a harvesting end-effector. It has four degrees of freedom and its weight capacity is 6 kg. Figure 11 shows the dimensions of this robot arm. Since the heights of tomato fruit clusters are similar in both plant training systems (high-wire and high-density systems), horizontal manipulator motions are more frequently conducted than vertical motions. This is the reason why SCARA was used instead of a vertically articulated robot. Its operational space on the horizontal plane has a 550 mm radius and 200 mm height. The maximum speed of SCARA is 1000mm/s and pneumatic actuators or hands are available for this robot. Fig. 12 Flow chart of the experimental method of harvesting tomato clusters using end-effector. 3. Experimental methods Figure 12 shows the experimental method of harvesting tomato clusters using the developed end-effector. First of all, the location data of a target tomato cluster was input. Second, the manipulator moves the end-effector to the main stem near the tomato cluster between the upper and lower fingers. If the photo sensor detects the main stem, the manipulator stops moving and the air cylinder acts to grasp the main stem. After that, the manipulator moves the
24 Engineering in Agriculture, Environment and Food Vol. 3, No. 1 (2010) end-effector upward until the lower fingers contact the peduncle and the peduncle detection sensor is ON. The upper fingers move down to grasp and cut the peduncle of the tomato cluster. The air cylinder acts to open the fingers. The manipulator moves the end-effector to release the tomato cluster into a harvest tray. Finally, the robot resumes the initial posture to complete the harvesting routine. Harvesting experiments were conducted on 20 tomato plants in the high-density system. IV Results and Discussion Experiments using the developed end-effector were conducted to harvest 20 tomato clusters. According to a harvesting algorithm, the end-effector performed properly. It took about 15 s for this end-effector to harvest a tomato cluster. However, node lengths in the high-density system were too short for the end-effector to grasp the main stem near the tomato cluster. Consequently, the end-effector could not extend into the plant to grasp the main stem as shown in Fig.13. improved if the overall height of the end-effector was small enough to extend into the plant. V Summary and Conclusions An end-effector for a tomato cluster harvesting robot was developed and experiments for harvesting tomato clusters were conducted in a high-density system. It took about 15 s for this end-effector to harvest a tomato cluster. Although the high-density plants have a high productivity, many fruit clusters, peduncles, and main stems are hidden by leaves and node lengths were too short for the end-effector to extend into the plant. It was observed that this size end-effector was suitable for the high-wire system. It is desirable to make the end-effector small and compact to easily harvest tomato clusters in the high-density system. Acknowledgement The authors would like to acknowledge the financial support of a project from NARO (National Agriculture and Food Research Organization). Fig. 13 End-effector cannot extend into the plant (top part of the end-effector hits the upper fruit cluster.). The success rate of harvesting the tomato clusters using the end-effector was 50% (10 tomato clusters out of 20). The cause of 35% (7 tomato clusters out of 20) failure rate was due to the inability of the end-effector to extend into the plant as described above. The failure rate of another 10 % was due to insufficient air pressure to grasp the peduncles repeatedly. It is expected that this problem can be solved by replacing the current air compressor with one that has a high capacity. The remaining 5% (1 tomato clusters out of 20) of the failure rate was caused by the loss of a fruit from the harvested cluster. It is important to control the manipulator to prevent impacts on the tomato cluster. Therefore, it was considered that the success rate could be References Kawamura, N., K. Namikawa, T. Fujiura, N. Ura, 1984 Agricultural Robot, Journal of JSAM, 46(3): 353-358. Kondo, N. and K.C. Ting: Robotics for Bioproduction Systems, ASAE, 1998 Kondo, N., Y. Nishitsuji, P. Ling and K.C. Ting, 1996, Visual feedback guided robotic cherry tomato harvesting, Transactions of the ASAE 39(6): 2331-2338. Kondo, N., K. Yamamoto, H. Shimizu, K. Yata, M. Kurita, T. Shiigi, M. Monta, T. Nishizu, 2009a, A Machine Vision System for tomato Cluster Harvesting Robot, EAEF, 2(2): 60-65. Kondo, N., K. Tanihara, T. Shiigi, H. Shimizu, M. Kurita, M. Tsutumi, V.K. Chong, S.Taniwaki, 2009b Path-Planning of Tomato-Cluster Harvesting Robot to Realize Low Vibration and Speedy Transportation, EAEF, 2(3):108-115. Mitsubishi Electric Corporation : http: //wwwf2.mitsubishielectric. co.jp/robot/lineup/work/sd/rh/rh6sdh_04.htm Monta, M., N. Kondo, K.C.Ting, 1998, End-effectors for Tomato Harvesting Robot, The Artificial Intelligence Review Journal, 12(1-3): 11-25. (Received: 23. June. 2009, Accepted: 9. September. 2009)