Multiple single-axis robots for automated grafting machine ......2 Plant description In this prototype, the grafting operation is performed by a core unit (Belforte & Eula, 2012) where

Proceedings International Conference of Agricultural Engineering, Zurich, 06-10.07.2014 – www.eurageng.eu 1/7

Ref: C0316 Multiple single-axis robots for automated grafting machine supplying system Lorenzo Comba, Paolo Gay, Davide Ricauda Aimonino and Cristina Tortia. Department of Agricultural, Forest and Food Sciences, Largo Paolo Braccini 2, 10095 Grugliasco (Turin-Italy)

Abstract

Recent European Community directives have increasingly limited use of pesticides and chemical fu-migants, such as methyl bromide, to counteract soilborne pathogens. An efficient option to these dis-couraged chemical treatments can be the adoption of disease resistant grafted plants (as e.g. toma-toes or peppers), even if this technique is labour consuming and involves tedious repetitive actions. These aspects encourage the development of automated machines, in order to increase the productiv-ity and the rooting success rate and to reduce costs, allowing grafting to be more economically sus-tainable. The design and implementation of the supplying and sorting system of a new prototype of a grafting machine is presented in this paper. Target of developing a mechanical structure simplest and cheapest possible was achieved basing the supplying system on a passive linear guide with six single-axis independent robot modules, relying on automation and control for the fulfilment of performance specifications. While modules translate along the guide, a grip can drop off seedlings with shoots picked up from the feeding trays and shoots can be classified, in terms of diameter, by a vision system positioned along the guide. The same system is also devoted to handling produced grafted seedling to the outlet grip. The optimal design of the system, in terms of number of single-axis robots and control strategy principles, has been obtained with the goal to maximize the throughput of the machine and minimize the occurrence of scions and rootstock mismatches.

Keywords: Robotics, Grafting, Automation, Arduino

1 Introduction

The use of pesticides or chemical fumigants, such as methyl bromide, to counteract soil-borne pathogens has been increasingly limited by recent European Community directives. In this scenario, a growing and promising option can be the adoption of disease resistant graft-ed plants (as e.g. tomatoes or peppers), even if this technique is labour consuming and in-volves tedious actions. Indeed the selection of compatible shoots, stems cutting and applica-tion of the best graft are repetitive operations that encourage the development of automated machines, in order to increase the productivity and the rooting success rate and to reduce costs, allowing grafting to be more economically sustainable. Few semi-autonomous prototypes for grafting were developed since 1990s (Honami, Taira, Murase, Nishiura, & Yasukuri, 1992; Kubota, McClure, Kokalis-Burelle, Bausher, & Rosskopf, 2008; Lee et al., 2010; Masayuki, 1994), but these kind of machines were able to perform only a limited number of the needed operations to obtain a complete grafted seedling and usually required the supervision of at least three expert workers to feed the machine and check the quality of the out coming products (Helper Robotech, 2013; Iseki, 2013; Kang, Han, Noh, & Choi, 2005). In Europe, also new fully-automated grafting robots have been re-cently developed: machines able to reach higher performances, producing even more than 1000 grafted seedlings per hour with a success rate up to 90% (Iso Groep, 2013). However, their cost and complexity make these machines not suitable to typical Mediterranean nurse-


ries. For this reason in almost all cases the operation is still carried out manually. A national research project (PRIN 2009) has been financed to develop an innovative machine well suit-ed for small and medium farms. The design and development of a supplying and sorting system, coupled with a new proto-type of grafting machine, is presented in this work. The mechanical structure has been creat-ed simplest as possible, relying on automation and control for the fulfilment of performance specifications. The design of the grafting unit (designated to cut, join, and fix rootstocks and scions) is not discussed in this paper, but can be found in Belforte & Eula (2012). Alveolate trays are adopted to fed the machine with seedling, used to obtain both scions and rootstock. In order to increase the effectiveness of grafting operation, diameters of the stems to be processed should be isometric. The supply system should be therefore able to classify, in two or more stem diameter classes, the shoots from the incoming trays, and then to pro-vide two fitting stems of the same class to the graft unit. This procedure is carried out by a computer vision system based on a camera that acquires the image of the stem at fixed dis-tance coupled with a backlighted illumination, like the similar system in (Ashraf, Kondo, & Shiigi, 2011). Unlike other machines, which typically uses two independent supply sub-systems (one for scions and one for rootstocks) and another for the grafted seedling, a single supply system which handles at the same time scions, rootstocks and the complete seed-lings was adopted. The system proposed in this paper is based on five single-axis independ-ent robot modules, hereafter referred as sliders, which move on a single rail. While sliders move along the guide, several operations can be performed on the seedlings in correspond-ence of each working stations. For example, a grip can drop off seedlings picked up from the feeding trays or stems can be classified by the vision system positioned along the guide. Each single-axis robot module, which is controlled to move along the guide, is predisposed to hold one seedling. In this way, the set of independent robots behave also as a buffer. Coor-dinating adequately the robots, it is possible to reach the size matching between scions and rootstock, even if the respectively incoming shoots have temporary different diameters.

2 Plant description

In this prototype, the grafting operation is performed by a core unit (Belforte & Eula, 2012) where a scion and a rootstock, conveniently provided by the automatic supply system de-scribed in this paper, are processed. The supply system is constituted by a single rail, on which six independent motorized modules (sliders) can move, allowing shoots to be pro-cessed at the several working station of the machine, as showed in Figure 1.

Figure 1: Scheme of the whole grafting machine. The main rail, on which sliders can shift, is repre-

sented in purple. The grafting module is sketched by the cyan box. Scions and rootstocks are supplied on the trails on the right, while the complete seedlings are delivered on the trays on the left.


First step of the grafting process is the machine fed with shoots, both for obtaining scions and rootstocks, disposed on trays. A handling system is devoted to pick seedlings by needle-pliers from trays and place them into the sliders, conveniently positioned at the receiving po-sition on the rail. Shifting along the rail, sliders make shoots available for classification, by the computer vision system, and then conveniently to the grafting module in two different posi-tions, one for scions and one for rootstocks. The same slider receives completed grafted seedlings and move them toward a second handling system devoted to place them on the outgoing tray, completing the entire process. Since incoming seedlings can differ in stems size, the computer vision system is adopted to classify the shoots in two (or more) size clas-ses. After classification, the supply system has to be able to move a scion and a rootstock of the same class to the loading position of the grafting unit. The possible temporary absence of shoots belonging to the same size class is faced with a (virtual) buffering system that allows the temporarily storing of incompatible seedling, while new ones are loaded and classified. This functionality has been achieved increasing to six the number of independent sliders moving on the rail. The pick&place operations in the first and last stations can be performed using the clips nowadays used in high-performances transplanting machines (see e.g. Urbinati, 2013) Each slider consists of a stand able to hold one seedling, coupled to the guide rail by a bracket and a recirculating ball chassis, as described in figure 2. A stepper motor that en-gages on a timing belt by a pinion provides the driving force to the module.

Figure 2: Motorized modules of the supply system. Stepper motor allows sliding movement along the guide rail. A single shoot stand is settled at the bracket bottom

The control system, schematized in figure 3, coordinates interactions between the supply system and the grafting core unit, planning properly the movements of each slider. The head of the control system is constituted by a real-time control algorithm implemented in Matlab® and running on a pc, which generates and provides motion instruction to a network of Ar-duino® microcontroller that, organized according to a master-slave structure, finally pilots the stepper motor drivers of each slider. More in detail, a master Arduino® Mega 2560 has been programmed to address command generated by the control unit to six slave modules, one for each slider, using an I2C bus communication. Each slave module consist of two linked Ar-duino® microcontroller, an Arduino UNO® and an Arduino Mini®, devoted to plan motion speed profiles and to pilot the motor driver respectively. Blending operations between two consecutive movements of a slider, to obtain fluid motion, are also performed by microcon-


trollers in the slave modules. Slider position is estimated counting the number of impulses supplied by the motor driver. In addition, estimated position are also updated with five addi-tional references detected by proximity/optical sensors.

Figure 3: Scheme of the control system

The use of a single rail reduces the hardware complexity and costs, but adequate control strategies have to be studied to manage sliders displacement in an efficient way. Sliders cannot pass over each other; therefore their movements are constrained being their relative order not invertible. This aspect, if not properly tackled by control strategies, could strongly limit the efficiency of the system.

3 Control algorithms and motion planning

The control algorithm is devoted to planning the best combination of sliders movement, in order to timely serve the stations (trays unloading/loading, grafting unit supplying and seed-lings classification) of the machine during grafting operations. This goal is achieved using an optimization procedure that generates, at each step, all possible movement combinations and discern among all these feasible solutions of the problem the most performing one. In more detail, the supply system has been modelled as a discrete-event system, where the state is constituted by the position of the sliders on the rail. Considering a 3 m long rail, it has been virtually discretized in 𝑛 = 50 slots 60 mm wide, which corresponds to the slider length. Starting from an initial configuration of the sliders on the rail, the strategy adopted consists in computing all the possible new configurations obtained with one-segment long movement of the sliders at each time step. Obviously, only feasible configurations are taken in to account, verifying that all physical constraints of the system are fulfilled, such as the impossibility of sliders to be located in the same place or to swap each other on the rail. Each feasible solu-tion within this domain is evaluated computing a cost function that mainly depends on the distances of the modules from target positions, i.e. the loading and unloading stations of in-put-output trays and of the grafting unit. Distances are properly weighted in order to assign different priorities to the operations (for instance, supplying shoots to the grafting module or retrieve grafted seedlings were considered as high-priority tasks) and to privilege some slid-ers for a specific operation. An improvement of the proposed approach consists in exploring possible sliders configura-tions at more than one step ahead, building a decision tree of feasible solutions. In this case, the best branch of the tree is selected evaluating the cost value of final configurations, but at each step only the first movements are performed. After that, the procedure is repeated computing a new tree of decisions on the basis of updated inputs and events that in the meanwhile could be modified.


4 Optimal system design

The optimal design of the system, in terms of number of single-axis robots and control strat-egy principles, is discussed in (Comba, Gay, & Ricauda Aimonino, 2013). Optimization has been performed with the goal of maximize the throughput of the machine and minimize the occurrence of scions and rootstock mismatches, with the aid of an ad hoc discrete-events simulation tool, developed in Matlab®. Thirty minutes long working sessions of the supply system has been simulated with a number of independent sliders ranged from 2 to 7.

One Step Ahead Two Steps Ahead

Number of Sliders

Grafting Rate

Unmatching Seedlings Rate Grafting Rate

Unmatching Seedlings Rate

No. Units∙min-‐1 Units∙min-‐1 Units∙min-‐1 Units∙min-‐1 2 5.5 10.3 5.5 10.3 3 7.1 8.5 7.1 8.5 4 9.1 5.4 9.1 5.4 5 10.5 3.8 11.1 3.1 6 12.1 2.0 -‐ -‐ 7 11.4 2.0 -‐ -‐

Table 1. Performances indexes of the supply system in terms of grafted seedlings production rate and occurrence of unmatching shoots on the modules

Table 1 shows performances indexes in terms of number of grafted seedlings processed per minute and number of shoots discarded, or brought back to the handling system, because of unmatching characteristic. Results of simulations adopting both control logic that explore one-step-ahead and two-step-ahead movements configurations were presented. Production rate increase with the number of sliders, achieving a maximum value of 12.1 grafted seed-lings per minute with six independent sliders.

One Step Ahead Two Steps Ahead Number of Sliders

Computing Time

Optimization Domain

Computing Time

Optimization Domain

No. ms No. ms No. 2 4.7 7.5 27.1 61.2 3 8.6 16.4 136.3 336.4 4 17.5 32.4 658.6 1565.1 5 41.4 73.0 3427.0 7689.2 6 92.4 125.8 38317.8 84009.7 7 221.3 200.5 139844.8 262968.2

Table 2. Average computing time of discrete steps of the simulations and mean size of the solutions domain (number of feasible solutions) of the optimization problem

With further enhancement of the number of sliders, performances begin to decrease because the advantage of a higher number of sliders is contrasted by the higher probability of conflicts in their movement. With six independent sliders, the supply system also achieves a good buffering capability, needful to perform the matching operation between different sizes of shoots. This can be noted in Table 1 where the rate of unmatching seedling decreases with an increasing number of sliders.


Average computing time of all the discrete event steps calculated during simulations are pre-sented in Table 2, together with their solution domain mean size, expressed in number of future feasible configurations of the sliders on the rail, at one or two ahead steps of move-ments. Values in Table 1, concerning simulations with the second control strategy governing 6 and 7 modules, are missing due to the excessive computing time (38 and 140 seconds respectively), incompatible with a real time control. Indeed the number of feasible configura-tion of 6 or 7 modules became extremely high, slowing down the simulation framework. The same events and inputs sequences has been adopted to perform all simulations, replaying the same conditions at each test. In (Comba et al., 2013), only single-position sliders are considered, but the design optimization can be extended to multi-position sliders.

5 Conclusions

In this paper, an innovative system for the supply of an automated module for vegetable grafting is presented. A path planning control algorithms, based on recursive optimization, allows to simplify the mechanics of the system still ensuring required performances. The adopted solution is modular and can be adapted to different system productivity targets. The preliminary results presented in this paper demonstrate how, under some circumstanc-es, mechanical complexity can be reduced adopting advanced control algorithms. Future developments will concern the use of sliders allowing more than one holding position, in or-der to enhance the buffering capability without increasing the number of sliders and an accu-rate analysis of the role of the speed of the sliders. Higher velocity leads to faster move-ments, but also reduces the duration of the time available to elaborate an optimal control strategy. For this reason, operative speed should be therefore set as a tradeoff between the-se two aspects.

6 References

Ashraf, A. A., Kondo, N., & Shiigi, T. (2011). Use of Machine Vision to Sort Tomato Seed-lings for Grafting Robot. Engineering in Agriculture, Environment and Food, 4(4). Belforte, G., & Eula, G. (2012). Macchina automatica per l’innesto erbaceo (Italian). Oleo-dinamica Pneumatica, 12, 43–47. Comba, L., Gay, P., & Ricauda Aimonino, D. (2013). Optimal Design of a Supply System of an Automated Grafting Machine. In CIOSTA XXXV - From effective to intelligent farming and forestry. Billund (Denmark). Helper Robotech. (2013). Helper Robotech. http://helpersys.co.kr/home/index.html. Honami, N., Taira, T., Murase, H., Nishiura, Y., & Yasukuri, Y. (1992). Robotiziation in the production of grafted seedlings. Acta Hort. (ISHS), 319, 579–584. I seki. (2013). Iseki. http://iseki.co.jp/english/. Iso Groep. (2013). Iso Groep. http://www.iso-group.nl/lees/55. Kang, C.-H., Han, G.-S., Noh, T.-H., & Choi, H.-G. (2005, September 30). Splice Grafting Robot for Fruit and Vegetable Plants. Retrieved from http://patentscope.wipo.int Kubota, C., McClure, M. A., Kokalis-Burelle, N., Bausher, M. G., & Rosskopf, E. N. (2008). Vegetable grafting: History, use, and current technology status in North America. HortScience, 43(6), 1664–1669.


Lee, J.-M., Kubota, C., Tsao, S. J., Bie, Z., Echevarria, P. H., Morra, L., & Oda, M. (2010). Current status of vegetable grafting: Diffusion, grafting techniques, automation. Scientia Hor-ticulturae, 127(2), 93–105. doi:10.1016/j.scienta.2010.08.003 Masayuki, O. (1994). New Grafting Methods for Fruit-Bearing Vegetables in Japan. New Grafting Methods for Fruit-Bearing Vegetables in Japan. Urbinati. (2013). Urbinati. http://en.urbinati.com/. Retrieved from http://en.urbinati.com/

Documents

Multiple single-axis robots for automated grafting machine ......2 Plant description In this prototype, the grafting operation is performed by a core unit (Belforte & Eula, 2012) where