leaf removal, trims the stems, andinserts the cuttings into plug trays.While the system has been shown toprocess effectively many plantsautomatically, the robot is notequipped to handle successfully thewide variety of cuttings that a trainedworker handles with aplomb. A keychallenge in greenhouse automationwill be to develop productive systemsthat can perform in a reliable andcost-effective way with highly variable

Automation inthe Greenhouse:Challenges,Opportunities,and a RoboticsCase Study

Ward Simonton1

Additional index words. mechaniza-tion, computers, material handling,machine vision

Summary. The commercial green-house operation, with a controlledand structured environment and alarge number of highly repetitivetasks, offers many advantages forautomation relative to other segmentsof agriculture. Benefits and incentivesto automate are significant andinclude improving the safety of thework force and the environment,along with ensuring sufficientproductivity to compete in today’sglobal market. The use of equipmentand computers to assist productionalso may be particularly important inareas where labor costs and/oravailability are a concern. However,automation for greenhouse systemsfaces very significant challenges inovercoming nonuniformity, culturalpractice, and economic problems. As acase study, a robotic workcell forprocessing geranium cuttings forpropagation has been developed. Therobot grasps randomly positionedcuttings from a conveyor, performs

1Assistant Professor. Dept. of Biological and Agricul-tural Engineering, Univ. of Georgia, Georgia Station,1109 Experiment Street, Griffin, GA 30223-1797.

HortTechnology • Apr./June 1992 2(2)

biological products.

T oday’s floriculture industry is avibrant, diverse segment of ag-riculture that is valued at more

than $2 billion per year in the UnitedStates (U.S. Dept. Agr., 1988). Inthese times of global competition,floriculture companies are looking atmany methods of improving produc-tivity. Automating labor intensive orhazardous tasks is one example. Ro-botic systems along with fixed auto-mation machines are being developedfor the floriculture industry in an at-tempt to increase competitiveness andalso to improve quality (Neal, 1991).

A robot or robotic system refersto a computer-controlled, general-purpose machine. These machines aredifferent from fixed automation in thatrobots are programmable and multi-functional. Robotic systems areprevalent in the automotive, elec-tronic, general fabrication, and otherindustries for conducting repetitive,precise, and/or hazardous operations.Robots are used for large tasks such ashandling castings and welding carframes to small tasks such as assem-bling printed circuit boards andtransferring machined parts. The samerobotic system, or even the same typeof robot, is not always used for suchtasks. However, each robot that per-

forms these operations can performmultiple tasks. For example, the samerobot that spray paints an automobilecan be programmed to paint a child’stoy. The term “flexible automation”comes from this proficiency: the abilityof a single machine to be flexibleenough to perform multiple tasks (Nof,1985).

With increased emphasis on au-tomation in commercial greenhouseoperations, interest in both roboticand fixed automation systems is ex-panding (Smith, 1991). With this in-terest in automation comes bothchallenge and opportunity. In terms ofchallenge, why is automation particu-larly difficult for this industry relativeto other similarly sized industries? Interms of opportunity, why might it beworthwhile for this industry to pursueautomation more vigorously? Whattypes ofgreenhouse tasks are engineerstargeting for automation?

This paper attempts to addressthese questions briefly. The scope in-cludes any system for the moving,processing, or grading of plants and/or containers, whether robotic or fixedautomation. The objectives of thispaper are: 1) to describe challengesengineers and horticulturists face indeveloping productive and cost-effec-tive automation systems, 2) to de-scribe the potential benefits and in-centives for further automation in thecommercial greenhouse, and 3) todescribe a greenhouse automation re-search case study on robotic process-ing of geranium cuttings for propaga-tion.

ChallengesCurrent challenges for increased

automation can be classified broadlyinto four groups: nonuniformity, cul-tural practice, marketing and eco-nomics, and funding equipment de-velopment.

Nonuniformity. Nonuniformity,or variability, refers to that in which“all entities are not the same.n In thefloriculture industry, we see tremen-dous diversity among greenhouse op-erations including production layout,trays, flats, pots, crops, and of courseeven within crops. This diversity makescost-effective, functional automationdifficult and time-consuming to de-velop (Moncaster, 1985). The greatmajority of current industrial roboticapplications, such as the spray paintingoperation mentioned above, are per-

231

formed with known items in a knownposition from day to day. In the green-house we frequently do not have thisluxury.

Even while performing the mostmonotonous duties, greenhouseworkers are making decisions andmodifying their tasks to fit the specificneeds of the plants. Whether handlingdifferent pot spacings on a bench orprocessing zonal geranium cuttings,which may vary 300% in stem caliper,600% in stem length, and have two to10 leaves, people have the ability toadjust their actions based on the re-quirements of the situation. Generally,neither fixed nor robotic automationsystems are equipped to modify theirown operations in this manner. Theprimary technical constraint for ad-vanced automation systems for thecommercial greenhouse is this highdegree of nonuniformity.

Cultural practice. Day-to-dayoperation within most commercialgreenhouses does not fit the traditionalfactory, where economic justificationfor automating may be simpler. Forexample, many of the labor-intensivetasks in a greenhouse are not performeddaily for ≥8 h, but rather are periodic.This includes work required 2 h·day-1,weekly, or even seasonally. An examplefor some greenhouses would be ship-ping. If an automated system per-formed shipping only once per week, itmay be sitting idle 80% of the time.Obviously, 20% usage on a machinehas a significant effect on break-evencosts. A flexible, multipurpose systemor a low-cost, single-use alternativemay be required in this case.

Other cultural practice challengesto automation include few generallyaccepted. standards (bench layouts,work areas, and procedures); multiplecrops grown within the same range;and the environment (heat, moisture,dust, and dirt). Disorder of propaga-

tion stock (e.g., geranium, chrysan-themum, and poinsettia cuttings) isyet another difficulty due to a lack ofsingularity in equipment for this mar-ket. Also, there is the need for retro-fitting. Most companies are under-standably reluctant to alter their entireoperation to accommodate a newmachine (L. Carney, personal com-munication). However, care must betaken to minimize so-called “islands ofautomation” and the bottlenecks ofmaterial flow due to isolated points ofautomation.

232

Marketing and economics . F a c -tors related to marketing and econom-ics play a significant role in the challengeto automate. The manner in whichdemand for a product varies from sea-son to season and year to year, with theaccompanying price fluctuations,makes growers very cautious (D. Busch,personal communication). This cau-tion leads to resistance to change. His-torically, the plant production indus-try has preferred operating costs overcapital spending. Cost and availabilityof labor play an important role. Evi-dence exists that capital intensity mayhave an inverse relationship to profit-ability for some industries (Smith,1991). In a survey in Georgia, nurseryand greenhouse growers cited capitalas the chieflimiting factor to expansion(Turner and Mixon, 1990).

Few economic studies have beenpublished in the last decade on green-house system automation. Fang et al.(1991) evaluated the effects of annualproduction volume, equipment cost,labor cost, and other characteristics onthe economic feasibility of robotictransplanting of bedding plants. Ex-pected payback varied from 2.6 to 6.4years and return on investment variedfrom 8% to 61%, depending on theassumptions. vanWaarde (unpublisheddata) examined the effects ofefficiency,robot number, equipment cycle time,labor cost, and other characteristics onrobotic processing of geranium cut-tings for propagation. The analysisprovided design goals for an eco-nomically feasible system with resultssuch as predicting a 100% increase inallowable cost per robot resulting froma 33% decrease in equipment cycletime.

Funding equipment develop-ment. Devising and testing new ideasand machines requires funding. De-velopment of new automation equip-ment for the greenhouse industry isperformed both by government re-searchers and private industry engi-neers. Unfortunately, all workings inthis area are woefully lacking in suffi-cient funds (Bolusky, 1989). Thegreenhouse industry has not beenwilling or able to fund research, asevidenced by the small grants given bythe Horticultural Research Instituteand the Bedding Plant Foundation.Federal and state government fundsfor applied research are dropping an-nually. Few machine/engineeringcompanies are willing to risk corporate

funds for innovative automationequipment when the market for spe-cialized equipment is relatively small.And lastly, there are few close grower/manufacturer and researcher/manu-facturer relationships, which are es-sential for producing cost-effective,productive equipment.

OpportunitiesCurrent equipment and re-

search. While there are many challengesto automate in commercial greenhouseoperations today, a number of re-searchers and equipment developersare busy readying new ideas and ma-chines. The most active area of devel-opment is transplanting (Neal, 1991).An estimate for 1989 bedding plantproduction was 60 million flats (Miller,1990). Equipment for plug or seedlingtransplanting within a greenhouse isbeing sold by several companies in-cluding Timmer Industries (Jenison,Mich., Plug Transplanter; 10,000 plugsper hour) and Visser (‘s-Gravendeel,The Netherlands, Plant-o-Mat; 10,000plugs per hour). Transplanting systemsunder development include those byBouldin and Lawson (McMinnville,Tenn., Coverplant, a unit initially de-veloped in France, sensors detect pooror empty plugs, 7 to 10,000 plugs perhour); Agrobotics (West Lafayette,Ind., Rotran; handles any standardtrays or flats and detects poor or missingplugs); and Rutgers Univ. (NewBrunswick, N. J., experimental roboticworkcell, innovative gripper for han-dling plugs). In-field transplantingmachines have been developed by theU.S. Dept. of Agriculture, AgriculturalResearch Service (USDA, ARS),Tifton, Ga., Univ. of Florida, Gaines-ville, and North Carolina State Univ.,Raleigh. Each of these machines hasdifferent production rates, efficiencies,support personnel requirements, andflexibility.

Most growers are aware of the im-pressive pot- and flat-handling equip-ment developed by Dutch companiessuch as Visser, Hawe, and Javo, andUnited States companies such asBouldin and Lawson and GleasonEquipment. Flexible container-han-dling systems are being researched byOhio State Univ. (Columbus, roboticsys tem) and Ecole Nat iona l d ’Ingenieurs des Travaux Agricoles(E.N.I.T.A) (France; table handlinggantry system). Other applications thatare under development include spray-

HortTechnology • Apr./June 1992 2(2)

ing by Bouldin and Lawson (mobile,self-guided unit) and cutting propaga-tion by the Univ. of Georgia (roboticworkcell for geraniums). Future po-tential robotic tasks in the greenhousethat may not require direct plant con-tact include packaging of the outgoingproduct, unpackaging ofincoming potsand trays, general materials handlingin the headhouse, and grading. Thoserobotic tasks that may require directplant contact include harvesting (to-matoes and cut flowers), pruning, andgrafting/budding.

Benefits and incentives. What arethe potential benefits of and incentivesfor the automation work describedabove?

1) Improved global marketpos i t ion th rough in -c reased p roduc t iv i ty(Bolusky, 1989).

2) Improved quality for amarketing advantage byproviding consistentgrading and marketingprocedures (U.S. Dept.Agr., 1989).

3) Increased standardizationthroughout the industryenabling faster machinedevelopment as the needsof the industry change.

4) Enhanced commercial-greenhouse operationstability through reducedlabor turnover, reducedlayoffs and rehires, andpromotion of worker re-training (Bolusky, 1989).

5) Improved safety of theworkplace through re-duction of chemical ex-p o s u r e - t o w o r k e r s(Moncaster, 1985).

6) Improved productionpractices as related to theenvironment through re-ductions in the amount ofchemicals required forproduction and elimina-tion of runoff (U.S. Dept.Agr., 1989).

Other incentives for automationare based on rapidly changing tech-nology. Faster and less expensivecomputers, more reliable computernetworks, improved sensing techniquesincluding machine vision, and im-proved motors and drives combine togive us better tools with which to

HortTechnology • Apr./June 1992 2(2)

design, test, and produce new equip-ment. Incorporating these tools withinnovative software may produce trulycost-effective and productive automa-tion equipment for the industry thatcan adapt to the changing needs andconditions of the commercial green-house.

A Robotics Case StudyCurrent research. Research in the

mechatronics laboratory at the Geor-gia Station is focused on methods ofaccommodating nonuniformity as wework to develop reliable, cost-effectiveflexible automation for agriculturalsystems. The mechatronics laboratoryuses a robotic system, a gripping sys-tem, and a supervisory computer. Thefirst case study task for the laboratoryhas been the preparation of geraniumcuttings for vegetative propagation.Geraniums were chosen due to therelatively simple and open structure ofthe cuttings, a nearby commercial co-operator in Oglevee Products, and themany technical challenges required forsuccessful processing. Our primary goalwith this task has been to perform alloperations required on the cuttingsfrom the time they are removed fromthe cooler until they are on thegreenhouse bench.

Workcell design for geraniumpropagation. A robotic workcell maybe defined as a unit consisting of ma-chines, sensors, and material transportmechanisms for performing work onone or more products. The roboticworkcell for processing geraniumcuttings is designed to perform theoperations of sizing (i.e., trimming themain stem to a particular size); strip-ping (i.e., removing those leaves hav-ing petiole branches within the root-ing zone); and inserting the cuttinginto a tray. Simple grading based onmain-stem caliper also is performedautomatically. The primary purpose ofthe robotic arm is to handle and ma-nipulate the cutting as other mecha-nisms placed in the robot workspaceperform required processing andsensing operations. The robot theninserts the cutting into a selected traybased on grade.

The workcell (Fig. 1) consists ofthe robot and gripper, a conveyor, apneumatic cutter, a pneumatic leafstripping mechanism, a positioningsensor, and a frame supporting threetrays. The cutter, leaf str ippingmechanism, and sensor are devices

developed specifically for the geraniumprocessing workcell. The positioningsensor measures the amount of bendin the stem and allows the robot tosuccessfully stick cuttings with severelybent stems. Trays with a 2 × 19 cellarrangement are used along withOglevee rubber dirt plugs.

Machine vision. For the roboticsystem to process a geranium cuttingsuccessfully, the plant parts of thecutting must be identified automati-cally. The system must locate the cut-ting in the robot workspace and beequipped to distinguish a leaf from astem, for example, and thus recognizethe different sizing, stripping, andsticking actions required for each in-dividual cutting. Handling the widevariability of shapes and sizes ofcuttingsis a difficult challenge.

Machine vision, the combinationof computer hardware and softwarethat allows a computer to “see,” can bea powerful tool for minimizing theeffects of variability in automatic agri-cultural processes. A technique usingmachine vision has been developed inthe mechatronics laboratory that ana-lyzes an image of a geranium cuttingon a conveyor surface and identifies allprimary plant parts including mainstem, petioles, and leaf blades.

We use the vision information todetermine a strategy for the processingof an individual cutting. A processingstrategy is based on an optimal grasplocation and orientation on the mainstem and location of leaves to be re-moved. The most difficult plant part to

233

Fig. 2. The position and force controlled robotic gripper can handlegeranium cuttings from ≈5 to15 mm in stem caliper.

Fig. 3. The robot can properly insert cuttings into plugs even if the main stem has a significantbend.

identify is also the most important: thegrowth tip. Significant errors duringlocation of the growth tip can causethe robot to damage a cutting duringprocessing.

Sequence of operations. A con-veyor brings cuttings into the work-space of the robot. A cutting is locatedusing machine vision, and the conveyoris stopped momentarily by the super-visory computer. Cuttings can be in asomewhat random position and ori-entation (±45 degrees) on the convey-or, but must be kept singular. The su-pervisory computer analyzes the vi-sion data and determines the grade,position, orientation of the robot grasppoint, and leaf removal requirements.This information is passed to the ro-bot, and processing of the cuttingbegins.

The robot grasps the cutting (Fig.2), performs leaf removal if required,and trims the main stem. From thetrimming mechanism, the robot movesthe cutting through a position sensorto determine the amount of bend inthe main stem. The robot then movesthe cutting to the next available plug inthe proper tray and inserts the cuttingwith an adjusting motion to accountfor the degree of stem bend (Fig. 3).While the robot processes a cutting,the conveyor is activated and the nextcutting is located and analyzed by thesupervisory computer. Mechanicalprocessing of a cutting by the robotoccurs concurrently with the visual

234

Workcell performance. Manycuttings have been processed in theworkcell. Samples from the zonal variet-ies ‘Bridgette’, ‘Veronica’, ‘Yours Truly’,‘Sincerity’, ‘Crimson Fire’, ‘Fame’, andseveral others have been prepared suc-cessfully for rooting by the robotic sys-tem. Rooting and growth have beenshown to be very comparable to manuallyprocessed cuttings (Simonton, 1990).The robot can handle a wide range ofsizes of geranium cuttings in a timeequal to that of one worker. Machinevision, which is used to locate the cut-ting and identify the details of thecutting’s structure, the gripper, which isused to sense and control the forceapplied to the stem, and the bend sen-sor, which is used to allow sticking ofbent stems, aids the robot in processingthe highly nonuniform plant material(Simonton, 1991).

analysis of the next cutting on theconveyor by the supervisory computer.

HortTechnology • Apr./June 1992 2(2)

Significant problems exist, how-ever, in the overall performance of theworkcell relative to commercial pro-duction requirements. Problems in-clude: 1) relatively frequent inaccu-racy (15%) of the identification of thegrowing tip that leads to overly longcuttings inserted into the plugs, 2)inability to selectively remove leavestoward the top of the cutting thatcauses an increase in foliage densityand a corresponding increase in diseaseand transpiration problems, 3) dam-age to petioles (3% to 5%) duringgrasping, and 4) ≈3% of cuttings notbeing properly inserted into the plugs.Each of these difficulties are due to therobotic system not being equipped toprocess successfully the wide variety ofcuttings that a trained worker handlessatisfactorily. Also, two manual pre-processing operations are required:1) removal of flower buds and 2) re-moval of leaves within 8 to 10 mm ofthe base of the unprocessed main stem.Automatic flower-bud removal is notfeasible due to difficulties in visualidentification and the frequently closeproximity to the growth tip. Petiolesbranching from the lowest portion ofan unprocessed main stem can causesignificant occlusion problems for thevision system.

ConclusionsThe commercial greenhouse sys-

tem, with a controlled and structuredenvironment and a large number ofhighly repetitive tasks, offers manyadvantages for automation relative tomany other segments of agriculture.Benefits and incentives to automateare significant, and perhaps none more

HortTechnology • Apr./June 1992 2(2)

so than ensuring sufficient productiv-ity to compete in today’s global mar-ket. However, there are many chal-lenges for the floriculture industry andfor equipment developers.

Automation in the greenhousefaces difficulties in nonuniformity,cultural practice, economics, andfunding of equipment development.Frequently, a key challenge will be todevelop productive robotic or fixedautomation systems that can performtasks dealing with biological productsofvariable size, shape, color, position,orientation, and/or firmness in a reli-able and cost-effective way. Just as inthe automobile industry, where differ-ent robots or fixed automation ma-chines are used for different types ofapplications, the commercial green-house operation will require a numberof different systems for its many needs.

The need for successful, near-fu-ture automation equipment most likelywill be driven by high volume tasksperformed frequently, labor shortages,and wage pressures from surroundingindustries. Robotics and other automa-tion systems will not be replacing thoseindividuals with significant plant exper-tise, yet these systems do offer oppor-tunities for improving productivity,quality, and safety. For the industry toprogress in automation, researchers andequipment manufacturers will need in-put on requirements, goals, and priori-ties from commercial greenhousegrowers, propagators, and managers.

Literature CitedBolusky, B. 1989. Testimony to the U.S.Congress by the Amer. Assn. Nurserymen,25 Apr. 1989, Washington, D.C.

Fang, W., K.C. Ting, and G.A. Giacomelli.1991. Engineering economics of SCARArobot-based plug transplanting workcell.ASAE Paper 917026, Amer. Soc. Agr.Eng., St. Joseph, Mich.

Miller, R. 1990. U.S. floriculture worth$2.43 billion at wholesale. Grower Talks54(3):106-112.

Moncaster, M.E. 1985. Application of ro-botics to agriculture. Proc. Farm Electron-ics and Computing Symp., 20 to 25 Oct.,Stratford-upon-Avon, Great Britain.

Neal, K. 1991. Automatic transplanters.Greenhouse Manager 9(11):99-101.

Nof, S.Y. 1985. Handbook of industrialrobots. Wiley, New York.

Smith, I.C. 1991. Automation or people:Learn the right balance for financial suc-cess. Grower Talks 54(12):22-31.

Simonton, W. 1990. Automatic geraniumstock processing in a robotic workcell.Trans. Amer. Soc. Agr. Eng. 33(6):2074-2080.

Simonton, W. 1991. Automated plant han-dling and processing in a robotic workcell.p. 213-218. In: Proc. of the First IFAC/ISHS Workshop on Mathematical andControl Applications in Agr. and Hort.,Sept. 1991, Matsuyama, Japan.

Turner, S.C. and B.J. Mixon. 1990. Amarket examination of Georgia’s orna-mental plant industry. Univ. of GeorgiaAgr. Expt. Sta. Res. Rpt. 593.

U.S. Dept. Agr. 1988. USDA/ERS Flori-cultural Survey, Washington, D.C.

U.S. Dept. Agr. 1989. U S D A / A R Ssponsored report of the greenhouse andnursery mechanization advisory commit-tee workshop. Boynton Beach, Fla., Nov.1988.

235