Autonomous Pesticide Spraying Robot for use in a Greenhouse

Philip J. Sammons, Tomonari Furukawa and Andrew BulginARC Centre of Excellence for Autonomous Systems

School of Mechanical and Manufacturing EngineeringThe University of New South Wales, Australia

t.furukawa@unsw.edu.au

September 9, 2005

Abstract

This paper presents an engineering solution to thecurrent human health hazards involved in sprayingpotentially toxic chemicals in the confined space ofa hot and steamy glasshouse. This is achieved bythe design and construction of an autonomous mobilerobot for use in pest control and disease preventionapplications in commercial greenhouses. The effec-tiveness of this platform is shown by the platformsability to successfully navigate itself down rows of agreenhouse, while the pesticide spraying system effi-ciently covers the plants evenly with spray in the setdosages.

1 Introduction

The function of a greenhouse is to create the optimalgrowing conditions for the full life of the plants [1].Achievement of the desired conditions often requiresthe use of pesticides, fungicides, high temperaturesand increased carbon dioxide and humidity levels [2].Prolonged exposure of greenhouse workers to theseconditions leads to an uncomfortable and hazardouswork environment [3, 4], contravening modern Aus-tralian occupational, health and safety principles.

Often, there are substantial risks involved in green-house work due to the hazardous environment or dan-gerous operations. Specific examples of these include:repetitive strain injury, toxic chemical exposure, ex-

treme heat exposure, extreme humidity (heatstroke)[3], and working at heights. Automating tasks withinthe greenhouse will enable the avoidance of unwantedor hazardous human exposure whilst potentially lead-ing to an increase in overall efficiency and productiv-ity.

The main application of robots in the commercialsector has been concerned with the substitution ofmanual human labour by robots or mechanised sys-tems to make the work more time efficient, accurate,uniform and less costly [5]-[10]. One may argue thesocial implications of such developments, for exam-ple, the effects on employment through loss of bluecollar jobs to the more efficient robotic counterpart;there are also ethical considerations that may be ar-gued. Whilst there may well be some validity tothe argument in some cases, this current project isunique in the number of stakeholders that are af-fected in a positive sense. The farmers benefits arefound in more efficient maintenance of the crops andeither less work for themselves or a decreased needfor the employment of others (arguably, an expen-sive process). Increased demand on growers has be-gun to be met with increased specific automation inmany fields, as producers believe that automation isa viable and sometimes necessary [7] method to en-sure maximum profits with minimum costs [6]. In-deed Hopkins [6] argues that automation enables theexpansion of a greenhouse without having to investmore financial resources on labour. Merchants maybenefit from increased sales due to a lower cost prod-

1

uct; the consumers will benefit, likewise, from a lowercost product of comparable quality. The stakeholdersthat benefit most, at least from an ethical or socialperspective, however, are the greenhouse workers.

This paper presents the design and constructionof an autonomous robot that seeks to address someof the human health concerns associated with green-houses. This robot is designed as a base for develop-ing systems to enable the automation of greenhouseprocesses such as the spraying of pesticides, picking offruit and the caring for diseased plants. The systemis designed to be as modular as possible, enablingthe development and/or modification of any of theindividual tasks.

The following sections of this paper are outlined asfollows, Section 2 describes the current greenhouseenvironment and technique of pesticide application,Section 3 outlines the design of the robot, Section IVoutlines the safety aspects considered in the design,Section V discusses the programs used in operation,Section VI provides a cost benefit analysis of automa-tion within a greenhouse and Section VII provides anoverview of the effectiveness and conclusions foundupon implementing the robot in a real greenhouseenvironment.

2 Pesticide Spraying

2.1 Overview

The optimisation of carbon dioxide enrichment, tem-perature, humidity, root moisture, fertiliser feed, pestand fungus control allow greenhouses to producefruits, vegetables out of season and ornamental flow-ers all year round. For example, carbon dioxide levelswithin a greenhouse are approximately five times thenormal atmospheric levels [11]. The optimal temper-ature and humidity levels of a greenhouse during thenormal working hours of the day can be quite high(up to 38◦C), making it very hot and uncomfortablefor someone wearing the heavy protective equipment.This can subject the worker to risks such as heatstroke and other health hazards associated with suchconditions.

The inhalation of pesticides used within a green-

house can, in some cases, be fatal or cause perma-nent damage to the lung tissue. Even relativelysmall amounts of exposure to many chemicals overlong periods of time can be damaging to the worker.Due the hazards of such exposure, protective cloth-ing and equipment must be used. However, researchhas showed that protective clothing does reduce thepossible amount of exposure but does not stop it[3].Many pesticides have been found to be able to pene-trate clothing and even rubber gloves within half anhour and a manual spraying task can take severalhours depending on the size of the greenhouse.

Figure 1 shows a typical crop of greenhouse toma-toes. Contemporarily, a human worker would walkdown these confined rows with a pesticide sprayinggun, in an attempt to cover the foliage of the plantswith an even coat of spray. An experienced workerwill attempt to coat the surface of the plants withthe appropriate calculated dosage. This manual ap-plication of pesticides is, as mentioned above, a timeconsuming, tedious and dangerous task, requiring theworker to wear protective clothing and breathing ap-paratus. Hence, this manual application technique islargely open for error. It is often difficult to knowexactly how much pesticide it really being applied toeach plant. In this harsh environment, it is possi-ble that the worker may also apply an inappropriateamount of spray to the plants through either fatigueor haste, resulting in the inefficient application of pes-ticide. The alternative method to manual spraying isFog spraying. This involves using a fogging sprayerto spray the whole greenhouse with a fine mist thatcovers every surface [12]. This is useful where thegreenhouse is infested with small flying pests and theentire greenhouse requires spraying. this method isnot always desirable and cannot cure all problems.

An autonomous pesticide spraying device wouldbe invaluable in the avoidance of human exposureto hazardous chemicals and to ensure the optimalcalculated amount of spray is applied to all plantsevenly, minimising waste due to increased accuracyand precision[13]. A further advantage of using anautomatic sprayer is that various concentrations ofpesticides can be applied at levels where there wouldnormally be an increased the risk of toxin exposureto the human applicator.

2

Figure 1: Typical crop of greenhouse tomatoes

The ideal time to spray plants within a greenhouseis in the early evening as a result of some chemi-cals used on plants adversely reacting to ultravioletlight and intense heat. An automatic spraying sys-tem could be set to begin operation at night avoidingout-of-hours work whilst ensuring that the plants aresprayed in conditions that cause the least amount ofdamage to the plants.

2.2 Cost Benefit analysis

For the future implementation of a greenhouse robotto be commercially viable, such a product would haveto be at a low cost, or have the potential to carryout all, or most, of the current manual tasks withina greenhouse. Some of these potential applicationsinclude monitoring plant health, plant growth rates,removing unwanted leaves and picking the produce.In principle, a system able to satisfy these require-ments seems viable, as the requirements may be metvia a common basic pathway. Whilst it is recognisedthat there is generally more than one solution to anengineering problem, an example of a system ableto achieve this would be that consisting of a mo-bile platform with a robot arm on an elevating plat-form, which would then satisfy the pruning, pickingand monitoring of produce, essentially adding to therobot this paper has outlined.

As mentioned, the considerable economic and so-cial costs related to the manual process of pesticideapplication within greenhouses (specifically labourcosts and the human health risks respectively) is faroutweighed by the economic and social benefits of anautonomous robot such as that discussed by this pa-per. The benefits include the decreased labour costs,reduced pesticide wastage costs and the obvious re-duction in human health hazards[14][10].

Whilst this particular project has commercial ex-pectations, the initial development model has beendesigned for use in the National Centre for Green-house Horticulture greenhouses in NSW, Australia.The commercial aspirations of this project requirethe total cost to be kept to a minimum. Such costeffective automation should therefore be relatively at-tractive to growers. In addition to the cost consider-ations, all parts and materials have been purchasedfrom Australian distributors, nearly all of them Syd-ney based. Whilst this has been done to simplify thefuture manufacturing of the commercial product, thediscerning Australian consumer was also considered.Finally, to enable the ease of reproduction, modifi-cation and replacement, almost all components areassembled from off the shelf parts.

3 Design

Gan-Mor et al.[10] and Michelini [15] realised the po-tential of steel pipes as a method of guidance for theirautonomous robots. The hot water piping shown inFigure 2, is a standard installation for most mod-ern greenhouses and can reach very hot temperatures(80-90◦C). To roll along the pipes, Nylon wheels arechosen over other types as they can withstand heavyloads and high temperatures with no deviation totravel height. Figure 3 shows the CAD model of thewheel arrangement which is one of the main designaspects of this robot. The two sets of wheels; oneset 100mm larger than the other (to accommodatefor the 50mm pipes) is arranged in such a way thatthere is a seamless transition in moving onto the railsfrom the work area at the end of the rail set, thenback again at the completion of that row. The in-duction sensor also visible, is connected directly to

3

the microcontroller; allowing the robot to sense thatit is indeed on the rails and on course. The arrangedwheel assembly keeps the robot on the tracks allow-ing the robot to drive along without the need for anyexpensive and complicated navigation ability.

Figure 2: Hot water piping

Figure 3: Precise placement and organisation of allcomponents

For ease of design and reproduction, as few parts aspossible require special machining or manufacturingin an attempt to minimise the cost of manufacturingthis and repeat robots. This choice to use as many

off-the-shelf products as possible carried with it sig-nificant challenges, but is mostly regarded as an im-portant reality check in the design process enablingthe optimisation of a simple and flexible design to berealised.

The robot is designed for simplicity and valuefor money, while still being able to effectively sprayplants in the greenhouse. The robot is capable ofdriving to the end and back along the hot water pip-ing rails of a row in greenhouse, requiring then to bemanually moved to the next row to continue spraying.This is then repeated for each row in the greenhouse.This limitation allows the first step of the robot pro-totype to be realised as a functional robot.

Figure 4 shows the CAD prototype design, whichis used to allow the precise placement and organi-sation of all components. From the micro-controllerplacement to aligning individual nut and bolts, eachcomponent is modelled into place to ensure the mostspace efficient design, minimising the physical dimen-sions of the robot, while keeping all simplicity andfunctionality.

Figure 4: CAD model of the wheel alignment

4 Safety Considerations

As this system is possibly working without supervi-sion, safety aspects must be considered to ensure the

4

safety of the robot, the people that may be within thegreenhouse and the greenhouse itself. Easily accessi-ble Emergency Stop buttons, anti-bump start button,separate enable switches for pump and motor, Keyactivated power switch, fast blow fuses and bumpsensors are the main safety components built intothe design of this robot. If an emergency stop buttonis triggered by either button or microcontroller, therobot will stop and user intervention is required torestart to robot. Onboard Ethernet connectivity al-lows for the future realisation of remote wireless con-trol of the robot. This allows for remote monitoringand control of robot functions.

The bump sensors are located at the front and backof the robot to stop the robot from bumping intoany objects that may be in the path of travel. Iftriggered, the microcontroller will promptly stop therobot within its buffer zone and stop any sprayingfunctions. Unlike the Emergency stop buttons, therobot will continue its path of travel when the ob-struction that triggered the sensor is cleared and it issafe to move off.

5 Control System

Figure 5 shows the control environment currently im-plemented. The user interfaces have control over therunning of the microprocessor and are fed back in-formation about the status of the robot. The mi-croprocessor reads the information and controls themovements of the robot and actions of the sprayingsystem.

The spray system consists of a large tank for hold-ing the pesticides, a pump and four valves to directthe allocated spray to the sections of plant eitherside of the robot as it moves past the desired sprayarea. The valves are electronically controlled by theon-board microprocessor which receives input signalsfrom infer-red sensors on the underside of the robot.As the robot passes over reflective markers placed onthe ground, the pump is turned on and off to en-able selective spraying of the greenhouse plants. Anadditional valve is used to recycle the fluid from thevalve manifold back into the tank, enabling continualmixing of pesticides within the tank.

Microcontroller

Spraying System

Motor Control

Induction Sensors

Infra Red Sensors

Bump Sensors

LCD¥ Keypad

Interface

Web Interface

Figure 5: Control environment of the developed sys-tem

The drive system consists of a worm drive motorcapable of a maximum speed of 0.26 m/s whilst driv-ing on the heating rails. The motor is driven by ahigh powered PWM (Pulse Width Modulation) con-troller board, taking in an analogue signal from themicrocontroller. The Motor controller has its ownsoft start and stop facilities allowing smooth stop-ing and starting with no processing time required forthe microprocessor. The 0-5v analogue signal inputalso allows the microprocessor to control the speed,acceleration and declaration.

The induction proximity sensors accurately detectthe presence of the metal rails beneath the robot.This allows for safe, hassle free operation of the robotwithin the greenhouse requiring no changes to the ex-isting greenhouse configuration. The SICK infra-redsensors accurately sense the placement of the reflec-tive sensors as they pass beneath the robot. Thereflective sensors are a simple and efficient method ofmarking a selected area. The sensors can be replacedwith a more advanced method if detecting a selectedarea as the technology advances, allowing the poten-tial for the robot to almost fully automated.

The LCD/Keypad module shows the user relevantinformation on the robots status and allows the userto control the robot directly with ease. The web host-ing capabilities of the microprocessor allow the po-tential for the robot to be monitored and controlledfrom a remote location. Keeping the user at a safedistance for the pesticide spray whilst enabling fullview and control of what is going on.

5

6 Control Algorithm

The on-board microprocessor controls all of the in-puts and outputs of the system. The software run-ning on the processor is Dynamic C, a variant of thecommon C language. Extra librarys and functionshave been added to the original C language that arespecific to the rabbit microprocessor. Of particularnotice is the multi-tasking functions. Both The Pre-emptive multitasking and Cooperative multitaskingis available. Pre-emptive, being the ability to time-slice the program code so that many processors canbe sharing the processor time to make it seem like theprocesses are running at the same time. in the Co-operative multitasking inviroment, each process vo-lentariliry passes control to the other processes. usu-ally done while waiting for someting to happen. forexample, a button press, or a delay to finish. Theprocessor uses this usually wasted ’waiting time’ toexecute other processes.

Figure 6 shows the current basic program struc-ture. The Platform Control program is used to con-trol the mobility of the base platform. It carefullymonitors the speed of the robot and monitors sensorinputs. The Sprayer Control module enables intelli-gent spraying of plants within the greenhouse on ei-ther side of the robot. Future developments may en-able the robot to decide which areas to spray, at thisstage the growers must make the decision on whichplants require spraying. The following modes of spraycan be used to operate the robot: Sensor spray In-dividual plants can be marked for spraying. Sectionspray Sections or groups of plants can be targeted forpesticide application. Continuous spray All plants inthe greenhouse will be sprayed with the appropriateamounts of pesticides.

7 Testing

The robot was tested in the research greenhouse atThe National Centre of Greenhouse Horticulture atGosford, NSW where cucumber plants are grown.Figure 7 shows the on-site experiment with the devel-oped robot where the robot is spraying on damagedplants. The tasks necessarily conducted prior to the

Start

Decide Spray Mode

Control spray as necessary

Reverse direction

Decide Spray mode

Inform user to change rows

Robot on rails?

End of rails reached?

All rows complete?

End

Yes

Yes

Yes

No

No

No

Figure 6: Control program structure

robot spraying is the placement of reflective markerson the ground by the greenhouse scientist only. Thetesting consisted of single runs down to the end ofa row of the greenhouse and back to the work areawhile spraying the plants with water. Along the run,sections of cucumber plants were marked out to besprayed on both sides of the robot. While in opera-tion, emergency stop buttons and bump sensors weretested for their effectiveness in a real situation.

7.1 Test 1

The platform was able to successfully and smoothlydrive up and back along a row in the greenhouse andreturn to the work area at the end of the row. The in-duction sensors worked sufficiently to turn the robotaround and stop the robot when it returned to thestart.

The Spraying system had a small lag to fill the

6

Figure 7: Control program structure

pipes with water before the nozzles began to workproperly. The spray was enough to penetrate the fo-liage and spray into the adjacent row. The identifica-tion of the reflectors was excellent until leaking waterbegan dripping from one of the Infra-Red sensors.The water dripping from the sensor lens interferedwith IR beam and the sensor begun to malfunction.After the leak was repaired the sensor began to func-tion properly again. Observations were made on theamount of over-spray while the robot was in action.the robot was stopped and the leaves were inspected.Most of the leaves were correctly sprayed with wa-ter. However, the underside of some leaves were notsufficiently sprayed. This problem was fixed whenthe robot sprayed the adjacent isle. The spray thencovered the patches that were missed by the initialspray.

Table 1: Results of Tests 1 and 2

Test 1 Test 2Run Success 90% 95%

Topside leaf coverage 95% 90%Underside leaf coverage 95% 80%Amount of over-spray 20% 10%

7.2 Test 2

Test 2 went much the same as Test 2, however thistime the foliage was more dense. In one section therobot stopped when it became stuck on some strayfoliage. The spray from the nozzles was still enoughto penetrate the foliage and cause some over-sprayinto the next row. The healthier and larger leavescaused some parts of the foliage to become shieldedfrom the full force of the spray, thus, not as muchcoverage as in Test 1. The leaking water problemwas solved by moving the sprayer boom to the otherend of the robot. Once again, observations were madeon the coverage of spray and amount of over-spray

7.3 Results

Table 1 shows the data collected by three separateobservers. Test 1 was in a greenhouse of old andsick cucumbers. The foliage was not very dense anddid not protrude past the heating rails. Test 2 was inanother cucumber greenhouse at the Gosford researchstation, However this time it was a healthy crop witha dense canopy with some leaves protruding past theheating pipes.

8 Conclusions

The results showed the robot was able to successfullyfulfil the physical specifications outlined by The Na-tional Centre for Greenhouse Horticulture so as to beable to function within their greenhouses. The robotalso met the economic and time constraints that itwas subject to. The robot was able to drive up andback along the tracks in the greenhouse. The Induc-tion Proximity Sensors detected the rails effectively

7

and operated satisfactorily. The spraying device de-signed by another thesis student was able to selec-tively spray designated groups of plants in the green-house whilst moving along the rails. The coverage ofthe spray coated the plants in adequate and consis-tent dosage.

Even though the results do not show a 0% over-spray and a 100% coverage, the results are very en-couraging. When manually spraying these problemsalso arise. The coverage of spray way comparableto that of an experienced manual applicator and therobot was able to spray the plants an a consistentand timely manner.

At the far end of the greenhouse the induction sen-sors successfully detected the end of the rail, enablingthe platform to change direction, moving back alongthe rails and returning to the start. The wheel as-sembly was successful in delivering a smooth tran-sition from the rails to the concrete slab where theinduction sensors identified the start of the rail andbrought the platform to a stop. The bump sensorswere tested for situations where people and obstaclessuch as branches were in the path of motion. Thebumpers were successfully activated and the robotwas able to be stopped within a safe distance so thatno damage or injury was or could be caused.

The design and successful construction of the robotusing off the shelf Australian components to under-take the task of autonomously spraying pesticideswithin a greenhouse shows that it is possible andcan be done via a simple and cost effective manner.Being a prototype, the robot has some components(namely motor and microcontroller) that are over de-signed for the simple task of spraying, resulting in therobot having surplus capacity. The use of this moresophisticated micro-controller enables greater cost ef-ficiency when reproducing a commercial product, asa dedicated microcontroller with less capability anda correctly designed motor for the final weight andperformance requirements would be correspondinglycheaper.

Conversely, as a result of this over design there is alot of further work that can be done to improve thisrobot without increasing the price by the additionalcost of a more sophisticated micro-controller. Theover-design can be justified because it can cope with

multi-tasking additional attachments. A number ofadditional features are considered under future work.In the meantime it would be prudent to seek advicefrom NSW Agriculture and the horticulture sectorgenerally to find out more about likely demands andspecific requirements for a range of crops.

9 Discussion

Within commercial greenhouses there is a perceivedneed for automation that ranges from planting andgrowing disease free plants to pruning and picking thefinal product. Surprisingly, relatively little time andlarge-scale research has been employed into green-house robotic systems to replace human effort in thisfairly hostile environment. The opportunities for in-creasing food production from outdoor into indoorhorticultural crops are especially attractive at thehigh cost end of the market. In addition to boost-ing growth rates, glasshouses and greenhouses can bebuilt near prime markets and do not require largetracts of fertile land to be productive.

Experience with this project places us in a goodposition to better understand what might be achiev-able in the near future in terms of viable modifica-tions to the design. Further automation of the drivesystem will enable to robot autonomously navigatethe entire greenhouse. Improvements to the sprayingsystem will allow the precise application of pesticidespray at varying dosages. Improvements to the spray-ing hardware will enable improved coverage with re-duced over-spray. Further work on the software willallow remote monitoring and operation.

10 Acknowledgment

This work is supported by The National Centrefor Greenhouse Horticulture, funded by NSW Agri-culture, Supported by the ARC Centre of Excel-lence programme, funded by the Australian ResearchCouncil (ARC) and the New South Wales State Gov-ernment

8

References

[1] Badgery-Parker, J. (1999), Agnote DPI/249, Edition1, pp. 1-2.

[2] Van Henten, E.J., Hemming J., Van Tuijl, B.A.J.,Kornet, J.G., Meuleman, J., Bontsema J., & Van Os,E.A. (2002), “An Autonomous Robot For harvestingCucumbers in Greenhouses”, Autonomous Robots,Vol 13, pp. 241-258.

[3] Riley B., Siemsen-Newman L., (2003) “Health Haz-ards Posed to Pesticide Applicators”, NorthwestCoalition For Alternatives to Pesticides, pp. 17-24.

[4] Tosaki K., Miyahara S., Ichikawa T, Taniai S.,Mizukura Y, Moriki H., and Miyashita S. (1998),“Development of Microcomputer-Controlled Un-manned Air Blast Sprayer” JARQ Japan Agricul-tural Research Quarterly, Vol 32, No. 3, pp. 83-90.

[5] Sezen, B. (2003), “Modelling Automated Guided Ve-hicle Systems in material Handling”, Dogus Univer-sitesi Dergisi, Vol 4, No. 3, pp. 207-216.

[6] Hopkins, M. (2000), “Automating in the 21st Cen-tury Career and Technical Education”, GreenhouseGrower, pp. 4-12.

[7] Giacomelli G.A., and Ting K.C. (1995) “Transporta-tion and Robotics for Greenhouse Crop ProductionSystems”, Acta Hort (ISHS), Vol 399, pp. 49-60.

[8] Van Henten, E.J., Hemming J., Van Tuijl, B.A.J.,Kornet, J.G., Meuleman, J., Bontsema J. (2003),“Collision-free Motion Planning for a CucumberPicking Robot”, Biosystems Engineering, Vol 86,No. 2 pp. 135-144.

[9] Pilarski, T., Happold, M., Pangels, H., Ollis, M.,Fitzpatrick, K., Stentz A. (2002), “The DementerSystem for Automated Harvesting”, AutonomousRobots, Vol 13, pp. 9-20.

[10] Gan-Mor S., Ronen B., Kazaz I., Josef S., BilankiY. (1997), “Guidance for Automatic Vehicle forGreenhouse Transportation”, ACTA Horticulture,Vol 443, pp. 99-104.

[11] Kellerman, D. (2002) “Pickling Cucumbers”, Com-mercial Vegetable - Production Guide, AgricultureMagazin, pp. 22-25.

[12] Austerweil M., Grinstein A., (1997), “AutomaticPesticide Application in Greenhouses”, Phytopara-sitica, Vol 25, pp. 37-42.

[13] Hetzroni A., Meron M., (2003) “Mapping TreeCanopy for Precision Spraying”, Technical Report,Department of Agriculture, Illinois, pp. 75-83.

[14] Haire, B. (2003), “UGA Scientists Develop ’row-bot’- Farming Robots not Science fiction but technicalfact”, Technical Report, Georgia Faces, Universityof Georgia.

[15] Michelini R. C., Razzoli R. P, Acaccia G. M, MolfinoR. M., (1998), ”Mobile Robots in greenhouse Culti-vation: Inspection and Treatment of plants” ACTAHorticulture, Vol 453, pp52-60.

9