英文文献原文.docx

英文文献原文.docx

《英文文献原文.docx》由会员分享，可在线阅读，更多相关《英文文献原文.docx（22页珍藏版）》请在冰豆网上搜索。

英文文献原文

编号：

毕业设计外文翻译

（原文）

学院：

机电工程学院

专业：

机械设计制造及其自动化

学生姓名：

何源

学号：

1201110310

指导教师单位：

机电工程学院

姓名：

郭中玲

职称：

高级工程师

2016年6月3日

RoboticsandComputer-IntegratedManufacturing

Duetoageneralshiftinmanufacturingparadigmfrommassproductiontowardsmasscustomization,reconfigurableautomationtechnologies,suchasrobots,arerequired.However,currentindustrialrobotsolutionsarenotoriouslydifficulttoprogram,leadingtohighchangeovertimeswhennewproductsareintroducedbymanufacturers.Inordertocompeteonglobalmarkets,thefactoriesoftomorrowneedcompleteproductionlines,includingautomationtechnologiesthatcaneffortlesslybereconfiguredorrepurposed,whentheneedarises.Inthispaperwepresenttheconceptofgeneral,self-assertingrobotskillsformanufacturing.Weshowhowarelativelysmallsetofskillsarederivedfromcurrentfactoryworkerinstructions,andhowthesecanbetransferredtoindustrialmobilemanipulators.Generalrobotskillscannotonlybeimplementedontheserobots,butalsobeintuitivelyconcatenatedtoprogramtherobotstoperformavarietyoftasks,throughtheuseofsimpletask-levelprogrammingmethods.Wedemonstratevariousapproachestothis,extensivelytestedwithseveralpeopleinexperiencedinrobotics.Wevalidateourfindingsthroughseveraldeploymentsofthecompleterobotsysteminrunningproductionfacilitiesatanindustrialpartner.Itfollowsfromtheseexperimentsthattheuseofrobotskills,andassociatedtask-levelprogrammingframework,isaviablesolutiontointroducingrobotsthatcanintuitivelyandontheflybeprogrammedtoperformnewtasksbyfactoryworkers.

1.Introduction

Inordertoremaincompetitiveinaglobalizedenvironment,manufacturingcompaniesneedtoconstantlyevolvetheirproductionsystemsandaccommodatethechangingdemandsofmarkets.Currently,productionisexperiencingaparadigmshiftfrommassproductiontomasscustomizationofproducts.Theimpactofthistrendonproductionsystemsisthattheyshouldadapttohandlemoreproductvariation,smallerlifecycles,andsmallerbatchsizes–ideallybatchsize1.Today,robot-basedproductionisanessentialpartoftheindustrialmanufacturingbackbone.However,theconceptofanindustrialrobotstaticallyplacedinacellandcontinuouslyrepeatingacarefullypredefinedsequenceofactionshasremainedpracticallyunchangedformanydecades.Notsurprisingly,typicalindustrialrobotsarenotflexibleandthussuchadegreeoftransformableproductionisbeyondthecapabilitiesofcurrentsystems.

AsimplifiedillustrationofthesituationisshowninFig.1.Traditionalmanufacturingsystemsareautomatedtoalargedegree,butcanonlybereconfiguredwithgreatdifficulty.Ontheotherhand,traditionalmanuallaborisveryflexible,butnoteconomicallyviableforlargescaleproduction,especiallysoinhighwagecountries.Inthefuture,masscustomizationwillmakeitnecessarytocombinehighreconfigurabilitywithalargedegreeofautomation.Thisgoalcanbeachievedintwodistinctways:

workerscaneitherbeequippedwithbetterautomationtools,suchasintuitiveon-the-flyprogrammingofrobots,therebyincreasingtheirproductivity,orthereconfigurabilityoftraditionalautomatedproductionlinescanbeimproved,e.g.throughtheuseofmultipurposerobots.

Fig.1.Aclearrequirementforthefactoriesoftomorrowisthatofatransformableproductionsystem,showingahighdegreeofflexibility,whilestillapplyingahighdegreeofautomation.

Roboticsisexpectedtobeoneofthemainenablersofthistransitiontothetransformablefactoryoftomorrow.Toreachthedemandedlevelofflexibilityrobots,ormoregenerallymobilemanipulators,needtobeabletomoveautonomously,copewithuncertaintyininteractionswithhumansandpartiallyknownenvironments,handleavarietyofdifferenttasks,andbeabletobereprogrammedfastbynon-robotexpertswhenanewtaskinthefactoryarises.

So,thequestionemerges:

Whatshouldbethecharacteristicsofaframeworkthatwouldallowrobotstobeflexibleenoughtohandleuncertaintyandchangingproductiontasks,whilebeingintuitiveenoughtobeusedandre-programmedbynon-robotexperts?

Wearguethattheanswerliesinthetightintegrationofsensingandactionwithinasmallsetofmodularandparametrizable“robotskills”.Contrarytotraditionalrobotmacros,ourskillsarecharacterizedasbeinggeneral,inthattheycanhandleavarietyofobjects,andself-asserting,inthattheycontainpre-andpostconditionchecks.Finally,thissetofskillsforagivenindustrycanbenaturallyextractedfromacarefulanalysisofindustrialstandardoperatingprocedures（SOP）.

Inthisworkwepresentaconceptualmodelofthe“skills”andhowthisconceptcanbeimplementedandsuccessfullyappliedinreal-worldindustrialscenarios.Theconductedexperimentsshowthatasmallsetofskillscanbeintuitivelyparametrizedbyeitherkinestheticteaching,pointinggestures,orautomatictaskplannerstosolvedifferentindustrialtasks,suchaspartfeeding,transportation,qualitycontrol,assembly,andmachinetending.

Themaincontributionsofthispaperarethat?

Weproposeaconceptualmodelofrobotskillsandshowhowthisdiffersfromtraditionalmacros.

Weshowhowthisapproachcanenablenon-expertstoutilizeadvancedroboticsystemsbyencapsulatingadvancedcapabilities,implementedaprioribyexperts,insideeachindividualskill.

Concreteapplicationsoftheapproacharepresented,inwhichadvancedrobotsystemshavebeenappliedinseveralrealindustrialscenarios.

1.1.Relatedwork

Thegoalofachievingeasyandquickreprogrammingofrobotsbynon-expertshasbeenconsistentlypursuedusingthetask-levelprogrammingparadigm.Task-levelprogrammingconstitutesahigherabstractionlevelthantraditionalrobotprogramming.Itspecifiesagiventaskasasequenceofactions,butavoidsdescribingeachactioninfulldetail.Thesequenceofactionsleadingtothefulfillmentofthistaskcaneitherbeplannedorspecifiedbyanoperator.SuchanapproachissimilartothesymbolicrepresentationoftasksoftheShakeyrobotthatusedaSTRIPSplannertoreasonaboutwhichactionsmayleadtoaccomplishingagoal,ortheconceptofObject-ActionComplexes（OACs）thatallowedcognitiverobotstoidentifypossibleactionsbasedonobjectaffordances.

Task-levelprogrammingisbasedonlowerlevelentities,usuallycalledrobotskills,thatinstantiateactions.Variousrepresentationsofrobotskillsthatwouldbesuitablefortask-levelprogramminghavebeenpursuedduringthelastdecades.Whatmostoftheaforementionedattemptshaveincommon,isthatrobotskillsare,inturn,composedofprimitivesetsofrobotmotions,calledactionormotionprimitives.Theseskillprimitivesaresimple,atomicrobotmovementsthatcanbecombinedtoformmorecomplexbehaviors.Asystematicsurveyoftherichrelevantliteraturerevealshowfragmentedtheconceptofskillsis,lackingawidelyaccepted,strictdefinition.

Incontrasttotheskillsthemselves,theskillprimitivesareratherwelldefinedintheroboticscommunity;althoughseveraldifferentdescriptionsexist,mostofthemlooselyfollowMason'sworkoncompliantcontrollers,whichpavedthegroundfortheTaskFrameFormalism（TFF）introducedbyBruyninckxandDeSchutter.Recentworkhasexpandeduponthisidea,forinstanceenablingtheuseofanysensorandcontrollerinthesametaskframe（e.g.visualservoingcombinedwithforcecontrol）,andoperatinginadynamictaskframe.

Ourdefinitionofskillsintegratespre-andpost-conditionschecksintheexecution,whichmakestheinterconnectionofskillsintotasksanintuitiveandcontrolledprocess.Inthatsense,ourapproachiscomparabletothemanipulationprimitivenets.However,thisapproachrequireslow-levelparametrizationofthemanipulationprimitives,makingitaccessibleonlytorobotexperts.Anotherconcept,similartoournotionofskills,isthatofActionRecipesasdefinedintheRoboEarthproject.Differenttoourskillsthough,ActionRecipeskeepsensingandactingseparateandmainlyfocusonknowledgesharingbetweenrobotswithsimilarcapabilities.Alltheaforementionedapproachesrequirearobotexperttoprogramandusethesystem.Thisisalsooneofthereasonswhytheseworks,incontrasttoours,wereneverdeployedinarealindustrialsetting.

Easyprogrammingofindustrialrobotsrequires,apartfromtheunderlyingconcepts,intuitiveHuman–RobotInteraction（HRI）mechanisms.HRIinindustrialsettingsfocusesonofflineprogrammingbyusingCADmodelsoronlineprogrammingusingaugmentedreality（AR）.Theapparentlimitationofsuchmethodsisthattheyassumethepresenceofanexpertforreprogramming,whichmakesthemunsuitablefortheenvisionedtransformablefactories.Incontrasttothemajorityofpublishworksandsimilartoourapproach,theauthorsofcombinedskillprimitivesandspeechrecognitionforprogramminggraspingtasksbyspeech,withotherwork.Thisworkincorporatestask-levelprogramming,however,theprogrammingisacombinationofprimitives,withnohigher-levelskillsbetweentheseandthetasklayer.Ingeneral,theliteratureofHRIinindustrialsettingsisratherlimited.Onepossiblereasonisthatindustrialroboticsapplicationstraditionallyfocusedonaccuracyandspeedwithouttheneedforexternalsensors.However,latelythistrendisgivingwaytomoreintuitiveanduser-friendlyinterfacesthatlargelyrelyonsensing.Furthermore,giventhenatureoffactoriesofthefuture,thatareonlysemi-structured,thereisahighneedforaccurate,externalsensorsaswellasintuitiveHRI.