机械 自动化 外文翻译 外文文献 英文文献 最小化传感级别不确定性联合策略的机械手控制.docx

机械 自动化 外文翻译 外文文献 英文文献 最小化传感级别不确定性联合策略的机械手控制.docx

《机械 自动化 外文翻译 外文文献 英文文献 最小化传感级别不确定性联合策略的机械手控制.docx》由会员分享，可在线阅读，更多相关《机械 自动化 外文翻译 外文文献 英文文献 最小化传感级别不确定性联合策略的机械手控制.docx（14页珍藏版）》请在冰豆网上搜索。

机械自动化外文翻译外文文献英文文献最小化传感级别不确定性联合策略的机械手控制

FusionStrategiesforMinimizingSensing-Level

UncertaintyinManipulatorControl

Abstract：

Humanoidroboticapplicationsrequirerobottoactandbehavelikehumanbeing.Follow-ingsoftcomputinglikeapproachhumanbeingcanthink,decideandcontrolhimselfinunstructureddynamicsurroundings,whereagreatdegreeofuncertaintyexistsintheinformationobtainedthroughsensoryorgans.Intheroboticsdomainalso,oneofthekeyissuesinextractingusefulknowledgefromsensorydataisthatofcopingwithinformationaswellassensoryuncertaintyatvariouslevels.Inthispaperageneralizedfusionbasedhybridclassifier（ANN-FDD-FFA）hasbeendevelopedandappliedforvalidatingongeneratedsyntheticdatafromobservationmodelaswellasfromrealhardwarerobot.Thefusiongoal,selectedhere,isprimarilytominimizeuncertaintiesinroboticmanipulationtasksthatarebasedoninternal（jointsensors）aswellasexternal（visioncamera）sensoryinformation.Theeffectivenessofpresentmethodologyhasbeenextensivelystudiedwithaspeciallyconfiguredexperimentalrobothavingfivedegreesoffreedomandasimulatedmodelofavisionguidedmanipulator.Inthepresentinvestigationmainuncertaintyhandlingapproachincludesweightedparameterselection（ofgeometricfusion）byatrainedneuralnetworkthatisnotavailableinstandardmanipulatorroboticcontrollerdesigns.Theseapproachesinhybridconfigurationhassig-nificantlyreducetheuncertaintyatdifferentlevelsforfasterandmoreaccuratemanipulatorcontrolasdemonstratedherethroughrigoroussimulationsandexperimentations.

Keywords:

sensor,fusion,FDD,FFA,ANN,computing,manipulators,repeatability,accuracy,covariance,matrix,uncertainty,uncertaintyellipsoid.

1.Introduction

applications（industrial,military,scientific,medicine,welfare,householdandamusement）areincreasinglycomingupwithrecentprog-ressinwhicharobothastooperateinlargeandunstructuredenvironment[3,12,15].Inmostcases,theknowledgeofhowthesurroundingsarechangingeveryinstantisfundamentallyimportantforanoptimalcontrolofrobotmotions.Mobilerobotsalsoessentiallyhavetonavigateandoperateinverylargeunstruc-tureddynamicsurroundingsanddealwithsignificantuncertainty[1,9,19].When-everarobotisoperatinginanaturalnondeterministicenvironment,therealwaysexistssomedegreeofuncertaintyintheconditionsunderwhichagivenjobwillbedone.Theseconditionsmay,attimes,varywhileagivenoperationisbeingcarriedout.Themajorcausesleadingtotheuncertaintyarethediscrepanciesarisingintherobotmotionparametersandinthevarioustask-defininginformation.Theamountbywhichtheydifferfromthosecalledforintheprocessspecificationsmaynotalwaysbeinsignificant.Thesedeviationsmaybeduetoinaccuraciesinanalyticaldesignorinreproductionsofprogrammedmotionsorbecauseofdeterministicaswellasrandomerrorsinthealgorithms,measurementdata,datatransmissionlinks,andotherfactors.Changesinthestatusoftherobotlikeinstancesofmalfunctions,failures,shiftintheframeofreference,etc.,alsoleadtouncertaintyintheoperatingconditionsoftherobot.Thepresenceofsubstantialuncertaintysignificantlyaffectstherobotinthevariousstepsofsensingthestateofatask;inadaptingtothechangesthroughthecontrolsystem;andinreasoningtoselecttheactionsneededtoachieveagoal.

Infact,itisfeltthatoneofthekeyissuesinextractingusefulknowledgefromdataisthatofcopingwithuncertaintyatalllevelsandespeciallyatthesensinglevel.Alongwiththequantityoftheobservedsensorymeasurements,thequal-ityinvolvedalsoneedtobeinvestigatedintermsoftheresidualuncertaintyitpropagatestothedesiredsensoryinformation.

Inroboticsdomain,theuncertaintyprobleminthesensoryinterpretationlevelisaverycrucialoneforspecifictaskslikerobotisedspacestructuremanipulators,robotisedsurgeryetc.wherebothhighlevelofmachineprecisionandhumanlikeprehensionareneededThekeyprobleminthesensingprocessisinmakingtheconnectionbetweenthesignaloutputsofallthesensorsandtheattributesofthethree-dimensionalworld.Oneoftherecenttrendsistosolvetheproblemthroughsensorfusionandtherearenumerousfusiontechniquescoveringaverybroadspectrumofapplicationareas[10,13].Underthebackdropofthestudyoftheseresearchworks,itwasfeltthatthereisagreatneedforevolvingageneralizedandeasilyapprehensiblesoftcomputingbasedsensorfusionstrategy（humanoidapproach）formultiplesensorysystems.Thehumanoidapproachmakesitavailableforversatileapplications.Theeasilyapprehensiblecharacterofthedevelopmentmakesitparticularlysuitableforprocessingcomplex,highlynonlinearfunctionalrelationshipsbetweenlow-levelsensorydataandhigh-levelinformation.Thefusionstrategieswouldbemostsuitabletoapplyindistributedfusionarchitecturesasitcaneffectivelyenableustominimizetheuncertaintiesatanydesiredlevel.Areviewofsomepapersonuncertaintyanalysisinthecontextofmanipulatorcontrol[4,14,16,20,23]showsthatacommonstepinvolvedinallthesesystemsistheinterpretationofidenticalinformationthathasbeenacquiredthroughmultiplesensoryunits.Thefusedinformationneedstoberepresentedwithminimizeduncertaintyandthelevelofthisminimizationdependsontaskspecificapplications.Theresearchstudydescribedinthispaperhasfocusedonthisobjectiveinthecontextofsensoryguidedroboticmanipulations.AsatokenapplicationherethechallengeofimprovingrepeatabilityofaveryordinaryRSCtyperobothasbeenundertaken.

Real-worldsystemsarestochasticinnaturehavingnonlinearityanduncertaintyintheirbehaviorsandhencehumanoidapproachofsolutionsareonlyacceptableoneinmanysuchtasks.Formultivariableinput–outputsystems,effectsofsuchnonlinearityanduncertaintyaresignificantandneedstobeaddressedproperlyinordertocontrolthemeffectively.Take,forexample,thecaseofadvancedroboticsystems（manipulatingrobotshavingredundantdegreesoffreedomormobilerobotshavingredundantsensorysystemswouldfallinthiscategory）.Thesesystemsrequirevariouskindsofsensorsforrespondingintelligentlytoadynamicenvironment.Theymaybeequippedwithexternalsensorssuchasforce-torquesensors,rangesensors,proximitysensors,ultrasonicandinfraredsensors,tactilearraysandothertouchsensors,overheadoreye-in-handvisionsensors,cross-fire,overloadandslipsensingdevicesetc.Inaddition,therearealsovariousinternalstatesensorssuchasencoders,tachometers,revolversandothers.Moreisthenumberofsensors,moreisthecomputationalcomplexityforcontrollingthesystemandmoreisitsintelligencelevel.Sincerecentindustrialaswellasnon-industrialapplicationsneedroboticsystemswithhighlevelofintelligence,thecomplexityassociatedwithithastobeaddressedproperly.Forthispurpose,systemsequippedwithmultiplesensorshavingdifferentrangesofuncertaintieshasbeentakenuphereforstudy.

Informationobtainedfromdifferentsensorsareinherentlyuncertain,impreciseandinconsistent.Occasionallyitmayalsobeincompleteorpartial,spuriousorincorrectandattimes,itisoftengeographicallyorgeometricallyincompatibleamongstthedifferentsensorviews.Ourknowledgeofthespatialrelationshipsamongobjectsisalsoinherentlyuncertain.Taketheexampleofaman-madeobject.Itmaynotmatchitsgeometricmodelexactlybecauseofmanufacturingtolerance,human/machineerrorsandothernaturaluncertainties.Evenifitdoes（inmacrolevel）,asensorcannotmeasurethegeometricfeaturesandlocatetheobjectexactlybecauseofmeasurementerrors.Evenifitcan（withincertainboundedtolerancelimit）,arobotusingthesensormaynotmanipulatetheobjectexactlyasintended,maybebecauseofallcumulativeerrorsaddedwiththeend-effectorpositioningerrors.Theseerrorscanbereducedtoaverysignificantlevelforsometasks,byreengineeringthesolution,structuringtheworkingenvironmentandusingspeciallysuitedhighprecisionequipment-butatgreatcostoftimeandequipment[20].Analternativesolutionmaybetodevelopsensorfusionstrategiesthatcanminimizeandeliminatetheuncertaintiesofanyengineeringsystemtoadesiredlevel,atamuchlessercost,incorporatingallinherentuncertainties.InthispaperwefocusondevelopingaFDD-FFA-ANNbasedhybridtypesensorfusionstrategy.

Theorganizationofthepaperhasbeenarrangedasfollows.Section2outlinesthecomputationalstepsthroughwhichtheoverallfusionalgorithmhasbeenformulatedanddeveloped.ThesedevelopmentsandpropositionshavebeenappliedinSection3forvalidatingonsyntheticdataofanobservationmodel.Section4isdedicatedtowardsapplyingthedevelopedhybridfusionstrategiesforimprovingrepeatabilityofahardwarerobotmanipulator.TheireffectivenesshasbeenextensivelystudiedwithaspeciallyconfiguredRCStypeexperimentalrobothavingfivedegreesoffreedom.Aneuralnetworkformulationofthefusionalgorithmisalsopresented.Finally,inSection5thesignificantresultsandinferenceshavebeenlisted.

2.FormulationoftheFusionAlgorithmStructure

Thefusionalgorithmstructureconsistsofthefollowingcomputationalsteps:

（i）Theuncertaintiesintheinformationderivedthroughprocessingofmultiplenoisysensorydataarerepresentedbyindividualuncertaintyellipsoids.

（ii）Theuncertaintyellipsoidsaremergedinamannersoastominimizethevolumeofthefuseduncertaintyellipsoidbyproperassignmentofoptimalweightingmatrices.

（iii）FusionintheDifferentialDomain（FDD）hasbeendevelopedtofurtherreducetheuncertai