The Berkeley Lower Extremity.pdf

《The Berkeley Lower Extremity.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

TheBerkeleyLowerExtremityExoskeletonThefirstfunctionalload-carryingandenergeticallyautonomousexoskeletonwasdemon-stratedattheUniversityofCalifornia,Berkeley,walkingattheaveragespeedof1.3m/sH.Kazerooni
2.9mphwhilecarryinga34kg
75lbpayload.Fourfundamentaltechnologiesasso-ciatedwiththeBerkeleylowerextremityexoskeletonweretackledduringthecourseofR.Stegerthisproject.Thesefourcoretechnologiesincludethedesignoftheexoskeletonarchitec-e-mail:exo@berkeley.eduture,controlschemes,abodylocalareanetworktohostthecontrolalgorithm,andaseriesofon-boardpowerunitstopowertheactuators,sensors,andthecomputers.ThisUniversityofCalifornia,Berkeley,papergivesanoverviewofoneofthecontrolschemes.TheanalysishereisanextensionBerkeley,CA94720oftheclassicaldefinitionofthesensitivityfunctionofasystem:theabilityofasystemtorejectdisturbancesorthemeasureofsystemrobustness.Thecontrolalgorithmdevelopedhereincreasestheclosed-loopsystemsensitivitytoitswearer’sforcesandtorqueswith-outanymeasurementfromthewearer(suchasforce,position,orelectromyogramsig-nal).Thecontrolmethodhaslittlerobustnesstoparametervariationsandthereforerequiresarelativelygooddynamicmodelofthesystem.Thetrade-offsbetweenhavingsensorstomeasurehumanvariablesandthelackofrobustnesstoparametervariationaredescribed.DOI:10.1115/1.21681641Introductionlabor.ItisourvisionthatBLEEXwillprovideaversatileandrealizabletransportplatformformission-criticalequipment.TheprimaryobjectiveoftheBerkeleyLowerExtremityExosk-Thecapabilityofthelowerextremityexoskeletonstemsfromeleton
BLEEXProjectattheUniversityofCalifornia,BerkeleythecombinedbeneÞtofthehumanintellectprovidedbythepilotistodevelopfundamentaltechnologiesassociatedwiththedesignandthestrengthadvantageofferedbytheexoskeleton;inotherandcontrolofenergeticallyautonomouslowerextremityexoskel-words,thehumanprovidesanintelligentcontrolsystemfortheetonsthataugmenthumanstrengthandenduranceduringlocomo-exoskeletonwhiletheexoskeletonactuatorsprovidemostofthetion.TheÞrstÞeld-operationallowerextremityexoskeletonisstrengthnecessaryforwalking.Thecontrolalgorithmensuresthatcomprisedoftwopoweredanthropomorphiclegs,apowerunit,theexoskeletonmovesinconcertwiththepilotwithminimalandabackpack-likeframeonwhichavarietyofheavyloadscaninteractionforcebetweenthetwo.Thecontrolschemeneedsnobemounted.Thissystemprovidesitspilot
i.e.,thewearerthedirectmeasurementsfromthepilotorthehuman-machineinter-abilitytocarrysigniÞcantloadsonhis/herbackwithminimalface
e.g.,noforcesensorsbetweenthetwo;instead,thecontrol-effortoveranytypeofterrain.BLEEXallowsthepilottocom-lerestimates,basedonmeasurementsfromtheexoskeletononly,fortablysquat,bend,swingfromsidetoside,twist,andwalkonhowtomovesothatthepilotfeelsverylittleforce.Thiscontrolascendinganddescendingslopes,whilealsoofferingtheabilitytoscheme,whichhasneverbeforebeenappliedtoanyroboticsys-stepoverandunderobstructionswhilecarryingequipmentandtem,isaneffectivemethodofgeneratinglocomotionwhenthesupplies.BecausethepilotcancarrysigniÞcantloadsforextendedcontactlocationbetweenthepilotandtheexoskeletonisunknownperiodsoftimewithoutreducinghis/heragility,physicaleffec-andunpredictable
i.e.,theexoskeletonandthepilotareincontacttivenessincreasessigniÞcantlywiththeaidofthisclassoflowerinavarietyofplaces.Thiscontrolmethoddiffersfromcompli-extremityexoskeletons.InordertoaddressissuesofÞeldrobust-ancecontrolmethodsemployedforupperextremityexoskeletonsnessandreliability,BLEEXisdesignedsuchthat,inthecaseof4Ð6andhapticsystems7,8becauseitrequiresnoforcesensorpowerloss
e.g.fromfuelexhaustion,theexoskeletonlegscanbetweenthewearerandtheexoskeleton.beeasilyremovedandtheremainderofthedevicecanbecarriedThebasicprincipleforthecontrolofBLEEXrestsontheno-likeastandardbackpack.tionthattheexoskeletonneedstoshadowthewearerÕsvoluntaryandinvoluntarymovementsquickly,andwithoutdelay.Thisre-BLEEXwasÞrstunveiledin2004,attheUniversityofCali-quiresahighlevelofsensitivityinresponsetoallforcesandfornia,BerkeleyÕsHumanEngineeringandRoboticsLaboratorytorquesontheexoskeleton,particularly,theforcesimposedbythe
Fig.11Ð3.Inthisinitialmodel,BLEEXofferedacarryingpilot.Addressingthisneedinvolvesadirectconßictwithcontrolcapacityof34kg
75lb,withweightinexcessofthatallowancescienceÕsgoalofminimizingsystemsensitivityinthedesignofabeingsupportedbythepilot.BLEEXÕsuniquedesignoffersanclosed-loopfeedbacksystem.IfÞttedwithalowsensitivity,theergonomic,highlymaneuverable,mechanicallyrobust,light-exoskeletonwouldnotmoveinconcertwithitswearer.Wereal-weight,anddurableoutÞttosurpasstypicalhumanlimitations.ize,however,thatmaximizingsystemsensitivitytoexternalforcesBLEEXhasnumerouspotentialapplications;itcanprovidesol-andtorquesleadstoalossofrobustnessinthesystem.diers,disaster-reliefworkers,wildÞreÞghters,andotheremer-Takingintoaccountthisnewapproach,ourgoalwastodevelopgencypersonneltheabilitytocarryheavyloads,suchasfood,acontrolsystemforBLEEXwithhighsensitivity.Wewerefacedrescueequipment,Þrst-aidsupplies,communicationsgear,andwithtworealisticconcerns;theÞrstwasthatanexoskeletonwithweaponry,withoutthestraintypicallyassociatedwithdemandinghighsensitivitytoexternalforcesandtorqueswouldrespondtootherexternalforcesnotinitiatedbyitspilot.Forexample,ifsomeonepushedagainstanexoskeletonthathadhighsensitivity,ContributedbytheDynamicSystemsDivisionofASMEforpublicationinthetheexoskeletonwouldmovethesamewayitwouldmoveinJOURNALOFDYNAMICSYSTEMS,MEASUREMENT,ANDCONTROL.ManuscriptreceivedMarch31,2005;ÞnalmanuscriptreceivedSeptember17,2005.Assoc.Editor:Sunilresponsetotheforcesfromitspilot.AlthoughthefactthatitdoesK.Agrawal.PaperpresentedattheIEEERoboticsandAutomationConference2005.notstabilizeitsbehavioronitsowninresponsetootherforces14/Vol.128,MARCH2006Copyright©2006byASMETransactionsoftheASME ertheless,variousexperimentalsystemsinourlaboratoryhaveprovedtheoveralleffectivenessofthecontrolmethodinshadow-ingthepilotÕsmovement.2PreviousWorkIntheearly1960s,theU.S.DefenseDepartmentexpressedin-terestinthedevelopmentofaman-ampliÞer,aÒpoweredsuitofarmorÓthatwouldaugmentsoldiersÕliftingandcarryingcapabili-ties.In1962,theU.S.AirForcehadtheCornellAeronauticalLaboratorystudythefeasibilityofusingamaster-slaveroboticsystemasaman-ampliÞer.Inlaterwork,CornelldeterminedthatanexoskeletonÑanexternalstructureintheshapeofthehumanbodythathasfarfewerdegreesoffreedomthanahumanÑcouldaccomplishmostdesiredtasks9.From1960to1971,GeneralElectricdevelopedandtestedaprototypeman-ampliÞer,amaster-slavesystem,calledtheHardiman10Ð13.TheHardimanwasasetofoverlappingexoskeletonswornbyahumanoperator.Theouterexoskeleton
theslavefollowedthemotionsoftheinnerexoskeleton
themaster,whichfollowedthemotionsofthehu-manoperator.Thesestudiesfoundthatduplicatingallhumanmo-tionsandusingmaster-slavesystemswerenotpractical.Addition-ally,difÞcultiesinhumansensingandsystemcomplexitykeptitfromwalking.SeveralexoskeletonsweredevelopedattheUniversityofBel-gradeinthe1960sand1970stoaidpeoplewithparaplegiaresult-ingfromspinalcordinjury14,15.AlthoughtheseearlydeviceswerelimitedtopredeÞnedmotionsandhadlimitedsuccess,bal-ancingalgorithmsdevelopedforthemarestillusedinmanybi-pedalrobots16.Currentcommerciallyavailablerehabilitationdevices,suchastheÒLocomat,ÓuseasimilarpredeÞnedmotionstrategytotrainmusclesandnervepathwaysforpatientswithlocomotionimpairment17.TheÒRoboKneeÓisapoweredkneebracedevelopedbyMITthatfunctionsinparalleltothewearerÕskneeandtransfersloadtothewearerÕsankle
nottotheground18.ÒHALÓisanorthosisdevelopedbytheUniversityofTsukubainJapanthatisconnectedtothepatientÕsthighsandshanksandmovesthepatientÕslegsasafunctionoftheEMGsignalsmeasuredfromthewearer19,20.InourresearchworkatBerkeley,wehaveseparatedthetech-nologyassociatedwithhumanpoweraugmentationintolowerex-tremityexoskeletonsandupperextremityexoskeletons.Therea-sonforthiswastwofold;Þrst,wecouldenvisionagreatmanyapplicationsforeitherastand-alonelowerorupperextremityex-oskeletonintheimmediatefuture.Second,andmoreimportantlyforthedivision,isthattheexoskeletonsareintheirearlystages,Fig.1Berkeleylowerextremityexoskeleton„BLEEX…andpi-andfurtherresearchstillneedstobeconductedtoensurethatthelotRyanSteger.„1…Loadoccupiestheupperportionoftheupperandlowerextremityexoskeletonscanfunctionwell,inde-backpackandaroundthepowerunit,„2…rigidconnectionofthependently,beforewecanventureanattempttointegratethem.BLEEXspinetothepilotsvest,„3…powerunitandcentralcom-Withthisinmind,weproceededwiththedesignsofthelowerandputeroccupiesthelowerportionofthebackpack,„4…semi-rigidupperextremityexoskeletonsseparately,withlittleconcernforthevestconnectingBLEEXtothepilot,„5…oneofthehydraulicdevelopmentofanintegratedexoskeleton.WewillÞrstgiveaactuators,and„6…rigidconnectionoftheBLEEXfeettothepilotsboots„morephotographscanbefoundatsummaryoftheupperextremityexoskeletoneffortsatBerkeleyhttp://bleex.me.berkeley.edu….andthenwillproceedwiththedescriptionoftheBLEEXproject.Inthemid-1980s,weinitiatedseveralresearchprojectsonup-perextremityexoskeletonsystems,so-calledhumanextendersmaysoundlikeaseriousproblem;ifitdid
e.g.,usingagyro,the4,5,21.Themainfunctionofanupperextremityexoskeletonispilotwouldreceivemotionfromtheexoskeletonunexpectedlyhumanpoweraugmentationformanipulationofheavyandbulkyandwouldhavetostrugglewithittoavoidunwantedmovement.objects.Thesesystems,whicharealsoknownasassistdevicesorThekeytostabilizingtheexoskeletonandpreventingitfromfall-humanpowerextenders,cansimulateforcesonaworkerÕsarmsinginresponsetoexternalforcesdependsonthepilotÕsabilitytoandtorso.Theseforcesdifferfromandareusuallymuchlessthanmovequickly
e.g.,stepbackorsidewaystocreateastablesitu-theforcesneededtomaneuveraload.Whenaworkerusesanationforhimselfandtheexoskeleton.Forthis,asufÞcientlywideupperextremityexoskeletontomoveaload,thedevicebearsthecontrolbandwidthisneededsotheexoskeletoncanrespondtobulkoftheweightbyitself,whiletransferringtotheuserasabothpilotÕsvoluntaryandinvoluntarymovements
i.e.,reßexes.naturalfeedback,ascaled-downvalueoftheloadÕsactualweight.ThesecondconcernisthatsystemswithhighsensitivitytoForexample,fora20kg
44lbobject,aworkermightsupportexternalforcesandtorquesarenotrobusttovariations,andthere-only2kg
4.4lbwhilethedevicesupportstheremaining18kgfore,theprecisionofthesystemperformancewillbeproportional
39.6lb.Inthisfashion,theworkercanstillsensetheloadÕstotheprecisionoftheexoskeletondynamicmodel.Althoughthisweightandjudgehis/hermovementsaccordingly,buttheforceisaseriousdrawback,wehaveaccepteditasunavoidable.Nev-he/shefeelsismuchsmallerthanwhathe/shewouldfeelwithoutJournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/15 Fig.3Theexoskeletonsangularvelocityisshownasafunc-tionoftheinputtotheactuatorsandthetorquesimposedbythepilotontotheexoskeletonthereforeconsideredunknownvaluesinthisanalysis.Infact,oneoftheprimaryobjectivesindesigningBLEEXwastoensureapilotÕsunrestrictedinteractionwithit.TheequivalenttorqueonFig.2Simple1-DOFexoskeletonleginteractingwiththepilottheexoskeletonlegresultingfromthepilotÕsappliedforcesandleg.Theexoskeletonleghasanactuatorthatproducesatorquesisrepresentedbyd.torqueTaboutthepivotpointA.ThetotalequivalenttorqueIntheabsenceofgravity,
1andtheblockdiagramofFig.3associatedwithallforcesandtorquesfromthepilotontheexoskeletonisrepresentedbyd.representthedynamicbehavioroftheexoskeletonlegregardlessofanykindofinternalfeedbacktheactuatormayhavev=Gr+Sd
1thedevice.Inanotherexample,supposetheworkerusesthede-vicetomaneuveralarge,rigid,andbulkyobject,suchasanwhereGrepresentsthetransferfunctionfromtheactuatorinputrexhaustpipeinanautomotiveassemblyline.Thedevicewillcon-totheexoskeletonangularvelocity

actuatordynamicsarein-veytheforcetotheworkerasifitwasalight,single-pointmass.cludedinG.Inthecasewheremultipleactuatorsproducecon-Thislimitsthecross-coupledandcentrifugalforcesthatincreasetrolledtorquesonthesystem,risthevectoroftorquesimposedthedifÞcultyofmaneuveringarigidbodyandcansometimesontheexoskeletonbytheactuators.TheformofGandthetypeofproduceinjuriousforcesonthewrist.Inathirdexample,supposeinternalfeedbackfortheactuatorisimmaterialforthediscussionaworkerusesthedevicetohandleapoweredtorquewrench.Thehere.AlsobearinmindtheomissionoftheLaplaceoperatorinalldevicewilldecreaseandÞltertheforcestransferredfromtheequationsforthesakeofcompactness.wrenchtotheworkerÕsarmsotheworkerfeelsthelow-frequencyTheexoskeletonvelocity,asshownby
1,isaffectedbyforcescomponentsofthewrenchÕsvibratoryforcesinsteadofthehigh-andtorquesfromitspilot.Thesensitivitytransferfunction,S,frequencycomponentsthatproducefatigue.representshowtheequivalenthumantorqueaffectstheexoskel-TheBerkeleylowerextremityexoskeleton
BLEEXisnotanetonangularvelocity.Smapstheequivalentpilottorquedontoorthosisorabrace;unliketheabovesystems,itisdesignedtotheexoskeletonvelocity
.Iftheactuatoralreadyhassomesortofcarryaheavyloadbytransferringtheloadweighttothegroundprimarystabilizingcontroller,themagnitudeofSwillbesmall
nottothewearer.BLEEXhasfournewfeatures.First,anovelandtheexoskeletonwillonlyhaveasmallresponsetotheim-controlarchitecturewasdevelopedthatcontrolstheexoskeletonposedforcesandtorquesfromthepilotoranyothersource.Forthroughmeasurementsoftheexoskeletonitself2.Thiselimi-example,ahigh-gainvelocitycontrollerintheactuatorresultsinnatedproblematichuman-inducedinstabilityduetosensingthesmallSandconsequentlyasmallexoskeletonresponsetoexternalhumanforce8.Second,aseriesofhighspeciÞc-powerandforcesandtorques.Also,non-back-drivableactuators
e.g.,largespeciÞc-energypowersuppliesweredevelopedthatweresmalltransmissionratiosorservovalveswithoverlappingspoolsresultenoughtomakeBLEEXatrueÞeld-operationalsystem22Ð24.inasmallS,whichleadstoacorrespondinglysmallresponsetoThird,abodyLAN
localareanetworkwithaspecialcommuni-pilotforcesandtorques.cationprotocolandhardwarewasdevelopedtosimplifyandre-Notethatd
resultingtorquefrompilotontheexoskeletonisducethecablingtaskforthesensorsandactuatorsneededfornotanexogenousinput;itisafunctionofthepilotdynamicsandexoskeletoncontrol25,26.Finally,aßexibleandversatileme-variables,suchaspositionandvelocityofthepilotandtheexosk-chanicalarchitecturewaschosentodecreasecomplexityandeletonlegs.Thesedynamicschangefrompersontopersonandpowerconsumption3.Thispaperfocusesonthecontrolarchi-withinapersonasafunctionoftimeandposture.Wewilladdtectureandgivesanoverviewoftheelectronicdesignandthethesedynamicstoouranalysislaterinthepaper,butitisunrelatedbiomimeticmechanicaldesignoftheexoskeleton.Forfurthertothepurposeofcurrentdiscussion.Wealsoassumethatdisonlydepthineachofthesefourareas,thereaderisreferredtothefromthepilotanddoesnotincludeanyotherexternalforcesandpublicationsreferencedabove.torques.Theobjectiveistoincreaseexoskeletonsensitivitytopilot3ControllerDescriptionforcesandtorquesthroughfeedbackbutwithoutmeasuringd.Inotherwords,weareinterestedincreatingasystemthatallowsthe3.1ASimpleOne-Degree-of-Freedom(1-DOF)Example.pilottoswingtheexoskeletonlegeasily.MeasuringdtocreateThecontroloftheexoskeletonismotivatedherethroughthesuchsystemsdevelopsseveralhard,butultimatelysolvableprob-simple1-DOFexampleshowninFig.2.ThisÞgureschematicallylemsinthecontrolofalowerextremityexoskeleton.Someofdepictsahumanlegattachedorinteractingwitha1-DOFexosk-thoseproblemsarebrießydescribedasfollows:eletonleginaswingconÞguration
nointeractionwiththeground.Forsimplicity,theexoskeletonlegisshownasarigid1.Dependingonthearchitectureanddesignoftheexoskel-linkpivotingaboutajointandpoweredbyasingleactuator.Theeton,oneneedstoinstallseveralforceandtorquesensorstoexoskeletonleginthisexamplehasanactuatorthatproducesameasureallforcesfromthepilotontheexoskeletonbecausetorqueaboutpivotpointA.thepilotisincontactwiththeexoskeletonatseveralloca-Althoughthepilotisattachedsecurelytotheexoskeletonatthetions.Theselocationsarenotknowninadvance.Forex-foot,otherpartsofthepilotleg,suchastheshanksandthighs,canample,wehavefoundthatsomepilotsareinterestedinhav-contacttheexoskeletonandimposeforcesandtorquesontheingbracesconnectingBLEEXattheshankswhilesomeareexoskeletonleg.Thelocationofthecontactsandthedirectionofinterestedinhavingthemonthethighs.Inclusionofsensorsthecontactforces
andsometimescontacttorquesvaryandareonalegtomeasureallkindsofhumanforcesandtorques16/Vol.128,MARCH2006TransactionsoftheASME whereOistheexoskeletonmaneuveringbandwidth.Inclassicalandmoderncontroltheory,everyeffortismadetominimizethesensitivityfunctionofasystemtoexternalforcesandtorques.Butforexoskeletoncontrol,onerequiresatotallyoppositegoal:maximizethesensitivityoftheclosed-loopsystemtoforcesandtorques.Inclassicalservoproblems,negativefeed-backloopswithlargegainsgenerallyleadtosmallsensitivityFig.4FeedbackcontrolloopisaddedtoblockdiagramofFig.withinabandwidth,whichmeansthattheyrejectforcesand3.Cisthecontrolleroperatingonlyontheexoskeletontorques
usuallycalleddisturbances.However,theaboveanalysisvariables.statesthattheexoskeletoncontrollerneedsalargesensitivitytoforcesandtorques.Fromtheperspectiveofthepilot,thishastheeffectmakingtheexoskeletonfeelandbehavelikeaverysmallmasswhenthesensitivityoftheclosed-loopsystemtoforcesandmayresultinasystemsuitableforalaboratorysettingbuttorquesishigh.notrobustenoughtobedeployedintheÞeld.Toachievealargesensitivityfunction,weusetheinverseofthe2.IftheBLEEXdesignissuchthattheforcesandtorquesexoskeletondynamicsasapositivefeedbackcontrollersothattheappliedbythepilotontheexoskeletonarelimitedtoaspeci-loopgainfortheexoskeletonapproachesunity
slightlylessthanÞedlocation
e.g.,thepilotfoot,thenthesensorthatmea-1.Assumingpositivefeedback,
2canbewrittenassuresthepilotforcesandtorqueswillalsoinadvertentlymeasureotherforcesandtorquesthatarenotintendedforvSlocomotion.ThisisamajordifferencebetweenmeasuringSnew==
5forcesfrom,forexample,thehumanhandsandmeasuringd1−GCforcesfromthehumanlowerlimbs.Usingourhands,weareIfCischosentobeC=0.9G−1,thenthenewsensitivitytransferabletoimposecontrolledforcesandtorquesonupperex-functionisSnew=10S
tentimesforceampliÞcation.Ingeneral,tremityexoskeletonsandhapticsystemswithveryfewun-werecommendtheuseofpositivefeedbackwithacontrollercho-certainties.However,ourlowerlimbshaveotherprimarysenasandnonvoluntaryfunctions,suchasloadsupport,thattakepriorityoverlocomotion.−1−1C=
1−G
63.OneoptionwehaveexperimentedwithwastheinstallationofsensingdevicesforforcesonthebottomofthepilotÕswhereistheampliÞcationnumbergreaterthanunity
fortheboots,wheretheyareconnectedtoBLEEX.Sincetheforce−1aboveexample,=10ledtothechoiceofC=0.9G.EquationonthebottomofthepilotÕsboottravelsfromheeltotoe
6simplystatesthatapositivefeedbackcontrollerneedstobeduringnormalwalking,severalsensorsarerequiredtomea-chosenastheinversedynamicsofthesystemdynamicsscaledsurethepilotforce.Ideally,wewouldhaveamatrixofforce−1downby
1−.Notethat
6prescribesthecontrollerinthesensorsbetweenthepilotandtheexoskeletonfeettomea-absenceofunmodeledhigh-frequencyexoskeletondynamics.Insurethepilotforcesatalllocationsandatalldirections.Inpractice,Calsoincludesaunitygainlow-passÞltertoattenuatepractice,onlyafewsensorscouldbeaccommodated:atthetheunmodeledhigh-frequencyexoskeletondynamics.toe,ball,midfoot,andtheheelyet.ThisoptionhasledtoTheabovemethodworkswellifthesystemmodel
i.e.,Gisthickandbulkysoles.wellknowntothedesigner.Ifthemodelisnotwellknown,then4.Thebottomsofthehumanbootsexperiencecyclicforcesthesystemperformancewilldiffergreatlyfromtheonepredictedandtorquesduringnormalwalkingthatleadtofatigueandby
5,andinsomecasesinstabilitywilloccur.Theabovesimpleeventualsensorfailureifthesensorisnotdesignedandiso-solutioncomeswithanexpensiveprice:robustnesstoparameterlatedproperly.variations.Inordertogettheabovemethodworking,oneneedstoFortheabovereasonsandourexperienceinthedesignofvari-knowthedynamicsofthesystemwell.Section3.2discussesthisouslowerextremityexoskeletons,itbecameevidentthattheex-trade-off.istingstateoftechnologyinforcesensingcouldnotprovidero-3.2RobustnesstoParameterVariations.Thevariationinbustandrepeatablemeasurementofthehumanlowerlimbforcethenewsensitivitytransferfunctionwhenpositivefeedbackisontheexoskeleton.Ourgoalthenshiftedtodevelopinganexosk-usedisgivenbyeletonwithalargesensitivitytoforcesandtorquesfromtheop-eratorusingmeasurementsonlyfromtheexoskeleton
i.e.,nosensorsonthepilotortheexoskeletoninterfacewiththepilot.SnewSGCGCreatingafeedbacklooponlyfromtheexoskeletonvariables,as=+
7SnewS1−GCGshowninFig.4,thenewclosed-loopsensitivitytransferfunctionisIfGCisclosetounity
whentheforceampliÞcationnumbervSislarge,anyparametervariationonmodelingwillbeampliÞedasSnew==
2well.Forexample,iftheparameteruncertaintyinthesystemisd1+GC10%,i.e.,Observationof
2revealsthatSnewS,andthereforeanynegativefeedbackfromtheexoskeleton,leadstoanevensmallersensitivitytransferfunction.Withrespectto
2,ourgoalistoGS=0.10and=0designacontrollerforagivenSandGsuchthattheclosed-loopGSresponsefromdtov
thenewsensitivityfunctionasgivenby
2isgreaterthantheopen-loopsensitivitytransferfunction
i.e.,Sthen
7resultsinwithinsomeboundedfrequencyrange.ThisdesignspeciÞcationisgivenbyinequalitySnewGC=0.10
8SnewS"

0,O
3Snew1−GCoralternatively−1NowassumeCischosensuchthatC=0.9G.Substitutinginto1+GC1"

0,O
4
8resultsinJournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/17 feedbackloopaffectstheexoskeleton.Whilethelowerfeedbackloopispositive
potentiallydestabilizing,theupperfeedbackloopstabilizestheoverallsystemofpilotandexoskeletontakenasawhole.3.4EffectofPilotDynamicsonClosed-LoopStability.HowdoesthepilotÕsdynamicbehavioraffecttheexoskeletonbehavior?InordertogetanunderstandingofthesystembehaviorFig.5Thisblockdiagramshowshowanexoskeletonmoves.inthepresenceofpilotdynamics,weuseour1-DOFsystemandTheupperloopshowshowitspilotmovestheexoskeletonassumeHisalineartransferfunction.Thestabilityofthesystemthroughappliedforces.Thelowerloopshowshowthecontrol-showninFig.5isdecidedbytheclosed-loopcharacteristiclerdrivestheexoskeleton.equation1+SH−GC=0
11Snew=0.90.
9IntheabsenceoffeedbackcontrollerC,thepilotcarriestheSnewentireload
payloadplustheweightoftheexoskeletontorso.TheEquation
9indicatesthatanyparametervariationdirectlyaf-stabilityinthiscaseisdecidedbythecharacteristicequationfectsthesystembehavior.Intheaboveexample,a10%errorinmodelparametersresultsinninetimesthevariationinthesensi-1+SH=0
12tivityfunction.Thisiswhymodelaccuracyiscrucialtoexoskel-Characteristicequation
12isalwaysstablesinceitrepresentsetoncontrol.thecoupledpilotandexoskeletonbehaviorwithoutanycontrollerTogettheabovemethodworkingproperly,oneneedstounder-
i.e.,whenGC=0.Providednoneuromuscularcontroldisordersstandthedynamicsoftheexoskeletonquitewell,asthecontrollerexist,ahumancoupledtoanentirelypassivesystemisnaturallyisheavilymodelbased.Onecanseethisproblemasatrade-off:stable.Forexample,ifoneweretoholdingapurelypassiveob-thedesignapproachdescribedaboverequiresnosensor
e.g.,ject,suchasapencil,thereislittlechancethattheinteractionwithforceorEMG20intheinterfacebetweenthepilotandthetheobjectwouldbecomeunstable.WhenfeedbackloopCisexoskeleton;onecanpushandpullagainsttheexoskeletoninanyadded,theclosed-loopcharacteristicequationchangesfrom
12directionandatanylocationwithoutmeasuringanyvariablesonto
11,andusingthesmall-gaintheorem,onecanshowthatthetheinterface.However,thecontrolmethodrequiresaverygoodclosed-loopstabilityisguaranteedaslongasinequality
13ismodelofthesystem.Atthistime,ourexperimentswithBLEEXsatisÞedhaveshownthatthiscontrolschemeÑwhichdoesnotstabilizeBLEEXÑforcestheexoskeletontofollowwide-bandwidthhu-GC1+SH"

0,
13manmaneuverswhilecarryingheavyloads.WehavecometoAccordingto
6,CischosensuchthatGC1,andtherefore,believe,torephraseFriedrichNietzsche,thatthatwhichdoesnotintheabsenceofuncertainties,
13isguaranteedaslongas1stabilize,willonlymakeusstronger.1+SH.Unlikecontrolmethodsutilizedinthecontrolofthe3.3PilotDynamics.Therearemanyapproachestomodelingupperextremityexoskeletons21,thehumandynamicsinthethedynamicsofahuman,rangingfromcompleteneuromuscu-controlmethoddescribedherehaslittlepotentialtodestabilizetheloskeletalmodels27,28toasimpliÞedspringdamperrepresen-system.EventhoughthefeedbackloopcontainingCispositive,tation.Inparticular,twotypesofhumanmusclemodelinghavethefeedbackloopcontainingHstabilizestheoverallsystemofbeenusedsuccessfullytoprovideinsightintohumandynamics.pilotandexoskeleton.OneisbasedontheinvestigationofthemolecularorÞberrangeExample.Fora1-DOFsystem,S=G=1/JS,visangularveloc-ofthemuscle,whilethesecondisbasedontherelationshipbe-ity,Jisthemomentofinertia,andsistheLaplaceoperator.Thetweentheinputandoutputpropertiesofthemuscle.See29,30humanimpedanceismodeledasH=MHs+CH,whereMHandCHforin-depthmodelingandanalysis.Wehavechosenthesecondarepositivequantities.If=10andthecontrollerischosenasapproachandreportedourpreliminaryworkasappliedtohapticC=0.9Js,thenewsensitivityfunctionistentimeslargerthanthesystemsandhumanpowerampliÞers.originalsensitivityfunctionInourcontrolscheme,thereisnoneedtoincludetheinternalcomponentsofthepilotlimbmodel;thedetaileddynamicsofvSnerveconduction,musclecontraction,andcentralnervoussystemSnew===10S
14d1−GCprocessingareimplicitlyaccountedforinconstructingthedy-namicmodelofthepilotlimbs.Thepilotforceontheexoskel-ThesystemcharacteristicequationwhenC=0isgivenby
15eton,d,isafunctionofboththepilotdynamicsHandthekine-andalwaysresultsinastablesystemmaticsofthepilotlimb
e.g.,velocity,position,oracombination
J+MHs+CHthereof.Ingeneral,Hisdeterminedprimarilybythephysical1+SH=
15propertiesofthehumandynamics.HereweassumeHisanon-JslinearoperatorrepresentingthepilotimpedanceasafunctionofTheclosed-loopcharacteristicequationwhenapositivefeed-thepilotkinematicsbackloopisusedisgivenby
16andalsoresultsinastablesystemd=−H
v
10ThespeciÞcformofHisnotknownotherthanthatitresultsin
0.1J+MHs+CHthehumanmuscleforceontheexoskeleton.Figure5represents1+SH−GC=
16Jstheclosed-loopsystembehaviorwhenpilotdynamicsisaddedtotheblockdiagramofFig.4.ExaminingFig.5revealsthat
5,Evenifischosenasalargernumber,thesystemintheab-representingthenewexoskeletonsensitivityfunction,isnotaf-senceofparameteruncertainties,isstable.NowsupposeJ/J=fectedbythefeedbackloopcontainingH.−20%,i.e.,S/S=G/G=20%,thenthevariationinnewsensi-Figure5showsanimportantcharacteristicforexoskeletoncon-tivityfunctionistrol.Onecanobservetwofeedbackloopsinthesystem.TheupperfeedbacklooprepresentshowforcesandtorquesfromthepilotSnewSGCG=+=200%
17affecttheexoskeleton.ThelowerloopshowshowthecontrolledSnewS1−GCG18/Vol.128,MARCH2006TransactionsoftheASME Fig.7Rigidattachmentbindingsbetween„a…thepilotbootand„b…theBLEEXfootingmechanismandacompliant,butloadbearing,toesectionthatbeginsatmidfootandextendstothetoe.TheBLEEXfoothasacompressiblerubbersolewithatreadpatternthatprovidesbothshockabsorptionandtractionwhilewalking.TherubbersoleoftheBLEEXfootcontainsembeddedsensors,asshowninFig.8thatdetectthetrajectoryoftheBLEEX-groundreactionforceFig.6ThepilotvestsshownhereandinFig.1aredesignedtostartingfromheel-striketotoe-off.ThisinformationisusedintheuniformlydistributetheBLEEX-pilotforceonthepilotsupperBLEEXcontrollertoidentifytheBLEEXfootconÞgurationrela-bodytivetotheground.AlthoughbiomechanicalstudiesofwalkingfrequentlyidentifysevenormoredistinctphasesofthehumanwalkinggaitcycleInthiscase,GC=
1/0.8Js0.9Js=9/8,S=1/0.8Js,andthe31,forsimplicityincontrolweconsiderBLEEXtohavethreeclosed-loopcharacteristicpolynomialisrepresentedbydistinctphases
showninFig.9,whichmanifesttothreedifferent
10MH−Js+10CHdynamicmodels:1+SH−GC=
188Js1.Singlesupport:OnelegisinthestanceconÞgurationwhileEquation
18statesthatthesystemisunstableifJ10MH.anotherlegisinswing.Thus,thesystemisvulnerabletomodelparameteruncertainties.2.Doublesupport:BothlegsareinstanceconÞgurationandInsummary,thecontrollerdiscussedhereisstablewhenwornbysituatedßatontheground.thepilotaslongasparameteruncertaintiesarekepttoaminimum.3.Doublesupportwithoneredundancy:BothlegsareinstanceconÞguration,butonelegissituatedßatonthegroundwhiletheotheroneisnot.4ImplementationonBLEEXTheabovediscussionmotivatedthedesignphilosophyusingaUsingtheinformationfromthesensorsinthefootsole,the1-DOFsystem.BLEEX,asshowninFig.1,isasystemwithcontrollerdeterminesinwhichphaseBLEEXisoperatingandmanydegreesoffreedomandthereforeimplementationofwhichofthethreedynamicmodelsapply.BLEEXcontrolneedsfurtherattention.EachBLEEXleghasInourinitialcontroldesignprocess,wedecoupledthecontrolthreedegreesoffreedomatthehip,onedegreeoffreedomattheoftheabduction-adductionDOFatthehipfromthecontrolofknee,andthreedegreesoffreedomattheankle.Boththeßexion-jointsinthesagittalplane.Thisisvalidbecausewenotedthroughextensionandabduction-adductiondegreesoffreedomatthehipmeasurementsthattheabduction-adductionmovementsduringareactuated.Thekneehasoneßexion-extensiondegreeoffree-normalwalking
0.9m/sor2mpharerathersmall.Incom-domthatisactuated.Theankleplantar/dorsißexion
inthesagit-parisontothemovementsinthesagittalplane,theabduction-talplaneisalsoactuated.Theotherthreedegreesoffreedomadductionmovementscanbeconsideredquasi-staticmaneuvers
i.e.,rotationandabduction-adductionattheankleandrotationatwithlittledynamicalaffectsontherestofsystem.Thisindicatesthehipareequippedwithpassiveimpedancesusingsteelspringsthattheexoskeletondynamicsinthesagittalplaneareaffectedandelastomers.Insummary,eachBLEEXleghasfourpoweredonlybytheabduction-adductionangleandnotbytheabduction-degreesoffreedom:hipjoint,kneejoint,andanklejointintheadductiondynamics.Forthesakeofbrevity,Secs.4.1Ð4.3de-sagittalplane,andahipabduction-adductionjoint.scribethecontrolmethodinthesagittalplaneforagivensetofThepilotandBLEEXhaverigidmechanicalconnectionsattheabduction-adductionangles.torsoandthefeet;everywhereelse,thepilotandBLEEXhave4.1SingleSupport.Inthesingle-supportphase,BLEEXiscompliantorperiodiccontact.Theconnectionatthetorsoismademodeledasthe7-DOFseriallinkmechanisminthesagittalplaneusingavest,twovariationsofwhichcanbeseeninFig.1andshowninFig.10.TheinversedynamicsofBLEEXcanbewrittenFig.6.Oneoftheessentialobjectivesinthedesignofthesecus-inthegeneralformastomvestswastoallowthedistributionoftheforcesbetweenBLEEXandthepilot,therebypreventingabrasion.Thesevestsaremadeofseveralhardsurfacesthatarecompliantlyconnectedtoeachotherusingthickfabric.TheadjustmentmechanismsinthevestsallowforasnugÞttothepilot.Thevestsincluderigidplates
withholepatternsontheirbacksforconnectiontotheBLEEXtorso.ThepilotÕsshoesorboots
Fig.7
aattachtotheBLEEXfeetusingamodiÞedquick-releasebindingmechanismsimilartosnowboardbindings
Fig.7
b.Aplatewiththequick-releasemechanismisattachedtotherigidheelsectionoftheBLEEXfoot.EarlyversionsoftheBLEEXsystemhadthepilotwearingaFig.8ThesensorysysteminoneprototypeBLEEXfootsolestandardbootthathashadamatingbindingcleatsecuredtotheiscomposedofpressuresensitivesemi-conductiverubberem-heel.ThecleatonthemodiÞedpilotbootdoesnotinterferewithbeddedinapolyurethanesole„Fig.7„b…….ThisfootmeasuresnormalwearwhenthepilotisunclippedfromBLEEX.Thethegroundreactionforceprofileatfourlocations:toe,ball,BLEEXfootiscomposedoftherigidheelsectionwiththebind-midfoot,andheel.JournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/19 Fig.9ThreephasesoftheBLEEXwalkingcycleM
¨+C
,˙˙+P
=T+d
19T=Pˆ
+
1−−1Mˆ
¨+Cˆ
,˙˙
20where=,,...,TandT=0,T,T,...,TT.127126Cˆ
,˙,Pˆ
,andMˆ
aretheestimatesoftheCoriolismatrix,Misa77inertiamatrixandisafunctionof,C
,˙isagravityvector,andtheinertiamatrix,respectively,forthesystemshowninFig.10.Notethat
20resultsina71actuatortorque.77centripetalandCoriolismatrixandisafunctionofand˙,SincethereisnoactuatorbetweentheBLEEXfootandtheandPisa71vectorofgravitationaltorquesandisafunctionofground,thetorqueprescribedbytheÞrstelementofTmustbeonly.Tisthe71actuatortorquevectorwithitsÞrstelementprovidedbythepilot.SubstitutingTfrom
20into
19yieldssettozerosincethereisnoactuatorassociatedwithjointangle1
i.e.,anglebetweentheBLEEXfootandtheground.disthe−1M
¨+C
,˙˙+P
=Pˆ
+
1−Mˆ
¨+Cˆ
,˙˙+deffective71torquevectorimposedbythepilotonBLEEXatvariouslocations.Accordingto
6,wechoosethecontrollertobe
21theBLEEXinversedynamicsscaledby
1−−1,whereistheInthelimitwhenM
=Mˆ
,C
,˙=Cˆ
,˙,P
=Pˆ
,ampliÞcationnumber.andissufÞcientlylarge,dwillapproachzero,meaningthepilotcanwalkasifBLEEXdidnotexist.However,itcanbeseenfrom
21thattheforcefeltbythepilotisafunctionofandtheaccuracyoftheestimatesCˆ
,˙,Pˆ
,andMˆ
.Ingeneral,themoreaccuratelythesystemismodeled,thelessthehumanforcedwillbe.Inthepresenceofvariationsinabduction-adductionangles,onlyP
inEqs.
19and
20needstobemodiÞed.4.2DoubleSupport.Inthedouble-supportphase,bothBLEEXfeetareßatontheground.Theexoskeletonismodeledastwoplanar3-DOFseriallinkmechanismsthatareconnectedtoeachotheralongtheiruppermostlinkasshowninFig.11
a.Theinversedynamicsfortheseseriallinksarerepresentedby
22and
23.ML
mTL,L¨L+CL
mTL,˙L,L˙L+PL
mTL,L=TL+dL
22MR
mTR,R¨R+CR
mT,˙R,R˙R+PR
mTR,R=TR+dR
23Rwhere=Tand=T.mandmareLL1L2L3RR1R2R3TRTLeffectivetorsomassessupportedbyeachleg,andmTisthetotaltorsomasssuchthatmT=mTR+mTL
24ThecontributionsofmToneachleg
i.e.,mTLandmTRarechosenasfunctionsofthelocationofthetorsocenterofmassrelativetothelocationsoftheanklessuchthatmTRxTL=
25mTLxTRFig.10SagittalplanerepresentationofBLEEXinthesingle-wherexTListhehorizontaldistancebetweenthetorsocenterofstancephase.ThetorsoincludesthecombinedexoskeletonmassandtheleftankleandxTRisthehorizontaldistancebetweentorsomechanism,payload,controlcomputer,andpowerthetorsocenterofmassandtherightankle.Forexample,ifthesource.centerofmassofthetorsoislocateddirectlyabovetherightleg,20/Vol.128,MARCH2006TransactionsoftheASME Fig.12Thecontrollerreliesonahigh-speedsynchronousringnetworktopologywhereseveralremoteinput/outputnet-worknodes„shownasI/OModule#1…collectlocalsensordataanddistributelocalactuationcommands5ControlImplementationFig.11Sagittal-planerepresentationofBLEEXin„a…theSinceallcomputationsrequiredtoimplementthecontrolaredouble-supportphaseand„b…thedouble-supportphasewithoneredundancyconductedonasinglecomputer,weneededacontrolplatformtominimizethenumberofsignalwiresinthesystem.Theexoskel-etonelectronicssystem,EXOLINK,wasdesignedtosimplifyandreducethecablingtaskofallthesensorsandactuatorsneededforexoskeletoncontrol.Itreliesonahigh-speedsynchronousringthenmTL=0andmTR=mT.Similartothesinglestancephase,thenetworktopologywhereseveralelectronicremoteinput-outputcontrollersarechosensuchthatmodules
RIOMresideinaring.EachRIOMisincommunica-T=Pˆ
m,+
1−−1Mˆ
m,¨+Cˆ
m,,˙˙tionwithseveralsensorsandoneactuatorincloseproximity,andLLTLLLTLLLLTLLLLincludeseightsixteen-bitanalog-to-digitalconverters
ADC,two
26quadraturecounters,eightbitsofdigitalinputandoutputports,twodigital-to-analogconverters
DACandanalogÞlters.EachT=Pˆ
m,+
1−−1Mˆ
m,¨+Cˆ
m,,˙˙RIOMalsoincludeslocalizedpowerregulationandisolationtoRRTRRRTRRRRTRRRRminimizesignalnoiseandsystemgroundloops,andabuilt-in
27FPGAmanagesallRIOMdatatransactionandÞltering.ThedataNeedless-to-say,
25isvalidonlyforquasi-staticconditions,gatheredbyeachmoduleareencodedandtransmitteddigitallytowheretheaccelerationsandvelocitiesaresmall.Thisis,infact,acentralcomputerthroughthering.TheEXOLINKhasfourthecase,sinceinthedouble-supportphase,bothlegsareontherings,twoofwhichareassociatedwiththetwolegs.EachringgroundandBLEEXÕsangularaccelerationandvelocitiesarequitecontainsthreeremoteinput-output
I/Omodules
Fig.12.Asmall.Thisallowsustosimplify
26and
27duringslowwalk-thirdringisconnectedtoagraphicaluserinterfacefordebuggingingbyremovingalltermsexcepttheestimatesofthegravitationalanddataacquisition.Afourthringisusedtoaccommodateothervectors.electronicandcommunicationgearswhicharenotrelatedtotheexoskeletonbutwhichthepilotmaychoosetocarry.Eachring4.3DoubleSupportWithOneRedundancy.Doublesup-canaccommodateuptoeightRIOMs
Fig.13.TheEXOLINKportwithoneredundancyismodeledasa3-DOFseriallinkconsistsofamicrocomputerandasupervisorIOmodule
SIOM.mechanismforthestancelegwiththefootßatonthegroundandTheSIOMincludesaFPGAprogramedtoserveasthecommuni-a4-DOFseriallinkmechanismforthestancelegthatisnotcom-cationhubforallfourrings.Atransceiverchipresidinginthepletelyontheground
Fig.11
b.EachseriallinksupportsaSIOMandalltheRIOMsallowfordatatransferatarateofportionofthetorsoweight.Theinversedynamicsfortheseserial1500Mb/s.Currently,a650MHzPentiumPC-104form-factorlinksarerepresentedby
28and
29,whereinthespeciÞcmo-microcomputerisusedtoimplementthecontrolalgorithm,andmentshowninFig.11
b,theleftleghasfourdegreesoffreedomthecurrentexoskeletonutilizes75%oftheI/Ocapabilityoftheandtherightleghasthreedegreesoffreedom.EXOLINK.Theuseofahigh-speedsynchronousnetworkinML
mTL,L¨L+CL
mTL,˙L,L˙L+PL
mTL,L=TL+dL
28placeofthetraditionalparallelmethodenablestheexoskeletontoreducetheover200sensorandactuatorwirestoonly24commu-nicationandpowerwires.WhilethesensorsarereadattherateofMR
mTR,R¨R+CR
mTR,˙R,R˙R+PR
mTR,R=TR+dR
2910KHz,thecontrolisupdatedat4KHz
controlsamplingtimeiswhereL=L1L2L3L4T,R=R1R2R3T,TL250s.Thedetailedimplementationisdescribedin26.=0TTTT,andT=TTTT.mandmaretheL1L2L3RR1R2R3TRTLeffectivetorsomassessupportedbyeachlegandarecomputed6ExperimentalHardwaresimilartothedouble-supportcasebyuseof
25.Utilizing
28Fundamentaltodesigningalowerextremityexoskeletonisse-and
29asdynamicmodelsoftheexoskeleton,
26and
27arelectingtheoverallstructuralarchitectureofthelegs.Manydiffer-usedascontrollersinthiscase.Clearly,theactuatortorquevectorentlayoutsofjointsandlimbscancombinetoformafunctioningassociatedwiththelegthathasfourdegreesoffreedom
e.g.,TLleg.Regardlessofwhetherlinearsliders,rotaryjoints,orgeneralinthecaseshowninFig.11
bisa41vector.Asinthesinglecompliancyareusedtoprovidethenecessarydegreesoffreedom,supportphase,thetorqueprescribedbytheÞrstelementofTmustthearchitecturegenerallyfallsintooneofafewcategories.beprovidedbythepilotbecausethereisnoactuatorbetweentheBLEEXfootandtheground.AsBLEEXgoesthroughthevarious6.1Anthropomorphic.Anthropomorphicarchitecturesat-phasesshowninFig.9,thesensorsshowninFig.8detectwhichtempttoexactlymirrorthehumanleg.Bykinematicallymatchingleghasfourdegreesoffreedomandwhichhasthreedegreesofthehuman,theexoskeletonÕslegpositionfollowsthehumanlegÕsfreedom.Thecontrollerthenchoosestheappropriatealgorithmposition.ThisgreatlysimpliÞesmanydesignissuesfromavoidingforeachleg.human/machinecollisionstopredictingtherequiredrangeofmo-JournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/21 Table1Exoskeleton-jointrangesofmotion.Exoskeletonflex-ibilitymustbelessthanthehumanflexibilitylimitsforsafety,butiskeptclosetothemaximumhumanflexibilitytomaximizemaneuverability.Fig.13EachRIOM„shownhere…isincommunicationwithseveralsensorsandoneactuatorincloseproximitytionfortheroboticjoints.However,manydifÞcultiesarrivewhendifÞculttodeveloparchitecturesigniÞcantlydifferentfromahu-tryingtomatchthehumanlegarchitecture.MajorissuesincludemanlegthatcanstillmovethefootthroughallthenecessarymatchingthehumanÕsknee,whichismoreofaslidingjointthanmaneuvers
i.e.fromturningtightcornerstodeepsquats.Safetyapurerotaryone.Also,fordifferentoperatorstoweartheexosk-issuesbecomemoreprominentwithnonanthropomorphicdesignseleton,itmustbehighlyadjustabletoensurethatalloftheexosk-becausetheexoskeletonmustbeabsolutelyprohibitedfromforc-eletonjointsalignwiththecorrespondinghumanjoints.Anthro-ingtheoperatorintoapositionhe/shecannotreach.Eventhoughpomorphicexoskeletonsareacommonlyseenarchitectureanthropomorphicexoskeletonsaremorecommon,aclevernonan-becauseitallowstheexoskeletontoattachtotheoperatorwher-thropomorphicarchitecturecouldleadtosimpleractuationoreverdesired.lowerenergyconsumption.6.2Nonanthropomorphic.Althoughnotascommoninex-6.3Pseudoanthropomorphic.TheBLEEXprojectchoseanoskeletons,manynonanthropomorphicdevicesarehighlysuc-architecturethatisalmostanthropomorphic.Iftheexoskeletoncessful
e.g.,bicycles.Nonanthropomorphicarchitecturesopenkinematicsareclosetohumankinematics,thenappropriaterangesupawiderangeofpossibilitiesforthelegdesignaslongastheofmotionforeachdegreeoffreedomcanbeeasilyapproximatedexoskeletonneverinterfereswithorlimitstheoperator.Oftenitisfromhumanphysiologicaldata.SimilarkinematicsalsomakeitFig.14BLEEXdegreesoffreedom22/Vol.128,MARCH2006TransactionsoftheASME Fig.15Humanpowerrequiredforwalking.Theflexion/extensiondirectionrequiresthemostpowerforallthreejoints„ankle,knee,andhip….Besidesthesesagittalplanedirections,thehipabduction/adductionrequiresthenextmostpower†31‡.easierfortheexoskeletontofollowthehumanthroughanyma-hasahip,knee,andankle.Generally,theexoskeletonkinematicsneuverandnotcollidewiththeoperator.However,attemptingtoaremodeledexactlyafterthehuman,buttheexoskeletondegreesexactlymatchhumankinematicscreatesmanydesignissues;thus,offreedomincludeafewkeysimpliÞcations.slightdifferencesaretoleratedforsimplicity,suchasapproximat-First,theexoskeletonkneejointissimpliÞedtoapurerotaryingthekneeasapurerotaryjoint.Sincethehumanandexoskel-joint.Ahumankneejointisacomplexcombinationofrollingandetonlegkinematicsarenotexactlythesame,thehumanandma-slidingbetweenthefemurandtibiathatallowsthejointÕscenterchineareonlyrigidlyconnectedattheextremitiesoftheofrotationtomoveasthekneebends31.Usingapurerotaryexoskeleton
feetandtorso.AnyotherrigidconnectionswouldjointatthekneesimpliÞesthedesignanddynamicmodeloftheleadtolargeforcesimposedontheoperatorduetothekinematicexoskeleton,butwillcausetheexoskeletonkneetonotexactlydifferences.However,compliantconnectionsalongthelegaremirrorthehumanknee.Also,themovingcenterofrotationofthetolerableaslongastheyallowrelativemotionbetweenthehumanhumankneeplaysanimportantroleinhelpingthelegslightlyandmachine.Iftheinertiasandmassesoftheexoskeletonleghyperextendtoanover-centerconÞguration.Thisfunctionwillbesegmentsaresimilartothecorrespondinghumanlimbs,thentheabsentintheexoskeletonkneeandthatcompromiseshouldbedesiredjointtorquesfortheexoskeletoncanbeestimatedusingacknowledged.humanclinicalgaitanalysis
CGAdata1.Usingapseudoan-AnotherkinematicsimpliÞcationintheexoskeletonisthelegthropomorphicarchitecture,theexoskeletonismucheasiertosizerotation.AhumanÕslegcanrotateasmallamountatmanydiffer-forvariousoperatorsandthejointrangesofmotionandtorquesentlocations:thehipjoint,kneejoint,alongtheshank,andintheareapproximatelyequaltothoseofahuman,buttherigidcon-anklejoint32.Tokeepthefunctionalityofthesemotions,butnectionspointsarelimitedtojustthefeetandtorso.simplifythedesign,theexoskeletonrotationiscondensedtojoints6.4DegreesofFreedom.SincetheBLEEXhasapseudoan-atthehipandankle.thropomorphicarchitecture,itsdegreesoffreedomneedtoap-Initially,thethreedegreesoffreedomatthehipweredesignedproximatelymatchahuman.Thus,likeahuman,theexoskeletontobecollocatedandalignedwiththehumanÕshipjoint.However,JournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/23 Fig.16Powerrequiredforascending/descendingStairs.Thekneerequirespowerwhenascendingstairsinsteadofabsorbingpowerasitdoesduringlevelwalking†35‡.allthesedesignssigniÞcantlylimitedthehiprotationbecauseof6.5RangeofMotion.TheBLEEXkinematicsaresimilartomechanicalinterferenceofthehiplinkageswiththemselvesorthehumanlegkinematics;thus,themotionlimitsforeachofthehuman.Therefore,therotationwasmovedsuchthatitnolongerexoskeletonjointsisdeterminedbyexamininghuman-jointrangesalignswiththehumanÕsrotation.Kinematically,theexoskeletonofmotion.Attheveryleast,theexoskeletonmustbeabletoleghassufÞcientdegreesoffreedomtoaccountforthemisalign-shadowthehumanthroughanormalwalkingmotion.CGAdatament.Similarallowancesweremadeattheankle,wheretheab-revealtheanglesofeachjointduringwalkingand,thus,themini-ductionandrotationaxesdonotalignwiththehumanÕsaxesofmumrangeofmotionforeachjoint30.Safetydictatesthattherotation.exoskeletonshouldnothavemotionlimitsgreaterthantheopera-tor.Therefore,themaximumrangeofmotionforeachjointisWiththesesimpliÞcations,theexoskeletonhassevendegreesofdeterminedbythehumanÕsmaximumßexibility32.Ideally,allfreedomperleg
Fig.14:thejointrangesofmotionwouldbeslightlysmallerthanthelim-¥threedegreesoffreedomatthehipitsofhumanßexibilitytopreventinjurybutallowmaximumma-¥onedegreeoffreedomattheknee
purerotationinthesag-neuverabilityoftheexoskeleton.However,asdiscussed,linearittalplaneactuatorsareusedinBLEEX,whichlimitssomeofthejoint¥threedegreesoffreedomattheanklerangesofmotion.Joints,suchastheknee,haveareducedßex-ibilitytopreventtheactuatorfromreachingsingularity.OtherAnadditionaldegreeoffreedomisaddedtotheexoskeletonlimitsonthejointsÕmotionweredeterminedthroughprototypefoot.Thefrontoftheexoskeletonfoot,underthehumanÕstoes,istesting.Forexample,theankleplantar-dorsißexionlimitsarecompliant.Thisallowstheexoskeletonfoottoßexwiththehu-outsidethenormalhumanßexibility.Mock-uptestingindicatedmanÕsfootastheygetupontheirtoes.thatthisadditionalßexibilityisnecessaryforfullmaneuverability24/Vol.128,MARCH2006TransactionsoftheASME sincethehumanÕsfootisnotheldcompletelyrigidrelativetotheoftheBerkeleyLowerExtremityExoskeleton
BLEEX,ÓIEEEInt.Conf.onexoskeletonfootTable1showstheexoskeleton-jointrangesofRoboticsandAutomation,April,Barcelona.3Zoss,A.,andKazerooni,H.,2005,ÒOntheMechanicalDesignoftheBerke-motion.leyLowerExtremityExoskeleton,ÓIEEEIntelligentRobotsandSystemsCon-ference,August,Edmunton.6.6WhichJointstoActuate?Eachexoskeletonleghas4Kazerooni,H.,1990,ÒHuman-RobotInteractionviatheTransferofPowerandsevendegreesoffreedom
eightcountingthetoeßexibility,butInformationSignals,ÓIEEETrans.Syst.ManCybern.,20
2,pp.450Ð463.arbitrarilydecidingtoactuateallofthemleadstounnecessarily5Kazerooni,H.,andGuo,J.,1993,ÒHumanExtenders,ÓASMEJ.Dyn.Syst.,highpowerconsumptionandcontrolcomplexity.Instead,amini-Meas.,Control,115
2B,pp.281Ð289.6Kazerooni,H.,andMahoney,S.,1991,ÒDynamicsandControlofRoboticmumnumberofjointsnecessarytomaintainfunctionalityshouldSystemsWornByHumans,ÓASMEJ.Dyn.Syst.,Meas.,Control,113
3,pp.beactuated.Fortheexoskeletontobefunctional,thehumanop-379Ð387.eratorshouldonlymovetheexoskeletonandpayloadwithamini-7Kazerooni,H.,andHer,M.,1994,ÒTheDynamicsandControlofaHapticmumamountofpower.Therefore,anydegreesoffreedomrequir-InterfaceDevice,ÓIEEETrans.Rob.Autom.,10
4,pp.453Ð464.8Kazerooni,H.,andSnyder,T.,1995,ÒACaseStudyonDynamicsofHapticingasubstantialamountofpowershouldbeactuated,orlimitedDevices:HumanInducedInstabilityinPoweredHandControllers,ÓJ.Guid.byanotherimpedance.ForthisÞrstgenerationofBLEEX,actua-ControlDyn.,18
1,pp.108Ð113.tionwasdesignedprimarilyforwalking;thus,onceagain,CGA9Mizen,N.J.,1965,ÒPreliminaryDesignfortheShouldersandArmsofadatawereusedtodeterminewhichdegreesoffreedomconsumedPowered,ExoskeletalStructure,ÓCornellAeronauticalLaboratoryReportNo.powerwhilewalking.VO-1692-V-4.10Groshaw,P.F.,GeneralElectricCo.,1969,ÒHardimanIArmTest,HardimanAsexpected,CGAdatashowthatthehighestamountofpowerIPrototype,ÓGeneralElectric,Schenectady,NY,ReportNo.S-70-1019.isusedforßexionandextensionattheankle,knee,andhip11GeneralElectricCo.,1968,ÒHardimanIPrototypeProject,SpecialInterim30,31,33,34
Fig.15.TheankleandhipbothrequiresigniÞcantStudy,ÓGeneralElectricSchenectady,NY,ReportNo.S-68-1060.positivepowerand,thus,needtobeactuated.Thekneemainly12Makinson,B.J.,GeneralElectricCo.,1971,ÒResearchandDevelopmentPro-totypeforMachineAugmentationofHumanStrengthandEndurance,Hardi-requiresnegativepower
itabsorbspowerwhilewalking;thus,itmanIProject,ÓGeneralElectric,Schenectady,NY,ReportNo.S-71-1056.couldbecontrolledwithdamper.Eventhoughtheactuationis13Mosher,R.S.,1960,ÒForce-ReßectingElectrohydraulicManipulator,Ómainlydesignedforlevelwalking,whenwalkingupsteps,anElectro-Technol.,pp.138Ð141.incline,orsquatting,thekneebecomesaverycriticaljointfor14Vukobratovic,M.,Hristic,D.,andStojiljkovic,Z.,1974,ÒDevelopmentofActiveAnthropomorphicExoskeleton,ÓMed.Biol.Eng.,12,pp.66Ð80.addingpositivepowertothesystem35
Fig.16.Therefore,the15Vukobratovic,M.,Ciric,V.,andHristic,D.,1972,ÒContributiontotheStudykneejointisalsoactuated.Besidesßexionandextension,hipofActiveExoskeletons,ÓProc.ofthe5thIFACCongress,Paris.abduction/adductionrequiresthemostpowerforwalkingbecause16Hirai,K.,Hirose,M.,Haikawa,Y.,andTakenaka,T.,1998,ÒTheDevelopmentitprovidesthelateralbalancingforces.TohelpwiththelateralofHondaHumanoidRobot,ÓProc.ofthe1998IEEEInternationalConferenceonRobotics&Automation,Leuven,Belgium,IEEE,NewYork,pp.1321Ðbalancingandmaneuverability,thehipabductionofBLEEXis1326.actuated.AccordingtoCGAdata,theotherdegreesoffreedomall17Colombo,G.,Jorg,M.,andDietz,V.,2000,ÒDrivenGaitOrthosistodohaveverysmallpowerconsumptionswhilewalkingand,thus,LocomotorTrainingofParaplegicPatients,Ó22ndAnnualInternationalConf.remainunactuated.oftheIEEE-EMBS,Chicago,July23Ð28.18Pratt,J.,Krupp,B.,Morse,C.,andCollins,S.,2004,ÒTheRoboKnee:AnExoskeletonforEnhancingStrengthandEnduranceDuringWalking,ÓIEEEIntl.Conf.onRoboticsandAutomation,NewOrleans.7Conclusion19Kawamoto,H.,Kanbe,S.,andSankai,Y.,2003,ÒPowerAssistMethodforAlthoughthereisstillsigniÞcantworktobedonebeforetheHAL-3EstimatingOperatorÕsIntentionBasedonMotionInformation,ÓProc.of2003IEEEWorkshoponRobotandHumanInteractiveCommunication,projectiscomplete,BLEEXhassuccessfullywalked,carryingitsMillbrae,CA,IEEE,NewYork,pp.67Ð72.ownweightandproducingitsownpower.ThismakesittheÞrst20Kawamoto,H.,andSankai,Y.,2002,ÒPowerAssistSystemHAL-3forgaitlowerextremityexoskeletoncapableofcarryingapayloadandDisorderPerson,ÓICCHP,July,Austria.beingenergeticallyautonomous.Currently,BLEEXhasbeen21Kazerooni,H.,1996,ÒTheHumanPowerAmpliÞerTechnologyattheUniver-sityofCalifornia,Berkeley,ÓInt.J.Rob.Autom.,19,pp.179Ð187.demonstratedtosupportupto70kg
exoskeletonweightplus22McGee,T.,Raade,J.,andKazerooni,H.,2004,ÒMonopropellant-DrivenFreepayload,walkatspeedsupto1.3m/s,andshadowtheoperatorPistonHydraulicPumpforMobileRoboticSystems,ÓASMEJ.Dyn.Syst.,throughmostmaneuverswithoutanyhumansensingorprepro-Meas.,Control,126
1,pp.75Ð81.gramedmotions.BLEEXisnotatypicalservomechanism.While23Raade,J.,andKazerooni,H.,2005,ÒAnalysisandDesignofaNovelPowerSourceforMobileRobots,ÓIEEE.Trans.Autom.Sci.Eng.,2
3,pp.226Ðprovidingdisturbancerejectionalongsomeaxespreventingmo-232.tioninresponsetogravitationalforces,BLEEXactuallyencour-24Amundsen,K.,Raade,J.,Harding,N.,andKazerooni,H.,2005,ÒHybridagesmotionalongotheraxesinresponsetopilotinterfaceforces.Hydraulic-ElectricPowerUnitforFieldandServiceRobots,ÓIEEEIntelligentThischaracteristicrequireslargesensitivitytopilotforces,whichRobotsandSystemsConference,August,Edmunton.25Kim,S.,Anwar,G.,andKazerooni,H.,2004,ÒHigh-SpeedCommunicationinvalidatescertainassumptionsofthestandardcontroldesignNetworkforControlsWithApplicationontheExoskeleton,ÓAmericanCon-methodologiesand,thus,requiresanewdesignapproach.ThetrolConference,Boston,June.controllerdescribedhereusestheinversedynamicsoftheexosk-26Kim,S.,andKazerooni,H.,2004,ÒHighSpeedRing-BasedDistributedNet-eletonasapositivefeedbackcontrollersothattheloopgainforworkedControlSystemforReal-TimeMultivariableApplications,ÓASMEInternationalMechanicalEngineeringCongress,Anaheim,November.theexoskeletonapproachesunity
slightlylessthan1.Ourcur-27Berstein,N.A.,1967,TheControlandRegulationofMovements,PergamonrentexperimentswithBLEEXhaveshownthatthiscontrolPress,London.schemehastwosuperiorcharacteristics:
iitallowsforthesame28Bizzi,E.,Hogan,N.,Mussa-Ivaldi,F.A.,andGiszter,S.,1992,ÒDoesthewidebandwidthmaneuversahumaniscapableofperformingandNervousSystemUseEquilibriumPointControltoGuideSingleandMultipleJointMovements?,ÓBehav.BrainSci.,15,pp.603Ð613.
iiitisunaffectedbychanginghumandynamics
i.e.,nochanges29Wilkie,D.R.,1950,ÒTheRelationBetweenForceandVelocityinHumantothecontrollerarerequiredwhenpilotsareswitched.Thetrade-Muscle,ÓJ.Physiol.
London,K110,pp.248Ð280.offisthatitrequiresarelativelyaccuratemodelofthesystem.A30Winters,J.M.,andStark,L.,1985,ÒAnalysisofFundamentalHumanMove-bodylocalareanetworktohostthecontrolalgorithmisdevelopedmentPatternsThroughtheUseonIn-DepthAntagonisticMuscleModels,ÓIEEETrans.Biomed.Eng.,BME32
10,pp.826Ð839.in25.Videoclipsthatdemonstratetheeffectivenessofthiscon-31Rose,J.,andGamble,J.G.,1994,HumanWalking,2nded.,Williams&trolschemecanbefoundathttp://bleex.me.berkeley.edu/Wilkins,Baltimore,p.26.bleex.htm.32Woodson,W.,Tillman,B.,andTillman,P.,1992,HumanFactorsDesignHandbook,McGraw-Hill,NewYork,pp.550Ð552.33Kirtley,C.,2005,ÒCGANormativeGaitDatabase:HongKongPolytechnicUniversity,10YoungAdults,ÓAccessedAugust,http://guardian.curtin.edu.au/Referencescga/data/.1Chu,A.,Kazerooni,H.,andZoss,A.,2005,ÒOntheBiomimeticDesignofthe34Linskell,J.,ÒCGANormativeGaitDatabase:LimbFittingCentre,Dundee,BerkeleyLowerExtremityExoskeleton
BLEEX,ÓIEEEInt.Conf.onRobot-Scotland,YoungAdult,ÓAvailableathttp://guardian.curtin.edu.au/cga/data/.icsandAutomation,April,Barcelona.35Riener,R.,Rabuffetti,M.,andFrigo,C.,2002,ÒStairAscentandDescentat2Kazerooni,H.,Racine,J.-L.,Huang,L.,andSteger,R.,2005,ÒOntheControlDifferentInclinations,ÓGaitandPosture,15,pp.32Ð34.JournalofDynamicSystems,Measurement,andControlMARCH2006,Vol.128/25