The Humanoid Robot ARMAR: Designand Control

TamimAsfour,KarstenBerns,andRudigerDillmann

ForschungszentrumInformatikKarlsruhe,Haid-und-Neu-Str. 10-14D-76131Karlsruhe,Germany�

asfour,dillmann � @ira.uka.de,berns@fzi.dehttp://wwwipr.ira.uka.de/ � asfour/armar

Abstract. Thispaperaddressesthemechanismdesignmethodologies,specification,andcontrolstrategiesof amobilemanipulationsystemfor thehumanoidrobotARMAR, thathasto work autonomouslyor interactivelyin cooperationwith humansin dynamicunstructuredenvironmentssuchasworkshopsor homes.

1 Intr oduction

Robotsof the currentgenerationhave beenusedin fields isolatedfrom the humansociety. They suffer majorshortcomingsbecauseof their limited abilities for manipulationandinteractionwith humans.Humanoidrobotsareexpectedto exist andwork togetherwith humanbeingsin the everydayworld suchashospitals,officesandhomesandto serve theneedsof elderlyanddisabledpeople[1]. Theserobotsmustbeableto copewith thewidevarietyof tasksandobjectsencounteredin dynamicunstructuredenvironments.In cooperationwith humanbeingshumanoidrobotsshouldsharethe sameworking spaceandshouldreacthumanfriendly. Therefore,they needalight-weightbody, high flexibility , many kindsof sensorsandhigh intelligence.They have to beadaptive to newsituationsandcapableof performingtasksin dynamicenvironments.Their designrequiresalsoa high extentofintegrationof mechanical,electronicalandcomputationaltechnologies.

Sincevery ancienttime, humanshave beentrying to replicatehumans.Between1495and1497LeonardodaVinci designedandpossiblybuilt thefirst articulatedanthropomorphicrobot[2]. Recently, humanoidroboticshasreceivedmuchinterestin theroboticresearchcommunity, andmany significantresultshavebeenachievedworld-wide [3–11]. The manipulationcapabilitiesand intelligenceof theserobotsare still far away from the humancapabilityin solvingcomplex servicetasks.At theForschungszentrumInformatik Karlsruhe(FZI) thehumanoidrobotARMAR is developedfor applicationslikeassistancein workshopsor homeenvironment[12] (seefigure1).Main focusof our researchis theprogrammingandexecutionof manipulationtasksof ARMAR by a directandreal-timemappingbetweentherobotandtheperson,whichdemonstratesthetask.

2 The Mechatronicsof ARMAR

Thehumanoidrobothastwenty-fivemechanicaldegrees-of-freedom(DOF). It consistsof anautonomousmobilewheel-drivenplatform,abodywith 4 DOF, two anthropomorphicredundantarmseachhaving 7 DOFs,two simplegripper and a headwith 3 DOF. The total weight of the upper-body of ARMAR is about45kg. This sectiondescribesthemechanics,sensorsystemandcomputerarchitectureof therobot.

2.1 Arms and Hand

In order to achieve a high degreeof mobility andto allow the simpleanddirect cooperationwith humans,thestructure(size,shapeandkinematics)of thearmandof thetorsoshouldbesimilar to thatof ahuman.Wedesignedtwo anthropomorphicarms,eachhaving 7 DOF anda lengthof 65 cm (including thegripper).Detailsaboutthemechanicsof the arm of ARMAR arereportedin [13]. Currently, a simpleparallel jaw gripperis implemented,howevera new humanoidfive-fingeredlightweighthandwith only oneactuatorand21 DOFis underconstruction[14]. The new handis designedfor anatomicalconsistency with the humanhand.This includesthe numberoffingers,theplacementandmotionof thethumb,theproportionsof thelink lengthsandtheshapeof thepalm.Thenew handaccommodateautomaticallyto theshapeof graspedobjects.It hasalsotheability of performingmostofhumanhandgraspingtypes.

2 TamimAsfour etal.

Fig.1. ThehumanoidrobotARMAR andits kinematicsmodel.The joint variables��� – ��� arethearmDOFs. �� – �� corre-spondto thedifferentjointsof thetorso: ��� describestherotation,while �� and ��� specifytheforward,backwardandsidewardbendingof of theupperbody. � representsthetelescopicjoint of thebody.

2.2 Mobil Platform

Thereareseveralrequirementsfor thelocomotionsystemof ahumanoidrobotto dealwith adynamicunstructuredenvironment.Mobility is necessaryto extendworkingspaceandto performcooperativetaskswith humans.Stabil-ity of themobilesystemis themostessentialto insurehuman’ssafety. Fromthis remarks,we useanautonomousmobilewheel-drivenplatform. It hasanoctagonalground-planwith a diameterof 70 cm anda differentialdriveconceptwith two activedrivenwheelsonthesides.Two passive,freerotatingwheelsarealsoused.Themaximumvelocityof theplatformis about1 � � . Theplatformis equippedwith ultrasonicsensors,aplanarlaser-scanner(seesection2.4), andsufficient batterypower to allow for autonomousoperation.Up to now it is not plannedto uselegs for the locomotionof the robot,sincein a workshopenvironmentit is not necessaryto have sucha flexiblelocomotionsystem.In factonefunctionnormallysupportedby legs is thechangeof the total height.This influ-encesthe workspaceof the robot.However, we installeda telescopicjoint ( ��� ) in the torsoof ARMAR to havethis degreeof freedomtoo.

2.3 Torso and Neck

Theupperbodyof ARMAR has4 DOF (figure1). It is placedon themobileplatformandsupportsa rotationofabout������� . It alsocanbebentforward,backwardandsideward(circa ������� ). To adapttheheightof therobot(180cm), a telescopicjoint is includedin the body. With this joint the total heightof the robot canbe increasedby40cm.TheNeckof ARMAR has3 DOF.

2.4 Sensors

The sensorsystemconsistsof angleencodersfor eachjoint with a resolutionof ������� . The currentas well asthe voltageof eachmotor aremeasuredanddeterminedby specialpower electroniccard.For gripping variouskindsof objectsanartificial skin is placedontheinnersideof thegripper. It is realizedby measuringtheelectricalresistanceof theconductingrubberthatis dividedinto severalfieldsof anarray. To detecttheenvironment,astereocamerasystemis fixed on the headof ARMAR. Additionally, it is plannedto includestaingaugeson different

Humanoids2000 3

partsof ARMAR, gyroscopesandaccelerationsensorsfor collision measurements,andfor the determinationofthe positionandorientationof the body of ARMAR. The sensorsystemof the mobile platform includeseightultrasonicsensorsanda planarlaser-scanner. Both typesof sensorsareusedfor a collision freenavigationof therobot.

2.5 Computer Ar chitecture

Thecomputerarchitectureof therobothasbeendesignedto bemodular. It is hierarchicallyorganizedanddividedinto computerarchitectureandsoftwarearchitecture.Thecomputerarchitectureconsistsof threelevels:themicro-controller level, the PC level andthe PC-network level. Currently, the robot is controlledby a clusterof C-167micro-controllersanda standardPC.The micro-controllersarecoupledwith specialpower cards,which controlfour motors.Themicro-controllerboardsareconnectedvia CAN-Buswith a maximumtransferrateof 1 Mbit/sto thePC.For real-timerequirementsa modularcontrolarchitectureis developed.As operatingsystemLinux aswell asReal-Time Linux areused.Thechoicewasmotivatedby theavailability of a high numberof devicesandof sourcecodes.The standardLinux kernelrunswith a lower priority asa taskof the RT-Linux kernel.For theefficient implementationof thedifferentcontrol levels,theobjectorientedmoduleMCA is implemented,sinceitenablesrapiddevelopmentandtheexchangeof controlalgorithmsat differentcontrol levels[16]. Figure2 showsthehardwareandsoftwarearchitecturesappliedto thehumanoidrobot.

C167 C167 C167 C167C167 C167

PC104Linux

RT−Linux

CAN−Bus

PWMA/D

Sensors Actors

PWMA/D

Sensors Actors

PWMA/D

Sensors Actors

PWMA/D

Sensors Actors

PWMA/D

Sensors Actors

PWMA/D

Sensors Actors

Fig.2. Thehardware(left) andsoftware(right) architectures.

3 Control

Manipulatorsareoneof themostimportanthardwarecomponentsof humanoidrobots.So,safetyandrobustcontrolis essentialrequirementfor successfulexecutionof cooperative manipulationtaskswith humans.Robustness,stability andsafetyareof greatestdegreeof importancein the caseof humanoidrobots.The implementationoffull dynamiccontrolon a robotstill remainsa challengeto robotscientistsandresearcherstoday. It is known thattheperformanceof a robotcanbeimprovedwith theincludingof therobotdynamicsinto its controller. However,thecomplexity and,moreimportant,thelackof knowledgeaboutthedynamicparametersof therobot,leadrobotsto becontrolledmostlyby PID control,wherethecontrolis doneindependentlyfor eachjoint.

SinceARMAR’s tasksarecurrentlylimited to thoserequiringlow speed,thedynamicseffectsfrom high-speedmotionscanbeneglected.Therefore,positionjoint controllersareused,becausethey canbetterdealwith nonlinearfriction. The purposeof a positioncontrolleris to drive themotor so that the actualangulardisplacementof thejoint will trackthedesiredangulardisplacementspecifiedby apreplanedtrajectory. Thejoint-anglemeasurementsof thearmsandbodyof ARMAR areobtainedby accurateencoders.A robustrobotcontrolrequiringonly positionmeasurementsis easyto implementandincreasesthedynamicperformanceof therobotmanipulator. Nevertheless,whenvelocity andforce sensorsareavailable,feedbackof the velocity andforcescanbe addedto improve theperformanceof thesystem.Figure3 shows thestructureof thecontrollerwe use.Thefuzzy-like modulechoosesa setof parameterof a classicalpositionjoint controllerdependingon the configurationof the arm.The setsofparameterareestablishedthroughexperiments.

4 TamimAsfour etal.

+

-

set of parameter n

Index

desiredposition

FuzzyModule

disturbance

Inverse kinematics

Position Controller

Robot Arm Is-postion

Position Sensors

set of parameter 1

Fig.3. Thestructureof thecontroller

For the control problemof the dual arm systemof ARMAR only the kinematiccontrol is considered.Thecontrol problemis solved in two stages:first, an inversekinematicproblemis solvedto transformtaskvariablesinto the correspondingjoint variablesfor the armsandbody of the robot.The obtainedjoint variablesareinputof a suitablejoint control scheme.The coordinationis thensolvedat the inversekinematicslevel while the arminteractioncanbeconsideredat joint controllevel.

4 Programming of manipulation tasks

Theprogrammingof manipulationtasksis doneby a directmappingof thehumanarmmovementsto the robot.Firstly, tasksaredemonstratedbyahumanoperatorandthemanipulationtrajectoriesarerecorded.In thefollowing,theprogrammingapproachof themanipulationtasksis described.

4.1 Motion Capturing

Many motioncapturingdevicesarecommerciallyavailablethesedaysanddifferenttechniquesareusedin orderto track the body motionof a person.Most of themsuffer drawbackssuchas:low updateratedueto the sensorcharacteristicsand pretty low communicationbandwidth.We usetwo commerciallyavailable position sensorscalledFasTraks.Thepositionsandorientationsof theelbow andthewrist aredirectly provided.Thehumanarmmovementsareonly kinematicallyrepresented,andthedynamicsfor humanmanipulationtaskscanbetakenintoaccountasapost-processingstep.It is notnecessaryto considerthedynamicsunlessrealisticvelocitydistributionfor manipulationmotionsis required.

4.2 Determination of the Arm Configuration

The transferof demonstratedmovementsto ARMAR is provided by an inversekinematicsalgorithm.This isnecessarybecausemost manipulationtasksare specifiedin termsof the object trajectories.The presenceof aredundantjoint in thearmof ARMAR resultsin infinite distinctarmconfigurationswith thesamehandpositionandorientation.Theredundancy of thearmcanbedescribedby therotationof thecenterof theelbow joint aboutthe axis, that passesthroughthe wrist andthe shoulderjoint (seefigure 4). The feasiblepositionsof the elbowaroundthis axisaredefinedby a curve.This curvecanbederivedfrom thefact that theendingpoint of theupperarm describesan ellipsoid centeredon the shoulderjoint and that the startingpoint of the forearmdescribesaspherecenteredon thewrist. Sincebothof thesepointshaveto bethesame,theredundancy curveresultsfrom theintersectionof theellipsoidandthesphere.Theelbow position,togetherwith thehandposition,formsa completerepresentationof thepostureof thearm.

For a givenpositionandorientationof theendeffectorandbasedon thearmgeometry, we calculatea possi-ble positionof the elbow, which is optimalwith respectto somecriteria (joint movementtime, mechanicaljointconstraints,singularityavoidance,redundancy resolutionresultingin human-likemotionsof therobotand”com-fortable” joint movements).Oncehaving theelbow position,theremainingjoint anglesaretheneasyto determine.For a completedescriptionof thealgorithmreferto [15]. So,insteadof usingtimeconsumingiterativesolutionofinversekinematics,ananalytical,geometrical,closedform solutionis provided.

Humanoids2000 5

Startingfrom the humanarm movementdetectedby the above mentionedtrackingsystem,the arm config-urationcanthenbe computedfrom the sensordatausingthe inversekinematicsalgorithm.In orderto computethe joint anglesof the robot arm correspondingto the operator’s currentarm configuration,we assumethat theshoulderpositionsarefixed.

Shoulder

Elbow

Wrist

r

f

s

Hand

α

rh

rTCP

r

ru

Fig.4. Determinationof thearmconfiguration:Theelbow position,togetherwith thehandposition,formsa completerepre-sentationof thepostureof thearm.Theredundancy of thearmis describedby a curve in thecartesianspace.

4.3 Motion Mapping to ARMAR

Basedon theelbow andwrist trajectoriesof thehumanarm,eachjoint angleof thearmof ARMAR is calculatedvia inversekinematics.The arm motion is mappeddirectly to the correspondinghumanoidarm motion: verticalmotion of the humanarm aremappedto the elbow joint andvertical shoulderjoint of the robot arm, whereashorizontalmotionsaremappedto thehorizontalshoulderjoint1. This mappingstrategy caneasilybeextendedtorealizealsoneckandtorsomotions.Thefinal stepof theprogrammingapproachwill beto adjustall joint anglesaccordingto thedescriptionof theobjectto bemanipulatedandtherobotenvironment.

5 Conclusionand Further Work

In this paper, the mechanismsand control schemeof the mobile manipulationsystemfor the humanoidrobotARMAR aredescribed.A closedform solutionof the inversekinematicsof the redundantarm of the robot isprovidedandan approachto transferthe humanarm movementsof typical manipulationtasksto the humanoidrobotARMAR is alsoproposed.Singlearmmotionplannersaredevelopedfor point-to-pointmotionandcurve-trackingmotion.TheInputsof themotionplannerarespecifiedby a humanoperator. TheOutputsof themotionplannersaresequencesof joint angles,andareexecutedin the simulationand thenwith the real robot via thecontrollerswe used.A simpledual-armmotionplannerfor coordinatedmotionis developed.Thismotionplannerconsiderstheclosedkinematicchainof botharmsandtheobject.Wealsoimplementedanobject-orientedsoftwaresystemthat allows a fastdebuggingof behavior-basedcontrol in a graphicalsimulationenvironment.It canbeusedsimultaneouslyto the controlof the real robot.Early manipulationtasksarealsoperformedto demonstratethe capabilitiesof the manipulationsystemof the humanoidrobot. Figure5 shows a typical manipulationtaskperformedby therobot.

Furtherwork will concentrateon the extensionof the sensorsystem,the integration of a humanoperator,the recognitionof the environment,the integrationof knowledgebasesandhuman-friendlyinterfaces,and the

1 Thevertical shoulderjoint allows theshoulderto move in thesagittalplane(forward/backward direction).Thehorizontalshoulderjoint allows theshoulderto move in thetransverseplane(outward/inwarddirection).

6 TamimAsfour etal.

Fig.5. ARMAR is performinga manipulationtask,whichwasdemonstratedby a humansupervisor.

implementationof servicetasks.For the control system,intelligent controllersable to performtasksinvolvingmultiple sensorinformationareto develop.Controlstrategiesfor thecoordinatedmotionof thewholehumanoidrobot (platform, torso,armsandhead)are also requiredfor the successfulexecutionof complex manipulationtasks.In addition,the control systemfor the navigation of the mobile platform will be implementedin ordertoprovideacollision freenavigationandto integratemobility in manipulationtasks.

References

1. Guglielmelli,E.,Laschi,C.,Dario,P.: Robotsfor PersonalUse:Humanoidsvs.DistributedSystems.The2ndInternationalSymposiumin HUmanoidRObots(HURO’99), Tokyo, Japan,October8-9,1999

2. Rosheim,M.: Leonardo’s Lost Robot. In: AchademiaLeonardiVinci. Journalof LeonardoStudies& Bibliogrphy ofVinciana,Vol. IX, 99–110,1996,CarloPedretti(ed.),Giunti Publishers

3. Brooks,R.A.: TheCogProject:Building a HumanoidRobot.The1st InternationalConferenceon HumanoidRobotsandHumanfriendly Robots,Tsukuba,Japan,Oktober26-27,1998.

4. Brooks, R.A., Cynthia, B., Brain, S., Una-May, O.: Technologiesfor Human/HumanoidNatural Interaction.The 2ndInternationalSymposiumin HUmanoidRObots(HURO’99), Tokyo, Japan,October8-9,1999,135–147

5. Hashimoto,S.etal.:HumanoidRobotsin WasedaUniversity– Hadaly-2andWABIAN –.The1stInternationalConferenceon HumanoidRobotsandHumanfriendly Robots,Tsukuba,Japan,26-27Oktober, 1998.

6. Hashimoto,S.: HumanoidRobotfor KanseiCommunication– Computermusthave body–. The2nd InternationalSym-posiumin HUmanoidRObots(HURO’99), Tokyo, Japan,8-9October, 1999,156–160

7. Tanie, K.: MITI’ s HumanoidRoboticsProject.The 2nd InternationalSymposiumin HUmanoidRObots(HURO’99),Tokyo, Japan,October8-9,1999,71–76

8. Hirai, K., Hirose,M., Haikawa, Y., Takenaka,T.: TheDevelopmentof HondaHumanoidRobot.Proceedingof the Inter-nationalConferenceonRoboticsandAutomation.Leuven,Belgium,May 1998,1321–1326

9. Konno,A. et al: Developmentof a HumanoidRobot Saika.Proceedingof the InternationalConferenceon IntelligentRobotsandSystems.Grenoble,France,September7-11,1997,805–810

10. Hwang,Y.K., Kang,S.C.,Park,S.M.,Cho,K.R.,Kim, H.S.,Lee,C.W.: HumanInterface,AutomaticPlanning,andControlof a HumanoidRobot.TheInternationalJournalof RoboticsResearch.Vol. 17,No. 11,November1998,1131-1149

11. Bergener, Th., Bruckhoff, C., Dahm,P., Janen,H., Joublin,F., Menzner, F.: Arnold: An AnthropomorphicAutonomousRobotfor HumanEnvironments.SOAVE’97, SelbstorganisationvonadaptivemVerhalten,1997

12. Asfour, T., Berns,K., Dillmann, R.: The HumanoidRobot ARMAR. The 2nd InternationalSymposiumin HUmanoidRObots(HURO’99), Tokyo, Japan,October8-9,1999,174–180

13. Berns,K., Asfour, T., Dillmann, R.: DesignandControlArchitectureof anAnthropomorphicRobotArm. The3rd Inter-nationalConferenceonAdvancedMechatronicsICAM 98,Okayama,Japan,August3-6,1998

14. Fukaya,N., Toyama,S.,Asfour, T., Dillmann,R.:Designof theTUAT/KarlsruheHumanoidHand.IEEE/RSJInternationalConferenceonIntelligentRobotsandSystems(IROS2000),Takamatsu,Japan,October30- November5,2000(to appear)

15. Asfour, T., Berns,K., Schelling,Dillmann,R.: Programmingof ManipulationTasksof theHumanoidRobotARMAR. The9th InternationalConferenceon AdvancedRobotics(ICAR’99), Tokyo, Japan,October25-27,1999,107–112.

16. Scholl,K.-U., Kepplin,V., Albiez, J.,Dillmann,R.: DevelopingRobotPrototypeswith anExpandableModularControllerArchitecture.The 6th InternationalConferenceon Intelligent AutonomousSystems(IAS-6), Venice,Italy, July 25-27,2000