USF Libraries
USF Digital Collections

Breaking the typecast

MISSING IMAGE

Material Information

Title:
Breaking the typecast revising roles for coordinating mixed teams
Physical Description:
Book
Language:
English
Creator:
Long, Matthew T
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
Taskable robots
Teamwork
Human-robot interaction
Multi-agent system
Social abstraction
Dissertations, Academic -- Computer Science and Engineering -- Doctoral -- USF   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Summary:
ABSTRACT: Heterogeneous multi-agent systems are currently used in a wide variety of situations, including search and rescue, military applications, and off-world exploration, however it is difficult to understand the actions of these systems or naturalistically assign these mixed teams to tasks. These agents, which may be human, robot or software, have different capabilities but will need to coordinate effectively with humans in order to operate. The first and largest contributing factor to this challenge is the processing, understanding and representing of elements of the natural world in a manner that can be utilized by artificial agents. A second contributing factor is that current abstractions and robot architectures are ill-suited to address this problem. This dissertation addresses the lack of a high-level abstraction for the naturalistic coordination of teams of heterogeneous robots, humans and other agents through the development of roles.Roles are a fundamental concept of social science that may provide this necessary abstraction. Roles are not a new concept and have been used in a number of related areas. This work draws from these fields and constructs a coherent and usable model of roles for robotics. This research is focussed on answering the following question: Can the use of social roles enable the naturalistic coordinated operation of robots in a mixed setting? In addition to this primary question, related research includes defining the key concepts important to artificial systems, providing a mapping and implementation from these concepts to a usable robot framework and identifies a set of robot-specific roles used for human-robot interaction. This research will benefit both the artificial intelligence agent and robotics communities. It poses a fundamental contribution to the multi-agent community because it extends and refines the role concept.The application of roles in a principled and complete implementation is a novel contribution to both software and robotic agents. The creation of an open source operational architecture which supports taskable robots is also a major contribution.
Thesis:
Dissertation (Ph.D.)--University of South Florida, 2007.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Matthew T. Long.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 141 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001917761
oclc - 182555427
usfldc doi - E14-SFE0002061
usfldc handle - e14.2061
System ID:
SFS0026379:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001917761
003 fts
005 20071128115953.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 071128s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002061
035
(OCoLC)182555427
040
FHM
c FHM
049
FHMM
090
TK7885 (ONLINE)
1 100
Long, Matthew T.
0 245
Breaking the typecast :
b revising roles for coordinating mixed teams
h [electronic resource] /
by Matthew T. Long.
260
[Tampa, Fla.] :
University of South Florida,
2007.
520
ABSTRACT: Heterogeneous multi-agent systems are currently used in a wide variety of situations, including search and rescue, military applications, and off-world exploration, however it is difficult to understand the actions of these systems or naturalistically assign these mixed teams to tasks. These agents, which may be human, robot or software, have different capabilities but will need to coordinate effectively with humans in order to operate. The first and largest contributing factor to this challenge is the processing, understanding and representing of elements of the natural world in a manner that can be utilized by artificial agents. A second contributing factor is that current abstractions and robot architectures are ill-suited to address this problem. This dissertation addresses the lack of a high-level abstraction for the naturalistic coordination of teams of heterogeneous robots, humans and other agents through the development of roles.Roles are a fundamental concept of social science that may provide this necessary abstraction. Roles are not a new concept and have been used in a number of related areas. This work draws from these fields and constructs a coherent and usable model of roles for robotics. This research is focussed on answering the following question: Can the use of social roles enable the naturalistic coordinated operation of robots in a mixed setting? In addition to this primary question, related research includes defining the key concepts important to artificial systems, providing a mapping and implementation from these concepts to a usable robot framework and identifies a set of robot-specific roles used for human-robot interaction. This research will benefit both the artificial intelligence agent and robotics communities. It poses a fundamental contribution to the multi-agent community because it extends and refines the role concept.The application of roles in a principled and complete implementation is a novel contribution to both software and robotic agents. The creation of an open source operational architecture which supports taskable robots is also a major contribution.
502
Dissertation (Ph.D.)--University of South Florida, 2007.
504
Includes bibliographical references.
516
Text (Electronic dissertation) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 141 pages.
Includes vita.
590
Advisor: Robin Murphy, Ph.D.
653
Taskable robots.
Teamwork.
Human-robot interaction.
Multi-agent system.
Social abstraction.
690
Dissertations, Academic
z USF
x Computer Science and Engineering
Doctoral.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.2061



PAGE 1

BreakingtheTypecast:RevisingRolesforCoordinatingMixedTeamsbyMatthewT.LongAdissertationsubmittedinpartialfulllmentoftherequirementsforthedegreeofDoctorofPhilosophyDepartmentofComputerScienceandEngineeringCollegeofEngineeringUniversityofSouthFloridaMajorProfessor:RobinMurphy,Ph.D.WilliamArmitage,Ph.D.LawrenceHall,Ph.D.KimonValavanis,Ph.D.DateofApproval:May15,2007Keywords:taskablerobots,teamwork,human-robotinteraction,multi-agentsystem,socialabstraction cCopyright2007,MatthewT.Long

PAGE 2

NotetoReader Theoriginalofthisdocumentcontainscolorthatisnecessaryforunderstandingthedata.TheoriginaldissertationisonlewiththeUSFlibraryinTampa,Florida.

PAGE 3

Dedication ThisworkisdedicatedtoHannah,Artonandthefamilyandfriendswhosesupportmadeitpossible.

PAGE 4

TableofContentsListofTablesiiiListofFiguresivAbstractviiiChapterOneIntroduction11.1NaturalisticCoordinationofHeterogeneousMixedTeams 1 1.2RolesasaSocialAbstraction 3 1.3ResearchQuestion 6 1.4Terminology 7 1.5Contributions 8 1.6OrganizationoftheDissertation 10 ChapterTwoADiscussiononRoles112.1RolesfromSocialScience 11 2.1.1Summary 15 2.2RolesfromComputerScience 16 2.2.1Modeling:KnowledgeRepresentation 16 2.2.2RolesinProgrammingLanguages 17 2.2.3RolesinSocio-TechnicalSystems 18 2.2.4AgentsUsingRoles 20 2.2.5RobotsUsingRoles 21 2.3Summary 23 ChapterThreeApproach243.1TheRoleModel 24 3.2PracticalConstraints 26 3.3ImplementationApproach 28 3.4Summary 29 ChapterFourImplementation304.1TheDistributedFieldRobotArchitecture 30 4.1.1DistributedLayer 34 4.2TheCONTEXTADAPTER 34 4.2.1TheROLEMANAGER 36 4.2.1.1SupportforRoles 38 4.2.1.2SupportforRoleDimensions 38 4.2.1.3SupportforRoleMechanisms 38 4.2.2CONTEXTPLUGINs 41 4.2.3TheEXPECTATIONMANAGER 43 4.2.4Messaging 45 4.2.5Scripts 45 i

PAGE 5

4.3ImplementedRoles 47 4.3.1DRONE 47 4.3.2COURIER 48 4.3.3SCOUT 49 4.3.4OPERATOR 49 4.3.5TELEFACTOR 50 4.3.6RELAY 50 4.3.7COMMANDER 50 4.3.8SICK 51 4.4Summary 51 ChapterFiveCaseStudies525.1ElementsofaFunctionalRole-BasedSystem 53 5.2Demonstrations 54 5.2.1RoleDesign:UrbanOperationsContext 55 5.2.1.1Scenario 56 5.2.1.2RolesIdentied 57 5.2.1.3ExampleScenarioTimeline 59 5.2.1.4ImplementationConsiderations 62 5.2.1.5Discussion 63 5.2.2SimulationDemonstration:LittoralWarfareContext 64 5.2.2.1Scenario 65 5.2.2.2RolesUsed 69 5.2.2.3ScenarioTimeline 70 5.2.2.4Summary 72 5.2.3PhysicalDemonstration:UncertainTerrainContext 86 5.2.3.1Scenario 86 5.2.3.2RolesUsed 88 5.2.3.3ScenarioExecution 89 5.2.3.4Summary 91 5.3LimitationsoftheDemonstrations 100 5.4Summary 101 ChapterSixSummaryandFutureWork1036.1ReviewofSalientPoints 103 6.2Contributions 106 6.3FutureWork 108 Bibliography111Appendices118AppendixA:Acronyms 119 AppendixB:RoleDetails 121 AppendixC:ImplementationClasses 126 AbouttheAuthorEndPage ii

PAGE 6

ListofTablesTable1:DescriptionoftheDRONEroleintheLittoralWarfarecontext.ThetabledescribestheDRONEroleandliststherolefunctions,positionandexpectations. 121 Table2:DescriptionoftheCOURIERroleintheLittoralWarfarecontext.Thetablede-scribestheCOURIERroleandliststheroledimensions:rolefunctions,positionandexpectations. 121 Table3:DescriptionoftheSCOUTroleintheLittoralWarfarecontext.ThetabledescribestheSCOUTroleandliststherolefunctions,positionandexpectations. 122 Table4:DescriptionoftheOPERATORroleintheLittoralWarfarecontext.Thetablede-scribestheOPERATORroleandliststheroledimensions:rolefunctions,positionandexpectations. 122 Table5:DescriptionoftheTELEFACTORroleintheLittoralWarfarecontext.Thetablede-scribestheTELEFACTORroleandliststheroledimensions:rolefunctions,positionandexpectations. 122 Table6:DescriptionoftheRELAYroleintheLittoralWarfarecontext.ThetabledescribestheRELAYroleandliststherolefunctions,positionandexpectations. 123 Table7:DescriptionoftheCOMMANDERroleintheLittoralWarfarecontext.Thetablede-scribestheCOMMANDERroleandliststheroledimensions:rolefunctions,positionandexpectations. 123 Table8:DescriptionoftheSICKroleintheLittoralWarfarecontext.ThetabledescribestheSICKroleandliststheroledimensions:rolefunctions,positionandexpectations. 123 Table9:DetailsoftheDRONEroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. 124 Table10:DetailsoftheSCOUTroleintheUncertainTerraincontext.Thetableliststherolefunctions,positionandexpectations. 124 Table11:DetailsoftheOPERATORroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. 125 Table12:DetailsoftheTELEFACTORroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. 125 Table13:Summaryofsourcecodemetricsfortheimplementation.Codelinesarethenumberoflinesofcodeinallsourceles,includingcommentlines.Thesizeisthesizeinkilobytesofthebyte-compiledcode.Theremaybesomeerrorduetorounding. 126 iii

PAGE 7

ListofFiguresFigure1:InteractionofpersonaandsocialcontextforroleassignmentadaptedfromAllport. 13 Figure2:Anexampleofaprogrammingmodelingproblemassumingrolesarenaturaltypes.Eachofthevepossiblesolutionshassubtleerrors.Theproblemisdifcultunlessrolesareconsideredseparatefromtypes.Rolesarelistedinitalics.AdaptedfromSteimann2000. 17 Figure3:Aspecial-purposerolemodelusedforrole-basedaccesscontroladaptedfromFer-raioloetal.. 18 Figure4:TheSensorFusionEffectshybriddeliberative-reactivearchitecturehasbeenex-tendedthroughtheadditionofapersonathatrepresentskeyportionsoftheunder-lyingsystemtodistributedpeers. 33 Figure5:DiagramshowingaportionsoftheoriginalSFXarchitecture. 33 Figure6:Therobot'spersonacanbeexpressedasthecapabilitiesoftherobot.Portionsofthepersonamaybeprotectedinsomemannertorestrictaccesstotheunderlyingcapabilities. 35 Figure7:High-levelblockdiagramfortheCONTEXTADAPTERservice. 36 Figure8:DiagramoftheCONTEXTADAPTERinteractingwiththreetypesofexternalcon-texts.Thethreecontextsdifferonrepresentationofroleswithinthecontext. 37 Figure9:BlockdiagramoftheCONTEXTADAPTERservicehighlightingtheinternalROLE-MANAGERcomponent. 37 Figure10:Finite-StateMachineshowingtherolelifecycle. 39 Figure11:Finite-StateMachineshowingtherolelifecycle.Highlightedportionsofthedia-gramrepresenttransitionsthatupdateexpectationsviatheEXPECTATIONMANAGERcomponent. 39 Figure12:Therearethreetypesofexternalcontextsconsidered. 41 Figure13:Syntaxtranslationenablesinter-systemcommunication. 42 Figure14:Semanticmappingtranslatesthecontentsofmessages. 42 Figure15:Context-specicexpectationsarepropagatedthroughtheCONTEXTPLUGINandmaintainedintheEXPECTATIONMANAGER. 43 iv

PAGE 8

Figure16:Eachteammemberhasadifferentrole,butmayperformteamoperationsthroughasharedscript.Knowledgeofthescriptandtherolesofothersallowsroletakingandcoordinatedactivity. 46 Figure17:Exampleofrolesinaroom-clearingscenario,withtheRIFLEMANR,TEAM-LEADERTL,GRENADIERG,andAUTOMATICRIFLEMANARenteringandclearingaroom.ashowstheagentsoutsidetheroom.bthrougheshowtheagentsenteringtheroominthespeciedorder.ConesshowtheareaofcoverageofindividualagentsAdaptedfromDepartmentoftheArmy2003. 60 Figure18:ScenariooverviewforthesimulationdemonstrationintheLittoralWarfareLWcontext.TheUSVbeginsatAcarryingtheUAV.TheUSVproceedstoB,wheretheUAVislaunched.TheUSVcontinuestoCandthenbeginsscoutingtowardsF,butencountersananomalyatD.TheUSVthensharesthescoutingresponsibilitieswiththehumanoperatoruntilE,whentheUSVisreturnedtofullautonomy.FcompletesthescoutingmissionandtheUSVreturnstobaseatG.SatelliteimagecourtesyofGoogleEarth. 65 Figure19:Thedifferencebetweenthesoftwarethatinteractswithrealhardwareandsimula-tion.ashowsthecaseofDistributedFieldRobotArchitectureDFRAsoftwareonarealrobot.bshowsthecaseofDFRAsoftwareinteractingwiththesimula-tion.Thekeydifferenceisthelow-levelsoftware;thedriversforsensors,effectorsandschemasaredifferentbetweenrealityandsimulation,higher-levelfunctionalitycanoperateunchanged. 66 Figure20:Thegureshowsthehardwarecongurationforthesimulationscenario.Fourma-chineswereusedduringthesimulation.AdedicatedPCwasusedforeachoftheUnmannedAerialVehicleUAV,UnmannedSurfaceVehicleUSVandMissionCommanderMCOperatorControlUnitOCUtomimictheexpectedcongura-tionduringfuturework.Thesimulationitselfranonalaptopcomputer.OnlytheUSVcommunicateddirectlywiththesimulation. 69 Figure21:Roletimelineforthesimulatedscenariohighlightingroletransitionsgleanedfromrobotlogles. 74 Figure22:TheviewoftheUSVatthestartofthescenariothroughtheSARGEsimulator. 75 Figure23:Thestartofthesimulationscenario.ashowstheroleassignmenttothethreeagentsatthestartofthesimulation;bshowstheactivationoftheCOMMANDERrole;cshowstheassignmentoftheRELAYrole;dshowstheactivationoftheCOURIERrole. 76 Figure24:TheCOURIERtoDRONEtransitionwiththeassignmentoftheRELAYroleontheUAV,viewedthroughtheSARGEsimulation. 77 Figure25:TheCOURIERtoDRONEtransitionwiththeassignmentoftheRELAYroletotheUAV.ashowstheRELAYroleassignmenttoUAV;bshowsthedeactivationandrelinquishingoftheCOURIERroleontheUSV;cshowstheassignmentoftheDRONEroleontheUSV;dshowstheactivationoftheDRONErole. 78 Figure26:TheDRONEtoSCOUTroletransitionontheUSV,viewedthroughtheSARGEsimulation. 79 v

PAGE 9

Figure27:TheDRONEtoSCOUTroletransitionontheUSV.ashowsthedeactivationoftheDRONEroleontheUSV;bshowstherelinquishingoftheDRONEroleontheUSV;cshowstheassignmentandactivationoftheSCOUTroleontheUSV. 80 Figure28:Ananomalyhasbeendetectedduringthescoutingmission,shownintheSARGEsimulatorandoperatorGUI.ashowstheSearch-and-RescueGameEnvironmentSARGEsimulatorview;bshowsanoperatorGraphicalUserInterfaceGUIelementpromptingforidentication. 81 Figure29:Ananomalyhasbeendetectedduringthescoutingmission.ashowstheOPER-ATORroleassignmenttotheMCandTELEFACTORassignmenttotheUSV;bshowstheroleactivationoftheOPERATORandTELEFACTORroles,withcascadedactivationoftheDRONEontheUSVandSCOUTontheMC.ThisstartsthesharingoftheSCOUTrolebetweenthetwoagents;cshowsthedeactivationandrelin-quishingoftheDRONEandTELEFACTORrolesontheUSVandtheSCOUTandOPERATORrolesontheMC,endingtherolesharing;dshowsthereactivationoftheSCOUTrole. 82 Figure30:ThecompletionofSCOUTroleontheUSV,shownthroughtheSARGEsimulator. 83 Figure31:ThecompletionofSCOUTroleontheUSV.ashowsthedeactivationoftheSCOUTroleontheUSV;bshowstherelinquishingoftheSCOUTroleontheUSV. 83 Figure32:ThereassignmentofDRONEroleontheUSVtoreturntothestartingpositionandcompletethemission,shownthroughtheSARGEsimulator. 84 Figure33:ThereassignmentofDRONEroleontheUSVtoreturntothestartingpositionandcompletethemission.ashowstheassignmentandactivationoftheDRONEroleontheUSV. 84 Figure34:TheDRONEUSVhasreturnedtothestartingpositionandcompletedthemission,shownthroughtheSARGEsimulator. 85 Figure35:TheDRONEUSVhasreturnedtothestartingpositionandcompletedthemission.ashowsthattheallrolesontheUSVandUAVhavebeenrelinquished. 85 Figure36:ScenariooverviewforphysicaldemonstrationintheUncertainTerraincontext.SatelliteimagecourtesyofGoogleEarth. 86 Figure37:Roletimelineforthesimulatedscenariohighlightingroletransitionsgleanedfromrobotlogles. 92 Figure38:ThestartofthephysicalscenarioshowingtheexternalviewofthephysicalATRV-Jrrobot. 93 Figure39:Thestartofthephysicalscenario.ashowstheroleassignmenttothetwoagentsatthestartofthesimulation;bshowstheassignmentandactivationoftheOP-ERATORroleontheMCandTELEFACTORroleontheUnmannedGroundVehicleUGV;cshowstheassignmentandactivationoftheDRONErole. 94 Figure40:ExternalviewoftheUGVasitbeginstheSCOUTroleinthephysicalscenario. 95 vi

PAGE 10

Figure41:TheUGVbeginstheSCOUTroleinthephysicalscenario.ashowsthedeactivationoftheDRONEroleontheUGV;bshowstheUGVrelinquishingtheDRONEroleattherequestoftheOPERATOR;cshowstheassignmentandactivationoftheSCOUTrole. 96 Figure42:ExternalviewoftheUGVasitnishesthescoutingmissionandreturnstobaseinthephysicalscenario. 97 Figure43:TheUGVnishesthescoutingmissionandreturnstobaseinthephysicalscenario.ashowsthedeactivationoftheSCOUTroleontheUGV;bshowstheUGVrelin-quishingtheSCOUTroleattherequestoftheOPERATOR;cshowstheassignmentandactivationoftheDRONEroletoreturntobase. 98 Figure44:AviewfromtheGUIattheendofthephysicaldemonstration.NotethatattherobotSCOUThasmappedpreviously-unknownterrain,shownintheupperrightofthelargercentralmap. 99 Figure45:Thephysicaldemonstrationscenarioiscomplete.ashowsthedeactivationoftheDRONEroleontheUGVasitreachestheendofthemission;bshowstheUGVrelinquishingtheDRONEroleattherequestoftheOPERATOR. 99 Figure46:High-levelblockdiagramfortheCONTEXTADAPTERservice. 127 Figure47:UniedModelingLanguageUMLdiagramshowingrelationshipsofkeyclassesintheCONTEXTADAPTERservice. 129 Figure48:UMLdiagramshowingtherelationshipoftheJiniserviceentriesfortheCONTEXT-ADAPTER,SCRIPTMANAGER,ROLE,andSCRIPTclasses. 129 Figure49:UMLfortheevent-relatedclasses.Theclassesrepresentrole-,expectation-,andscript-relatedevents. 130 Figure50:UMLfortheexception-relatedclasses.Theclassesrepresentrole-,expectation-,andscript-relatedexceptions. 130 Figure51:UMLdiagramshowingrelationshipsofimportantclasseswithintheROLEMANAGERintheCONTEXTADAPTERservice. 134 Figure52:UMLdiagramshowingrelationshipsofimportantclasseswithintheEXPECTATION-MANAGERintheCONTEXTADAPTERservice. 136 Figure53:UMLdiagramshowingrelationshipsofimportantclassesintheCONTEXTADAPTERservice. 139 Figure54:Finite-StateMachineFSMshowingtheobjectlifecycleforaSCRIPT. 140 vii

PAGE 11

BreakingtheTypecast:RevisingRolesforCoordinatingMixedTeamsMatthewT.LongABSTRACT Heterogeneousmulti-agentsystemsarecurrentlyusedinawidevarietyofsituations,includingsearchandrescue,militaryapplications,andoff-worldexploration,howeveritisdifculttounderstandtheactionsofthesesystemsornaturalisticallyassignthesemixedteamstotasks.Theseagents,whichmaybehuman,robotorsoftware,havedifferentcapabilitiesbutwillneedtocoordinateeffectivelywithhumansinordertooperate.Therstandlargestcontributingfactortothischallengeistheprocessing,understandingandrepresentingofelementsofthenaturalworldinamannerthatcanbeutilizedbyarticialagents.Asecondcontributingfactoristhatcurrentabstractionsandrobotarchitecturesareill-suitedtoaddressthisproblem.Thisdissertationaddressesthelackofahigh-levelabstractionforthenaturalisticcoordinationofteamsofheterogeneousrobots,humansandotheragentsthroughthedevelopmentofroles.Rolesareafundamentalconceptofsocialsciencethatmayprovidethisnecessaryabstraction.Rolesarenotanewconceptandhavebeenusedinanumberofrelatedareas.Thisworkdrawsfromtheseeldsandconstructsacoherentandusablemodelofrolesforrobotics.Thisresearchisfocussedonansweringthefollowingquestion:Cantheuseofsocialrolesenablethenaturalisticcoordinatedoperationofrobotsinamixedsetting?Inadditiontothisprimaryquestion,relatedresearchincludesdeningthekeyconceptsimportanttoarticialsystems,providingamappingandimplementationfromtheseconceptstoausablerobotframeworkandidentiesasetofrobot-specicrolesusedforhuman-robotinteraction.Thisresearchwillbenetboththearticialintelligenceagentandroboticscommunities.Itposesafundamentalcontributiontothemulti-agentcommunitybecauseitextendsandrenestheroleconcept.Theapplicationofrolesinaprincipledandcompleteimplementationisanovelcontributiontobothsoftwareandroboticagents.Thecreationofanopensourceoperationalarchitecturewhichsupportstaskablerobotsisalsoamajorcontribution. viii

PAGE 12

ChapterOneIntroduction1.1NaturalisticCoordinationofHeterogeneousMixedTeamsHeterogeneousmulti-agentsystemsarecurrentlyusedinawidevarietyofsituations,includingsearchandrescue,militaryapplications,andoff-worldexploration,howeveritisdifculttounderstandtheactionsofthesesystemsornaturalisticallyassignthemtotasks.Theseagents,whichmaybehuman,robotorsoftware,havedifferentcapabilitiesbutwillneedtocoordinateeffectivelywithhumansinordertooperate.Thiscoordination,themanagingofdependenciesbetweenactivitiesMaloneandCrowston1994,hasbeenrecognizedasakeyresearchareaforunmannedsystemsbytheUnitedStatesUSmilitaryforaerialvehiclesOfceoftheSecretaryofDefense2005,pg.76andotherunmannedsystemsNationalResearchCouncilU.S.2005,pg.161.Agentsthathavelittleautonomytendtooperateinwell-known,xedpatternsthatcanbeaccountedforbyotheragents.Forexample,searchandrescuerobotsaretoolsusedbyrescuepersonnelandhavelimitedtonoautonomy;evenroversdeployedonMarshaveveryconstrainedautonomyWrightetal.2006.Hencethereisnorealneedtocoordinate.Insteadagentscanplantheiractionsbasedonknowledgeoftheothersentitieswithwhichtheymayinteract.ThisproblemhasbeenstudiedindetailGage2004,andisknownbyseveralnamesincludingmulti-robottaskallocation,recruitment,orplanning.Withincreasedautonomy,robot,human,andotheragentsmustbeabletoactinacoordinatedfashiontoaccomplishjointgoals.RecentresearchbyFeltovichetal.,Kleinetal.,Diasetal.,HoffmanandBreazealandothershaveidentiedanumberofchallengesinthisareaincludingmutualpredictabilityamongpeers,theabilitytomodelintentionandactionsofothers,andadaptabilitytochangingteamsituations.Thislackofautonomyislikelytochangeastheneedforroboticsystemsincrease.Thesesystemswillbecomeincreasinglyautonomousandcomplex.Therstandlargestcontributingfactortothesechallengesistheprocessing,understandingandrepresentingofelementsofthenaturalworldinamannerthatcanbeutilizedbyarticialagents.Thisisafundamentalprobleminboththearticialintelligenceandroboticselds.Toworkaroundthisproblemall 1

PAGE 13

articialsystemsneedtomakesimplifyingassumptions.Roboticsisnoexception.Theseassumptionsareoftenusedrepeatedlytohelpsolvesimilarproblems.Tofacilitatethereuseofthesolution,itandtheassumptionsonwhichthesolutionisbasedonarecodiedintoanarchitecture.Asecondcontributingfactoristhatcurrentabstractionsandrobotarchitecturesareill-suitedtoaddressthisproblem.TheserobotarchitecturesaretraditionallyegocentricdevelopedwithrespecttotherobotitselfandnotdesignedwithrespecttothesocialenvironmentinwhichtherobotwillinteractMurphy2000,chs.2,4,7.Thisisanaturaldevelopmentgiventhehistoryofrobotics.Earlyrobotarchitectureswerefocusedonhowtousesensorseffectively,howtoprocessandinterpretsensordataandwhatstructuresandabstractionswouldallowtheselow-levelcomponentstobereused.Thisapproach,whilecriticalandnecessaryfortheadvancementofroboticsasaeld,treatedtherobotasasingleentity.Asmulti-robotsystemsdeveloped,researchersbegantoconsidertheinteractionofmultiplerobotsinasharedenvironment.Muchofthisresearchwasonsystemswithlimitedornocommunicationbetweenindividualrobotsaswithassemblagesoflow-levelbehaviors,complexjointbehaviorwasexpectedtoemergefromtheinteractionsofindividuals.Toanextentthisworks,althoughasasystemdesigneritisdifculttopredicttheemergentbehaviorofthesesystems.Inotherwords,itispossibletodevelopbehaviorsonindividualrobotsthatproducecomplexsystem-widebehaviorbutitismuchmoredifculttostartwithacomplexjointbehaviorinmindanddeveloptherobot-specicbehaviorstoproducethedesiredemergentjointactivity.ArchitecturessuchasAylluWerger2000,ALLIANCEandL-ALLIANCEParker2004;Parker2000,andVaughanetal.2000areofthisform.Multi-robotsystemscanalsotakeadvantageofcommunicationbetweenrobotswhendevelopingbehaviors.Thesesystemsusesignalingandmessagingmechanismstocoordinateandcontrolthebehaviorsofdifferentrobotsallowingsystemdesignersmorecontroloverwhatthejointbehaviorofthesystemaccomplishes.However,thiscommunicationwasinitiallyadhocandoftenthehigh-levelbehaviorhastoberedesignedandredevelopedtoaccommodatechangestothejointscenario.SwarmrobotsDudeketal.1993exemplifythisapproachtomulti-robotsystems.Astructuredabstractionforhandlingmixedteamsisstillmissing.Recentrobotarchitectureshavetriedtoaddresstheadhocnatureofmulti-robotcommunication,andhaveusedabstractionsthatmakecontrollingmultiplesystemssimplertodevelop.OneexampleofthisispriorworkdevelopingDFRALong2004.CoordinationmechanismsbasedontheContract-NetProtocolSmith1980suchasMurdochGerkeyandMataric2002areexamplesofusingcommunicationtoorganizejointrobotactivity.Eventhoughtheseabstractionsmakestructuredmessagingbetweendiscretesystemssimpler, 2

PAGE 14

moreexibleandmoreextensible,theydonotaddresstheunderlyingegocentricassumption.Themerefactthatindividualrobotscaneasilycommunicatedoesnotmeanthatthearchitecturalconstructsdealwithteamsorgroupsofrobotsandhowtheyinteractatasociallevelbothamongmembersofthegrouporwithhumanorsoftwareagentsinthesurroundingsocialsystem.1.2RolesasaSocialAbstractionThisdissertationaddressesthelackofahigh-levelabstractionforthenaturalisticcoordinationofteamsofheterogeneousrobots,humansandotheragentsthroughthedevelopmentofroles.Dautenhahnhasdiscussedrobotsinsociety,particularlyindealingwithchildrenwithautism.Whatisparticularlynoteworthy,isthatsocialskillsareidentiedasabsolutelyvitalfortherobot'sfunctionalitiesandsuccessinsituationswherearobotrequiressomeformofcontinuedsocialinteraction.Humansocialsystemsuserolesasakeymechanismforsocialinteroperability.Theintroductionofrolesintoarticialsystemsisthusbothcognitivelyplausibleandanaturalisticapproach.Anaturalisticapproachisimportantbecauseifthestateoftherobotcanberepresentedinawaythatisfamiliartoahuman,thenthehumancanunderstandthisstateinamoreintuitivemanner.CoordinationrevolvesaroundmanagingdependenciesbetweenactivitiesMaloneandCrowston1994.Traditionalmulti-agentsolutionstothecoordinationproblemhaveconsideredonlysituationswheretheseagentshavebeenofthesameclass,e.g.articialagents.Boutiliersplitsolutionstothecoordinationproblemintothreegeneralclasses,withsolutionsbasedoncommunication,convention,orlearning.Communication-basedapproachesinvolveinter-agentmessages,thecontentofwhichfollowsacoordinationprotocolWeiss1999.ExamplesofthistypeapproachincludethecontractnetprotocolSmith1980,BrowncoordinationBeaumontandChaib-draa2007,andsolutionstothemulti-robottaskallocationproblem,suchasGageandMurphy2004.Convention-basedtechniquesinvolvetheintroductionofconventionsorothersimilarsocialconstructstocoordinateaction.CoordinationtechniquesthatincorporatesocialnormsandpoliciesincludeBradshawetal.,ShohamandTennenholtz,ConteandCastelfranchiandthezonedefencecoordinationinBeaumontandChaib-draa.Finally,learning-basedtechniques,whichmaythemselvesincorporatecommunicationorconventions,derivecoordinationmechanismsthroughtherepeatedinteractionofagents.Quinnetal.usedneuralnetworkstoevolvejustsuchacoordinationmechanismforajointmobilitytask.However,thesecoordinationmechanismsarenotsufcienttocoordinatebotharticialagentsandhumans.Feltovichetal.,Kleinetal.,Diasetal.,HoffmanandBreazeal 3

PAGE 15

haveidentiedanumberofresearchchallengesthatmustbeaddressedtoenablethissortofcoordination.Forexample,Kleinetal.noteTenchallengesformakingautomationateamplayer.Thesechallengesincludethefollowing:agroundedsocialcompact;teammembersmustbeabletomodeltheintentionsofothers;teammembersmustbemutuallypredictable;agentsmustbetaskable;agentsmustbeabletocommunicatetheirstatusandintentionandinterpretthatofothers;teammembersmustbeabletonegotiatewithrespecttogoalsinachangingenvironment.Itiswiththesechallengesinmindthattheuseofrolestocoordinateactivitybecomesincreasinglyvaluable.SocialsciencehasidentiedageneralsetofvepropositionsthatrelatetorolesBiddle1979: 1. Atleastsomebehaviorsarepatterned,andarecalledroles.Theserolesadheretoindividualswithinasocialcontextandareboundtothatsocialcontext. 2. Rolesaretypicallynotboundtospecicindividualsthoughandmorethanoneindividualmayplayagivenrole.Inasense,theindividualisinterchangeablefromaroleperspective. 3. Theawarenessofrolesmakesindividualsconsciousoftheiractionsrelatingtotheroleinthesocialcontext.Thisawarenessisofsocialexpectationsandtheseexpectationsinuencetheactionsoftheindividual. 4. Rolesareimportantfortworeasons.First,theyhaveconsequences.Thatis,rolesareassociatedwithactionsthatcanhelpattaingoalsorotherbenecialresults.Second,therolesareembeddedinasocialsituationandtheactionsofanindividualinaroleimpactotheragents. 5. Rolesarealsonotstatic;theychangeovertime.Ingeneral,socialrolesforhumansarelearnedaspartofasocializationprocess.Rolesforarticialagentsmustpresentlybecreatedexternallytotheagentratherthanlearned.Rolesdenetheinteractionsofindividualswithinasocialcontext,andareinseparablefromthecontextitself.Theuseofrolesallowhumanstoproducethecorrectbehaviorinawidevarietyofsocialsituations.Forrobotics,thisiskeytoallowingarobotsystemtobeusableinawidevarietyofsituations.Rolesarelearned;peoplearenotbornwiththedirectknowledgeoftherolestheywillneedinlife.Thus,rolesallowhumanpersonstoadapttochangingornewsocialenvironments.Thisisanadmirableability,andasrobotsystemsbecomemorelong-lived,theytoowillneedtoadaptandchangetonewsituations. 4

PAGE 16

Thefollowingscenarioservestomotivatetheimportanceofrolesinthiscoordinationproblem.UnderfundingfromDefenseAdvancedResearchProjectsAgencyDARPAtheroleconceptisbeingappliedtomixedhumanrobotteamsperformingcooperativetasksinanurbancombatsituation.Thisuseofrobotsinaheterogeneousteamservesastheprimarymotivationforthiswork.AsamplescenariointheUrbanOperationsUOcontextinvolvesclearingaroominapossiblyoccupiedbuilding.Thisisonecommontaskthatcanbeassignedtoamilitaryreteamandisassignedwhenthesituationrequiresroom-by-roomclearingofarelativelyintactbuildinginwhichbothenemiesandnon-combatantsmaybelocated.Thereisincreasedrisktohumansoldiersinthisscenarioandthisriskisamotiveforreplacingallorpartoftheteamwitharticialsystems.Thescenarioisstandardprocedureforhumanteams,andservesasabaselinemeasureofhowhigh-levelsocialcoordinationcanoccurinteamsofarticialagents.Rolesinthisscenarioareimportantforthreereasons.First,therolesplayedbyrobotsinthisscenariocanbedrawnfromexistinghumanroles,sotheytheactionsoftherobotteammembersareunderstandabletotheirhumanpeers.Second,therolesarealsoheavilyinterdependentbutdonotchangeoverthecourseofthescenario.Onceroleshavebeenassignedtoeachagent,coordinationbetweenteammembersismanagedbythedenedinteractionofthevariousroles.Finally,becauseteamoperationisdenedintermofrolesitshouldbepossibletointerchangehumanandarticialteammemberscapableofplayingagivenroleandstillcompletetheoverallteamgoals.Whileroleshavebeenusedpreviouslyinrobotics,theiremploymentisnotcommonandmostusesareinformal.Manyoftheseworks,suchasStoneandVeloso1999;MartinsonandArkin2003;Quinnetal.2003;Sty,Shen,andWill2002,useroleasasynonymfortask,withoutconsideringanystrongsocialaspectsandimplications.Others,suchasChaimowicz,Kumar,andCampos2004;Gupta,Messom,andDemidenko2004,denealimitedrolemodel,butthismodelisnotstronglygroundedintheoryorisapplicabletoasinglecontext,suchasrobotsoccer.ThemostcompleteuseofrolesinaroboticenvironmentisintroducedbySattereld,Choxi,andHoustenandisdetailedbySattereldetal..Whilethisworkhasamuchstrongergroundingintheory,andhasthegoalofcoordinatingheterogeneousteams,itdoesnotaddressseveralofthemorecontext-dependentaspectsofroles.Forsimplesocialenvironments,thismaynotbealargedifference,butforalong-livedrobotsystemthesemechanismshavethepotentialtoradicallyalterthedisplayedbehavioroftheteamwithoutchangingtheunderlyingfunctionsoftheroles.Withoutthese,thecorerolefunctionsmustberedevelopedfornewsocialcontexts. 5

PAGE 17

Roleshavealsobeenusedinsoftwareagents,andtheseexplorationsareusefulinprovidingspecicguidelinesandrecommendationsintherolemodel.Forexample,BoellaandLesmo2001;Bradshawetal.2003;BoellaandvanderTorre2005;Felicissimoetal.2005presentsworkexploringtheuseofsocialexpectationsandenforcementonbehavior,andDastani,Dignum,andDignum2003;Dastanietal.2005;Boissieretal.2005haveprovideinsightintodynamicaspectsofrolesandtheimpactonagentgoals.1.3ResearchQuestionTheprimaryresearchquestionthatthisworkaddressesisthefollowing:Cantheuseofsocialrolesenablethenaturalisticcoordinatedoperationofrobotsinamixedsetting?Rolesareafundamentalconceptofsocialsciencethatmayprovidethenecessaryabstraction.Thisisalsocognitivelyplausibleandanaturalisticsolution;asolutionthatisbasedonstudyofexistingnaturalsocialsystems.Thisquestionsuggeststhefollowingadditionalissues: Whatroleconceptsfromnaturalsocialsystemsareimportant?Roleshavebeenstudiedinsocialscienceforover70yearsandalargebodyofliteraturehasbuiltup.Findingwhatpartofthiscorpusisimportantforarticialsystemsiskey.ChapterTwoidentiesasubsetofthiseldthatappliestorobotandagentsystems,andinvestigateswhathasbeenaddressedinpreviousliterature. Isthereaplausiblemappingfromtheseroleconceptstoarticialsystems?Findingasetofcoreroleconceptsisjusttherststep.Inordertobeusefulasapracticalsolutionforheterogeneousteams,theremustbesomewaytomaptheseconceptstoatechnologicalframework.TheapproachtakeninthisdissertationisdescribedinChapterThreeandChapterFourpresentsanimplementationthatmapstherolemodelintoarobotarchitecture.Whiletheroleconceptandapproacharegeneralenoughtoapplytoanyhybriddeliberative/reactivearchitecture,thisimplementationbuildsontheDistributedFieldRobotArchitecturedescribedinLong2004. Howarerolesidentiedordecomposedfromtheproblemdomain?Forcaseswheretherobotisreplacingahumanrole,thenthisisatrivialmapping.Becauserolesaresuchanimportantpartofhowhumansinteractinasocialsetting,suchrolesareeasilyidentiable.Itismoredifcultto 6

PAGE 18

describeorconstructrolesinanentirelynewcontextoronewherearobotisperformingarolethathasnodirectanalogueinahumansocialsystem.Thisdissertationfocusesonacorerolemodelandexcludesthefollowingfromconsideration: Learnedsocialinteractionsanddynamicroledenitionaretwoaspectsthatarenotimplementedinanyform.ChapterFivepresentsanexamplescenarioinwhichlearningandadaptationcanbeused,butafullexplorationofthistopicisleftforfuturework. Severalimplementationdecisionshavebeenmadeintheinterestoftime.Forexample,asimpleroleconicthandlingmechanismhasbeenimplementedtoillustratethatportionoftherolemodel,howevermuchstrongerandmorecompleteimplementationisleftforfuturework.Roleexpectationshavebeenstaticallydened.SoftwareagentframeworkssuchasKnowledgeableAgent-orientedSystemKAoScanbeusedforamorerobust,exibleanddynamicexpectationmechanism.Indeed,thishasbeenimplemented,butthedemonstrationshaveusedthesimplermechanismfortheresultsinthisdissertation. Likewise,eightroleshavebeenimplemented.Inseveralcases,theimplementationisofaminimalnaturetosupportthedemonstrations.Morecompletetreatmentsarealsoleftforfuturework.ThesearenotedinthetextinChapterFour.1.4TerminologyMostterminologyusedinthisdissertationisdenedordescribedwhenitisrstused.However,thetermsystemisusedinseveralmannersanddeservessomeclarication.Systemcantakeonseveraldifferentmeaningsdependingonthecontextinwhichitisusedandthedisciplinefromwhichtheusageisdrawn.Thisdissertationusestheterminthefollowingways: Fromthesocialscienceusage,asocialsystemconsistsofapluralityofindividualactorsinteractingwitheachotherinasituationwhichhasatleastaphysicalorenvironmentalaspect,actorswhoaremotivatedintermsofatendencytothe`optimizationofgratication'andwhoserelationtotheirsituations,includingeachother,isdenedintermsofasystemofculturallystructuredandsharedsymbolsParsons1951,pp.5.Thatis,asocialsystemconsistsofagentspursuingindividualgoalsbutinteractingwithotheragentsinacontext-dependentmanner.Thisdissertationusestheterminologyofsocialsystemorsocialcontexttorefertothisdenition. 7

PAGE 19

Fromthegeneralcomputerscienceusage,asystemisasetofinter-relatedentitiesthattogetherservesomecommongoal.Thisoftenconsideredasoftwareframeworkorplatform.Thisdissertationusestheterminologyroleframeworkandrole-basedsysteminterchangeably. Engineeringdisciplinesusethetermsysteminamannerconcernedwiththeoutputofaprocessortaskforagiveninput.Analysisofthesecontrolsystemsareinterestedwithcertainproperties,namelyresponse,stability,controllabilityandobservabilityNise1992.Thisdenitionisnotconsideredoraddressedinthisdissertation.1.5ContributionsThisresearchwillbenettheroboticsandagentcommunities,articialintelligence,andmakecontributionstohuman-robotinteraction.Itposesafundamentalcontributiontothemulti-agentcommunitybecauseitextends,renes,andpresentsapracticalimplementationoftheroleconcept.Theapplicationofrolesinaprincipledandcompleteimplementationisanovelcontributiontobothsoftwareandroboticagents.Thecreationofanopensourceoperationalarchitecturewhichsupportstaskablerobotsisalsoamajorcontribution.Thisdissertationmakesatleastthefollowingninecontributions: Enablesacognitivelyplausiblenaturalisticapproachtothecoordinationofheterogeneousteams:Thisworkenablestheuseofanaturalisticapproachtothecoordinationofheterogeneousteams.ThisisoutlinedinChapterThree,andhasbeenimplementedusingtheDistributedFieldRobotArchitectureasdescribedinChapterFour.Theapproachhasbeenusedtodesignrolesforateamofrobotsinanurbanoperationsscenario.Ithasalsobeendemonstratedinsimulationandonrealrobots.Theuseofanaturalisticmethodalsoenablesfutureworkintomoreadvancedandintuitivemeansofcontrollingandinteractingwithteamsofrobots,suchasnaturallanguageprocessing. Fundamentalcontributionofarealizableframeworkthatusesrolesforcoordinatingamixed,heterogeneous,multi-agentteam:Fromapracticalrobotdesignandarticialintelligencestandpoint,rolesareanabstractionwhichgroupactionandconstraintsonactionwithinaparticularcontext.Thisallowsafocusofattentiononaparticularsubsetofrobotactions;itisamethodtoprunethestatespaceofallpossiblerobotstatesandactionstothosethatareusefulinaparticularsetting. Renesthescriptmechanismforcoordinatingagentsinasocialcontext:ScriptsSchankandAbelson1977,aconceptfromarticialintelligencethathasbeenusedextensivelyinrobotics,have 8

PAGE 20

beenusedfortheimplementationofarole'sactionsandhowaroleinteractswithotherroles.Indoingso,theuseofscriptshasreturnedtomorecloselyfollowtheoriginalconceptwhichwasconcernedwiththeinteractionofrolesinaparticularsituation.Inaddition,scriptshavebeenexpandedtoconsiderbothbasicinstrumentalscripts,usedforcontrollingsimple,repetitivetaskswithlittlevariability,andsituationalscripts,orcomplexscriptsforhandlingsituationswithmanyinterdependentrolesandactions. Renementoftheroleconceptsuitablefordistributedagent-basedsystems:Thisworkpresentsarenementoftheroleconceptformulti-agentsystems.Thisrenementnotonlyallowstheuseofrolestocoordinateheterogeneousteamsofhumansandrobots,butitisalsoapplicabletoothermulti-agentsystems.Inaddition,thisworkidentiesareasofresearchintheapplicationofontologiesandsemanticmappingthatarenecessaryforallowingarticialagentstoautomaticallyjoin,understand,andtakepartinnewsocialcontexts,usingrolesasakeystartingconcept. Theuseofrolestoshareresponsibilityforataskbetweenhumanandrobotagents:ThescenarioinSection5.2.2sharesarolebetweenarobotandahumanoperator.Thisrolesharinghastheeffectofenablingbothagentstotakeresponsibilityforthecompletionofthejointtask.Formanyrobot-relatedscenarios,thistypeofjointactionisnecessary,particularlyiftherobotdoesnothavethecognitivecapabilitytocompletethetaskforitself.Intheexamplescenario,therobotdetectsananomalousobjectduringitsscoutingmission,butsharesthescoutrolewiththehumanforanalidentication.Duringthistime,thehumanusesthesensorsandeffectorsontherobotbutprovidesthecognitiveprocessingfortherole. Concisereviewofrobotliteraturerelatedtoroles:ChapterTwopresentsaliteraturesurveythatrstdescribesrolesfromasocialscienceperspectiveandisolatesasetofkeypropositionsthatdescribefundamentalpropertiesthatareimportantforrobotics.Roleshavebeenusedincomputersciencebeforebutmuchofthisworkhasbeeninareasthatarenotdirectlyrelevanttorobotsystems.Theliteraturesurveythenlooksattheuseofrolesinsoftwareagentsystemsandelsewhereinroboticsandexplainswhytheseusesofrolesarenotcompleteorsufcientforthisdissertation. Initialpresentationoffundamentalrobot-specicroles:Thisworkalsoidentiestwofundamentalrobotrolesthatcutacrossallsocialcontextsandwillbefoundinanyscenariowheretherobotsarenotfullyautonomous.ChapterThreealsopresentsthreekeyinteractionlevels,passiveobservation, 9

PAGE 21

sharedcontrol,anddirectcontrol,thatarenecessaryforfulluseoftheTELEFACTORandOPERATORroles. Roleconceptsandapproacharetransferrabletootherarchitectures:TheapproachoutlineinChapterThreediscussesthefundamentalrolemodelandadditionalrequirementsinamannerthatisapplicabletomanyarchitectures.Theprimaryconstraintisthatthearchitecturebeahybriddeliberative/reactivearchitecture.PurelyreactivearchitecturesdonotbuildworldmodelsorconsiderplanningMurphy2000,pg.108,whichisarequirementformuchoftherolemodel.TheimplementationinChapterFourusesDFRAasanunderlyingrobotarchitecturebuttheoveralldesignshouldbeusableinotherarchitecturesaswell. Contextadapterandplugins:AsnotedinSection1.1,robotteamsareorcanbeusedinawidevarietyofsetting.Inthesesettings,softwareexternaltotherobotwillmostlikelywantorneedtoreadandwritedataontherobot,includingaccesstosensors,effectors,andtherobot'scognitivestateandroles.ThecontextadapterandcontextpluginsdescribedinSection4.2andSection4.2.2areintendedtoalleviatethisintegrationproblem.Thecontextpluginshavebeenusedtointerfacewithtwodifferentagentsystemsthateachuseadifferentunderlyingcommunicationprotocolandagentframework.1.6OrganizationoftheDissertationTheremainderofthisdocumentisorganizedasfollows:ChapterTwopresentsrelatedworkonrolesfromboththesocialscienceandtechnicalperspectives.ChapterThreethenpresentstheapproachtakenbythisdissertationandadiscussionoftheimplementationoftheapproachispresentedinChapterFour.ChapterFivediscussesthecasestudiesandscenariosusedtodemonstratetheimplementation,aswellaspresentsseverallimitationsofthiswork.ChapterSixsummarizesthedissertationandlistsseveraldirectionsforfuturework.Theappendicescontainsupportinformationandmaterialthatdidnottinthebodyofthedissertation.AppendixAlistsandexpandsacronymsthatareusedinthebodyofthetext.AppendixBprovidesadditionaldetailontherolesusedinthedemonstrations,andAppendixClistsimplementationclassesanddesigndiagrams. 10

PAGE 22

ChapterTwoADiscussiononRolesThepurposeofthischapteristoidentifyhowhumansocialsystemsmakeuseofrolesandwhatfundamentalaspectsofthistheorybearconsiderationinanarticialsocialsystem.Inadditiontounderstandingthehumanelement,wearealsointerestedinhowtheseideashavebeenusedintechnologicalsystems,particularlywithregardtorobotics.Sincerolesarefundamentaltohumansocialsystems,thischapterrstprovidesabackgroundonrolesandroletheoryfromasociologicalperspectiveSection2.1.Second,Section2.2examineshowtheroleconcepthasbeenappliedintechnicalsystemssuchassoftwareagentsandrobotics.2.1RolesfromSocialScienceThetermrolehasacommonplaceintheEnglishlanguageanditsusageinsociologicaltheoryhasbeendrawninpartfromthiscommonusage.OriginallythetermwasLatinrotula,meaninglittlewheel.TheOldFrenchwordrolledenotedarollofparchmentrolledaroundawoodenrod.InFrenchtheater,thisrollofparchmentcontainedthescriptforanactor.Commonusagechangedthemeaningandeventuallyrollecametodescribetheparttheactorplayed.ItisthismeaningthatmigratedtoEnglishintheearly1600sPickett2000,Role.Inmodernsocialscienceroleisaspecicationofthecommonusagemeaningabehavioralrepertoirecharacteristicofapersonoraposition;asetofstandards,descriptions,norms,orconceptsheldforthebehaviorsofapersonorsocialposition;orlessoftenapositionitselfBiddle1979,pg.9.Thissociologicaldenitionisbasedinlargepartontheworkofthreeindividualsinthe1930s:RalphLinton,GeorgeHerbertMeadandJacobLevyMoreno.Oneimportantcharacteristicofsocialrolesisthattheyareembeddedinalargersocialsystem.Morespecically,asocialsystemconsistsinapluralityofindividualactorsinteractingwitheachotherinasituationwhichhasatleastaphysicalorenvironmentalaspect,actorswhoaremotivatedintermsofatendencytothe`optimizationofgratication'andwhoserelationtotheirsituations,includingeachother,isdenedintermsofasystemofculturallystructuredandsharedsymbolsParsons1951,pp.5.That 11

PAGE 23

is,asocialsystemconsistsofagentspursuingindividualgoalsbutinteractingwithotheragentsinacontext-dependentmanner.Meadfocusedonindividualswithinsociety,andconsideredrolesafundamentalmechanismforsocialinteractionandcooperativeability.Indeed,themimicryandplayofsmallchildrenistheprocessthroughwhichtheylearntherolesofothersandlearnhowtoactinalargersocialcontextorsocialsystem. Itisgenerallyrecognizedthatthespecicallysocialexpressionsofintelligence,ortheexerciseofwhatissocialintelligence,dependonthegivenindividual'sabilitytotaketherolesof,orputhimselfintheplaceof,otherindividualsimplicatedwithhimingivensocialsituations;anduponhisconsequentsensitivitytowardtheirattitudestowardhimselfandtowardoneanother.Mead1934LintonLinton1936,ch.8examinedthelinkbetweenroleandstatus,oftentermedposition.Statusrepresentsapositioninaparticularsocialpattern,whetheritisaformalstructuresuchasthatabusinessoramoreinformalpositioninaninformalsocialsetting.Associatedwitheachpositionisasetofrightsandresponsibilitiesthatwhenactivearetheposition'sroleintheoverallsocialpattern.Anindividualmayhavemanystatuses,andthusmanyroles,buttheaggregateofanindividual'sstatusesisthestatusoftheindividual. Arolerepresentsthedynamicaspectofastatus.Theindividualissociallyassignedtoastatusandoccupiesitwithrelationtootherstatuses.Whenheputstherightsanddutieswhichconstitutethestatusintoeffect,heisperformingarole.[...]Everyindividualhasaseriesofrolesderivingfromthevariouspatternsinwhichheparticipatesandatthesametimearole,general,whichrepresentsthesumtotaloftheserolesanddetermineswhathedoesforhissocietyandwhathecanexpectfromit.Althoughallstatusandrolesderivefromsocialpatternsandareintegralpartsofpatterns,theyhaveanindependentfunctionwithrelationtoindividualswhooccupyparticularstatusesandexercisetheirroles.Tosuchindividualsthecombinedstatusandrolerepresenttheminimumofattitudesandbehaviorwhichhemustassumeifheistoparticipateintheovertexpressionofthepattern.Linton1936,pp.114Lintonalsodescribedtwotypesofstatus,ascribedandachieved.Inhumansociety,almostanyindividualcanperformalmostanyrole.Theserolesandstatusesareascribedtoanindividualandformthe 12

PAGE 24

basisofhabit;rolesthatcanbeplayedwithlittlethoughtorattention.Ascribedrolesmakeupalargeportionofthesocialfabricofsociety.Achievedstatusesaredifferentinthattheyhavesomephysicalormentalrequirementorsomeotherspecialqualityandarelledthroughsomespecialeffortortraining.Laterworkinsocialsciencerenestheterminologysurroundingstatusforclarity.Socialpositionismostcommonlyusedforstatus,andthesumofthesocialpositionsofanindividualisthepositionsetBiddle1979.Inthisrenement,thestatusisreassociatedwithrank,wherehigherstatuspositionshavemoreauthority,autonomyorsomeotherdesirablecharacteristic.Morenowrotethatrolesarekeyforprovidinginsightintoindividualsandthereforetherolesapersonplaysareanaccuratereectionoftheinnerself.Morenowasparticularlyinterestedinrole-playingandfocusedontheaspectsofspontaneityandcreativitythatallowedanindividualtocorrectconictsbetweenthesubject'sinnerselfandofcialroles. Thetangibleaspectsofwhatisknownasegoorselfaretherolesinwhichitoperates.Roleandrelationshipsbetweenrolesarethemostsignicantdevelopmentwithinanyspecicculture.Workingwiththeroleasapointofreferenceappearstobeamethodologicaladvantageascomparedwithpersonality,self,orego...Role-emergenceispriortotheemergenceoftheself.Rolesdonotemergefromtheself,buttheselfmayemergefromroles.Morenoetal.1960,pg.81 Figure1:InteractionofpersonaandsocialcontextforroleassignmentadaptedfromAllport. 13

PAGE 25

Apsychologiststudyingthegrowthofpersonalityinindividuals,Allport,pp.181alsosawanunquestionedimportancetorolesandnotedthatlifeisreallyasuccessionofrolesthatenableanindividualtorelatetocomplexandchangingsocialsystems.Allportwasprimarilyconcernedwithhowrolesinuencepersonality,anddistinguishedfourkeyaspectsofroles,showninFigure1:expectation,conception,acceptanceandperformance.However,Allportnotedaproblemwiththeviewthatlifeismerelyasetofprescribedroles.Societyallowsagreatdealofexibilityinhowaroleisperformedandthisvariabilityallowsindividualswithwide-rangingvaluesandpreferencestoplaythesamerolessuccessfully.Earlyworkinroleswaspopularandspawnedalargebodyofworkinthesocialsciences.However,eachoftheearlyviewsofroleshadslightdifferencesintheory,approach,orterminology.Whilethegeneralconceptsremainedcompatible,comparingliteraturebecamedifcult.ThomasandBiddleprovidedanearlysurveyoftheeldandbodyofworkandadecadelater,Biddleprovidedanupdatedviewofthestateofroletheoryinthesocialsciences.Itisthislatterworkthatprovidesarevisedterminologyandconceptmapaswellasatheoreticalfoundationforthiswork.Biddlenotedtheexistenceofunderlyingdifferencesinroletheoryresearch,butofferedasetofvepropositionsforwhichthereisgeneralscienticagreement: 1. Roletheoristsassertthatsomebehaviorsarepatternedandarecharacteristicofpersonswithincontextsi.e.,formroles. 2. Rolesareoftenassociatedwithsetsofpersonswhoshareacommonidentityi.e.,whoconstitutesocialpositions. 3. Personsareoftenawareofroles,andtosomeextentrolesaregovernedbythefactoftheirawarenessi.e.,byexpectations. 4. Rolespersist,inpart,becauseoftheirconsequencesfunctionsandbecausetheyareoftenimbeddedwithinlargersocialsystems. 5. Personsmustbetaughtrolesi.e.,mustbesocializedandmayndeitherjoyorsorrowintheperformancesthereof.[...]Roleconceptsarewidelyacceptedwithinthesocialsciences,andformanyreaderstheveprecedingpropositionswouldappeartobeself-evident.Biddle1979,pg.8ThesevegeneralstatementsincorporatetheworkofMead,LintonandMoreno,aswellasmanyothersinthesocialscienceeld,andthisdenitionprovidesaminimumsetofconceptsthatmustbe 14

PAGE 26

addressedbyanytechnicaladaptationofrolestoarticialsystems.Namely,itmustdenerolesrelativetoasocialpositioninthesocialcontext.Theserolesmustbeassociatedwithsomefunctionorbehavior,andthisbehaviormustbeassociatedwithexpectationsofhowtheroleoperatesinthesocialcontext.Finally,theremustbesomeconsiderationofhowtheroleschangeovertimeassocializationintonewcontextsoccurs.Almostadecadelater,Biddleagainlookedatthebroadbodyofroleliteratureandgroupeditintovecategoriesofroletheory:functional,symbolicinteractionist,structural,organizational,andcognitive. FunctionalroletheoryisderivedinlargepartfromLintonandothersinthesametradition.Asitsnamesuggests,functionalroletheoryseesrolesassharednormsthatdescribethecharacteristicbehaviorofindividualsthatoccupyagivenpositioninasocialsystem. Thesymbolicinteractionistperspective,typiedbyMead,stressesrolesfromtheperspectiveoftheindividual.Thisperspectivestressesthatrolesevolvethroughinteractionwithinasocialsystemanduidlychangeinresponsetonorms,situationalcontextandotherdemands. StructuralroletheoryisalsobuiltonLinton'swork,andisprimarilyfocusedonsocialstructuresandpositionswithinthestructures.Hererolesareareectionofthecharacteristicbehaviorsofmembersofthepositions,butunlikethefunctionaltradition,thefocusofthisbranchofroletheoryismoreonthesocialnetworkitselfratherthanindividualswithinthesystem. Organizationalroletheoryismorerigidlydenedthanotherapproaches,lookingatroleswithinthecontextofformalorganizations.Theseformalsystemstendtobepre-dened,task-orientedandrigidlyhierarchical.Rolesinthissensearestronglyidentiedwithsocialpositionandinuencedbygroupexpectation,althoughindividualvariationisstillpossible. FinallycognitiveroletheoryisinuencedtheworkofMorenoandbycognitivepsychologyandisprimarilydirectedtowardtheeffectofexpectationonbehavior.ThelatterviewisthatofBiddlehimselfandistoutedashavingabroaderempiricalbasethantheothervariantsofroletheory.2.1.1SummaryThegeneralconsensusprovidesaminimumsetofconceptsthatmustbeaddressedbyanytechnicaladaptationofrolestoarticialsystems.Namely,anytreatmentofrolesmustdenerolesrelativetoa 15

PAGE 27

socialpositioninasocialcontext.Withinthatsocialcontext,theserolesmustbeassociatedwithsomefunctionorbehavior,andthisbehaviorissubjecttoasetofexpectationsregardinghowtheroleoperates.Finally,theremustbeaconsiderationofhowtheroleschangeovertimeassocializationintonewcontextsoccurs.2.2RolesfromComputerScienceRoleshavebeenusedinseveralcommunitieswithincomputerscience,withdifferentdisciplinesmappingtotheunderlyingsocialsciencemodelatvaryingdegreesofdelity.Inparticular,rolesformapartoffourmaincommunities:knowledgerepresentation,programminglanguages,socio-technicalsystemsandagentsystems,includingsoftwareagentsandrobotics.2.2.1Modeling:KnowledgeRepresentationTherstdiscipline,knowledgerepresentation,derivesfromabranchofarticialintelligence,focusedonpresentinginformationinaformthatcanbestoredandprocessedbyacomputer.Inparticular,thisknowledgeisparticularlydesirableforhuman-likereasoningandinference.Knowledgerepresentationiscloselytiedtotheeldoflogic,whichprovidesaformalstructuretoknowledgeaswellasrulesofinference,andtothatofontology,aformaldescriptionoftheelementsofaparticulardomain.Ontologyisaformalmechanismforrepresentingknowledge.Therearetwobroadclassesofontologies:upperontologiesandspecial-purposeontologies.Ageneral-purposeorupperontologyisaframeworkthatattemptstodescribeknowledge,startingwiththemostgeneralconcepts.AnexampleofanupperontologyisCycknowledgebaseLenatetal.1990;Lenat1995;GuhaandLenat1990,whichstartswithagenericThingandthenrenestheThingintosub-categoriessuchasIndividualObject,Intangible,Event,andsoon.Thisiscontrastedtoaspecial-purposeontology,whichisadetailedrepresentationofaspecicdomain,wheretheconceptsintheontologyareonlythoserequiredtomodelthedomain.Upperandgeneralpurposeontologiesoperateatdifferentlevelsofknowledgeandhavespecicpropertiesofinterest.Upperontologiesaregeneric,andthustheirconceptsshouldbeapplicabletoanyspecialpurposedomain.Inaddition,ageneral-purposeontologycanunifytheconceptsfrommultipledomains,allowingmorecomplexproblem-solvingcapabilities.However,general-purposeontologiesaremorecomplex,containingmoreinformationdescribingtheworldandmorestatementsaboutthestateoftheworld. 16

PAGE 28

Rolesappearasaconceptinseveralupperontologies,includingthatofSowa,GuhaandLenatandMasoloetal..Otherworkhasinvestigateddifferenttypesofroles:Masoloetal.alsoexaminedrelationalrolesandhowtheymaybeusedtosolvetheclassicalcountingprobleminlogicandLoebeexaminedhowactorsplayingarolecouldparticipatenotonlyinsocialenvironments,butalsoaspartofarelationorprocess.2.2.2RolesinProgrammingLanguages Figure2:Anexampleofaprogrammingmodelingproblemassumingrolesarenaturaltypes.Eachofthevepossiblesolutionshassubtleerrors.Theproblemisdifcultunlessrolesareconsideredseparatefromtypes.Rolesarelistedinitalics.AdaptedfromSteimann2000. Roleshavealsobeenusedasatooltoaidinmodelingandrepresentingcertaintypesofproblemsinprogramminglanguagesandsoftwaredesign,wherestandardusageofclassesandinterfacesisnotsufcienttoadequatelydescribeormodeltheproblem.Evenasimplemodelingproblemsuchastheorganizationoftypesrelatingtocustomers,suppliers,people,andcompaniesFigure2isnon-trivialifrolesarediscounted.SteimannandMayerconsiderrolesascontext-specicinterfacesthatcanbeusedtoproperlyrepresentthedomain.TheadditionofrolesasafundamentalconceptinmodelingmakestheseandotherproblemstractableSteimann2000.Roleshavealsobeenaddedtoprogramminglanguagesasconstructsatthelevelofaclassorinterface.Object-levelimplementationsinhaveappearedinObjectTeams/JavaHerrmann2005,PowerJavaBaldoni,Boella,andvanderTorre2005,andSelfPaesschen,Meuter,andD'Hondt2005.Whiletheuseofrolesinprogramminglanguagesandmodelingshowsthatrolescanbeapowerfulabstraction,thisuseataprogrammingobjectlevelistoone-grainedtoapplytoagentsdirectly. 17

PAGE 29

Figure3:Aspecial-purposerolemodelusedforrole-basedaccesscontroladaptedfromFerraioloetal.. 2.2.3RolesinSocio-TechnicalSystemsSocio-technicalsystemsaresystemsthatarecomposedofbothtechnicalandsocialsystems.Thesecanbeverywideranginginscope,fromlargesocialsystemssuchasauniversitywhereworkisdoneusingtechnicalsystems,totechnicalsystemswhichfacilitateorcoordinatesocialinteractionsuchasonlineelectroniccommunities.Roleshavebeenusedinseveralcontextswithinsocio-technicalsystems.Acommonsocio-technicalsystemandspecial-purposeontologythatincludestheroleconceptisexempliedbyRole-BasedAccessControlRBAC.Role-BasedAccessControlisanaccesscontrolmethodcenteredaroundroles.InRBAC,permissionsareassociatedwithroles.UsersarethenassignedtorolesandinheritthepermissionsassociatedwiththeroleSandhuetal.1996;Zhu2003.Figure3showsanexampleofsucharolemodel.Inthisparticularexample,roleslinkuserswithpermissionstiedtoobjectsandoperationsFerraioloetal.2001.Herrmann,Jahnke,andLoserandJahnke,Ritterskamp,andHerrmannviewrolesintermsofastaticdescriptionroledimensionsandafunctionaldescriptionrolemechanisms,basedonastudyofanonlineweb-basedcollaborativelearningsystem.Inthislight,rolesaredenedintermsoffourstaticdimensionsposition,functions,expectationsandinteractionsandsixdynamicmechanismsroleassignment,rolechange,rolemaking,roletaking,inter-roleconict,androledenition.Thesetencharacterizationsformthebasisforthisdiscussionofroles.Thefourroledimensionsaredescribedbrieybelow: 1. Position:Rolesreectastaticviewofanorganizationalstructure.Thepositionofaroleinasocialstructuredenesthefunctionsandtasksthatarerequiredfortherole. 18

PAGE 30

2. Functionsandtasks:Eachroleisassociatedwithaformalsetofpermissions,obligationsandactivitiesthataredenedbythesocialorganization,andassociatearolewithhowitisperformed. 3. Behavioralexpectations:Whilethefunctionsandtasksofaroleareaformaldescriptionoftherole,thebehavioralexpectationsareinformalorprescribedconventionsboundinthesocialinteractionsofarole.Violatingconventionscanearnanagentnegativesanctionsfromotheragentsinthecommunity.Indeed,insomesituationsthesebehavioralconventionsareasformalandbindingasthefunctionsandtasks. 4. Socialinteractions:Finally,aroleplayer'sabilitytoplaytherolecanshapetheroledenitionitself,whetherthroughmodifyingtherole'sbehavioralexpectations,corefunctions,orevenpositioninthesocialsystemitself.Ifthefourroledimensionsareastructuraldescriptionofaroleanditsrelations,thefollowingsixdynamicrelations,knownasrolemechanisms,areafunctionaldescriptionofhowanindividualinteractswithroles.Thesemechanismsdenearolelifecyclewhichcontrolstheoperationandinteractionofarolewithinasocialcontext: 1. Roleassignment:Roleassignmentistheprocessofassigningaroletoanagentinasocialsystem.Roleassignmentisastatementofdesire,notaguaranteethattheagentwillacceptorbeabletoplaytherole.Notethatanagentcanassignaroletoitselftofulllinternalgoals. 2. Roletaking:Roletakingoccurswhenanagentusestheknownroleplayedbyanotheragenttobuildamodeloftheotheragent'sactions.CoutuCoutu1951notedthatroletakingisoftenerroneouslyconfusedwithroleassignment.Roletakingrequiresasharedunderstandingoftheroledimensionsforconstructionofameaningfulmentalmodel. 3. Rolechange:Rolechangeistheprocessofenablinganagenttoplayoneormorerolessimultaneouslyorinsequence.Anyindividualplaysanumberofrolesindailylife. 4. Rolemaking:Inahumansociety,eachpersonthatplaysaroledoessointheirownuniquemanner,consequentlytransformingthebehavioralexpectationsintoconcreteactionindifferentways. 5. Inter-roleconict:Anagentmayholdmultipleroles,andthegoalsofeachmayconict. 6. Roledenition:Arolemayberadicallychangedoranewrolecreatedduetochangingcircumstancesoranegotiatedsocialchangeinbehavioralexpectations. 19

PAGE 31

2.2.4AgentsUsingRolesThebulkofpriorworkwithrolesinmulti-agentsystemshasappearedoutofthesoftwareagentcommunity,andmuchisrelevanttomulti-robotsystems.WorkinuencingthestaticroledescriptioninagentsystemshasincludedsocialnormsandpoliciesBoellaandvanderTorre2005;Felicissimoetal.2005aswellassocialenforcementBoellaandLesmo2001;Bradshawetal.2003.Wooldridge,Jennings,andKinny,inparticular,observethatmulti-agentsystemdesignismorecomplicatedthantraditionalsoftwareengineering,whichfailstocaptureanagent'sproblemsolvingbehaviorandsocialinteractions.TheyproposetheGAIAmethodologyforagent-orienteddesign,usingrolestomodeltheresponsibilities,permissions,activitiesandprotocolsofanagent.Inasense,though,thismethodologyandextensionsbyZambonelli,Jennings,andWooldridgetoincorporateorganizationalstructureonlydescribethersttwostaticdimensionsofroleanalysis:positionandfunction.Inopenmulti-agentsystems,whereagentscanenterorleavethesystem,itisnotenoughtosimplydescribethestaticroledimensions.Theseagentsneedtoincorporateroledynamicsaswell.Dastani,Dignum,andDignumandDastanietal.havestudieddynamicroleassignmentinsuchopensocieties,particularlyhowagentsenactanddeactroles,andtheeffectthishasonagentgoals.Boissieretal.lookathowanindividual'scognitiveframeworkandbehaviorcanchangewhenplayingarole,anddescribehowtherolecaninuencetheindividual'sgoals,desiresandbeliefs,aswellascausetheindividualtodynamicallygainorloseinuenceorpower.Areasoningagentmayverywellagreetoplayaroletogainaccesstoinformationthatwillhelpwithprivategoalsevenifitdoesmeanacceptingnewrestrictionsorobligations.ColmanandHanhaveinvestigatedautonomyandhowitcanrelatetorolesandagency.Theyproposeve-levelsofautonomy:noautonomy,processautonomy,system-stateautonomy,intentionalautonomy,andautonomyfromconstraints.However,thegenerallevelofimplementationoftheroledimensionsandrolemechanismsisunclear,andautonomyhasnotbeeninvestigatedinthiscontext.Sims,Corkill,andLesserbandSims,Corkill,andLesserapresentarole-basedcoalition-formationframeworkthatusesmatchesrolecapabilitieswithagentsabletoplaytheroleandusestheseagentstomeetsystemgoals.Theprimaryfocusofthisworkisonbuildinganorganizationalstructureratherthanoninvestigatingroles;therolesareprimarilyasynonymfortask,andtheframeworkisnotbasedonanyparticularbranchofroletheory.Otheragentsystemsandarchitecturesdonotdirectlyencapsulatetheconceptofarole,butcanoftenincludeconceptsthataresimilaroroffersimilarfunctionality.TheKAoSBradshawetal.1997,for 20

PAGE 32

example,denesbasicontologiesforactions,actors,groups,places,variousentitiesrelatedtoactionse.g.,computingresources,andpoliciesBradshawetal.2003.However,theseontologiesdonotdirectlyaddresstheroleconceptsinSection2.2.3,butcouldbeusedtoimplementordescribetheroledimensions.Inparticular,KAoSusestheconceptofpolicytoexpressauthorizationsorobligationsrelatedtoactionsinagivensituation.Thesecanbebothpositiveornegative;authorizationscanallowordenyactionandobligationscanrequireorforbidactionsUszok,Bradshaw,andJeffers2004.Thisuseofpolicytomodifybehaviorisaformofexpectationandcanbeusedassuchinarolecontext.2.2.5RobotsUsingRolesMulti-robotsystemsisalessdevelopedeldthengeneralmulti-agentsystems.Asaresult,researchonstrongsocialinteractionsbetweenrobotsandwell-denedmulti-robotdomainshavebeenslowertodevelop.However,therehasbeensomerecentworkinthisarea.Aseldroboticsisinherentlyfailure-proneCarlson,Murphy,andNelson2004andisthusbydenitionanopenagentenvironment,itisnaturalthatresearchwillleadinthedirectionofthedynamiccharacterizationofroles.Rolesappearinlimitedforminpreviousroboticsliterature,buttheydosoprimarilyasasynonymfortask,withoutmanyofthestrongersocialaspects;theseusesofrolesappeartobestrictlylimitedtorole-assignmentandrole-change.StoneandVelosousedrolesinthismanner,buildingformationscomposedofasetnumberofspecicrolesforrobotsoccer.Asthesoccergameprogressed,formationsandthusroleswouldchangedynamically.RolesalsoappearedinMartinsonandArkinasavehicletotestaQ-learning-basedrole-assignmentmechanismusingaforagingtaskinahostileenvironment.Quinnetal.usedneuralnetworkstoevolvecontrollersforateamofthreesmallrobotsinateamlocomotiontask.Therobotsasetofcontrollersthatallowedmobility,andtheauthorsarguethateachrobotplaysaroleontheteam.Whiletherobotsdoappeartohaveaformofroleassignmentandalearnedrole,othersocialaspectsaremissing.Finally,Sty,Shen,andWillusedrolesinamorerecursivemannerforself-recongurablerobots;eachrecongurablemodulecouldassumearolewithintherobotsuchaslegorspine.Roleselectionandassignmentprovidedtherequiredbehaviorforthemodulewithinthesocialcontextoftherobot,butdidnotinuencehowtherobotinteractedwithotheragents.Chaimowicz,Kumar,andCampospresentarobotteamthatdoesconsidertheuseofrolestocoordinateateamofrobotsinasinglecooperativetask.Thisrolemodelconsidersonlyabasicroleassignmentmechanismcapableofallocatingandreassigningroles.Onenovelfeatureofthisworkisthat 21

PAGE 33

robotscanexchangetheirrolewithanotherrobotifnecessarytocompletetheoveralltask.Onelimitationofthisworkisthat,aswiththeabovesystems,theuseofthetermroleisagainasynonymforataskorsub-task.TherobotsoccersystemsdescribedinGupta,Messom,andDemidenkodenethreerolesforrobotsoccer:thedefender,attacker,andgoalkeeper.Theroleselectionmechanismswitchestheattackeranddefenderrolesoncertainrobotswhiletheassignmenttogoalkeeperisstatic.Aninterestingcharacteristicofthisworkisthateachsoccerteamhastworobotsintheattackerordefenderroles,andthepaperconsidersbothrobotsinthedefenderroleasaseparaterole.Thisallowsalimitedformofroletaking,aseachrobotcanbasedecisionsaboutappropriatebehaviorsbasedontheknownbehaviorofitsteammate.PerhapsthemostcompleteuseofrolesinaroboticenvironmentisintroducedbySattereld,Choxi,andHoustenandisdetailedbySattereldetal..ThisworkpresentedaRole-BasedOperationsArchitecturedesignedformultiagentcoordinationinamilitaryoperationsdomain.ThearchitecturedenesaRoleandZoneSystemresponsibleformanagingmissiontasksandhandlinginter-agentcoordination.Inaddition,thearchitecturespeciesaSalienceEngineordistributedeventprocessinglayerandandActionEngineforexecutingandtrackingthestateofactions.ThissystemhasbeentestedonrealrobotsandwasdevelopedusingreasoningsimilartothatpresentedinChapterOneandSection2.1.Thebiggestlimitationwithrespecttothevepropositionslistedaboveisthatwhilethearchitectureshowstheutilityofarole-basedsystemforcoordinatingheterogeneousteams,itdoesnotaddressthemorecontext-dependentaspectsofrolessuchasroleexpectationsandroleposition.Forsimpledomainsandmissions,thismaynotbealargedifference,howeverforalong-livedrobotsystem,thesemechanismshavethepotentialtoradicallyalterthedisplayedbehavioroftheteamwithoutchangingtheunderlyingfunctionsoftheroles.Withoutthese,thecorerolefunctionsmustberedevelopedfornewsocialcontexts.Noneoftherole-basedsystemspresentedherehavebeentestedordemonstratedonoutdooreldrobots,althoughSattereldetal.2005hasbeendemonstratedoniRobotMagellanProplatformsandChaimowicz,Kumar,andCampos2004usedTRCLabmateandNomadXR4000robots,bothinalaboratoryorindoorenvironment.Emotionhasbeenusedinaffectiveroboticstoinuencethebehaviorofindividualagentsinateamsetting.However,thiseldisprimarilyfocusedonsociallyinteractiverobotics,thatis,directhuman-robotinteraction.Whiletherehasbeensomeefforttoinvestigatehowhumansandrobotsshould 22

PAGE 34

interact,suchasBartneckandForlizzi,Lisettietal.andMoshkinaandArkin,theseeffortsaremostlyfocusedononhowtoenablearobotthatisperformingsomeactiontodosoinamorebelievableorhuman-likemanner.Inthissense,affecttsnicelyintoarole-basedsystem,asitcanbeusedtomaketheperformanceofarolemorebelievable.Forexample,Bruceetal.havetakenmotivationfromactinganddramatobuildthePlayarchitecture.Whilenotdirectlyrole-related,thearchitecturedoestreatrobotsasactorsinanimprovisationalsetting.Eachactorhasinternalandexternalgoalsandusesemotiontomodifyhowtherobotsactandchoosebehaviorsforexecution.Howevertheassignmentoftheoverallrolee.g.`hero'or`villain'isstaticandthearchitecturedoesnotexploretheroleconceptinanydetail.Inadifferentvein,affectcanalsobeusedtoinuencetheassignmentoftasks.Gageusedtheemotionshameinanaffectiverecruitmentprotocoltoinuencewhenindividualrobotswouldrespondtorequestsforhelpfromotheragents.Inthissense,affectcouldmodifyroleassignmentorotherwiseinuencetherolelifecycleinarole-basedsystem.2.3SummaryThischapterpresentedrelatedworkonhowhumansocialsystemsmakeuseofrolesandwhatfundamentalaspectsofthistheorybearconsiderationinanarticialsocialsystem.Rolesdenetheinteractionsofindividualswithinasocialcontext,andareinseparablefromthecontextitself.Theuseofrolesallowhumanstoproducethecorrectbehaviorinawidevarietyofsocialsituations.Socialscienceisclearthattherevegeneralpropositionsthatmustholdinarolesystem:namely,itmustdenerolesrelativetoasocialpositioninthesocialcontext;theserolesmustbeassociatedwithsomefunctionorbehavior,andthisbehaviormustbeassociatedwithexpectationsofhowtheroleoperatesinthesocialcontext;nally,theremustbesomeconsiderationofhowtheroleschangeovertimeassocializationintonewcontextsoccurs.Forrobotics,thisiskeytoallowingarobotsystemtobeusableinawidevarietyofsituations.Inadditiontorelatedworkonthehumanelement,thischapteralsopresentedhowtheseideashavebeenusedintechnologicalsystems,particularlywithregardtorobotics.Thisrelatedworkhasadirectinuenceontheapproachtakentoenablerole-basedrobotics,andispresentedinthenextchapter. 23

PAGE 35

ChapterThreeApproachThischapterdescribeskeyaspectsoftheapproachtakentointroducesocialrolestorobotics.First,Section3.1providesthetheoreticalapproachandrolemodelthatservesasthefoundationforthiswork.Second,itlistsseveralpracticalconstraintsthatguidetheuseofrolesinpractice.TheseconstraintsaredrawnfromexperiencewithmobilerobotsandarelistedinSection3.2.3.1TheRoleModelThetheoreticalapproachdrawsinlargepartfromtwosources.TherolecharacterizationofJahnke,Ritterskamp,andHerrmannservesastheprimarymodel,modiedbasedonrecommendationsfromDastanietal.andsupportedbyrelatedworkinagentsystemsBradshawetal.2003;Uszok,Bradshaw,andJeffers2004.Thismodelprovidesfourroledimensionsandsixrolemechanisms.Therolemechanismsrequirethatarolehavesomepositionrelativetootherroles,supportasetoffunctionsandtasks,belinkedtosocialexpectations,andprovidesupportforsocialinteractiontosupportlearningandadaptation.Thefourroledimensionsaredescribedbelow: 1. Position:Rolesreectastaticviewofanorganizationalstructure.Thepositionofaroleinasocialstructuredenesthefunctionsandtasksthatarerequiredfortherole.Inamilitaryscenario,forexample,eachindividualsoldierhasarank,orstatus,andthisdenesthefunctionsthatarepermittedforthatsoldier. 2. Functionsandtasks:Eachroleisassociatedwithaformalsetofpermissions,obligationsandactivitiesthataredenedbythesocialorganization.Thisroledimensionassociatesarolewithactionsthatcanormustbeperformed.Asoldierthatispartofateamclearingabuildingmayhavecertainactionshecanperform,suchasbreachingadoororusingweapons.Arolemyalsoincorporaterestrictionsorqualications;inthisexample,thesoldiermaybeprohibitedfromstartingthemissionuntilallordersaregivenandallteammembersareready. 24

PAGE 36

3. Behavioralexpectations:Ifthefunctionsandtasksofaroleareaformaldescriptionoftherole,thebehavioralexpectationsareinformalorprescribedconventionsboundinthesocialinteractionsofarole.Violatingconventionscanearnanagentnegativesanctionsfromotheragentsinthecommunity.Indeed,insomesituationsthesebehavioralconventionsareasformalandbindingasthefunctionsandtasks.Inthemilitaryscenario,asoldierisexpectedtousecertainformsofaddresswhencommunicatingwithothersoldiers,dependingontherelativeposition. 4. Socialinteractions:Finally,aroleplayer'sabilitytoplaytherolecanfeedbackandshapetheroledenitionitself,whetherthroughmodifyingtherole'sbehavioralexpectations,corefunctions,orevenpositioninthesocialsystemitself.Thisapproachdivergesfromtheoriginalrolemodelinthefollowingway.Theoriginalmodeldenedsixmechanisms:roleassignment,roletaking,rolemaking,rolechange,inter-roleconictandroledenition.However,Dastanietal.2005pointoutthatroleassignmentmaysubtlyinuencethegoalsofanagentevenifitisnotactivelypursuingthefunctionsoftherole.Theysuggestthatassignedrolesmayalsobeactiveorinactive,correspondingtowhetherthefunctionsoftheroleareactivelypursued.Thissuggeststheneedforamorecomprehensiverolelifecyclethatcanreplaceroleassignmentandrolechangeintheoriginalmodel.Thenalverolemechanismsareasfollows: 1. Rolelifecylemechanisms:Roleassignmentistheprocessofassigningaroletoanagentinasocialsystem.Roleassignmentisastatementofdesire,notaguaranteethattheagentwillacceptorbeabletoplaytherole.Anagentmayassignaroletoitselftofulllinternalgoals.Rolechangeistheprocessofenablinganagenttoplayoneormorerolessimultaneouslyorinsequence.Anyindividualplaysanumberofrolesindailylife. 2. Roletaking:Roletakingoccurswhenanagentusestheknownroleplayedbyanotheragenttobuildamodeloftheotheragent'sbehavior.CoutuCoutu1951notedthatroletakingisoftenerroneouslyconfusedwithroleassignment.Roletakingrequiresasharedunderstandingoftheroledimensionsforconstructionofameaningfulmentalmodel.Ifonesearchspecialistencountersanothercomingoutofasearcharea,therstsearcher'smodeloftheotherwouldexpecttheareatobeclearedandmarked. 3. Rolemaking:Inahumansociety,eachpersonthatplaysaroledoessointheirownuniquemanner,consequentlytransformingthebehavioralexpectationsintoconcreteactionindifferentways.A 25

PAGE 37

searchspecialistmighthavespecicknowledgeofcertaintypesofstructures,suchasthoseinacommercialbusinessdistrict.Thisspecialistwouldplayhisroleinadifferentmannerthanonewithknowledgeofsingle-storydwellings. 4. Inter-roleconict:Anagentmayholdmultipleroles,andthegoalsofeachmayconict.Asearchspecialistmightnoteastructuralproblemwithabuildingmakingitdangeroustosearch.Thisinformationwouldconictwiththerole'scoresearchfunctionandmustberesolved. 5. Roledenition:Arolemayberadicallychangedoranewrolecreatedduetochangingcircumstancesoranegotiatedsocialchangeinbehavioralexpectations.Thisisparticularlythecaseinthefaceofchange.AsnotedinHollnagelandWoods2005,pg.102,Newtoolsalterthetasksforwhichtheyweredesigned,indeedalterthesituationsinwhichthetasksoccurandeventheconditionsthatcausepeopletowanttoengagethetasks.Carroll&Campbell,1988,p.4.Theadditionoftechnologytoasocialsystemmayrequirethatroleschange,forexample,sincesomeactivitiescouldnolongerbenecessaryormayrequiredifferentinteraction.Thesefourroledimensionsandverolemechanismsdeneasetofcriteriauponwhichwecanjudgetheresultsofthisdissertation.Asolutionmustbedevelopedthatcanaccountforallofthesemeasures,eitherthroughdirectimplementationandtesting,orallowancesforfuturework.3.2PracticalConstraintsThebasicrolemodeldenesageneralframeworkthatcanbeusedtocoordinateheterogeneousteams.Therolemodelallowsthedenitionofroles,specifyinghoweachrolerelatedtoothersthroughtheposition,whateffecttherolehaswhenitisplayedthroughtherolefunctions,howtherolecanbemodiedandalteredbychangingroleexpectations,andageneralhookforenablinglearningandadaptationthroughsocialinteraction.Inaddition,theverolemechanismsdescribeasetoftoolsforusingrolesinaheterogeneoussystem.However,thismodeldoesnotdirectlyaddresscommoncoordinationissuesthatresultfrommorecomplexsocialinteractions.Therearemanyconstraintsrelatingtocomplexsocialbehavior,butthissectionfocusesonthreethatoccurineldroboticsandaredirectlyinuencedbyexperience.Thefollowingadditionalthreecriteriaareimportant: 26

PAGE 38

MultipleRoles Multiplerolescanbeassumedatthesametime.Sometimesthisleadstoconict,butoftenmultiplerolescanbeaccommodated.Thiscanhappenwhenmultiplerolehavedisjointfunctionsoroverlappingfunctionswithcompatibleexpectations. RoleSharing Multipleagentsshouldbeabletocoordinatetosharearole.Thisisparticularlynecessaryineldroboticswhenahumanoperatorwishesorneedstoassumeallorpartoftherolearobotwasplaying.Asanexample,inLong,Murphy,andHicinbothom2006,robotswereinvolvedinasimulateddeminingtask.Duringthemission,theywererequiredtosearchforminesandwhenaminewasdetected,thehumanoperatorsharedthetasktoprovidethatnalvisualconrmation. MultipleSocialContexts Themechanismshouldsupportmultiplesocialcontextsorsocialsystems.Thisisimportantforreusabilityofrolesandotherrobotfunctionality.Sinceveryfewresearchopportunitiesareprovidedwithentirelynewequipmentandoftenrequireinteractionwithcustomsoftwarepackages,itisnecessarytoreuserobotsystemsandsoftwareacrossmanydifferenttasksandintegrateeasilyintonewenvironments.Inaddition,theautonomoussystemsmustbeabletobecontrolledoroperatedremotelybyahumansincetherearemanyreal-worldsituationsthatrequirehuman-levelintelligence.Traditionally,thehumanoperatorhassimplyoverriddentheautonomyoftherobotandtakendirectcontrolofthesensorsandeffectors.However,theuseofrolessuggeststhefollowingthreecasesofhumanandrobotinteraction: Passiveobservation:Thehumanoperatorisnotexpectingtotakecontroloftherobot,butdoesdesireaccesstoadditionalinformation. Sharedcontrol:Thehumanoperatorisexpectingtocontroltherobotinsomemanner,butisallowingautonomousprocessesontherobottobeactive.Thishasthreebenets.First,thehumanoperatorcandirectandcontroltherobot.Second,thereactiveknowledgegeneratedontherobotisnotlost.Forexample,theoperatorcoulddirectarobottoproceedtoaparticularlocation,butallowtherobottodyanmicallyavoidobstacleenroute.Third,thiscombinationhelpsovercomepracticalissuessuchascommunicationlag,limitedbandwidthandcommunicationlossthatmightproscribedirect,data-intensivecontroloftherobot.Sharedcontrolisoftentermedguardedmotion,althoughsharedcontrol,ingeneral,connotesmorethanjustrobotmotion. 27

PAGE 39

Directcontrol:Finally,directcontrolisimportantforcaseswherefullhumancontrolisrequiredorautonomouscontrolwouldbeineffectiveordangerous.Anapplicationofthiscouldbedirectcontroltodiagnosehardwareorsoftwarefailures.Itisimportantthattheappropriatelevelofdesiredcontrolisvisibleattherolelevel,becauseprovidingtheservicesrequiredmayimpactotherrolesbeingplayed.Inparticular,providingtoomuchdataduringpassiveobservationcanreduceavailablebandwidthformission-criticaltasks,andallowingaccesstoservicessuchasmotorcontrollerscanimpacttheexecutionofhigher-levelrolesthatmayalsousethosecontroller.Providingtheintentofthehumanoperatorcanallowtheinter-roleconictmechanisminformationthatcanreducetheimpactofthiscontention.3.3ImplementationApproachTherolemodelandpracticalconstraintsrequiresomeformofimplementation.Thegeneralapproachtothisimplementationbeginsfromasocialperspective.Thelargeststructurefromthisperspectiveisthesocialcontext.Roleposition,function,expectation,andsocialinteractionarealldenedwithinthisconcept.Therealizationoftherolemodelsoftwarestartswiththisasthekeyconcept.TheimplementationdetailedinChapterFourdenesacontextadapter.Thisstructuretracksthesocialcontextsthatarerelevantatanygiventime.Thisisnotjustlimitedtoasinglecontext;itenablesaccesstomultiplecontexts.Withineachsocialcontext,rolesarethenextmostimportantconcept,andareboundtoaspeciccontext.Theserolesarethenassociatedwitharolepositionelement,arolefunction,andasetofexpectations.Expectationscancutacrossseveralroles,sotheabilitytotrackexpectationsfrommanyrolesmustbeexternaltotheroleitself.Thiscomponentiscalledtheexpectationmanagerandmaintainstheactivesetofexpectationsforallactiveroles.Inasimilarmanner,rolefunctionsneedasimilarcomponenttomaintainthestateandhandletheexecutionofrolefunctions.Forhistoricalreasonsthisiscalledthescriptmanager.Thiscomponentisnotnew,asithasbeenusedinpastwork.However,ithasbeenupdatedtousethescriptsneededforthisdissertation.Scriptsareusedtoimplementrolefunctionsandarecapableofmanagingadenedsocialinteractionwithotherroles.Theabovemeettherequirementsfortheroledimensions,butthedynamicaspectsoftherolearestillundened.Toaddressthesearolemanagerisintroducedtotrack,maintain,andmodifythestateof 28

PAGE 40

therolesforallactivesocialcontexts.Thiscomponentmanagestherolelifecycleandresolvesinter-roleconictinconjunctionwiththeexpectationmanager.Therolemanageralsoextractsinformationaboutavailablerolesandmakesitavailablethroughthecontextadaptertofacilitatetheroletakingofotheragents.Therolemanageralsoenablestheuseofmultiplesimultaneousroleinconjunctionwiththeexpectationmanager.Rolemakingandroledenitionisnotpresentlydened,butwilltinwiththiscomponentaswell.Thelastrequirement,sharedroles,usetheabovecomponentsbutarenotdenedwithinthematpresent.Thissharingispresentlyaccomplishedthroughacombinationofscriptsanddenedsocialinteractionbetweeninstancesoftheroleonmultipleagents.3.4SummaryThischapterdescribeskeyaspectsoftheapproachtakentointroducesocialrolestorobotics:thetheoreticalapproachandpracticalconstraintsthatinuencetheapproach.Italsopresentsseveralpracticalconsiderationsthataredrawnfrompastexperiencewithmobilerobotics.ThetheoreticalapproachmergesimportantconceptsfromtheuseofrolesintechnicalsystemspresentedinChapterTwo,andisultimatelybasedonobservationsfromthehumanuseofrolesinsocialsystems.Therolemodeldescribesfourroledimensions:roleposition,rolefunction,roleexpectations,andsocialinteraction.Allfourdescribestaticaspectsofhowarolefunctions.Therolemodelalsobuildsonverolemechanisms:therolelifecycle,roletaking,rolemaking,handlinginter-roleconict,androledenition.Thesevemechanismsdescribedynamicaspectsofhowrolesareused.Thepracticalconsiderationsaremeta-levelrequirementsfortheroleframework.Thefourconstraintsarethefollowing:multiplesimultaneousroleswillbeneededinmultiplesocialcontexts,humanswillneedtooperateandinteractdirectlywithautonomoussystemsandwillalsoneedtosharerolesandroleresponsibilitiesbetweenthetwo. 29

PAGE 41

ChapterFourImplementationThissectiondescribesthemethodstakentotranslatethetheoreticalapproachinChapterThreetoapracticaltechnicalsystemimplementation.Thegoalsoftheimplementationaretwofold.Therstgoalistoprovideatechnicalframeworkthatcanenablethecoordinatedoperationofmixedteamsusingrolesbasedontheorygroundedinsocialscience.ThesecondgoalistointegratethisframeworkwiththeDistributedFieldRobotArchitectureinamannerthatisconsistentwiththearchitecture'soriginaldesigngoal.Thisintegrationmustallowtheuseofpreviouslydevelopedrobotsoftware.Theimplementationdescribedinthissectionsupportsbothgoals.Thissectionrstdescribesthegeneralframeworkuponwhichthetechnicalsystemisbased.Thisframework,DistributedFieldRobotArchitectureDFRA,isdescribedinSection4.1.ThenSection4.2presentsthesatisfactionofthetheoreticalconstraintswithinthecontextofDFRAbyintroducinganewrobotservice,theCONTEXTADAPTERthatisresponsibleforsocial-context-relatedactivitiesoftherobot,includingroles,expectations,andothercriteriadescribedabove.4.1TheDistributedFieldRobotArchitectureThissectiondiscussesadistributedroboticsarchitecture,DFRA,designedforcomplexroboticstasks.DFRAwasdesignedandbuiltbasedonpriorworkinrobotics,distributedsystemsandsoftwareagents,extendedintoacohesive,usefuldistributedrobotarchitecture.AcompletediscussionofDFRAmaybefoundinLong2004.DFRAitselfisfocusedaroundthreekeyareas.First,thedistributedarchitecturebuildsonexistinghybriddeliberative/reactivearchitecturesusedforindividualrobotsratherthancreatingadistributedarchitecturethatrequiresre-engineeringofexistingrobots.Themajorityofintelligentgroundrobotsareprogrammedusingthehybridarchitecturalparadigm,whichdividestherobotcontrolsoftwareintoalow-levelreactiveorbehaviorallayerandahigherdeliberativelayerMurphy2000.DFRAincorporatesadistributedlayerthatservesasanevenhigherlayer.Thisisconsistentwiththesoftwareengineeringprinciplesofmodularity,informationhiding,andsecurity.Italsomeansthattherobotscanfunctionontheirownevenifthedistributedlayerhasasoftware 30

PAGE 42

failure.Anotheradvantageofthedistributedlayeristhatitisexpectedtobeapplicabletoanysingle-robotarchitecture.Second,thedistributedlayerofthearchitectureincorporatesconceptsfromarticialintelligenceandsoftwareagents.Thedistributedlayerislooselybasedonthepersonaconceptfrompsychology.Eachpersonhasapersona,orwayheorsheinteractswiththeworld,underdifferentcircumstances.Thepersonaprovidesametaphorforthinkingabouttherobotandhowitinteractswithotherrobotsandintelligentagents.Thepersonaconceptisalsoconsistentwithgoodsoftwareengineering,especiallyinformationhidingandsecurity,sinceitimpliesthatallagentsdonothaveaccesstoallaspectsofeachrobot.Thearchitectureallowsthedevelopmentofdistributedalgorithmsanddecisionmakingaswellasinterfaceagentsandagent-orientedcommunicationfromsoftwareagencyanddistributedsystems,allowingthearchitecturetobenetfromadvancesinthosedomainsandtoensurethattherobotswillbeabletoworkwithsoftwareandcognitiveagentsinthelargerinformationalsystem.Third,thearchitectureisdesignedaroundSun'sJinimiddlewarelayerKumarandCohen2000;Li2000,ratherthancreatingamiddlewarelayerfromscratchorattemptingtoadaptasoftwareagentarchitecturesuchasControlofAgent-BasedSystemsCoABSKahnandCicalese2002.Thiswasdoneforthefollowingvereasons: Creatingarobust,functionalmiddlewarelayerisacomplex,time-consumingprocessandnotthemaingoalofthiswork. Jiniisacommerciallydeveloped,stable,openproduct. Jinihasathrivingusercommunitythathasavestedinterestinseeingbugsfoundandxed. Jiniiswellsuitedtothesecurity,networkingandeventmodelsoftheJavalanguage. Large-scaledistributedsystemssuchasCoABSarebuiltonJini;itisaviablechoiceforamiddlewarelayer.TherearesevenkeyconstraintsthathaveinuencedtheapproachtakentothedesignofDFRA,drawnfrompreviousworkineldroboticsCarlson,Murphy,andNelson2004;Gageetal.2004;Zimmeletal.2004,rescueroboticsBurkeetal.2004;Casper2002;Micire2002,anddistributedsystemsStroustrup1994;Tanenbaum1995;Long,Murphy,andParker2003.WhileoriginallyidentiedforworkonDFRA,theseconstraintsequallyapplytothedesignofthissystem. 31

PAGE 43

Behavior-basedanddeliberativesupport:Thearchitecturesupportscommon,provenroboticparadigms.Behavior-basedcontrolhashistoricallyworkedwellforlow-level,time-sensitivecontrol,whilethedeliberativeapproachisgearedtowardlearning,articialintelligenceandprocesseswithweakertimeconstraints. Openstandards:Robothardwareandsoftwareplatformsdonottypicallyhaveanimmenselylonglifeseveraldifferentmodelsofrobotscurrentlyusedhavebeendiscontinuedbythemanufacturerorthemanufacturerhasgoneoutofbusiness.Becauseofthis,itisimportanttobuildonabasethatisopen,exibleandextensible. Faulttolerant:Boththeoverallsystemandindividualmodulesshouldbereliableinthefaceofhardwarefaults,softwareerrorsandnetworkproblems. Adaptable:Thesystemshouldbeabletoadapttoitsoperatingenvironment.BecausethesystemisbasedonaJavafoundation,softwareportabilityisnotanissueaslongasallservicescorrectlyimplementspeciedinterfaces.However,modulesneedtoadapttoallowthenetworkenvironmentasawholetofunctiontobegoodnetworkcitizens. Longevity:Anultimategoalofthissystemislongevity.Arobotshouldnothavetobetakenoutofservicefortheinstallationofchangesandupdates.Tosupportthis,componentsneedtobemodied,administered,loggedandmaintainedatruntime. Consistentprogrammingmodel:Theimplementationshouldabstractthelocalityofobjects.Thesamemethodshouldbeabletoaccesslocalorremoteserviceswithoutsacricingerrorhandlingorperformance. Dynamicsystem:Thesystemshouldbedynamicratherthanstaticandshouldbeabletoexiblyaccommodatenewsensors,effectors,orothercomponents.Thisalsoimpliesthatclientswillbeabletodiscovertheservicesthatareneededatruntimeandadapttotheadditionandremovalofservicesovertime.Thisisparticularlyimportantforarole-basedsystem,whereitmaynotbeimportantwhichagentllsaparticularrole,aslongasanagentisfoundthatcansuccessfullyperformtherole.Figure4showsageneralarchitecturalviewoftheDFRA.Theuppermostportionofthediagramrepresentsthepersonaoftherobot.ThisandthemiddlelayerofthediagramcomprisethedistributedlayerofDFRA.Thebottomportionofthegurerepresentsthefoundationalsingle-agentarchitecturechosenforthiswork,SensorFusionEffectsSFX.TheSFXhybriddeliberative-reactivearchitecturehas 32

PAGE 44

Figure4:TheSensorFusionEffectshybriddeliberative-reactivearchitecturehasbeenextendedthroughtheadditionofapersonathatrepresentskeyportionsoftheunderlyingsystemtodistributedpeers. Figure5:DiagramshowingaportionsoftheoriginalSFXarchitecture. 33

PAGE 45

beenextendedthroughtheadditionofapersonathatrepresentskeyportionsoftheunderlyingsystemtodistributedpeers.4.1.1DistributedLayerThedistributedlayerbuildssquarelyonthehybriddeliberative/reactivelayerofSFX.Thedistributedlayerisformulatedonthenotionofapersona.IntheJungiansense,apersonaisapersonalfacadeonepresentstotheworldJung1928.Inthecaseofarobotarchitecture,arobot'spersonaisthewaytorepresenttherobot'srole,goals,capabilitiesandlimitationstoanotheragent.TheapproachtakeninDFRAistoextendeachprimaryrobotsoftwarecomponentwithaportionthatwillinteractwithadistributedsystem.Thecollectionofthecapabilitiesprovidedbythedistributedportionsofeachcomponentistherobot'spersona.Figure4showsthisvisually.ThegureshowstheSFXfoundationinFigure5asthebaseforthecognitivemodel,withportionsofeachoftherelevantcomponentsextendedtothedistributedrealm.Theportionsarearepresentationofthecapabilities,attributesandknowledgeoftherobottherobotpersona.Figure6ashowsanexampleofthepersonaofarobotthathasanumberofcapabilitiesexposed,suchasthesensorandeffectorpayloadsandtherobot'sskillset.Additionally,otherinformationandattributes,suchasamappingorcartographycapabilityalsohavearepresentationinthepersona.Thesecapabilities,nowincludingroles,canbeusedbyotheragentsinthesysteminanintelligentmanner.Amissionplannercanexaminethecapabilitiesofallavailablerobotsandgenerateanoptimalplanbasedonthedistributionofskills.LikewisearecruitmentagentsuchastheagentdetailedinGage2004canlterrequestsbasedontherequestedcapabilitesconstrainedtotheavailablepossibilities.Oramappingagentcouldrequestlocalmapsfromrobotsthathaveamappingcapabilityintheirpersonaandstitchtheselocalmapsintoaglobalmapandreturntheresulttothecartographers.Inthismannerglobalmapscanbedynamicallycreatedanddistributedbyanagentthatisonlyinterestedinotheragentsthathaveaparticularcapability.4.2TheCONTEXTADAPTERTheCONTEXTADAPTERisusedbothtoimplementroleswithintheDFRAandallowagentsusingDFRAtointeractinavarietyofexternalapplicationdomainsthatcorrespondtosocialcontextsorindividualsocialsystems.TheCONTEXTADAPTERisdesignedasaservicewithintheDFRAarchitecture,andassuchisinuencedbythedesigndescribedabove.Inparticular,theCONTEXTADAPTERisprimarily 34

PAGE 46

abFigure6:Therobot'spersonacanbeexpressedasthecapabilitiesoftherobot.Portionsofthepersonamaybeprotectedinsomemannertorestrictaccesstotheunderlyingcapabilities. concernedwiththedistributedanddeliberativelayersofthearchitecture,usingotherservicesattheselevelsforexecutionofrobotactions,discoveryofneededservices,anddynamicadaptationtochangingnetworkandrobotconditions.TheCONTEXTADAPTERisaDFRAserviceandservesatwofoldpurpose.First,itprovidesastructurebywhichtherobotservicesareabletoutilizeroles.Tothisend,theCONTEXTADAPTERdenesaROLEMANAGER,discussedinfurtherdetailinSection4.2.1,andinterfaceswithapolicymanagementsystemthroughanEXPECTATIONMANAGER,examinedinSection4.2.3.Second,theCONTEXTADAPTERallowsaccesstoexternalsocialcontextsorapplicationdomains.Tothisend,itprovidesthecapabilitytointegrateanarbitrarynumberofCONTEXTPLUGINs,describedinSection4.2.2,thatprovideamappingfromDFRAmessagingtothatoftheexternalcontext.Tosupportthesetwokeygoals,theCONTEXTADAPTERhasthefollowingresponsibilities: DFRAserviceresponsibilities:TheCONTEXTADAPTERisresponsibleforproperoperationwithintheDFRAframework.Thisincludesservicelifecyclemanagementtaskssuchasinitialization,activationanddeactivation,properlypersistingrelevantstate,eventgenerationandprocessingandexceptionhandling. ConstructtheROLEMANAGER:Atinitialization,theCONTEXTADAPTERisrequiredtoconstructandactivatetheROLEMANAGERforaccesstorole-basedfunctionality. ConstructtheEXPECTATIONMANAGER:Atinitialization,theCONTEXTADAPTERisrequiredtoconstructandactivatetheEXPECTATIONMANAGERtomanagepolicyandsocialexpectation. 35

PAGE 47

Figure7:High-levelblockdiagramfortheCONTEXTADAPTERservice. Loadcontextplugins:Atinitialization,theCONTEXTADAPTERisalsorequiredtoloadneededCONTEXTPLUGINs.Thissteploadsinformationabouttheexternalcontexts.Thisinformationincludestherolesthatarespeciedinthatdomainaswellasinitialexpectationsforthedifferentroles. Messagerouting:Finally,theCONTEXTADAPTERensuresthatallkeycomponentscanproperlycommunicatewitheachother.Figure7showsablock-leveldiagramoftheCONTEXTADAPTER.Thefollowingsectionsdiscussthemajoraspectsofthisdesign.Section4.2discussesthestructureandmakeupoftheservice,includingtheintegrationwithrolesthroughtheROLEMANAGERSection4.2.1andexternalcontextsthroughtheCONTEXTPLUGINmechanismSection4.2.2.4.2.1TheROLEMANAGERTheROLEMANAGERisprimarilyresponsiblefortheproperoperationofroles.TheROLEMANAGERcontainsarepositoryofroleclassesprovidedbythecontextplugins.Eachoftheseroleclassesisassociatedwithaparticularcontext,andisanunplayedrole.Whenaroleclassisactivated,itis 36

PAGE 48

Figure8:DiagramoftheCONTEXTADAPTERinteractingwiththreetypesofexternalcontexts.Thethreecontextsdifferonrepresentationofroleswithinthecontext. Figure9:BlockdiagramoftheCONTEXTADAPTERservicehighlightingtheinternalROLEMANAGERcomponent. 37

PAGE 49

instantiatedandparameterizedwiththecharacteristicsoftheparticularroleassignment.Thisallowsaroletobeplayedmultipletimessimultaneously.4.2.1.1SupportforRolesAroleisimplementedasaJavaclass,witharoleobjectinstantiatedforeachroleineachknowncontext.Eachroleinstanceprovidesaccesstoroledimensions;theROLEMANAGERitselfhandlestherolemechanisms.4.2.1.2SupportforRoleDimensionsTheROLEMANAGERenablessupportforthefourroledimensionsdescribedinSection3.1.Eachroleisassociatedwithdatadescribingtherole'spositioninthesocialcontext.Theimplementationcurrentlydescribestherolepositionintermsofave-pointscale.Thepositioncantakeononeofthefollowingvalues:LOWEST,LOW,NORMAL,HIGH,orHIGHEST.Thesevaluescanbeusedbytherolemechanisms,particularlywhenhandlingroleconict,togeneratearelativerankofeachrole.Inadditiontopositioninformation,eachroleisassociatedwithascriptSchankandAbelson1977thatenablestheroletoperformafunction.ThesescriptshavebeenusedinpriorworkBarnes,Murphy,andCraighead2005;Longetal.2005toenablecomplexrobotbehaviorsandaresuitableforrepresentingrolefunctions.Section4.2.5providesfurtherinformationonhowscriptsareusedtocoordinateheterogeneousteamsinthesocialcontexts.Thethirdroledimension,expectation,isenabledthroughtheuseofanEXPECTATIONMANAGER,whichisresponsiblefortracking,analyzingandprocessingbasicexpectationsorpoliciesthatareenabledwhencertainrolesareinuse.TheEXPECTATIONMANAGERdenesageneralinterfaceforexpectations,andprovidestwosampleimplementations.Therstisasimple,permissiveEXPECTATIONMANAGERthatallowsmostroleassignments.ThesecondEXPECTATIONMANAGERisaninterfacetoKAoSBradshawetal.1997.WhenrequestedbytheROLEMANAGERduringrolelifecyclestatechanges,thisEXPECTATIONMANAGERqueriesKAoSastowhetherthenewstateispermittedgiventhecurrentcongurationofroles.Thisismostusefulduringroleassignment.Thenaldimension,socialinteraction,isnotfullysupportedinthiswork.Basicsocialinteractionisdenedbyscripts.However,truesocialinteractionmustbeadaptive,sinceheterogeneousteamswilllikelybecomposedofhumanagentswithdifferingabilitiesandmethodsofinteracting.Thislattersupportisleftforfuturework.4.2.1.3SupportforRoleMechanismsTheROLEMANAGERis,ingeneral,abletosupporttheverolemechanismsdescribedinSection3.1.TheROLEMANAGERsupportstherolelifecycleandallowsassignmentofrolestotheagent.Thismechanismalsoallowsmultipleroleassignment,enablinganagent 38

PAGE 50

Figure10:Finite-StateMachineshowingtherolelifecycle. Figure11:Finite-StateMachineshowingtherolelifecycle.HighlightedportionsofthediagramrepresenttransitionsthatupdateexpectationsviatheEXPECTATIONMANAGERcomponent. 39

PAGE 51

toassumedifferentroles,potentiallyacrossmultiplecontexts.Asnotedpreviously,supportfortheroleassignmentmechanismdoesnotguaranteethatanygivenassignmentwillbesuccessful.Theroleassignmentmayfail,forexample,iftheagentisaskedtoplayarolethatconictswithotheractiverolesoriftheagentdoesnothavethecapabilitytosupportthedesiredcharacteristicsoftherole.Ateachrolestatechange,therolelifecyclemechanismsquerytheEXPECTATIONMANAGERastowhetherthenewstateispermitted.Figure10showsthegenerallifecyclestateforeachrole,andtheallowedstatetransitions.ThehighlightedportionofFigure11indicatesthetransitionsthattriggeracheckwiththeEXPECTATIONMANAGER.Iftheassignmentissuccessful,therolelifecyclesupportsarolechangemechanismtohandlethemechanicsofsequencing,activatinganddeactivatingthevariousroles.Iftheassignmentisnotsuccessful,however,theROLEMANAGERmediatestheroleconict.TheROLEMANAGERpresentlyusesasimple,priority-based-resolutionmechanismwheretherolepositionisexaminedandthehigher-priorityrolewins,howeverthisisabletouseotherheuristicssuchasaffectiveresolutionwheretheemotionalstateoftherobotcaninuenceroleselection,policy-basedresolutionwhereacontext-specicsetofpoliciesconstrainroleselection,orhuman-in-the-loopresolutionwhereahumanagentdecidestheroleselectionandresolvesconict.LikesocialinteractioninSection4.2.1.2,roletakingispartiallysupported.Simpleroletakingisenabledthoughinastaticmanner.Thisisthroughtheknowndenitionandimplementationoftherolesandfunctionsoftherole.However,foragentstoimproveinautonomy,theymustbeabledynamicallymodifyaninternalmodelofotheragentsastheirknowledgeofotheragentsincreases.Thislatteraspectofroletakingisalsoleftforfuturework.Althoughtheremainingtwomechanismsarenotpresentlysupported,theROLEMANAGERisdesignedtobeextensibletosupportrolemaking,androledenition.Rolemakingrequiressomeformofself-evaluationandfeedback.Thismechanismenablesanupdatetotheexpectationsassociatedwitharoletoreecthowanagentactuallyplaysarole.Forexample,theexpectationforthetimerequiredforagivenrolemayvaryfromagenttoagent.Thismechanismwouldpersonalizetherole.Thismechanisminuencesroletaking,whereanagentcanuseitsknowledgeoftheroleofotheragentstopredicttheiroverallbehavior.Finally,roledenitionistheconstructionofnewrolesorchangingthebehaviorofexistingrolesovertime.Thismechanismispresentlyperformedbyahumanprogrammerwhodesignsthescriptsand 40

PAGE 52

behaviorsthatimplementtheroleinapurelymanualmanner.Automationofbehaviorconstructionisstillanactiveareaofresearch.4.2.2CONTEXTPLUGINs Figure12:Therearethreetypesofexternalcontextsconsidered. ACONTEXTPLUGINisakeycomponentoftheCONTEXTADAPTER.ACONTEXTPLUGINallowsinteractionbetweenDFRAandanarbitraryexternalcontextthroughtwomethods.Thetwomethodsaresyntactictranslationandsemanticmapping.Syntactictranslation,showninFigure13allowsbasicconnectivitybetweentheJini-basedDFRAandthecontextofinterest.Thistranslationenablesacommunicationchannelbetweenthetwosystemsandisentirelyimplementation-dependent,mappinghowthetwosystemscommunicate.Forexample,thislayermightsendmessagesusingtheJavaRemoteMethodProtocolJRMPononesideandviacustomsocketsontheother.Semanticmapping,ontheotherhand,isconcernedwiththecontentofthemessagesandtranslateswhatisbeingpassedoverthecommunicationchannel.Forexample,thislayermighttranslateDFRAmethodcallstospecicJointArchitectureforUnmannedSystemsJAUSmessages. 41

PAGE 53

Figure13:Syntaxtranslationenablesinter-systemcommunication. Figure14:Semanticmappingtranslatesthecontentsofmessages. 42

PAGE 54

Dependingonthecontext,zero,oneorbothmethodsmaybeused.Thus,eachCONTEXTPLUGINcanvaryincomplexity.Section4.2.3examinesthreesuchcasesrelatedtopolicyandSection4.2.4describesthemessagingsupportrequiredinfurtherdetail.4.2.3TheEXPECTATIONMANAGER Figure15:Context-specicexpectationsarepropagatedthroughtheCONTEXTPLUGINandmaintainedintheEXPECTATIONMANAGER. TheEXPECTATIONMANAGERisprimarilyresponsibleforthepropertrackingofexpectation.TheEXPECTATIONMANAGERcontainsarepositoryofexpectationsthatareinitiallysetandperiodicallyupdatedbyacontext'sCONTEXTPLUGIN.TheEXPECTATIONMANAGERisqueriedbytheROLEMANAGERduringrolestatechangetoprovidetheactivesetofroleexpectations.Theseexpectationsareimplementedintwoways:asetofkey-valuepairsdenotingspecicpropertiesrelevanttoactiverolesandaspolicies.Theuseofpolicyprovidestwobenetstothediscussionoftheactionsassociatedwithroles.First,policycanprovideameta-leveldescriptionoftheactionsthatanagentperformsandcanbeadjustedatruntimetomodifyaspectsofagentbehavior.Second,theseadjustmentscanbemadeexternaltotheactionspecicationitself.Thus,inthecontextofroles,thefunctionsandtaskscanbespeciedorevenencodedintheagentconstruction,butpolicycanbeusedtodynamicallyadjustorparameterizethe 43

PAGE 55

actionsoftheagent.Thepolicydenitioncanbeabstractedfromthedetailsofthebehaviortodescribeconstraintsontheagent'sactionwithouttheknowledgeofhowthoseconstraintswillbehonored.Unfortunately,theprocessofconstructingandusingpolicyisnotstraightforwardwhendealingwitharbitraryexternalcontexts.Weconsiderthethreefollowingspecialcasesofexternalcontextsandagents,showninFigure15: Externalpolicyagentsarecompatible:Inthesimplestcase,thepoliciesdenedbytheexternalpolicyagentandthepoliciesusedinternallytoCONTEXTADAPTERaresemanticallyidenticalortriviallyconverted.Inthiscase,policyadjustmentsdenedbytheexternalagentarepropagatedthroughtheCONTEXTPLUGINandtransferredtotheinternalpolicyset.Thiscaseisidealsincethecomputationalcomplexityofthemessagingiskepttoaminimum. Externalpolicyagentsareincompatible:Whentheexternalpoliciesexist,butaresemanticallydifferentfromthepoliciesimplementedintheCONTEXTADAPTER,thenaprocessofsemanticmappingmustoccurLongandMurphy2005.Thisisanon-trivialprocess,andautomatedsemanticmappingispresentlythesubjectofintenseresearchbytheSemanticWebcommunity.However,ifautomatedmappingisunavailable,thenamanualmapcanbeconstructedintheCONTEXTPLUGIN.Thiscaseisdiscussedinmoredetailbelow. Externalagentsdonotusepolicies:Inthiscontext,agentsintheexternaldomaindonotusepolicies.However,weassumethateventheseagentswillincludesomemechanismtosendmessagesthroughthecontextadapter,andthatthesemessageswillincludesomemannerofinstructiondetailingthedesiredbehavior.ThespecicCONTEXTPLUGINforthisdomainmustnotonlytranslatethemessageandgeneratetheappropriatesetofrolemessages,butitmustalsocreateorupdatetheappropriatepolicieswithintheCONTEXTADAPTER.TheCONTEXTPLUGINsimplementedtosupportthethreedemonstrationsconsistofacontextwhereexternalagentsdonotusepoliciestwocontextsandlimitedsupportforacontextwithacompatiblepolicyagent.Thislimitedsupporthasbeendemonstratedintesting,butfullimplementationofthismappingisleftforfuturework.Inbothcases,theCONTEXTPLUGINisconguredwithabasesetofcontext-specicpoliciesdenedasaparameterlist.TheEXPECTATIONMANAGERdenesageneralinterfaceforexpectations,andtheworkinthisdissertationprovidestwosampleimplementations.Therstisasimple,permissiveEXPECTATIONMANAGERthatallowsmostroleassignments,andsimplyreturnstherole'scurrent 44

PAGE 56

key-valuepropertylistwhenqueried.ThesecondEXPECTATIONMANAGERisaninterfacetoKAoSBradshawetal.1997.WhenrequestedbytheROLEMANAGERduringrolelifecyclestatechanges,thisEXPECTATIONMANAGERqueriesKAoSastowhetherthenewstateispermittedgiventhecurrentcongurationofroles.TheformerEXPECTATIONMANAGERisusedforthedemonstrationlaterinthiswork,butthelatterhasbeenimplementedtosupportongoingworkonaelddemonstration.4.2.4MessagingTheCONTEXTADAPTER,usingtheROLEMANAGER,EXPECTATIONMANAGERandCONTEXTPLUGINmechanismsupportsthefollowingtypesofmessages: Messagesaboutroles:Therst,andperhapsmostimportant,typeofmessagearethosespecicallyrelatedtothekeymechanismsdescribedinSection4.2.1.3.Thesemessagesdirecttheagenttoassign,activate,deactivate,relinquish,andchangeroles,andassuchtheyarenotdirectedtoaparticularrole,butrathertotheROLEMANAGERitself.Messagesthatareusedtoresolveinter-roleconictalsofallinthiscategory. Role-specicmessages:Thesecondclassofmessagesarethosethathavemeaningtoaparticularrole.Thesemessagestypicallymodifyorrequestinformationaboutsomeaspectoftherole'sfunctions,tasksorbehavioralexpectations. Role-agnosticmessages:Thethirdclassofmessagesarerole-agnosticanddonottargetaspecicrole.Instead,thesemessagesareusedtobroadcastinformationtoallroles.Thisthirdclassismostusefulforbroadchangestobehavioralexpectationandwilltypicallyinvolvechangestopolicythatinuencehowtheagentperformsallofitsactions.Inaddition,thesemessagesalsoprovideacommunicationwiththerobotitself,lteredthroughthelensoftheactiveroles.4.2.5ScriptsScriptsarecurrentlyusedtohandlecomplexbehavioronindividualrobots,butthismechanismcanbeextendedtocoordinatetheactivityofheterogeneousagents.Indeed,theoriginalviewofscriptswasintendedtoprovideawayforanagenttounderstandandcoordinatewithothers. Ascriptmustbewrittenfromoneparticularrole'spointofview.Acustomerseesarestaurantoneway,acookseesitanotherway.Scriptsfrommanyperspectivesarecombinedtoform 45

PAGE 57

whatmightbeconsideredthe`wholeview'oftherestaurant.Sucha`wholeview'israrely,ifever,neededorcalledupinactualunderstanding,althoughitmightwellconstitutewhatwemayconsidertobeone's`concept'ofarestaurant.SchankandAbelson1977Iftheyweresolelyconcernedwithcoordinatingactions,scriptswouldbelittlemorethannite-stateautomata.Scriptsalsodenemetadatathatdescribethecontextandscenarioinwhichthescriptappliestheentryconditionsandprops,therolesthatareinvolvedinthescript,theresultsofthescript,thecausalchainofeventsorcuesthattakethebehaviorofthescriptfromtheentryconditionstotheendresult.Scriptsalsodisplayarecursivenatureexceptionalconditionswithinonescriptcantriggertheentryconditionsofasubscripttailoredtohandletheproblemencounteredinthemainscript.Scripts,likelife,canhavevaryingdegreesofcomplexity.Themostbasicformsofscripts,instrumentalscripts,arenotableinthattheydenesequencesofactionsbuthavelittlevariability.Thesescriptstendtoinvolvefewactorsandareusuallysimpleinnature.Morecomplexscriptsareknownassituationalscripts,whichgenerallyhavethreenotablecharacteristics:1thesituationiswell-knownandspecied;2therearemultipleactorswithinterdependentrolesandactions;and3thereisasharedunderstandingofgoalsandactions.Figure16showsthisconceptvisually. Figure16:Eachteammemberhasadifferentrole,butmayperformteamoperationsthroughasharedscript.Knowledgeofthescriptandtherolesofothersallowsroletakingandcoordinatedactivity. Thescriptsthatimplementtherolesforthisworkareprimarilyoftheformervariety,instrumentalscripts.Forexample,theDRONErolebelowisusedforwaypoint-basednavigation.Thescriptforthis 46

PAGE 58

rolesequenceslower-levelDFRAbehaviorstoaccomplishthis,butdoesnothaveinterdependentactionwithotherroles.Ontheotherhand,theCOMMANDERrolealsobelow,isasituationalscript.Itsequencestheinteractionofmultipleinterdependentrolesandactorsinawell-knownandspeciedsituationandsharesanunderstandingofthegoalsoftheotheragentsifonlybecauseitassignedthegoalsintherstplace.4.3ImplementedRolesAsetofroleshavebeendenedtosupportoperationsinthreesocialcontexts.ThesocialcontextsandasetofdemonstrationsareexpandedinChapterFive.Thissectiondescribestherolesinageneralsenseandbrieydescribesthegeneralfunctionsandexpectationsofeach;thespecicvaluesofcontext-specicexpectationsarepresentedinthenextchapterduringthediscussionofeachdemonstrationscenario.Tablesdescribingtherolesandsummarizingthefunctions,expectationsandpositionofeachmaybefoundinAppendixB.AppendixCliststheimplementationclassesfortheseroles,includingthescriptsthathandletherolefunctions.4.3.1DRONETheDRONEroleisusedforbasic,waypoint-basednavigation.Adronehasnoinitiative,butisdirectedtomovetogoalsprovidedwhentheroleisassigned.ThefollowingarethecurrentlyimplementedfunctionsandexpectationsoftheDRONErole: Storesalistofwaypointgoals,passedinasexpecations. OptionallyusesaCARTOGRAPHERserviceforpathplanning,locatedthroughthestandardDFRAlookupmechanism.ThecurrentCARTOGRAPHERimplementationgeneratesalistofwaypointsfromtherobot'slocationtothegoal,routingaroundknownobstaclesusingtheTrullaalgorithmHughes,Tokuta,andRanganathan1992;Murphy,Hughes,andNoll1996.UseoftheCARTOGRAPHERiscontrolledbyexpectation. TheDRONEusesaMOVETOWAYPOINTservicetodrivetothewaypointsinsequence.WithnoCARTOGRAPHER,thisissimplythesetofwaypointgoals,otherwiseitincludestheplannedwaypointsub-goals. Whenagoalisreached,theDRONEsendsanoticationlistingallcurrentgoalsandsubgoalstoregisteredlisteners. 47

PAGE 59

Whenallgoalshavebeenreached,theDRONEsendsanoticationthatthegoalshavebeencompleted. Additionally,theDRONEissensitivetotheexpectationofvaryinglevelsofcaution.CautionmaybeNOT SET,HIGH,NORMAL,orLOW,withthelevelsdenotingaspeed25%,50%,75%,or100%ofmaximum,respectively.Thevalueswerechosentoexhibitavisuallynoticeablespeeddifferenceinthesimulationdemonstrationdescribedinthenextchapter.4.3.2COURIERArobotplayingtheCOURIERroleisintendeddeliverapayloadtoaparticulargoallocation.ThefollowingarethecurrentlyimplementedfunctionsandexpectationsoftheCOURIERrole: Storesalistofwaypointgoals,passedinasexpecations. OptionallyusesaCARTOGRAPHERserviceforpathplanning,locatedthroughthestandardDFRAlookupmechanism.ThecurrentCARTOGRAPHERimplementationgeneratesalistofwaypointsfromtherobot'slocationtothegoal,routingaroundknownobstaclesusingtheTrullaalgorithm.UseoftheCARTOGRAPHERiscontrolledbyexpectation. TheCOURIERusesaMOVETOWAYPOINTservicetodrivetothewaypointsinsequence.WithnoCARTOGRAPHER,thisissimplythesetofwaypointgoals,otherwiseitincludestheplannedwaypointsub-goals. Whenagoalisreached,theCOURIERsendsanoticationlistingallcurrentgoalsandsubgoalstoregisteredlisteners. Whenallgoalshavebeenreached,theCOURIERsendsanoticationthatthegoalshavebeencompleted. Additionally,theCOURIERissensitivetotheexpectationofvaryinglevelsofcaution.CautionmaybeNOT SET,HIGH,NORMAL,orLOW,withthelevelsdenotingaspeed25%,50%,75%,or100%ofmaximum,respectively.Thevalueswerechosentoexhibitavisuallynoticeablespeeddifferenceinthesimulationdemonstrationdescribedinthenextchapter. WhentheCOURIERreachesthelastwaypoint,itismayautomaticallyreleaseitspayload,subjecttocontext-specicexpectation.Inthesimulationscenario,thishastheeffectoflaunchingthepayload,aUAV. 48

PAGE 60

4.3.3SCOUTArobotplayingtheSCOUTroleisintendedtogatherinformationandreportthisinformationtootherinterestedagents.ThefollowingarethecurrentlyimplementedfunctionsandexpectationsoftheSCOUTrole: Storesalistofwaypointgoals,passedinasexpecations. OptionallyusesaCARTOGRAPHERserviceforpathplanning,locatedthroughthestandardDFRAlookupmechanism.ThecurrentCARTOGRAPHERimplementationgeneratesalistofwaypointsfromtherobot'slocationtothegoal,routingaroundknownobstaclesusingtheTrullaalgorithm.UseoftheCARTOGRAPHERiscontrolledbyexpectation. TheSCOUTusesaMOVETOWAYPOINTservicetodrivetothewaypointsinsequence.WithnoCARTOGRAPHER,thisissimplythesetofwaypointgoals,otherwiseitincludestheplannedwaypointsub-goals. Whenagoalisreached,theSCOUTsendsanoticationlistingallcurrentgoalsandsubgoalstoregisteredlisteners. Whenallgoalshavebeenreached,theSCOUTsendsanoticationthatthegoalshavebeencompleted. Additionally,theSCOUTissensitivetotheexpectationofvaryinglevelsofcaution.CautionmaybeNOT SET,HIGH,NORMAL,orLOW,withthelevelsdenotingaspeed25%,50%,75%,or100%ofmaximum,respectively.Thevalueswerechosentoexhibitavisuallynoticeablespeeddifferenceinthesimulationdemonstrationdescribedinthenextchapter. TheSCOUTmustalsosendanoticationifconditionsofinterestareencountered.Forthesimulationscenario,forexample,theSCOUTsendsanoticationifanexplosivedeviceisfoundalongthepier.Inthephysicalscenario,theSCOUTsendsanoticationifitencountersterrainthatcausesittodeviatefromaplannedroute.ThisnoticationupdatestheinternalCARTOGRAPHERmaps.4.3.4OPERATORTheOPERATORistheroleassumedbyanagentthatisusingthefunctionalityofanotheragent.ThisroleistheconverseoftheTELEFACTORrole,andistherolethatisassumedwhenusinganotherrobotasa 49

PAGE 61

TELEFACTOR.TherearenoovertactionsprovidedbytheOPERATORrole,howevertheadoptionoftherolemaycausevariousinteractionswithotheractiveroles.Thefollowingexpectationisimportant: Thedesiredoperatortypeshouldbeprovided.ThisshouldbepassedtothecorrespondingTELEFACTOR.PossiblevaluesarePASSIVE,SHARED,orDIRECT.AfurtherdiscussionofthisissuemaybefoundinSection3.2.Thecurrentimplementationofthisroleisminimal,andisprimarilyusedasamarkerroletotriggerroleconictinthedemonstrationscenarios.Completeimplementationisleftforfuturework.4.3.5TELEFACTORWhenplayingtheTELEFACTORrole,therobotactsastheeyesandearsofanotheragent,theOPERATOR,andmayneedtoprovidedatafromanarbitrarycapabilityoftherobot.TherearenoovertactionsprovidedbytheTELEFACTORrole,howevertheadoptionoftherolemayinteractwithotheractiveroles.Thefollowingexpectationisimportant: Thedesiredoperatortypeshouldbeprovided.ThisshouldbeprovidedbythecorrespondingOPERATOR.PossiblevaluesarePASSIVE,SHARED,orDIRECT.AfurtherdiscussionofthisissuemaybefoundinSection3.2.Thecurrentimplementationofthisroleisminimal,andisprimarilyusedasamarkerroletotriggerroleconictinthedemonstrationscenarios.Completeimplementationisleftforfuturework.4.3.6RELAYAcommunicationRELAYtriestoprovidetwoormoreagentswithastablecommunicationlink.Ingeneral,aRELAYwillmovetomaximizethebandwidthorotherdesiredcharacteristicofthenetworkconnection.Forthisdemonstration,however,theRELAYissimplyamarkerrole,sincetheUAVisnotcapableofautonomousight.4.3.7COMMANDERAnagentplayingtheCOMMANDERroleisresponsiblefordirectingandcoordinatingtheoverallmission.Tofullltheseresponsibilities,theCOMMANDERmayassign,activate,deactivateorrelinquishrolestootheragents.TheCOMMANDERhasthehighestsocialpositioninthesimulationscenario.TheCOMMANDERhasthefollowingfunctionsandexpectations: 50

PAGE 62

TheCOMMANDERhasthemodeloftheoveralldesiredscenarioandtakesactiontodirectotheragentstofulllgoalsofthescenario. TheCOMMANDERusesthestandardDFRAlookupanddiscoverymechanismstondservicesonotherrobots. TheCOMMANDERcanassign,activate,deactivateorrelinquishrolesonotheragents. TheexpectationsfortheCOMMANDERconsistofelementsthatmustbesenttootherroles.Thatis,theyarethescenariogoalsthatarethensentasroleexpectationstotheagentandrolesthatfulllthegoals.4.3.8SICKTheSICKroleisaspecialrolethatdenotesimpairedfunctionality.Thisrolecanrelieveotherobligations,butrequiredthesickrobottoseekhelp.However,whileassumingtheSICKrolecanrelieveormodifyexpectation,itaddsexpectationsaswell.Inparticular,thereisanexpectationthattheagenttrytogetwell,andifthisexpectationisnotmet,thentheagentmayfacesocialsanctions.Parsons,pp.436discussestheseexpectationsinmoredetail.TheSICKrolehasbeenidentied,butonlyaskeletonimplementationisprovidedinthiswork.Fulldenitionisleftforfuturework.4.4SummaryThissectiondescribedthemethodstakentotranslatethetheoreticalapproachinChapterThreetoapracticaltechnicalsystemimplementation.Thissectionrstdescribedthegeneralframeworkuponwhichthetechnicalsystemisbased.Thisframework,DistributedFieldRobotArchitectureDFRA,wasdescribedinSection4.1.ThenSection4.2presentedthesatisfactionofthetheoreticalconstraintswithinthecontextofDFRAbyintroducinganewrobotservice,theCONTEXTADAPTERthatisresponsibleforsocial-context-relatedactivitiesoftherobot,includingroles,expectations,andtheothercriteriadescribedabove.AsetofadditionalcomponentshelpedtheCONTEXTADAPTERwiththistask.TheROLEMANAGERhandledrolemechanismsandmaintainedthestateoftheactiveroles.TheEXPECTATIONMANAGERenabledtheserolestousesocialexpectationtomodifythefunctionsoftherole.ThefunctionsthemselvesaremanagedbyaseparateSCRIPTMANAGERserviceandareimplementedthroughtheuseofscriptsfromarticialintelligence.Thecombinationofthesecomponentsimplementstherolemodelandadditionalconstraintsforthiswork. 51

PAGE 63

ChapterFiveCaseStudiesThischapterpresentsthecasestudiesanddemonstrationsusedtosupporttheoverallresearchquestionofthiswork:Cantheuseofsocialrolesenablethecoordinatedoperationofrobotsinahigher-ordersocialorteamsetting?ThesedemonstrationsshowthattheapproachandimplementationdescribedinChapterThreeandChapterFourcanbeusedtodesignandimplementrole-basedsolutionstopracticalchallenges.AllworkinthischapterwasperformedtosupportthreedifferentfundedprojectsfromtheArmyResearchInstitute,OfceofNavalResearchandDARPA.Therewerefourprimaryobjectivesforthesedemonstrations: Demonstratethatrolescanbeusedtocoordinateheterogeneousteamsofagents.Thisistheprimarythrustoftheresearchquestion,andthekeydemonstrationgoalofthecasestudies. Demonstratethecoreroledimensions:position,functionandexpectation. Demonstratethecorerolemechanisms:lifecycle,taking,making,handlinginter-roleconict. Demonstratetheadditionrequirements:multiplesimultaneousroles,rolesharing,multiplesocialcontexts.Roledenitionisnotaddressedinthesedemonstrations,androlemakingispresentedinthecontextofadesigndiscussion,butisnotimplemented.Bothrolemakinganddenitioninvolveonlinelearningandadaptation,andbothoftheseareasarebeyondthescopeofthiswork.Theroledenitionusedforthesedemonstrationsmanualdenitionthroughhuman-developedroles.ThreecasestudiesanddemonstrationscenariosarepresentedinSection5.2.TherstcasestudySection5.2.1issolelyadesignscenariothatdescribeshowthelearningandadaptationmechanismsthatwerenottestedcouldbeappliedtoaheterogeneousteam.Thiscasestudydescribesacoordinatedhuman-robotteamdesignedtomimicthestandardbuilding-clearingbehaviorofastandardUSArmyreteam.ThesecondcasestudySection5.2.2concernsthecoordinatedoperationofaheterogeneousteaminalittoralenvironment.ThisdemonstrationusesasimulatedUAV,UGVandrealhumanoperatorthatcoordinatetoclearaportfacilityofexplosivedevices.Simulationisusedtoduetologisticalconstraints, 52

PAGE 64

andallowsdemonstrationofcorerolemechanismsanddimensionsinacomplexcoordinatedscenario.ThenalcasestudySection5.2.3describestheapplicationofrolestocoordinatingtheinteractionofasingleUGVandhumanoperatorusinganOCUinanoutdooreldenvironment.Inthisdemonstration,therobotisusedasaremotesensingplatformtodiscoverpreviously-unknownterrainfeatures.Thiscasestudygroundstherolemodelandimplementationandshowsthatbasicroledimensionsandmechanismsoperateasdesigned.5.1ElementsofaFunctionalRole-BasedSystemChapterThreeofthisdissertationpresentedanapproachtomeettheaboveobjectives.Thischapterdemonstratesasubsetoftherolerequirementsduetotimeandequipmentconstraints.Theroledimensionsandmechanismsthatarenotbemetarelargelyconcernedwithlearningandadaptation,whichisalargeareaofresearchinitsownright.Thusthataspectofthisworkisleftforfutureinvestigation.Theremainderofthissectionenumeratestheroledimensionsandmechanismsthatwillbedemonstratedandthosethatwillnot.Thefollowingroledimensionsareshowninthiswork: Positionisshownthroughuseofasimplehierarchyintheroledenitionforbasicconicthandling.Whenassigningmultipleroleswithdifferentpositionsandconictingfunctionsorexpectations,thehigherpositionwins.BasicpositionisusedinSection5.2.2andSection5.2.3tohelpresolveinter-roleconict.Section5.2.1showshowpositionisdenedinamorecomplex,situationally-dependentmanner. FunctionisshownthroughstaticallydenedSCRIPTs.Onroleassignmentthescriptisloaded.Onactivation,thescriptisstarted,andthisscriptexecutionsequenceslower-levelrobotbehaviortoperformtherolefunction.Allthreedemonstrationsdescribeorshowthesatisfactionofrolefunction. Expectationsaresupportedthroughsimplecontext-dependentproperties.Section5.2.2showshowthesesimpleexpectationscanmodifyrolefunctiontoproducedifferentbehaviorthroughvariedexpectationsfortheDRONEandCOURIERrolesinthesimulation.Section5.2.1discussesmoreadvancedusesofexpectation. SocialInteractioniscurrentlystaticallydenedthroughSCRIPTS.Section5.2.2presentsandexampleofasharedrolethatrequirescoordinationandinteractionbetweenmultipleagents. 53

PAGE 65

AdvancedapplicationsoflearningandadaptationarediscussedinSection5.2.1,butimplementationisleftforfuturework.Thefollowingrolemechanismsareshowninthiswork: LifecycleiscontrolledthroughtheROLEMANAGER,whichassigns,activates,deactivates,relinquishesandotherwisechangesofrolestateonanagent.Rolelifecycleisrequiredtosupportallbasicroleactivitiesandisdiscussedorshowninallthreedemonstrations. RoleTakingsupportssharedroles,whereeachactormusttaketheroleoftheothertooperateproperly.InSection5.2.2,therobotandhumancansharetheSCOUTroletojointlyidentifyanunknownobject. RoleMakingoccurswhenexecutionofaroleonarobotupdatestheexpectationsfortherole.ThisisdiscussedinSection5.2.1inthecontextofarepeatedroom-clearingscenario. Inter-roleConictishandledaspartoftherolelifecyclemechanismintheROLEMANAGER.Conictmaybehandledinanumberofways,asmentionedinSection4.2.1.3.However,thedemonstrationdescribedinSection5.2.2showsthehandlingofsimpleroleconictduringtheassignmentoftwoconictingrolesbyresolvingtheconictinfavoroftherolewithahigherposition. RoleDenitionisthecreationofnewrolesautomaticallythroughasocializationprocess.Thisnalrolemechanismisleftforfuturework.5.2DemonstrationsThissectiondescribesthedemonstrationsrequiredtosubstantiatetheapproachinChapterThree.Therearethreetypesofdemonstrations,correspondingdirectlytothreecontexts,thatwilljointlyvalidatetheobjectives.Notethattheuseofmorethanonecontextsupportsthemultiplesocialcontextscriteria.Thethreedemonstrationsareasfollows: 1. AdesigndemonstrationSection5.2.1oftheexpectedscenarioandrolesfortheUrbanOperationscontext.Thispresentsanopportunitytodesignasetofrolesandascenariotoexplorecomplexcoordinationissuesinamixedteamofhumansandrobots.Inaddition,thisscenarioshowshowlearningandadaptationmechanismsmaybeincorporatedintotheroledesign.Thedesign 54

PAGE 66

demonstrationistheprimarymotivationfortheuseofrolesinthiswork,andpresentsthemostcomprehensivedescriptionofhowrolesmaybeusedtocoordinatetheoperationofheterogeneousteamsofagents. 2. AsimulationdemonstrationSection5.2.2ofanUSVandUAValongwithahumanoperatorusinganOCUfortheLittoralWarfarecontext.Thisrequirestheuseofsimulationsincetheactualrobothardwareispresentlyunderdevelopment.Howevertheuseofsimulationallowscoreroledimensionsandmechanismstobeshown.Thisincludesroleposition,function,andexpectation,aswellastherolelifecycle,roletaking,handlinginter-roleconict,androlesharing. 3. AphysicaldemonstrationSection5.2.3ofrolesintheUncertainTerraincontexttoshowbasicrolefunctionalityonrealrobots.Thiscorefunctionalityincludesbasicrolefunction,roleexpectations,rolelifecycle,handlingbasicinter-roleconict,andmultiplesimultaneousroles.Thefollowingassumptionsaremadeonthebasisofavailablehardwareandtimeconstraints: Theminimum,criticalroledimensionsandmechanismswillbetestedonrealrobotstoafrmthebasicconcepts.TheATRV-JrrobotswillphysicallydemonstratebasicrolefunctionsfortheUncertainTerrainUTcontext. Thedemonstrationscanalsousesubstitutehardwareandasimplied,simulationsoftwareenvironment.TheLittoralWarfarecontextcallsforautonomyontheUSVandaUAV.However,thishardwarewillnotbecompleteintimefortestingofthiswork,andsimulationcanhelpshowamorecomplexinteractionbetweenagents. Someaspects,namelythelearningandadaptationportionsofthisworkcanbeleftforfutureresearch,butadesigndiscussioncanshowhowtheseaspectscanbeapplied.5.2.1RoleDesign:UrbanOperationsContextTheprimarymotivationforintroducingrolestorobotsystemsistoapproachtheproblemofcoordinatingheterogeneoushuman,robotandagentteams.Thiscoordinationisfundamentallyimportantforrobotsandhumanteammemberstoworkinsmallunitsonconstrainedtasks.Thesetasks,suchasclearingorsearchingaroom,requiredynamicteamworkforallmembersoftheseheterogeneousteams.Thelevelofinteractionrequiredfordynamicteamwork,particularlybetweenhumansandarticialagents,isadifcultproblemandonethatissuitedtoarole-basedsystem.Sincerolesbindactionandsocialexpectationsfor 55

PAGE 67

theactionswithinacontext,theunderstandingofrolescanhelpagentsreasonaboutthebehaviorofotheragentsandreactaccordingly.ThedesignwasperformedaspartofaprojectfundedbyDARPA.ThisdesignappliestheroleconcepttotheUrbanOperationssocialcontextandshowshowtherolesystemdescribedinChapterThreeandChapterFourcanbeusedtoprovidethislevelofinter-agentinteraction.Inparticular,thissectiondescribesasetofrolesthatwillallowarobotteamtooperateinthesamemannerasastandardUSArmyreteamandcansupportsuchdynamicteamwork.Thisdesignmakesanumberofassumptionsthatmustbemetbeforeitcanbeimplemented.Theseassumptionsareasfollows: Therstassumptionisthateachteammembercanrecognizeandinterprettheactionsoftheothermembers.Forpurelyhumanorpurelyarticialteams,thisisatractableproblem.Humanteamsalreadydothis,andarticialteamscanusedirectcommunicationorsensorprocessingtoidentifyteammates.Andsincetherobotsareboundbytheirassignedrolesandonlyhavealimitedabilitytodeviatefromtheirassignedroles,humansshouldbeabletopredicttherobotactions.However,automatedrecognitionofhumantargetsandrecognitionoftheactionsofhumansismoredifcult.Thisabilityisassumedinthefollowingdesign.Thisassumptionisreasonablesincearststeptoalleviatethisproblemcouldbeforeachhumantocarryasmall,tactical,wearablecomputerthatcouldprovidenecessaryroleandstateinformationtotherestoftheteam. Thesecondassumptionisthatthehardwaretosupportthisprojectinarealeldenvironmentexists.Unfortunately,robothardwareplatformsthatcurrentlyexistarelimitedinsize,payload,durabilityandmobilityinwaysthatahumanisnot.Thisdesignassumesthatthereisarobotdesignthatcouldperformalloftherequiredactions,however,inpracticetherolesthatarobotmightactuallyplayintheteamarelimited.Theremainderofthissectionisorganizedasfollows.Section5.2.1.1describesthegeneralscenarioofinterestinmoredetail.Section5.2.1.2andSection5.2.1.3describetherolesthathavebeenidentiedandoneexamplescenariothatusestheserolesdrawnfromArmytrainingmanuals.Finally,Section5.2.1.5closesthesectionanddescribeshowthescenariorelatestotherolerequirementsidentiedabove.5.2.1.1ScenarioThegeneralscenariofortheUrbanOperationscontextisfocusedaroundcombatoperationsinanurbanenvironment.Thisincludesmissionssuchasateamofrobotandhumansoldiers 56

PAGE 68

clearingabuildingofhostileenemyforces,suchasthosedescribedinDepartmentoftheArmy2003.However,thisdiscussionconsiderstheuseofrobotsasasubstituteforallorpartoftheteam.Theroom-clearingscenarioisonlyoneofmanypossibilities,however,itservesasabaselinemissionfortheuseofrolesforcoordinatingrobotteamsinurbanenvironments.AstandardUSArmyfour-manreteamconsistsoffourteammembersinfourroles:theRIFLEMAN,TEAMLEADER,GRENADIER,andAUTOMATICRIFLEMANDepartmentoftheArmy1992;DepartmentoftheArmy2004.Thisteamsizeischosenduetosizeandsafetyconstraintsaslargerteamsbecomeunwieldyinindoorenvironments.Allsoldiershavestandardtraining,butareassignedtodifferentrolesintheteam.Thisroleassignmentconstrainsandinuenceshowtheyactindifferentsituations.Clearingaroominabuildingisonecommontaskthatmightbeassignedtosuchareteam.Thistaskisassignedwhenthesituationrequiresroom-by-roomclearingofarelativelyintactbuildinginwhichbothenemiesandnon-combatantsmaybelocated.Thereisincreasedrisktohumansoldiersinthisscenario,andthisriskprovidesmotivationforreplacingallorpartoftheteamwitharticialsystems.Therearethreeprinciplesthatguidethistypeofmission:surprise,speedandcontrolledaction.Operatingproceduresandtrainingaredesignedtomaximizetheseprinciples.Theoperatingproceduresareproceduralknowledgethatdictatescertainactionsforeachteammember.Inreturn,though,thisproceduralknowledgeallowseachteammembertopredicttheactionsoftheothers.Thetrick,then,forintroducingarticialagentsintothisscenarioistocorrectlydenethefunctionsoftherolessothisproceduralknowledgeisnotlost.5.2.1.2RolesIdentiedThissectionbrieydescribesthekeyrolesofthisscenario.TherstsixrolesROBOTSOLDIER,SQUADLEADER,RIFLEMAN,TEAMLEADER,GRENADIER,andAUTOMATICRIFLEMANareinuencedbythebuildingclearingscenariodescribedinSection5.2.1.1andoutlinedinDepartmentoftheArmy2003.Thediscussionoftherolesbelowareintendedtoaddresstheroom-clearingscenarioabove.Theserolesmayalsobeappropriateforothermissions,buttherolefunctionsandexpectationswouldnecessarilybedifferentduringthosesituations.Forpracticalconsiderationsoutsidethescopeofthespecicmission,additionalroleswillberequired.Theseroles,particularlytheOPERATOR,TELEFACTOR,RELAY,andSICKrolesareneededinhazardoussituationse.g.SICK,orwillbeneededduetotheuseofrobotsystemsthatwillneedhumanoversighte.g.TELEFACTORandOPERATOR.AdescriptionoftheserolesmaybefoundinSection4.3.TheSQUADLEADERisresponsibleforhigh-levelcoordinationanddirectionofteamsofagents.Theseteams,calledreteamsarestandardizedtoconsistoffourteammembersinfourroles: 57

PAGE 69

RIFLEMAN,TEAMLEADER,GRENADIER,andAUTOMATICRIFLEMAN.ThefunctionsoftheSQUADLEADERareasfollows: TheSquadLeaderprovidesmissiongoalstotheteamsandmayassignteamstocompletesquadgoals. TheSQUADLEADERallocatesandassignsrolestoteammembers.TodothistheSQUADLEADERmustbeawareofthecapabilitiesandstatusofteammembers.Thisinformationisupdatedperiodicallyastheteamscompleteportionsoftheassignedtasksandreporttheirstatus. TheSQUADLEADERmayassignotherreteamstosupportingpositions. Otheragentsupthechainofcommandmustbenotiedofthesuccessorfailureoftheoverallgoals.Thediscussionoftheseotheragentsandrolesisbeyondthescopeofthisscenario. TheSQUADLEADERalsoprovidestheteammemberswithsetsofexpectations.Inthiscase,theexpectationsrepresenttheRulesofEngagementthatmodifytheactionsoftheteam.AllfourteammembersandtheSQUADLEADERshareabasefunctionalitydenedbytheROBOTSOLDIERrole.Thisrolemaybeconsideredanabstractsuper-rolefromwhichotherrolesarespecialized.ForarticialROBOTSOLDIERs,thefunctionsofthisrolearethefollowing: Proceduresforhowtosignal,recognizeandcommunicatewithpeersoldiers.IfthepeerisanotherROBOTSOLDIER,thantheseprocedurescanusesoftware-basedmechanisms,butifthepeerteammemberisahuman,thantherobotmustbeabletorecognize,reactandcommunicateinreal-time.ThesefunctionsaremodiedbyacoresetofdomaindependentexpectationstheRulesofEngagement.Theseexpectationsprovideacoreunderstandingofthesituation,andcanallowindividualrobotstopredicttheactionsofpeersbasedontheirunderstandingofthesituation.Theseexpectationsmayincludethefollowing: Denednoreareaswhereweaponsmaynotbeused. Prohibittheuseofweaponsunlesscertainconditionshold.Forexample,ahumanOPERATORmayberequiredtoauthorizetheuseofforce.TheTEAMLEADERisapointofcommunicationbetweentheSQUADLEADERandtherestoftheteam.TheTEAMLEADERisthesecondagenttoentertheroomduringaroom-clearingmissionandhasthefollowingfunctionsinadditiontothoseoftheROBOTSOLDIER: 58

PAGE 70

TheTEAMLEADERmustmonitorthestateoftheremainderoftheteam.ThisstateistransmittedtotheSQUADLEADERwhenthemissiongoalisachieved.TheRIFLEMAN,GRENADIER,andAUTOMATICRIFLEMANaretherst,thirdandfourthagentstoenteraroomduringaroom-clearingmission.AllthreehavesimilarrolefunctionsinadditiontothefunctionscommontoallROBOTSOLDIERs: Eachteammemberisabletousethepayloadwithwhichtheyareequipped.Humanteammemberstypicallycarryweaponsofvarioustypes,howeverthismayormaynotbeappropriateforarticialteammembersduetorulesofengagementandsafetyconsiderations.Articialsystemsmayinsteadcarrysensors,additionalarmortodrawenemyattention,orsomeotherpayload. AllROBOTSOLDIERsmustreporttheiroperationalstatuswhenrequested.Thisincludesthehealthofthesystem,payloadstatusandotherrelevantcharacteristics.TheTEAMLEADER,RIFLEMAN,GRENADIER,andAUTOMATICRIFLEMANalsohavethefollowingfunctionsthatspecicallysupportthisscenario: Theteammembersmustbeabletoapproachtheentrypointoftheroom.ThiswillusetheDRONEroledenedinSection4.3.TheDRONEroleinheritsexpectations,suchastherulesofengagement,fromthehigher-levelrole. Theteammembersmustsynchronizeactivitiespriortomajoreventstoensureallagentscanproperlypredicttheactionsofothers.Forexample,allrolesmustbesynchronizedbeforetherstagententerstheroom. Eachroleisabletofollowthestandardroom-entryscript,modiedwithrespecttotheroleoftheagent.Thisscriptdelineateshowtheagentineachroleenterstheroom,withrespecttothepreviousagent. Onceallteammembershaveenteredtheroom,theTEAMLEADERcandirecttheothermemberstodifferentpositionsasneededtocovertheroom.5.2.1.3ExampleScenarioTimelineThefollowingscenario,showninFigure17andadaptedfromDepartmentoftheArmy2003,isaconcreteexampleofhowtheaboverolesinteractinabuilding-androom-clearingscenario: 59

PAGE 71

a b c d eFigure17:Exampleofrolesinaroom-clearingscenario,withtheRIFLEMANR,TEAMLEADERTL,GRENADIERG,andAUTOMATICRIFLEMANARenteringandclearingaroom.ashowstheagentsoutsidetheroom.bthrougheshowtheagentsenteringtheroominthespeciedorder.ConesshowtheareaofcoverageofindividualagentsAdaptedfromDepartmentoftheArmy2003. 60

PAGE 72

1. TheSQUADLEADERassignsagentstoanassaultteamandassignstherolesfortheteam.Thisassignmentisbasedonthestateoftheavailableagentsandtheneedsofthemission.TheSQUADLEADERalsoprovidesthegoalsfortheteamandotherexpectationsforthemission. 2. Theteammembersproceedtothedesignatedentrypoint.ThisoperationusestherolespeciedforeachagentRIFLEMAN,TEAMLEADER,GRENADIER,orAUTOMATICRIFLEMAN,theROBOTSOLDIERroletohandlebasiccontextfunctions.TheagentswillalsoneedtheDRONEroletoenablebasicmobility.Figure17ashowsthefourreteammembersstationedoutsidearoom.Theinterioroftheroomisunknown. 3. Theentrywayisbreachedifnecessary.Anumberofmethodsmaybeuseddependingonthetypeofdoortotheroom.TheassessmentismadebytheTEAMLEADERandthetaskassignedtotheappropriateteammember.Inhumanteams,theTEAMLEADERplacesandemolitionchargeforaexplosivebreachandtheAUTOMATICRIFLEMANuseshisweapontodestroythedoorinaballisticbreach. 4. Theteammovesthroughtheentrypointinamannerdesignatedbyprotocol.First,theRIFLEMANenterstheroomandmovesalongthepathofleastresistancetothecorneroftheroomFigure17b. 5. ThentheTEAMLEADERenterstheroomandmovesintheoppositedirectionFigure17c. 6. TheGRENADIERfollows,againalternatingdirectionandstopsbetweentheRIFLEMANandtheentryFigure17d. 7. FinallyAUTOMATICRIFLEMANentersasindicatedinFigure17e.Eachagentmustbeawareofthemotionofthepreviousagentinordertomoveinthecorrectmanner.Whilethegureshowsthescenarioindiscretesteps,inrealitysuccessiveteammembersentertheroomimmediatelyfollowingtheprecedingmemberandtheagentsmustreactuidlytounknownorchangingcircumstances. 8. Allagentscooperatetocleartheroom.Humanagentsdothisthroughnon-verbalcommunication.Robotagentshaveanadvantageinthisregardastheymaycommunicatesilentlythroughwirelesscommunication. 9. Oncetheroomisclear,theTEAMLEADERsignalstheSQUADLEADERthattheroomisclear.TheprocessrepeatsuntiltheSQUADLEADERhascompleteditsmissiongoals. 61

PAGE 73

5.2.1.4ImplementationConsiderationsAsnotedabove,thisdesignwasperformedaspartofaprojectfundedbyDARPA.Asfollow-onworkwillusethisdesigntoenablethedesiredhuman-robotcoordinatedactivity,itisimportanttooutlinehowthisdesigncanbeimplemented.TherobotsoftwarewilluseDFRAandbuildtherolesdescribedabovewithintheframeworkprovidedintheCONTEXTADAPTER.TheCONTEXTADAPTERandROLEMANAGERaregenericcomponentsandenableintegrationwithotherDFRAservicesandimplementthecorerolemechanisms,respectively.Context-specicmodicationsareprimarilymadeinaCONTEXTPLUGIN,andtheremainderofthissectiondiscussesthisinmoredetail,alongwithadditionsthatmayberequiredtosupportthecontextexpectationsmorefully.ACONTEXTPLUGIN,describedinSection4.2.2,providesthefundamentalinterfacebetweentheCONTEXTADAPTERandexternalagentsinthesocialcontext.ForthisprojectthekeyexternalagentisacognitiveagentdevelopedbyCHISystems.Forthisscenario,thecognitiveagentiscapableofperformingtheSQUADLEADERrole,andusestheCONTEXTPLUGINtocommunicatewithindividualrobots.ThiscommunicationdoesnotusestandardDFRAmessaging,butinsteadusesJAUSmessages.TheCONTEXTPLUGINmusttranslatetheseJAUSmessagesandroutethemtotheCONTEXTADAPTER,ROLEMANAGER,EXPECTATIONMANAGER,individualroleorotherserviceasrequired.TheCONTEXTPLUGINallowsthecognitiveagenttobuildandmaintainsituationawarenessbyprovidingitwithsystemandenvironmentstatefromtherobotandonboardsensors.TheroledescriptionforeachrequiredrolebuildsoffoftherolediscussionintheaboveSection5.2.1.2andinthepreviouschapterSection4.3.Inparticular,therolepositionisdenedbystandardmilitaryhierarchy,althoughthisisnotatotalordering.Forinstance,duringthelimitedmaneuverofaballisticbreach,theAUTOMATICRIFLEMANmaybeabletogivealimitedsetoforderstotheTEAMLEADERe.g.Standclear!,butinallothersituationsthepositionisreversed.TherolefunctionswilluseSCRIPTsandtheSCRIPTMANAGER.AsnotedinSection4.2.5,mostcurrentscriptsareinstrumentalinnature.Theactionsoftheteammembersinthisscenarioareheavilyinterdependentandwillrequireasituationalscripttocorrectlyfullltherolefunctions.Considerthefollowingdescription,andnotethatthescenarioiswell-knownandspecied,thereareinterdependentactions,andasharedmodelofgoalsandactions: Thedirectioneachmanmovesinshouldnotbepre-plannedunlesstheexactroomlayoutisknown.Eachmanshouldgoinadirectionoppositethemaninfrontofhim[Thereareinterdependentactions].Everyteammembermustknowthesectorsanddutiesofeachposition[Thereisasharedmodel]. 62

PAGE 74

Astherstmangoesthroughtheentrypoint,hecanusuallyseeintothefarcorneroftheroom.Heeliminatesanyimmediatethreatandcontinuestomovealongthewallifpossibleandtotherstcorner,whereheassumesapositionofdominationfacingintotheroom.Thesecondmanteamleader,enteringalmostsimultaneouslywiththerst,movesintheoppositedirection,followingthewallandstayingoutofthecenter.Thesecondmanmustcleartheentrypoint,cleartheimmediatethreatarea,clearhiscorner,andmovetoadominatingpositiononhissideoftheroom.Thethirdmangrenadiersimplygoesoppositeofthesecondmaninsidetheroomatleastonemeterfromtheentrypointandmovestoapositionthatdominateshissector.ThefourthmanSAWgunnermovesoppositeofthethirdmanandmovestoapositionthatdominateshissector.[Theentirescenarioforeachteammemberisspecied,althoughvariationbasedonconditionsisallowed.] AdaptedfromDepartmentoftheArmy2003Thesesituationalscriptsalsoimpactthesocialinteractioncomponentofeachrole.Thescriptsdenehoweachteammemberissupposedtointeract,butreal-worldeventsmayinterfereandchangetheactualinteractions.Forcertaintypesofevents,suchastemporarycommunicationfailuresorenvironmentalsituationsthatconstrainorrequireaction,thisdifferencemustbeaccommodatedthroughdynamic,reactiveadaptation.Forexample,duringroomentry,ateammembermayneedtoalteritsroutetoavoidobstacles,triggeringdelaysintheentryofsuccessiveteammembers.Asone-timeevents,thesecanbeprocessedandforgotten.However,abuilding-clearingmissionisreallyasequenceofroom-clearingscenariosandthereisapotentialforlonger-termlearning,particularlytoadaptthegeneralscriptwithexpectationsoftheactionsofotherteammembers.Forexample,TheGRENADIERwaslategettingintopositioninthelasttworooms,sodon'tworryaboutalatecompletionnoticationinthenextroom.5.2.1.5DiscussionThescenarioisstandardprocedureforhumanteams,andservesasabaselinemeasureofhowhigh-levelsocialcoordinationmayoccurinteamsofarticialagents.Becausetherolesaremappedinadirectcorrelationwithhumanrolesinthesamescenario.Itshouldbepossibletointerchangehumanandarticialteammembersinamannerthatrequireslessadjustmentthandevelopingahuman-robotteamfromscratch. 63

PAGE 75

Theroom-clearingscenario,throughitsrepetitivenature,offersanopportunitytoimplementrolemaking.Attheendofeachroomcleared,eachagentreportsitsstatetotheTEAMLEADERandSQUADLEADER.Inparticular,eachagentisre-evaluatedonitsabilitytoplaycertainrolesbasedonpastperformanceinthatroleanditsabilitytoplaytherolewhenrolesarereassignedandnewgoalsforanewroomareassigned.Thissuggeststwocharacteristicsthatareimportantforrolemaking:longevityofsystemsandrepetitionofroles.Faulttoleranceandadaptabilitywillbekeyfortherolesinthiscontext,particularlybecausetheenvironmentiscomplexandchaotic.Thechanceofamemberoftheteamsufferingdamageishigherduetoclosequartersindoorsandthepossibleadditionofsomeformofweaponrytotherobots.ThisimpliesthattheSICKrolewillbeparticularlynecessary,andtherolesoftheotherinvolvedagentsmustbeabletoaccountfordamagedrobots.Forexample,thefunctionsoftherolesonateammaychangemoresignicantlyiftheTEAMLEADERisdamagedthaniftheGRENADIERisdamaged,howeverineithercasetheotherteammembersmustadapttothenewconguration.Section5.2.3andSection5.2.2haveshownthatrolesandroledimensionsandmechanismscancoordinatetheoperationsofheterogeneousteams.Thissectionextendsthoseresultswithadiscussionofhowrolesmaybeappliedtoanewdomain,UrbanOperations.Thissectionthenpresentedaplausiblescenariobasedoncurrentoperatingproceduresformilitaryunitsclearingabuildingandshowshowrobotmaybeusedforthesametask.Finally,whiletheimplementationofrolelearningandadaptationisbeyondthescopeofthiswork,thisdesigndemonstrationshowshowthiscanbeincorporatedintoaworkingrolesystemthroughtheuseofanalrolemechanism,rolemaking.5.2.2SimulationDemonstration:LittoralWarfareContextTheprevioussectiondescribedanidealusageofrolesforcoordinatingheterogeneousteams.However,itcouldnotbeimplementedduetotimeconstraintsandtheotherassumptionslistedintheprevioussection.However,itisnecessarytotesttherole-basedsystemanddemonstratethatitisabletosuccessfullycoordinateaheterogeneousteam.Thissectiondescribesthesecond,simulated,demonstrationofthiswork.ThesimulationwasperformedaspartofaprojectfundedbytheOfceofNavalResearch.ThissimulationappliestheroleconcepttotheLittoralWarfaresocialcontextandhastheobjectiveofvalidatingkeyroledimensionsandmechanisms.ThescenariorequiresanUSV,anUAVandahuman 64

PAGE 76

operatorusinganMCOCU,andsatisesthefollowingdemonstrationobjectives:roleposition,function,expectation,rolelifecycle,roletaking,handlinginter-roleconict,androlesharing. Figure18:ScenariooverviewforthesimulationdemonstrationintheLWcontext.TheUSVbeginsatAcarryingtheUAV.TheUSVproceedstoB,wheretheUAVislaunched.TheUSVcontinuestoCandthenbeginsscoutingtowardsF,butencountersananomalyatD.TheUSVthensharesthescoutingresponsibilitieswiththehumanoperatoruntilE,whentheUSVisreturnedtofullautonomy.FcompletesthescoutingmissionandtheUSVreturnstobaseatG.SatelliteimagecourtesyofGoogleEarth. 5.2.2.1ScenarioTheLWcontextutilizesacombinationofunmannedsurface,aerial,andgroundvehiclesoperatingtosecureaportfacility.Thesurfaceandaerialvehiclescoordinatetodetectminesandintrudersalongapier,whilethegroundvehicles,aerialvehiclesandhumanscooperatetosecureadockarea.Inadditiontothephysicalagents,humanoperatorsinacontrolstationinteractwiththerobotsanddirectoveralloperationsfromacommandtent.ThesimulationusestheYetAnotherRobotSimulatorYARSsimulationenvironmentCraigheadb,adomain-specicsimulationbuiltusingtheSARGEgamelearningenvironmentCraigheada.ThissimulatorisbuildusingtheUnity3Dgameengine,whichincludestheAgeiaphysXphysicsengine.Thisallowsaphysics-based3Dmodeloftheenvironmenttobebuilt,androbotssimulatedinthisenvironmentcaninteractintheworldwithahigherdegreeofdelitythanmostrobotsimulators.YARStsthecriteriaforaclassCrobotsimulatorCraigheadetal.2007. 65

PAGE 77

a bFigure19:Thedifferencebetweenthesoftwarethatinteractswithrealhardwareandsimulation.ashowsthecaseofDFRAsoftwareonarealrobot.bshowsthecaseofDFRAsoftwareinteractingwiththesimulation.Thekeydifferenceisthelow-levelsoftware;thedriversforsensors,effectorsandschemasaredifferentbetweenrealityandsimulation,higher-levelfunctionalitycanoperateunchanged. 66

PAGE 78

Thesimulationusedforthisscenarioisprimarilyfocusedatahigh,cognitivelevel,andthusmostphysics-leveldetailscanbeabstractedout.Forexample,theUAVdoesnotneedtoyinacompletelyrealisticmannersincetherelevantaspectofthesimulationisthatitisassignedaparticularrole,notthatitoccupiesaparticularlocationinspace.Figure19showsthedifferencesbetweenoperatinginthissimulationenvironmentandonarealrobot.TheCONTEXTADAPTERisahigh-leveldeliberativeserviceanddoesnotdirectlyuseanyreactivecomponents.Thesimulatorwasconstructedtomatchtherealworldascloselyasfeasible.Thefollowingvepointsinuencedthissimulation: Thesimulatedlocationwasconstructedusingsatellitemapsandisananaloguetotheactualphysicallocation. TheUAVandUSVhavebeenmodeledtothescaleandproportionsofthecompletedrobothardware. Thesensorsprovidedatathatissimilartothatofrealhardwareinthesamelocation.Forexample,thelaserrangenderprovidesrangereadingsthatwouldbesimilartorealrangenderreadings.Thisallowsmuchoftherobotsoftwarerunningthesimulationtobeunchangedfromrealrobottosimulation.Figure19showsthedifferencebetweenthesoftwareasitexecutesonrealhardwareandthesoftwarethatinteractswiththesimulation. Partofthephysicsmodelisonlylooselybasedonreality.Forexample,theeffectsofthewater,wavesandtideareonlyapproximate,andarealUSVwouldlikelymoveinadifferentmanner.ThiswouldbeimportantifthegoalofthisworkweretobuildacorrectUSVcontrolmodel,howeverthehigh-levelscriptsandrolesareinterestedinwhethertheUSVcanreachitsgoals,notthatitdoessointhemostoptimalmanner. ThedetectingofanomaliesandtheightoftheUAVarecompletelysimulated.TheUAVsimplyiesabovetheUSVandneitherprovidesanydatanorprocessesanyuserinput.Likewise,thesimulationsimplysetsaagiftheUSViswithinathresholddistancesetto8metersofanexplosivedevice.Thisisanabstractionofaperceptualschemaprocessingvideoimagesthatisnotrelevanttothiswork.SuchaschemahasbeendemonstratedinLongetal.2005.ThesimulatedscenariousesasingleUSV,asingleUAV,andahumanoperatorataMCOCU.TheUSVrunsthefollowingsetoftenservicesdenedthroughDFRA: 67

PAGE 79

ACARTOGRAPHERserviceholdsanapriorimapoftheenvironment,andcanusethismapinconjunctionwithaTrulla-basedpathplannertogeneratearoutetoagivenlocation. AMOVETOWAYPOINTbehaviorreceivesawaypointandmovestothiswaypointwhileavoidingobstacles. ADRIVEEFFECTORservicethatusestheUSVdrivethrusterstomovethevehicle. Acomplementofsensors:HEADINGSENSOR,GPSSENSOR,IMUSENSOR,LASERRANGESENSOR,andANOMALYSENSOR.Thesesensorsprovidedatatohigher-levelbehaviorsandscripts. TheSCRIPTMANAGERisresponsibleforruntimemanagementofscripts,includingscriptlifecycleandeventmanagement. TheCONTEXTADAPTER,whichisresponsibleforproperrolemanagementaccordingtotheapproachinChapterThree,andisdescribedinChapterFour.TheCONTEXTPLUGINfortheLWcontexttheCONTEXTADAPTERdenesverolesfortheUSV:DRONE,COURIER,SCOUT,TELEFACTORandSICK.TheUAVandMCOCUaresimpler.TheUAVdenestwoservices,asdoestheMCOCU.TheUAVdenesaSCRIPTMANAGERandCONTEXTADAPTER.TheUAVisentirelycontrolledbythesimulationenvironment,sothesetwoservicesactasaplaceholderorproxyforthevehicle.SincethehardwareUAVhasnoonboardprocessing,thisabstractionisactuallynodifferentthanthesystemsetupthatwillberunforthephysicaldemonstrationinJuneof2007.TheCONTEXTPLUGINfortheLWcontextontheUAV'sCONTEXTADAPTERdenestworoles:RELAYandSICK.TheMCOCUrunstheoperatorcontrolGUIandrunsthesamecomplementofDFRAservicesthattheUAVdoes:aSCRIPTMANAGERandCONTEXTADAPTER.TheCONTEXTPLUGINfortheLWcontextontheMCOCU'sCONTEXTADAPTERdenesfourroles:OPERATOR,COMMANDER,SCOUTandSICK.TheMCOCUintheCOMMANDERroleistheprimarycontrolsourceforthemission.TheCOMMANDERisresponsibleforcoordinatingtherolesoftheotheragents.Figure20showsthehardwaresetupforthisdemonstration.ThesoftwarefortheSARGEsimulation,MCOCU,USVandUAVeachranonaseparatecomputerconnectedviaanetwork.Services,particularlytheCONTEXTADAPTERs,onMCOCU,USVandUAVwereindirectcommunication,butonlytheUSVwasdirectlyconnectedtothesimulation.Thischoiceofhardwarecongurationwasmade 68

PAGE 80

Figure20:Thegureshowsthehardwarecongurationforthesimulationscenario.Fourmachineswereusedduringthesimulation.AdedicatedPCwasusedforeachoftheUAV,USVandMCOCUtomimictheexpectedcongurationduringfuturework.Thesimulationitselfranonalaptopcomputer.OnlytheUSVcommunicateddirectlywiththesimulation. tofacilitatefuturework,wheresoftwareforeachrobotwillexecuteontheonboardcomputerandtheMCOCUwillberunfromacommandstation.5.2.2.2RolesUsedTherearesixrolesusedinthisscenariobytheUSV,UAV,andMCOCU:DRONE,SCOUT,RELAY,COURIERTELEFACTOR,OPERATOR,andCOMMANDER.ThecorefunctionsoftheserolesaredescribedinSection4.3,howevertheLWcontextexpectationsareasfollows.TheUSVhasthelargestsetofavailableroles.TheDRONEfunctionsrevolvearoundsimplemovementtodenedwaypoints.TheexpectationsfortheLWcontextarethefollowing:theDRONEshouldusethelocalCARTOGRAPHERservicetoplanitsroutetoeachgoal;itcanmoveasfastasisneeded.IntheSCOUTroletherobotissubjecttothefollowingexpectations:theSCOUTmusthaltandsendanoticationifananomalyisdetected;itmustusehighcautionwhenmoving;itshouldusetheCARTOGRAPHERforrouteplanning.AsaCOURIERtherobotmustobeythefollowingexpectations:theCOURIERshouldreleaseitspayloadwhenthenalwaypointisreached;itmustmovewithnormalcaution;itshouldusetheCARTOGRAPHERforrouteplanning.OntheUSV,neithertheTELEFACTORnortheSICKrolesaresubjecttoadditionalconstraints.TherolesfortheMCOCUaresubjecttothefollowingexpectations.AsaCOMMANDERtheMCOCUisresponsiblefortheoverallmission.Forthesimulationscenario,theexpectationsconsistofasetof 69

PAGE 81

waypointsthatmustbepassedtotheUSVasaDRONE,SCOUTorCOURIER.AsanOPERATOR,theMCOCUissubjecttothefollowingexpectation:theOPERATORshouldinteractwiththerobotinasharedmodethatallowstheOPERATORcontroloftherobotbutdoesnotsubsumelow-levelobstacleavoidance.NeithertheSCOUTrolenortheSICKroleuseanyadditionalexpectations.Finally,asbothoftherolesavailabletotheUAVaremarkerrolesandarenotassociatedwithanyparticularactions,neithertheRELAYrolenortheSICKrolehaveanyadditionalassociatedexpectations.5.2.2.3ScenarioTimelineFigure21showsatimelineoftherolechangesthroughoutthisscenario,extractedfromtheruntimeloggingofeventsbytheservicesontheMCOCU,USV,andUAV.Theremainderofthissectiondiscussesthistimelineindetail,andFigure5.2.2.4throughFigure35showrelevantimagesfromtheSARGEsimulationandrolevisualizationsfromtheMCOCUGUI.Anoteabouttherole-visualizationimagesisinorder.Eachdiagramshowstherolestateofeachagent.Aconcentriccircleisdrawnaroundeachagentwhenaroleiseitherassignedoractive.Assignedrolesaredrawninorange,whileactiverolesaredisplayedingreen.Whenmultiplerolesareassignedtoanagent,thehigherpositionroleisdrawnoutsidealowerpositionrole.Ifthepositionisequal,thentheorderofringsisarbitrary.Thegreyarcsshownintheguresshowinuence.Theagentorroleattheendofthearclabeledwithacircleisresponsiblefortheroleactionattheunadornedend.Certainactionsintheguresmayappearslightlydifferentthanexpectedforthreereasons.First,duetothedistributednatureofthescenario,noticationsfromdifferentagentmayappearoutoforder.Second,therecursivenatureofsomeroleoperationsdisplaysthesameeffect.Third,ifnoticationsofmultipleeventsarereceivedinclosesuccession,theGUIupdatewillshowthemostrecentvalidstateandwilldiscardintermediateevents.ThehumanoperatorisabletointeractwithboththeUSVandUAVtomonitor,assignandotherwiseinteractwiththerolesonbothsimulatedrobots.Allthreeagentsinitiallystartwithnoroleassignmentsandarenotassociatedwithanysocialcontext.AstheCONTEXTADAPTERoneachagentinitializes,eachagentjoinstheLWcontext.FortheMCOCU,thishastheeffectofautomaticallyassigningtheCOMMANDERroletotheMCOCU.Figure23ashowstheroleassignmentsatthestartofthescenario.Atthisstage,theUAVbeginswithnoroleassignment,theUSVbeginswithnoroleassignment,andtheMCbeginsassignedtotheCOMMANDERrole.ThemissionbeginswhentheCOMMANDERroleactivatesontheMCOCU.TheCONTEXTPLUGINfortheLWcontextisconguredtodothisthroughtworoutes.TherstisDFRA-based,wherearequestforactivationissentthroughtotheCONTEXTADAPTER.ThesecondisarequestfromtheexternalsocialcontextthroughtheCONTEXTPLUGIN.Forthisscenario,theactivationoftheCOMMANDERroleistriggeredbyhumanoperatorthroughtheOCUGUI 70

PAGE 82

Figure23b.ThisstageofthedemonstrationshowstherolelifecycleastheCOMMANDERchangesrolestate.TheactivationoftheCOMMANDERstartstherole'sprimaryfunctionofmanagingtheactivitiesoftheotheragentsinthescenario.TheCOMMANDERrstassignsandactivatestheCOURIERroleontheUSVFigure23candFigure23d.AstheCOURIERroleisassigned,itassignstheRELAYroletotheUSV.Followingthisassignment,theCOURIERactivates.ThisisanexampleoftherecursivenatureoftheseinteractionscausingunexpectedeffectsintheGUI.TheassignmentoftheCOURIERroleontheUSVcausestheassignmentoftheRELAYroleontheUAV.However,theassignmentoftheRELAYmustcompletebeforetheassignmentoftheCOURIERisnished.InFigure23ctheGUIhasupdatedtoshowtheRELAYassignmentbeforethenoticationoftheCOURIERhasbeenreceived.ThenextupdateFigure23dhasreceivedtheassignmentandactivationnoticationsfortheCOURIERsimultaneouslyandonlyshowsthenalvalidstate.Thisstageofthedemonstrationshowstherolelifecycleandtheeffectsofrolefunction.TheUSVthenproceedsfollowingthefunctionsoftheCOURIERrole,transportingtheUAVtoalocationdesignatedbytheMC.Atthisstage,theUSVisintheCOURIERactive,thehumanisintheroleofCOMMANDERactive,andtheUAVisintheRELAYassignedrole.Theactiverolesfollowtheirdenedfunctionandaresubjecttotheroleexpectations.WhentheUSVreachesthenalwaypointgoaloftheCOURIERrole,itobeystheexpectationthatitlaunchtheUAV,andimmediatelyactivatestheRELAYroleontheUAVanddeactivatestheCOURIERroleFigure25a.TheCOMMANDERisnotiedoftherolestatechanges,andhastheUSVrelinquishtheCOURIERroleFigure25b.TheCOMMANDERthenassignsandactivatestheDRONEroleontheUSVFigure25candsetsexpectationsdenotingthedesiredwaypointgoals.Figure24showsbothrobotsthroughtheSARGEsimulation.Thisstagecontinuestoshowtherolelifecycle,functions,andexpectationsmodifyingrolebehavior.InFigure26,theUSVhasreachedthenalwaypointgoaloftheDRONErole.TheCOMMANDERisnotiedoftherolestatechanges,andhastheUSVdeactivatetheDRONEroleFigure27a.Next,theCOMMANDERhastheUSVrelinquishtheDRONEroleFigure27bandassignsittotheSCOUTrole.Lastly,theCOMMANDERactivatestheSCOUTroleontheUSV,whichbeginstocautiouslyinvestigatetheseawallFigure27c.Onceagain,thisisanexampleofrolelifecycle,functions,andexpectationsmodifyingrolebehavior. 71

PAGE 83

Atthispointinthescenario,theUSVisintheroleofSCOUTactive,theroleoftheUAVisaRELAYactiveandtheMCOCUisintheroleofCOMMANDERactive.Asthescoutingmissioncontinues,ananomalyisdetected.Asperexpectation,theUSVSCOUTnotiestheCOMMANDERandMCOCUoftheevent.Inturn,theMCassignstheOPERATORroletoitself.Afterassignment,theMCactivatestheOPERATORrole.AspartoftheOPERATORroleassignment,theOPERATORassignstheTELEFACTORroletotheUSV.ThistriggersaconictbetweentheSCOUTandTELEFACTOR,whichisresolvedbasedonroleprioritybeforetheTELEFACTORroleassignmentcompletes.ThisresolutiondeactivatestheSCOUTroleontheUSVFigure29a.Thisstageofthescenario,showsmultiplerolesonanagent,rolefunction,rolelifecycle,theuseofroleexpectation,andhandlinginter-roleconictthroughroleposition.TheMCthenactivatesOPERATOR,whichtheninturnactivatestheTELEFACTORroleontheUSV.TheOPERATORassignsitselftheSCOUTrole,takingoverandsharingtherolewiththeUSV.ThisnallyassignsandactivatestheDRONEroleontheUSVtollthemobilityexpectationoftheMCSCOUTrole.TheOPERATORthenhascontroloftheUSVandmaytheninvestigatetheanomaly,makingadecisionastowhetheritisdangerousornotFigure29bandFigure28a.Thisstageofthescenario,showsmultiplerolesonanagent,rolefunction,rolelifecycle,theuseofroleexpectation,rolesharingwiththeMCperformingthecognitiveaspectsoftheSCOUTroleandtheUSVthephysicalextensionofthehuman.AftercompletingthediagnosisFigure28b,theOPERATOR,SCOUT,TELEFACTOR,andDRONEaredeactivatedandrelinquishedFigure29c.TheMCthenreactivatestheSCOUTroleontheUSVandthescoutingmissionresumesFigure29d.USVisnowintheroleoftheSCOUTactive,MCOCUisintheroleoftheCOMMANDERactive,andtheUAVisstillintheroleoftheRELAYactive.TheUSVcompletestheSCOUTrole.TheMCintheCOMMANDERrolereceivesthisnoticationanddeactivatestheSCOUTroleontheUSVFigure31a,Figure31b.Figure5.2.2.4showstheSARGEsimulationatthispoint.Next,theCOMMANDERassignsandactivatestheDRONEroleontheUSVwiththeexpectationthattheUSVwillreturntothebase,inthiscasetheinitialstartinglocationFigure33a.Finally,Figure5.2.2.4showsthesimulationastheUSVreachesthelastgoalwaypointintheDRONErole.ThemissionendsastheCOMMANDERdeactivatesandrelinquishestheRELAYroleontheUAVandtheDRONEroleontheUSVFigure35a.5.2.2.4SummaryFigure21showsatimelineofroleeventsconstructedfromloggingdatageneratedontheUSV,UAVandMCduringanexecutionofthescenario.ThistimelineandunderlyingdatavalidatesthattheimplementationoftheCONTEXTADAPTER,ROLEMANAGER,EXPECTATIONMANAGERand 72

PAGE 84

SCRIPTMANAGERoperateasdesignedforthisscenario.Thisentiresimulationscenariohasshownthefollowingrolemechanisms,dimensions,andothercriteria: Rolepositionwasshownwhenusedtoresolveinter-roleconictduringtheassignmentoftheTELEFACTORroletotheUSV. Rolefunctionandexpectationwereconsistentlyshownthroughthesuccessfulcompletionofthevariousroles.RoleexpectationmodiedthespeedoftheUSVandtriggeredthereleaseoftheUAVatthecompletionoftheCOURIERroleandtriggeredthenoticationoftheCOMMANDERregardingthedetectedanomalyintheSCOUTrole. Rolelifecyclewasconsistentlyexercisedtosupportallofthebaseroleactivities. TheuseofmultipleroleswasshowninseveralinstanceswiththeUSVsupportingtheDRONEandTELEFACTORroles,andtheMCOCUplayingtheCOMMANDER,OPERATOR,andSCOUTrolessimultaneously. Finally,rolesharingwasevidencedwhenthehumanandrobotsharedtheSCOUTrole,withthehumanprovidingtheanomalyidenticationandtherobotthephysicalaction.Thissectionhaspresentedthesimulationdemonstrationusedtoshowtheutilityoftherolesystem.Thisdemonstrationshowseightofthekeyroledimensionsandmechanisms:rolefunctions,roleexpectations,therolelifecyclethroughthechangingrolesonthevariousagentsofthecourseofthesimulation,roletakingsupportingrolesharingoftheSCOUTrolebetweentwoagents,handlinginter-roleconictusingtherolepositiontoresolveconict,andmultipleconcurrentroles.Thattherolefunctionsworkisshownbythesuccessfulcompletionofthedemonstrationscenario.Thatis,thebehaviorsandtasksassociatedwitheachoftheroleswereexecutedatthepropertimesandproducedthedesiredbehaviorontherobot.MultipleconcurrentrolesareshownbytheassignmentoftheTELEFACTORandDRONEorSCOUTrolesontheUGVduringthemission.Inaddition,theassignmentofmultiplerolesrequirestheuseoftheinter-roleconictmechanismtodeterminewhethertherolefunctionsconict. 73

PAGE 85

Figure21:Roletimelineforthesimulatedscenariohighlightingroletransitionsgleanedfromrobotlogles. 74

PAGE 86

Figure22:TheviewoftheUSVatthestartofthescenariothroughtheSARGEsimulator. 75

PAGE 87

a b c dFigure23:Thestartofthesimulationscenario.ashowstheroleassignmenttothethreeagentsatthestartofthesimulation;bshowstheactivationoftheCOMMANDERrole;cshowstheassignmentoftheRELAYrole;dshowstheactivationoftheCOURIERrole. 76

PAGE 88

Figure24:TheCOURIERtoDRONEtransitionwiththeassignmentoftheRELAYroleontheUAV,viewedthroughtheSARGEsimulation. 77

PAGE 89

a b c dFigure25:TheCOURIERtoDRONEtransitionwiththeassignmentoftheRELAYroletotheUAV.ashowstheRELAYroleassignmenttoUAV;bshowsthedeactivationandrelinquishingoftheCOURIERroleontheUSV;cshowstheassignmentoftheDRONEroleontheUSV;dshowstheactivationoftheDRONErole. 78

PAGE 90

Figure26:TheDRONEtoSCOUTroletransitionontheUSV,viewedthroughtheSARGEsimulation. 79

PAGE 91

a b cFigure27:TheDRONEtoSCOUTroletransitionontheUSV.ashowsthedeactivationoftheDRONEroleontheUSV;bshowstherelinquishingoftheDRONEroleontheUSV;cshowstheassignmentandactivationoftheSCOUTroleontheUSV. 80

PAGE 92

a bFigure28:Ananomalyhasbeendetectedduringthescoutingmission,shownintheSARGEsimulatorandoperatorGUI.ashowstheSARGEsimulatorview;bshowsanoperatorGUIelementpromptingforidentication. 81

PAGE 93

a b c dFigure29:Ananomalyhasbeendetectedduringthescoutingmission.ashowstheOPERATORroleassignmenttotheMCandTELEFACTORassignmenttotheUSV;bshowstheroleactivationoftheOPERATORandTELEFACTORroles,withcascadedactivationoftheDRONEontheUSVandSCOUTontheMC.ThisstartsthesharingoftheSCOUTrolebetweenthetwoagents;cshowsthedeactivationandrelinquishingoftheDRONEandTELEFACTORrolesontheUSVandtheSCOUTandOPERATORrolesontheMC,endingtherolesharing;dshowsthereactivationoftheSCOUTrole. 82

PAGE 94

Figure30:ThecompletionofSCOUTroleontheUSV,shownthroughtheSARGEsimulator. a bFigure31:ThecompletionofSCOUTroleontheUSV.ashowsthedeactivationoftheSCOUTroleontheUSV;bshowstherelinquishingoftheSCOUTroleontheUSV. 83

PAGE 95

Figure32:ThereassignmentofDRONEroleontheUSVtoreturntothestartingpositionandcompletethemission,shownthroughtheSARGEsimulator. aFigure33:ThereassignmentofDRONEroleontheUSVtoreturntothestartingpositionandcompletethemission.ashowstheassignmentandactivationoftheDRONEroleontheUSV. 84

PAGE 96

Figure34:TheDRONEUSVhasreturnedtothestartingpositionandcompletedthemission,shownthroughtheSARGEsimulator. aFigure35:TheDRONEUSVhasreturnedtothestartingpositionandcompletedthemission.ashowsthattheallrolesontheUSVandUAVhavebeenrelinquished. 85

PAGE 97

5.2.3PhysicalDemonstration:UncertainTerrainContextThenaldemonstrationisaphysicalexperimentwiththeobjectiveofvalidatingbasicroledimensionsandmechanismsonrealrobots.Thescenarioforthisdemonstrationissimple,onlyrequiringasinglerobotandhumanoperator.Asitissimple,itisanidealmethodtoshowtheeffectivenessofbasicroleelements.ThedemonstrationwasperformedaspartofaprojectfundedbytheArmyResearchInstitute.ThisphysicaldemonstrationappliestheroleconcepttotheUncertainTerrainsocialcontextandhastheobjectiveofdemonstratingcoreroledimensionsandmechanisms.ThescenariorequiresanUGVandahumanoperatorusinganMissionCommanderOCU,andsatisesthefollowingdemonstrationobjectives:function,expectation,rolelifecycle,multipleroles,andbasichandlingofinter-roleconict.Section5.2.3.1summarizedthescenariothatwillserveasaframeworkforthedemonstration,whileSection5.2.3.2liststherolesneededtocompletethescenario.Finally,Section5.2.3.4summarizesthesection. Figure36:ScenariooverviewforphysicaldemonstrationintheUncertainTerraincontext.SatelliteimagecourtesyofGoogleEarth. 5.2.3.1ScenarioTheUTcontextinvolvesarobotexploringanunknownenvironmentatthedirectionofhumanoperatorsoracognitivesoftwareagent.Therobotsprovidesensoryandterraininformationby 86

PAGE 98

usingsensingandmobilitycapabilities,andhavetheabilitytoplanlocalpathsbasedonhigh-leveldirectivesfromotheragents.Figure36showsahigh-leveldiagramofthegoalofthescenario.First,theUGVdrivestothestartofthemissionarea,fromlocationAtolocationBinthediagram.Thenitperformsitsmaintask,scoutingthetargetareaandgeneratingnewinformationaboutthephysicallocation.ThisisrepresentedbythepathfromlocationBtolocationCinthediagram.Finally,uponcompletionofthemission,theUGVreturnstobasealongthepathbetweenlocationCandlocationD.AllrolechangesontheUGVaredirectedbytheOCUnotshownonthediagram.TheOCUtracksandmonitorsallrolestatechangesanddirectstherobotthroughitstask.Section5.2.3.3presentsatracefromasamplerun,showingtherolechangesovertime.Theexternalsocialcontextconsistsofacognitivesoftwareagentwhichprovidesaccesstoterrainanalysisandplanningtoolsandprovidesdirectivestotherobot.Asthegoalofthisworkistoshowtheefcacyofrolesforthecoordinationofheterogeneousteams,thisdemonstrationwillfocusonthecoordinationofrolesbetweenasinglerobotandtherobotoperator.ThisscenarioutilizedasingleATRV-JrrobotandahumanoperatorstationedatanOCU.Therobothasfourwheelsthataredrivenbytwoindependentmotorsandsufcientgroundclearanceforoutdooroperationsintameenvironments.TherobothasaLinux-basedPCwitha1GHzPIIIprocessorand2GbofRandom-AccessMemoryRAM,runningRedHatLinux9.x.Theonboardcomputerincludes10auxiliaryserialports,rewireandUSBtoaccommodatesensorsandotherequipment.TherobotsarecapableoflocallyrunninganysoftwarethatwillrunonastandardPCandareaccessiblevia802.11bwirelessEthernet.Therobotareequippedwithasetoffoursensorsusedforthisdemonstration:aSICKLMS200scanningplanarlaseranddigitalcompassmountedonthefrontofthevehicle,aninertialgyroscopeandGarminGlobalPositioningSystemGPSmountedonthecenter-topequipmentrack.Thegyroscope,GPS,compass,andinternalodometryproviderobotlocalization,whilethelaserrangenderprovidesobstacledetection.TheOCUwasoperatedonaMacintoshMacBooklaptopcomputerwitha2GHzIntelCoreDuo2processorand1GbofRAM.Thelaptopranoperatorcontrolsoftwarethatallowedmonitoringtherolesofboththeoperatorandtherobot.ThelaptopwasalsoconnectedtothesamewirelessEthernetastherobot,allowingdirectcommunication. 87

PAGE 99

ThephysicalscenariousesasingleUGVandahumanoperatorataMCOCU.TheUGVrunsthefollowingsetofthirteenservicesdenedthroughDFRA: ACARTOGRAPHERserviceholdsanapriorimapoftheenvironment,andcanusethismapinconjunctionwithaTrulla-basedpathplannertogeneratearoutetoagivenlocation.TheCARTOGRAPHERalsoupdatesthismapbasedondynamicallydiscoveredchangesintheenvironment. AMOVETOWAYPOINTbehaviorreceivesawaypointandmovestothiswaypointwhileavoidingobstacles. ADRIVEEFFECTORservicethatusestheUSVdrivethrusterstomovethevehicle. Acomplementofsensors:HEADINGSENSOR,GPSSENSOR,ODOMETRYSENSOR,andLASERRANGESENSOR.Thesesensorsprovidedatatohigher-levelschemasandbehaviors. ALASERFILTERSCHEMAtocleannoisyrangenderdata;aSELFPOSESCHEMAthatgeneratesposeinformationfromcompass,GPSandodometrydata;twoservicestosetandndthedistanceanddirectiontoastaticwaypointgoal. TheSCRIPTMANAGERisresponsibleforruntimemanagementofscripts,includingscriptlifecycleandeventmanagement. TheCONTEXTADAPTER,whichisresponsibleforproperrolemanagementaccordingtotheapproachinChapterThree,asisdescribedinChapterFour.TheCONTEXTPLUGINfortheUTcontexttheCONTEXTADAPTERdenesthreerolesfortheUGV:DRONE,SCOUTandTELEFACTOR.TheMCOCUrunstheoperatorcontrolGUIandrunstheSCRIPTMANAGERandCONTEXTADAPTERDFRAservices.TheCONTEXTPLUGINfortheUTcontextontheMCOCU'sCONTEXTADAPTERdenesonerole:OPERATOR.TheMCOCUintheOPERATORroleistheprimarycontrolsourceforthemission.TheOPERATORisresponsibleforcoordinatingtherolesoftheitselfandtheUGV.5.2.3.2RolesUsedTherearefourrolesusedinthisscenariobytheUGVandOCU:DRONE,SCOUT,TELEFACTOR,andOPERATOR.ThecorefunctionsoftheserolesaredescribedinSection4.3,howevertheUTcontextexpectationsareasfollows. 88

PAGE 100

TheUGVhasthelargestsetofavailableroles.TheDRONEfunctionsrevolvearoundsimplemovementtodenedwaypoints.TheexpectationsfortheUTcontextarethefollowing:theDRONEshouldusethelocalCARTOGRAPHERservicetoplanitsroutetoeachgoal;itcanmoveasfastasisneeded.IntheSCOUTroletherobotissubjecttothefollowingexpectations:theSCOUTmusthaltandsendanoticationifananomalyisdetected;itmustusehighcautionwhenmoving;itshouldusetheCARTOGRAPHERforrouteplanning.TheSCOUTroledoesnothaveanyadditionalroleexpectations.TherolefortheOCUissubjecttothefollowingexpectations.AstheOPERATORforthemission,theOCUisresponsiblefortheoverallmission.TheOPERATORhasthefollowingexpectations:asetofwaypointsthatmustbepassedtotheUGVwhenthevehicleisaDRONEorSCOUT;theOPERATORshouldinteractwiththerobotinapassivemodethatallowstheOPERATORtoobserveanddirecttherobotatahighlevel,butdoesnotallowdirectintervention.5.2.3.3ScenarioExecutionFigure37showsatimelineoftherolechangesthroughoutthisscenario,extractedfromtheruntimeloggingofeventsbytheservicesontheMCOCUandUGV.Theremainderofthissectiondiscussesthistimelineindetail,andFigure38throughFigure45showrelevantimagesoftheUGVandrolevisualizationsfromtheMCOCUGUI.Anoteabouttherolevisualizationimagesisinorder.Eachdiagramshowstherolestateofeachagent.Aconcentriccircleisdrawnaroundeachagentwhenaroleiseitherassignedoractive.Assignedrolesaredrawninorange,whileactiverolesaredisplayedingreen.Whenmultiplerolesareassignedtoanagent,thehigherpositionroleisdrawnoutsidealowerpositionrole.Ifthepositionisequal,thentheorderofringsisarbitrary.Thegreyarcsshownintheguresshowinuence.Theagentorroleattheendofthearclabeledwithacircleisresponsiblefortheroleactionattheunadornedend.Certainactionsintheguresmayappearslightlydifferentthanexpectedforthirdreasons.First,duetothedistributednatureofthescenario,noticationsfromdifferentagentmayappearoutoforder.Second,therecursivenatureofsomeroleoperationsdisplaysthesameeffect.Third,ifnoticationsofmultipleeventsarereceivedinclosesuccession,theGUIupdatewillshowthemostrecentvalidstateandwilldiscardintermediateevents.ThehumanoperatorisabletointeractwiththeUGVtomonitor,assignandotherwiseinteractwiththerolesonthephysicalrobot.BoththeUGVandtheMCOCUinitiallystartwithnoroleassignmentsandarenotassociatedwithanysocialcontext.AstheCONTEXTADAPTERoneachagentinitializes,eachagentjoinstheUTcontext.FortheMCOCU,thishastheeffectofautomaticallyassigningtheOPERATORroletotheMCOCU.Beforethisassignmentcompletes,theOPERATORassigns 89

PAGE 101

theTELEFACTORroletotheUGV.Figure38showstheexteriorviewoftheUGVandFigure39ashowstheroleassignmentsatthestartofthescenario.ThescenariobeginswhentheMCOCUactivatestheOPERATORrole.TheCONTEXTPLUGINfortheUTcontextontheMCOCUisconguredtodothisthroughtworoutes.TherstisDFRA-based,wherearequestforactivationissentthroughtotheCONTEXTADAPTER.ThesecondisarequestfromtheexternalsocialcontextthroughtheCONTEXTPLUGIN.Forthisscenario,theactivationoftheOPERATORroleistriggeredbyahumanthroughtheOCUGUIFigure39b.TheOPERATORactivatestheTELEFACTORroleontheUGV.TheactivationofthemissionwithacommandfromtheexternalcognitiveagentthroughtheCONTEXTPLUGINwasshowninasimilardemonstrationonApril25,2006.ThisstageofthedemonstrationshowstherolelifecycleastheOPERATORandTELEFACTORchangesrolestate.Next,theOPERATORassignsandactivatestheDRONEroleontheUGVFigure39c,settingsocialexpectationsandwaypointgoalsfortherole.TheUGVisnowplayingtheDRONEandTELEFACTORroles,whiletheMCOCUisintheroleofOPERATOR.Atthisstage,thescenariohasshownmultiplesimultaneousroles,trivialprocessingofinter-roleconictnoconictcase,rolefunctions,basicroleexpectations,androlelifecyclemanagement.TheUGVthenreachesthenalwaypointgoalspeciedbytheDRONErole.TheMCdeactivatestheDRONEandhastheUGVrelinquishtheroleFigure41aandFigure41b.TheOPERATORassignsandactivatestheSCOUTroleontheUGVandsetssocialexpectationsandwaypointgoalsfortheroleFigure41c.Figure40showsanexternalviewoftherobotasitbeginstheSCOUTrole.TheUGVisnowplayingtheSCOUTandTELEFACTORroles,whiletheMCOCUisintheroleofOPERATOR.AstheUGVcompletesthescoutingmissionspeciedbytheSCOUTrole,theOPERATORdeactivatestheSCOUTandhastheUGVrelinquishtheroleFigure43aandFigure43b.Subsequently,theOPERATORassignsandactivatestheDRONEroleontheUGVandsetssocialexpectationsandwaypointgoalsfortheroletoreturntothestartingareaFigure43c.Figure42showsanexternalviewoftherobotatthispointintime.TheUGVisagainplayingtheDRONEandTELEFACTORroles,whiletheMCOCUisstillintheroleofOPERATOR.Tocompletethescenario,theUGVreachesthelastofthewaypointgoalsspeciedbytheDRONErole.TheMCOCUthendeactivatestheDRONEandhastheUGVrelinquishtheroleFigure45aandFigure45b.Figure45bshowsaviewfromtheGUIattheendofthephysicaldemonstration.NotethatattherobotSCOUThasmappedpreviously-unknownterrain,showninorangeattheupperrightofthelargercentralmap. 90

PAGE 102

5.2.3.4SummaryFigure37showstimelineofroleeventsconstructedfromloggingdatageneratedontheUGVandMCduringanexecutionofthescenario.ThistimelineandunderlyingdatavalidatesthattheimplementationoftheCONTEXTADAPTER,ROLEMANAGER,EXPECTATIONMANAGERandSCRIPTMANAGERoperateasdesignedforthisscenario.Thisentiresimulationscenariohasshownthefollowingrolemechanisms,dimensions,andothercriteria: Atrivialcaseofinter-roleconictwashandledduringtheassignmentoftheDRONEandTELEFACTOR.Noconictwasdetectedandnoworkneededtobedone. Rolefunctionwasconsistentlyshownthroughthesuccessfulcompletionofthevariousroles.RoleexpectationmodiedthespeedoftheUGVandmodiedtheDRONEandSCOUTrolewithwaypointgoals,butwasmorestronglyshowninthesimulationscenario. Rolelifecyclewasconsistentlyexercisedtosupportallofthebaseroleactivities. TheuseofmultipleroleswasshowninseveralinstanceswiththeUGVsupportingtheDRONE/SCOUTandTELEFACTORroles. ThedemonstrationofrolesinboththeLWandUTcontextsshowshowrolescanbeusedinmultipledomains.TheDRONEroleanditSCRIPTimplementationisexactlythesameforbothcontexts.Thissectionhaspresentedthephysicaldemonstrationusedtoshowthecorefunctionalityoftherolesystemasithasbeendeveloped.Thisdemonstrationshowsthreeofthekeyroledimensionsandmechanisms:rolefunctions,roleexpectations,handlinginter-roleconict,andmultipleconcurrentroles.Thattherolefunctionsworkisshownbythesuccessfulcompletionofthedemonstrationscenario.Thatis,thebehaviorsandtasksassociatedwitheachoftheroleswereexecutedatthepropertimesandproducedthedesiredbehaviorontherobot.MultipleconcurrentrolesareshownbytheassignmentoftheTELEFACTORandDRONEorSCOUTrolesontheUGVduringthemission.Inaddition,theassignmentofmultiplerolesrequirestheuseoftheinter-roleconictmechanismtodeterminewhethertherolefunctionsconict. 91

PAGE 103

Figure37:Roletimelineforthesimulatedscenariohighlightingroletransitionsgleanedfromrobotlogles. 92

PAGE 104

Figure38:ThestartofthephysicalscenarioshowingtheexternalviewofthephysicalATRV-Jrrobot. 93

PAGE 105

a b cFigure39:Thestartofthephysicalscenario.ashowstheroleassignmenttothetwoagentsatthestartofthesimulation;bshowstheassignmentandactivationoftheOPERATORroleontheMCandTELEFACTORroleontheUGV;cshowstheassignmentandactivationoftheDRONErole. 94

PAGE 106

Figure40:ExternalviewoftheUGVasitbeginstheSCOUTroleinthephysicalscenario. 95

PAGE 107

a b cFigure41:TheUGVbeginstheSCOUTroleinthephysicalscenario.ashowsthedeactivationoftheDRONEroleontheUGV;bshowstheUGVrelinquishingtheDRONEroleattherequestoftheOPERA-TOR;cshowstheassignmentandactivationoftheSCOUTrole. 96

PAGE 108

Figure42:ExternalviewoftheUGVasitnishesthescoutingmissionandreturnstobaseinthephysicalscenario. 97

PAGE 109

a b cFigure43:TheUGVnishesthescoutingmissionandreturnstobaseinthephysicalscenario.ashowsthedeactivationoftheSCOUTroleontheUGV;bshowstheUGVrelinquishingtheSCOUTroleattherequestoftheOPERATOR;cshowstheassignmentandactivationoftheDRONEroletoreturntobase. 98

PAGE 110

Figure44:AviewfromtheGUIattheendofthephysicaldemonstration.NotethatattherobotSCOUThasmappedpreviously-unknownterrain,shownintheupperrightofthelargercentralmap. a bFigure45:Thephysicaldemonstrationscenarioiscomplete.ashowsthedeactivationoftheDRONEroleontheUGVasitreachestheendofthemission;bshowstheUGVrelinquishingtheDRONEroleattherequestoftheOPERATOR. 99

PAGE 111

5.3LimitationsoftheDemonstrationsChapterFivedescribedasocialcontextthatrequiresusingheterogeneousteamsinanurbanenvironmentthatisanunderlyingmotivationforthiswork.ThechapteralsopresentedtwoadditionalscenariosthatdemonstratethatrolescanbeusedtocoordinatetheactionsofateamconsistingofahumanoperatorandautonomousUSVandUAVinalittoralenvironment,aswellademonstrationthatgroundedtheroleframeworkinademonstrationonaphysicalUGVwithahumanoperator.However,therearelimitationstothesedemonstrationsthatshouldbepresented.Therstlimitationisthelimitedteamsizeusedforphysicalandsimulateddemonstrations.Morerobotswouldhaveprovidedastrongerargumentforthepracticalutilityoftherolemodel.However,thisisnotaseriouslimitation.Eachagentmaintainsitsownrolestate,anddoesnotprocessthestateofotheragentsunlessactivelycoordinatingwiththosesystems.Thus,theoverheadforanysystemgoalrequiringasetofnrolesisproportionaltonforanyagent.GagehasshownthescalabilityofanintelligentaffectiverecruitmentagentusingthesameDFRAarchitecturethatunderliestheimplementationinChapterFourto53agents,showingthattheteamsizelimitationisnotduetounderlyingarchitecturalscalabilityconstraints.Inaddition,thisistherstdemonstrationofarole-basedsystemforcoordinatingheterogeneousagentsonphysicalhardwareinanoutdooreldenvironment.TheclosestcomparableimplementationisthatofSattereldetal.2005,whichusedalargerteamofthreerobotsandthreehumans,butdidsoinacontrolledindoorenvironment.Thesecondlimitationisthatthescenariosproceededinalinearfashion.Thiswasdonetoallowvericationthatthescenariowasproducingthedesiredrolebehavior,butdidnotilluminatethetrueexibilityoftherolemodelforenablingcoordinationinvolvingahumanoperator.Theoperatorwaslimitedtoinitiatingthescenarios,andidentifyingthemineinthesimulation.Duringtheremainderofthesimulationandphysicalscenariotheactionsofthehumanwereautomaticallyperformedbyascriptexecutingthefunctionsoftherole.Thisisalimitation,butwasdonefordemonstrationpurposes.TheeventsthattriggeredstatechangesintheCOMMANDERandOPERATORscriptsweregeneratedbyotheragents,butgivenaGUIandtheproperinterfacecouldjustaseasilyhavebeentriggeredbyahumanoperator.Third,learningandadaptationmechanismsweredescribed,butnottested.Section5.2.1.3,inparticular,describeshowateammustadaptbasedonthestateofteammembers.Forthesimplescenarios 100

PAGE 112

demonstratedinSection5.2.3andSection5.2.2,neitherlearningoradaptationwereimportantduetothesmallnumberofagentsinvolvedandthelinearnatureofthetasks.However,learningandadaptationwillplayalargerroleinscenariosthathavesomedegreeofrepetition,suchasthedesignscenarioisSection5.2.1orinlong-livedsystemsthatareabletoadapttoandlearnnewsocialcontextsthroughexperience.Thislimitationisnotamajorlimitationasthisworkisfocusedonthefoundationalframeworkthatwillenablefutureresearchintotheseareas.Fourth,thisroleframeworkisparticularlysuitedtoteamswheretherulesofthesocialcontextareadheredtobythemajorityofthemembersParsons1951,pp.27.Thisdoeslimittheapplicabilityofthisapproachtowell-denedcontexts,butitisnotacriticallimitationbecauseitdoesnotprecludemanyformsofcompetition.Robotsoccer,forexample,isasocialsystemwheretherulesofthesocialsystemsarefollowedbymembersoftwoopposingteams,eventhoughthegoalofoneteamistowinattheexpenseoftheotherteam.Thishasseveralimplications.First,agentsmustnotdisplaymuchdeviantbehavior,althoughacertainamountshouldberegulatedinternallythroughsocialmechanisms.Second,thesocialsystemmustdeneaminimumofsocialorderandnotplacetoomanyorimpossibledemandsonmembers.Failuretofollowthesetwocriteriamaynothavethesameconsequencestoarticialagentsastheywouldtohumans,butmaystillimpacttheoverallfunctionalityofthesocialsystem.Finally,ChapterThreespeciedthreetypesofexternalcontexts,butthesewerenotextensivelyimplementedortestedinthisdissertation.Inparticularthecaseofanexternalagentthathasdifferentconceptsforroles,tasks,orexpectationsprovidesaparticularlychallengingcase.Thisisanontologicalmismatch,andtheautomatedmatchingofconceptsinsimilarproblemsiscalledthesemanticmatchingproblem.ThisuncoversaweaknessintheunderlyingDFRAarchitecture.Thearchitecturewasnotdesignedwithontologicaldescriptivenessasacriteria,soveryfewelementscanprovidethedescriptivenessrequiredforthissortofontologicalmatching.Thisisamajorconstraintimpactingfuturedevelopmentofthisrolesystem,andshouldbeaddressed.5.4SummaryThethreecasestudiespresentedinthischapterservetosupporttheoverallresearchquestionofthedissertationandpresentsuccessfuldemonstrationsoftherolesystemcoordinatingtheoperationofaheterogeneousteamofrobots,softwareagentsorsimulatedrobotsandhumanoperators.Thedesigndemonstrationshowedhowrolescanbeusedtosolvecoordinationtasksforahighly-interdependentscenarioinvolvinghumansandrobotsinurbancombatoperations. 101

PAGE 113

Thesuccessfulsimulationresultsshowaheterogeneousmulti-agentteamcoordinatingrole-basedactivityinalittoralportlocation,withtheUAV,USV,andhumanoperatorcoordinatingtodetectexplosivesalongapier.Thesuccessfulphysicaldemonstrationgroundstherolemodelandshowsthatoperationispracticalonrealrobotsinaeldenvironment.Combined,thethreescenariosmeetthedemonstrationgoalsandcollectivelysupporttheresearchquestion. 102

PAGE 114

ChapterSixSummaryandFutureWorkThisdissertationisprimarilyconcernedwiththecoordinationofheterogeneousteamsorrobotsusinganaturalisticrole-basedapproach.Thishasbeenidentiedasakeylong-termchallengeforunmannedsystemsNationalResearchCouncilU.S.2005;OfceoftheSecretaryofDefense2005,andisspurringactiveresearchintoteamworkandteamprocessesBradshawetal.2005;Feltovichetal.2004;Kleinetal.2004;Sierhuisetal.2003;Tate,Bradshaw,andPechoucek2002.Thisdissertationhasintroducedamethodforcoordinatingheterogeneousteamsofhumans,robotsandsoftwareagentsbasedonthenaturalisticuseofsocialroles.Thisapproachhasadaptedamodelofrolesbasedonroledimensionsandmechanismsidentiedinrelatedworkintohumansocio-technicalsystemsJahnke,Ritterskamp,andHerrmann2005andinuencedbyusesofrolesinagentsystemsDastanietal.2005;Bradshawetal.2003;Uszok,Bradshaw,andJeffers2004.ThismodelisultimatelybasedonaconsensusofpropositionsaboutrolesfromsocialscienceBiddle1979,pg.8.Thisapproachhasledtoasuccessfuldemonstrationofrolescoordinatingaheterogeneousteamofrobotsandhumanoperatorsinsimulationandusingphysicalrobotsinaeldsetting.Theremainderofthischapterreviewshowtheapproachandimplementationpresentedinthisdissertationbuildsaworkingtechnicalframeworkforrobotsystemsthatisabletocoordinateheterogeneousteamsofrealrobotsandinsimulation.6.1ReviewofSalientPointsTheresearchquestionstatedinChapterOnewasthefollowing:Cantheuseofsocialrolesenablethenaturalisticcoordinatedoperationofrobotsinamixedsetting?Socialsciencehasbeenobservinghumansocialsystemsformanyyearsandtheimportanceofroleshasbeennoticedandstudied.Thisstudyhasrevealedkeypropositionsthataregenerallyagreedtoholdregardingrolesthatcanserveasaminimumbenchmarkfortheapproachinthiswork: 1. Atleastsomebehaviorsarepatterned,andarecalledroles.Theserolesadheretoindividualswithinasocialcontextandareboundtothatsocialcontext.Theapproachandimplementationofthiswork 103

PAGE 115

ispredicatedaroundintroducingtheroleconcepttoarticialsystems,andusingrolesforcoordinatingteamactivity. 2. Rolesaretypicallynotboundtospecicindividualsthough,andmorethanoneindividualmayplayagivenrole.Inasense,theindividualisinterchangeablefromaroleperspective.InthedemonstrationsinChapterFive,thespecicrobotplayingeachoftherolesisnotimportantanyrobotwiththeappropriatehardwareandsoftwarecouldplaytheroleandhelptheoverallteamcompletetheassignedmission. 3. Theawarenessofrolesmakesindividualsconsciousoftheiractionsrelatingtotheroleinthesocialcontext.Thisawarenessisofsocialexpectations,andtheseexpectationsinuencetheactionsoftheindividual.Inthesimulationdemonstration,theUSVisawareofitsrole,andtheexpectationsoftheDRONE,COURIER,andSCOUTrolesinuencethemaximumspeedoftherobot. 4. Rolesareimportantfortworeasons.First,theyhaveconsequences.Thatis,rolesareassociatedwithactionsthatcanhelpattaingoalsorotherbenecialresults.Second,therolesareembeddedinasocialsituation,andtheactionsofanindividualinaroleimpactotheragents.Thisisakeybenetthatdirectlyaddressesthecoordinationproblem. 5. Rolesarealsonotstatic;theychangeovertime.Ingeneral,socialrolesforhumansarelearnedaspartofasocializationprocess.Rolesforarticialagentsmustpresentlybecreatedexternallytotheagentratherthanlearned.Thisdissertationhasfocusedprimarilyonthepreviousfourpropositions.Section5.2.1introducesaleveloflearningandadaptation,butleavesmostissuesrelatedtolearningtherolesinasocialcontextforfuturework.Roleshavebeenusedinseveralcommunitieswithinthecomputersciencediscipline,withdifferentdisciplinesadheringtotheunderlyingsocialsciencemodelatvaryingdegreesofdelity.Inparticular,rolesformapartoffourmaincommunities:knowledgerepresentation,programminglanguages,socio-technicalsystemsandagentsystems,includingsoftwareagentsandrobotics.Multi-robotsystemsisalessdevelopedeldthengeneralmulti-agentsystems.Asaresult,researchonstrongsocialinteractionsbetweenrobotsandwell-denedmulti-robotdomainshasbeenslowertodevelop.Rolesappearinalimitedforminpreviousroboticsliterature,buttheydosoprimarilyasasynonymfortask,withoutmanyofthestrongersocialaspects;theseusesofrolesappeartobestrictlylimitedtorole-assignmentandrole-change. 104

PAGE 116

Thespecicrolemodelthatmeetstheabovevegeneralpropositions,presentedinChapterThree,isadaptedfromrelatedworkinsocio-technicalsystemsandtheuseofrolesinagentsystems.Thischapterpresentedtherolemodelanddiscussedthetechnicalapproachtakentoimplementthemodel.ACONTEXTADAPTERserviceintheDFRAarchitecturewasdevelopedtomanagesocialcontextsandtheagent'sinteractionwithinthem.Todothis,theCONTEXTADAPTERinteractedwithaROLEMANAGERthatimplementedtherolemodel'smechanisms,aEXPECTATIONMANAGERtoencapsulatetheexpectationssurroundingarole,andaSCRIPTMANAGERthatsupervisedtheexecutionandlifecycleofthescriptscontrollingrolefunction.ChapterFourprovidedmoredetailregardingtheimplementationofthesesoftwarecomponents.Finally,ChapterFivedocumentedthreedemonstrationsoftheroleframework.Therstdemonstrationwasadesigndiscussion.ThegeneralscenarioisanUrbanOperationssocialcontextfocusedaroundcombatoperationsinanurbanenvironment.Thisincludesmissionssuchasateamofrobotandhumansoldiersclearingabuildingofhostileenemyforces,suchasthosedescribedinDepartmentoftheArmy2003.Howeverthisdiscussionconsiderstheuseofrobotsasasubstituteforallorpartoftheteam.Theroom-clearingscenarioisonlyoneofmanypossibilities,however,itservesasabaselinemissionfortheuseofrolesforcoordinatingrobotteamsinurbanenvironments.Thisscenarioalsoacontainsrepetitiveelementsthatallowtheintroductionoflearningandadaptationelementsoftherolemodel.Theseconddemonstration,intheLittoralWarfaresocialcontext,utilizesacombinationofunmannedsurface,aerial,andgroundvehiclesoperatingtosecureaportfacility.Thesurfaceandaerialvehiclescoordinatetodetectminesandintrudersalongapier,whilethegroundvehicles,aerialvehiclesandhumanscooperatetosecureadockarea.Inadditiontothephysicalagents,humanoperatorsinacontrolstationinteractwiththerobotsanddirectoveralloperationsfromacommandtent.ThisdemonstrationwasperformedinaclassCrobotsimulatorthatallowedtheintroductionofamorecomplexscenarioandfurtherdemonstrationofrolefunctionality.Finally,thethirddemonstrationwasintheUncertainTerrainsocialcontextandinvolvesarobotexploringanunknownenvironmentatthedirectionofahumanoperatororacognitivesoftwareagent.Therobotprovidessensoryandterraininformationbyusingsensingandmobilitycapabilities,andhastheabilitytoplanlocalpathsbasedonhigh-leveldirectivesfromotheragents.Thephysicaldemonstrationvalidatedcorerolefunctionalityonrealrobothardware. 105

PAGE 117

6.2ContributionsThisresearchwillbenettheroboticsandagentcommunities,articialintelligence,andmakecontributionstohuman-robotinteraction.Itposesafundamentalcontributiontothemulti-agentcommunitybecauseitextends,renes,andpresentsapracticalimplementationoftheroleconcept.Theapplicationofrolesinaprincipledandcompleteimplementationisanovelcontributiontobothsoftwareandroboticagents.Thecreationofanopensourceoperationalarchitecturewhichsupportstaskablerobotsisalsoamajorcontribution.Thisdissertationmakesatleastthefollowingninecontributions: Enablesacognitivelyplausiblenaturalisticapproachtothecoordinationofheterogeneousteams:Thisworkenablestheuseofanaturalisticapproachtothecoordinationofheterogeneousteams.ThisisoutlinedinChapterThree,andhasbeenimplementedusingtheDistributedFieldRobotArchitectureasdescribedinChapterFour.TheapproachhasbeenusedtodesignrolesforateamofrobotsinanurbanoperationsscenarioSection5.2.1.IthasalsobeendemonstratedinsimulationSection5.2.2andonrealrobotsSection5.2.3inanoutdooreldenvironment.Theuseofanaturalisticmethodalsoenablesfutureworkintomoreadvancedandintuitivemeansofcontrollingandinteractingwithteamsofrobots,suchasnaturallanguageprocessing. Fundamentalcontributionofarealizableframeworkthatusesrolesforcoordinatingamixed,heterogeneous,multi-agentteam:Fromapracticalrobotdesignandarticialintelligencestandpoint,rolesareanabstractionwhichgroupsactionandconstraintsonactionwithinaparticularcontext.Thisallowsafocusofattentiononaparticularsubsetofrobotactions;itisamethodtoprunethestatespaceofallpossiblerobotstatesandactionstothosethatareusefulinaparticularsetting.Thisdissertationshowsthatrolesarepracticalandcanworkonrealrobots. Renesthescriptmechanismforcoordinatingagentsinasocialcontext:ScriptsSchankandAbelson1977,aconceptfromarticialintelligencethathasbeenusedextensivelyinrobotics,havebeenusedfortheimplementationofarole'sactionsandhowaroleinteractswithotherroles.Indoingso,theuseofscriptshasreturnedtomorecloselyfollowtheoriginalconceptwhichwasconcernedwiththeinteractionofrolesinaparticularsituation.Inaddition,scriptshavebeenexpandedtoconsiderbothbasicinstrumentalscripts,usedforcontrollingsimple,repetitivetaskswithlittlevariability,andsituationalscripts,orcomplexscriptsforhandlingsituationswithmanyinterdependentrolesandactions.Scriptswereusedtoimplementtherolefunctionsforthe 106

PAGE 118

demonstrations,andwerecapableofsequencingnotonlyrobotbehaviors,butalsosequencingandcontrollingrolesondifferentagents. Renementoftheroleconceptsuitablefordistributedagent-basedsystems:Thisworkpresentsarenementoftheroleconceptformulti-agentsystems.Thisrenementnotonlyallowstheuseofrolestocoordinateheterogeneousteamsofhumansandrobots,butitalsoapplicabletoothermulti-agentsystems.Inaddition,thisworkidentiesareasofresearchintheapplicationofontologiesandsemanticmappingthatarenecessaryforallowingarticialagentstoautomaticallyjoin,understand,andtakepartinnewsocialcontexts,usingrolesasakeystartingconcept.Inasense,eachrobotinthesimulationdemonstrationcanbeconsideredasoftwareagentinteractingusingroles,providinganotherexampleoftheutilityofthiswork. Theuseofrolestoshareresponsibilityforataskbetweenhumanandrobotagents:ThescenarioinSection5.2.2sharesarolebetweenarobotandahumanoperator.Thisrolesharinghastheeffectofenablingbothagentstotakeresponsibilityforthecompletionofthejointtask.Formanyrobot-relatedscenarios,thistypeofjointactionisnecessary,particularlyiftherobotdoesnothavethecognitivecapabilitytocompletethetaskforitself.Intheexamplescenario,therobotdetectsananomalousobjectduringitsscoutingmission,butsharesthescoutrolewiththehumanforanalidentication.Duringthisthehumanusesthesensorsandeffectorsontherobot,butprovidesthecognitiveprocessingfortherole. Concisereviewofrobotliteraturerelatedtoroles:ChapterTwopresentsaliteraturesurveythatrstdescribesrolesfromasocialscienceperspective,andisolatesasetofkeypropositionsthatdescribefundamentalpropertiesthatareimportantforrobotics.Roleshavebeenusedincomputersciencebefore,butmuchofthisworkhasbeeninareasthatarenotdirectlyrelevanttorobotsystems.Theliteraturesurveythenlooksattheuseofrolesinsoftwareagentsystemsandelsewhereinrobotics,andexplainswhytheseusesofrolesarenotcompleteorsufcientforthisdissertation. Initialpresentationoffundamentalrobot-specicroles:Thisworkalsoidentiestwofundamentalrobotrolesthatcutacrossallsocialcontexts,andwillbefoundinanyscenariowheretherobotsarenotfullyautonomous.ChapterThreealsopresentsthreekeyinteractionlevels,passiveobservation,sharedcontrol,anddirectcontrol,thatarenecessaryforfulluseoftheTELEFACTORandOPERATORroles. 107

PAGE 119

Roleconceptsandapproacharetransferrabletootherarchitectures:TheapproachoutlineinChapterThreediscussesthefundamentalrolemodelandadditionalrequirementsinamannerthatisapplicabletomanyarchitectures.Theprimaryconstraintisthatthearchitecturebeahybriddeliberative/reactivearchitecture.PurelyreactivearchitecturesdonotbuildworldmodelsorconsiderplanningMurphy2000,pg.108,arequirementformuchoftherolemodel.TheimplementationinChapterFourusesDFRAasanunderlyingrobotarchitecture,buttheoveralldesignshouldbeusableinotherarchitecturesaswell. Contextadapterandplugins:AsnotedinSection1.1,robotteamsareorcanbeusedinawidevarietyofsettings.Inthesesettings,softwareexternaltotherobotwillmostlikelywantorneedtoreadandwritedataontherobot,includingaccesstosensors,effectors,andtherobot'scognitivestate,includingroles.ThecontextadapterandcontextpluginsdescribedinSection4.2andSection4.2.2areintendedtoalleviatethisintegrationproblem.Thecontextpluginshavebeenusedtointerfacewithtwodifferentagentsystemsthateachuseadifferentunderlyingcommunicationprotocolandagentframework.6.3FutureWorkWhilethisworktriedtoaddressandimplementasmuchoftherolemodelandpossible,thereareanumberofareasthatbearfurtherinvestigation.Thissectionpresentsseveraloutstandingissuesthatarepotentialdirectionsforwardinresearch.WhilethecoreroledimensionswereimplementedandtestedduringthesimulationandphysicaldemonstrationscenariosinChapterFive,amorecompleteimplementationofsocialinteractionandadaptationisneeded.Inparticular,investigatingwhichelementsofsocialinteractionshouldbespeciedinanapriorimannerandwhichshouldbelearnedduringtherolelifecycleisapossibleareaoffuturework.Thisisimportantwithheavilyinterdependenthumanandrobotinteraction,sincethehumanelementsmaynotbehaveinpreciselytheexpectedmannereverytime.Articialagentsshouldbeabletoadapttheirsideoftheinteractioninadynamicmanner.Therearetwokeyelementsoftherolemechanismsthathavebeenleftforfuturework.First,handlinginter-roleconictisperformedusingasimple,role-position-basedtechnique.Whilethiswassufcienttoshowinitialsuccess,additionalmodesofroleconictresolutionwillbeneededinmorecomplexscenarios.Affectivetechniquesusingarobot'semotionalstateandotherinternalregulatory 108

PAGE 120

mechanisms,policy-basedtechniquesthatusepredenedsetsofpolicies,orhuman-in-the-loopresolutionwhichfallsuptoahumanoperatorinparticularlydifcultcasesarethreepossibleadditionalmethodstoinuencetheconicthandling.Second,automatedroledenitionisanareathatcouldbeexploredaswell.Currentroledenitionisafunctionofhumandesign.Theprogrammer,usingananalysisofthetargetcontext,createstheroles,implementstherolefunctions,anddetermineshowthosefunctionsareimpactedbypolicy.Inthelongerterm,amethodforautomaticallycreatingrolesbasedonfeedbackfromthesocialcontextwouldhelpalleviatethisprogrammerburden.ChapterFourdenedeightrolesthathavebeenimplementedinthiswork.TheroleimplementationshavebeensufcienttosupportthedemonstrationscenariosinChapterFive,butseveralexistingrolesshouldberevisedandexpanded.TheTELEFACTORandOPERATORrolesserveasmarkerroles;theycurrentlyhavefunctionalimplementations.ThisallowselementsoftheCONTEXTADAPTERtoreasonoverrolestate,buttheydonotimplementanyspecichuman-robotinteraction.Futureworkshouldprovideamoresolidbasistosupportthepassiveobservation,sharedcontrol,anddirectcontrolasdetailedinChapterThree.Additionally,theSICKrolehasalimitedimplementation.Itisalsolimitedtoamarkercapacity,butshouldencompassthefunctionsandexpectationsarenotedinParsons1951.Thiscouldbeapartofaresearchagendaincorporatingfaulttoleranceandtherecoveryfromsystemerrors.TheCONTEXTPLUGINframeworksuggeststhattherearethreetypesofexternalsocialcontexts;thosethatsupportcompatibleexpectations,incompatibleexpectations,orthosewithnounderstandingofexpectationatall.Thesecondofthesethreetypescouldbenetfromautomaticsemanticmappingofexpectations,wherestatementsofexpectationintheexternalcontextareautomaticallyidentiedandmappedtocomparablestatementintheinternalEXPECTATIONMANAGER.Thisareaofworkisrelatedtoresearchinthesemanticweb,andcoulddrawfromworkinthatareasuchasStuckenschmidtandHarmelen2005,AntoniouandvanHarmelen2004andNoy2004.However,thismayalsorequireaddingamoreformalelementtotheDFRAarchitecturetoextractrelevantstateintoanappropriaterobotontology.ArelatedissueisthelackofaformalrepresentationofconceptswithinDFRA.ThishasabearingontherepresentationofentitiesthatusetheCONTEXTPLUGINmechanismtointeractwiththeDFRA-basedagents.DFRAcanonlyutilizetherolesasrepresentedbytheCONTEXTADAPTERservices 109

PAGE 121

oneachoftheseagent;itdoeshaveaseparatemodelforagentsthatdonotuseDFRA.ThiswillbeneededtoinferortrackrolesforagentsthatdonothaveaCONTEXTADAPTERrolerepresentation.Rigorouslytestinghuman-robotinteractionissuesisaareaofresearchthatisparticularlyimportant.Thisworkhasshownthatabstractingagentcapabilitiesintermsofsocialrolescanbeusedtocoordinateandcontrolheterogeneousteamsofagents.However,itwillbenecessarytobegintoquantifyhowwellthisapproachworksinreality,andinvestigateactualteamexperienceswithrobotsusingroles.Toconclude,thisdissertationhasraisedthequestionofwhethertheuseofsocialrolesisanappropriatemechanismforthecoordinationofheterogeneousteamsofagents.Ithasansweredthisquestionthroughdesignofaroleframeworkthathasbeendemonstratedinsimulationandonrealrobots.Itisexpectedthatthisabstractionofrobotactionwillserveasthebasisfortheemergingrevolutioninnaturalisticinteractionsinjointhuman-robotsystems. 110

PAGE 122

Bibliography Allport,GordonW.1961.PatternandGrowthinPersonality.Holt,RinehartandWinston. Antoniou,Grigoris,andFrankvanHarmelen.2004.ASemanticWebPrimer.TheMITPress. Baldoni,Matteo,GuidoBoella,andLeendertvanderTorre.2005.IntroducingOntologicallyFoundedRolesinObjectOrientedProgramming:powerJava.AAAIFallSymposiumonRoles-aninterdisci-plinaryperspective,2005. Barnes,LauraE.,RobinR.Murphy,andJeffreyD.Craighead.2005.Human-RobotCoordinationUsingScripts.TechnicalReport,CenterforRobot-AssistedSearchandRescue. Bartneck,C.,andJ.Forlizzi.2004.ADesign-CentredFrameworkforSocialHuman-RobotInteraction.ProceedingsofRo-Man2004.591. Beaumont,Patrick,andBrahimChaib-draa.2007.MultiagentCoordinationTechniquesforComplexEnvironments:TheCaseofaFleetofCombatShips.Systems,ManandCybernetics,PartC:Appli-cationsandReviews,IEEETransactionson37:373. Biddle,B.J.1986.RecentDevelopmentinRoleTheory.AnnualReviewofSociology12:67. Biddle,BruceJ.1979.RoleTheory:Expectations,Identities,andBehaviors.AcademicPress. Boella,G.,andL.Lesmo.2001.Normsandcooperation:Twosidesofsocialrationality.Proc.ofIJCAIWorkshoponAutonomy,Delegation,andControl:InteractingwithAutonomousAgents. Boella,Guido,andLeendertvanderTorre.2005.Enforceablesociallaws.AAMAS'05:ProceedingsofthefourthinternationaljointconferenceonAutonomousagentsandmultiagentsystems.NewYork,NY,USA:ACMPress,682. Boissier,Olivier,CosminCarabelea,CristianoCastelfranchi,JordiSabater-Mir,andLucaTummolini.2005.TheDialecticsbetweenanIndividualandHisRole.Roles,anInterdisciplinaryPerspective:Ontologies,ProgrammingLanguages,andMultiagentSystems:PapersfromtheAAAIFallSymposium2005. Boutilier,Craig.1996.Planning,LearningandCoordinationinMultiagentDecisionProcesses.Theo-reticalAspectsofRationalityandKnowledge.195. Bradshaw,J.,A.Uszok,R.Jeffers,N.Suri,P.Hayes,M.Burstein,A.Acquisti,B.Benyo,M.Breedy,M.Carvalho,D.Diller,M.Johnson,S.Kulkarni,J.Lott,M.Sierhuis,andR.VanHoof.2003.Rep-resentationandReasoningforDAML-BasedPolicyandDomainServicesinKAoSandNomads.ProceedingsoftheAutonomousAgentswithMulti-AgentSystemsConference. Bradshaw,JeffreyM.,StewartDuteld,PeteBenoit,andJohnD.Wooley.1997.ChapterKAoS:TowardanIndustrial-StrengthOpenAgentArchitectureofSoftwareAgents,375418.AAAIPress/TheMITPress. Bradshaw,JeffreyM.,HyuckchulJung,ShriKulkarni,MatthewJohnson,PaulFeltovich,JamesAllen,LarryBunch,NathanaelChambers,LucianGalescu,ReniaJeffers,NiranjanSuri,WilliamTaysom, 111

PAGE 123

andAnUszok.2005.TowardTrustworthyAdjustableAutonomyinKAoS.Volume3577.SpringerBerlin/Heidelberg. Bruce,Allison,JonathanKnight,SamuelListopad,BrianMagerko,andIllahR.Nourbakhsh.2000,April.RobotImprov:UsingDramatoCreateBelievableAgents.Proceedingsofthe2000IEEEInternationalConferenceonRobotics&Automation. Burke,J.J.,R.R.Murphy,M.Coovert,andD.Riddle.2004.MoonlightinMiami:AFieldStudyofHuman-RobotInteractionintheContextofanUrbanSearchandRescueDisasterResponseTrainingExercise.Human-ComputerInteraction19-2:85. Carlson,Jennifer,RobinR.Murphy,andAndrewNelson.2004,April.FollowupAnalysisofMobileRobotFailures.ProceedingsIEEEInternationalConferenceonRoboticsandAutomation,Volume5.4987. Casper,Jenn.2002.Human-RobotInteractionsduringtheRobot-AssistedUrbanSearchandRescueResponseattheWorldTradeCenter.Master'sthesis,UniversityofSouthFlorida. Chaimowicz,Luiz,VijayKumar,andMarioF.M.Campos.2004.AParadigmforDynamicCoordinationofMultipleRobots.Auton.Robots17:7. Colman,AlanW,andJunHan.2005.OrganizationalRolesandPlayers.Roles,anInterdisciplinaryPerspective:Ontologies,ProgrammingLanguages,andMultiagentSystems:PapersfromtheAAAIFallSymposium2005. Conte,Rosaria,andCristianoCastelfranchi.1999.Fromconventionstoprescriptions.Towardsanintegratedviewofnorms.ArticialIntelligenceandLaw7:323. Coutu,Walter.1951.Role-Playingvs.Role-Taking:AnAppealforClarication.AmericanSociologicalReview16:180Apr. Craighead,Jeff,RobinMurphy,JennyBurke,andBrianGoldiez.2007.ARobotClassicationSys-temforHRI.Proceedingsofthe2007InternationalSymposiumonCollaborativeTechnologiesandSystemsCTS. Craighead,Jeffrey.SARGE:TheSearchandRescueGameEnvironment&RobotSimulator. .YARS:YetAnotherRobotSimulator. Dastani,Mehdi,VirginiaDignum,andFrankDignum.2003.Role-assignmentinopenagentsocieties.AAMAS'03:ProceedingsofthesecondinternationaljointconferenceonAutonomousagentsandmultiagentsystems.NewYork,NY,USA:ACMPress,489. Dastani,Mehdi,M.BirnavanRiemsdijk,JorisHulstijn,FrankDignum,andJohn-JulesCh.Meyer.2005.EnactingandDeactingRolesinAgentProgramming.Volume3382.Springer-VerlagGmbH. Dautenhahn,Kerstin.2003.Rolesandfunctionsofrobotsinhumansociety:implicationsfromresearchinautismtherapy.Robotica21:443. DepartmentoftheArmy,ed.1992.Chapter2,OperationsofInfantryRiePlatoonandSquadFM7-8,editedbyDepartmentoftheArmy.DepartmentoftheArmy. ,ed.2003.Chapter3,UrbanCombatSkillsofCombinedArmsOperationsinUrbanTerrainFM3-06.11FM90-10-1,editedbyDepartmentoftheArmy.DepartmentoftheArmy. ,ed.2004.InTheSoldier'sGuideFM7-21.13,editedbyDepartmentoftheArmy.DepartmentoftheArmy. 112

PAGE 124

Dias,M.Bernardine,T.K.Harris,B.Browning,E.G.Jones,B.Argall,M.Veloso,A.Stentz,andA.Rud-nicky.2006.DynamicallyFormedHuman-RobotTeamsPerformingCoordinatedTasks.EditedbyTerryFong,ToBoldlyGoWhereNoHuman-RobotTeamHasGoneBefore:Papersfromthe2006SpringSymposium,TechnicalReportSS-06-07.MenloPark,California:AmericanAssociationforArticialIntelligence. Dudek,G.,M.Jenkin,E.Milios,andD.Wilkes.1993.Ataxonomyforswarmrobots.Volume1.441vol.1. Felicissimo,Carolina,CarlosLucena,GustavoCarvalho,andRodrigoPaes.2005.NormativeOntologiestoDeneRegulationsOverRolesinOpenMulti-AgentSystems.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Feltovich,PaulJ.,JeffreyM.Bradshaw,ReniaJeffers,NiranjanSuri,andAndrzejUszok.2004.SocialOrderandAdaptabilityinAnimalandHumanCulturesasAnaloguesforAgentCommunities:TowardaPolicy-BasedApproach.EngineeringSocietiesintheAgentsWorld:4thInternationalWorkshops,Volume3071.SpringerBerlin/Heidelberg,21. Ferraiolo,DavidF.,RaviSandhu,SerbanGavrila,D.RichardKuhn,andRamaswamyChandramouli.2001.ProposedNISTstandardforrole-basedaccesscontrol.ACMTrans.Inf.Syst.Secur.4:224. Gage,A.,R.Murphy,E.Rasmussen,andB.Minten.2004.Shadowbowl2003:LessonsLearnedfromaReachbackExercisewithRescueRobots.IEEERoboticsandAutomationMagazine,September. Gage,A.,andR.R.Murphy.2004.AffectiveRecruitmentofDistributedHeterogeneousAgents.toappearinNineteenthNationalConferenceonArticialIntelligenceAAAI-04. Gage,Aaron.2004,August.Multi-RobotTaskAllocationUsingAffect.Ph.D.diss.,UniversityofSouthFlorida. Gerkey,BrianP.,andMajaJ.Mataric.2002.Sold!:Auctionmethodsformulti-robotcoordination.IEEETransactionsonRoboticsandAutomation18:758October. Guha,RamanathanV.,andDouglasB.Lenat.1990.CYC:AMidtermReport.AIMagazine11:32. Gupta,G.Sen,C.H.Messom,andS.Demidenko.2004,December.StateTransitionBasedSTBRoleAssignmentandBehaviourProgramminginCollaborativeRobotics.Proceedingsofthe2ndInternationalConferenceonAutonomousRobotsandAgents. Herrmann,Stephan.2005.ProgrammingwithRolesinObjectTeams/Java.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Herrmann,Thomas,IsaJahnke,andKai-UweLoser.2004.TheRoleConceptasaBasisforDesigningCommunitySystems.InCooperativeSystemsDesign,editedbyFrancoiseDarses,RoseDieng,CarlaSimone,andManuelZackland,163.Amsterdam:IOSPress. Hoffman,Guy,andCynthiaBreazeal.2006.WhatLiesAhead?ExpectationManagementinHuman-RobotCollaboration.EditedbyTerryFong,ToBoldlyGoWhereNoHuman-RobotTeamHasGoneBefore:Papersfromthe2006SpringSymposium,TechnicalReportSS-06-07.MenloPark,California:AmericanAssociationforArticialIntelligence. Hollnagel,Erik,andDavidD.Woods.2005.JointCognitiveSystems:FoundationsofCognitiveSystemsEngineering.CRCPress. 113

PAGE 125

Hughes,K.,A.Tokuta,andN.Ranganathan.1992.trulla:AnAlgorithmForPathPlanningAmongWeightedRegionsByLocalizedPropagations.IntelligentRobotsandSystems,1992.,Proceedingsofthe1992lEEE/RSJInternationalConferenceon,Volume1.469. Jahnke,Isa,CarstenRitterskamp,andThomasHerrmann.2005.SociotechnicalRolesforSociotechnicalSystemsAPerspectivefromSocialandComputerSciences.Roles,anInterdisciplinaryPerspective:Ontologies,ProgrammingLanguages,andMultiagentSystems:PapersfromtheAAAIFallSymposium2005. Jung,C.G.1928.TheRelationsbetweentheEgoandtheUnconscious. Kahn,MarthaL.,andCynthiaDellaTorreCicalese.2002.TheCoABSGrid.TechnicalReport. Klein,G.,D.D.Woods,J.M.Bradshaw,R.R.Hoffman,andP.J.Feltovich.2004.Tenchallengesformakingautomationa"teamplayer"injointhuman-agentactivity.IntelligentSystems,IEEE[seealsoIEEEIntelligentSystemsandTheirApplications]19:91. Kumar,Sanjeev,andPhilipR.Cohen.2000.Towardsafault-tolerantmulti-agentsystemarchitecture.ProceedingsofthefourthinternationalconferenceonAutonomousagents. Lenat,DouglasB.1995.CYC:alarge-scaleinvestmentinknowledgeinfrastructure.Commun.ACM38:33. Lenat,DouglasB.,R.V.Guha,KarenPittman,DexterPratt,andMaryShepherd.1990.Cyc:towardprogramswithcommonsense.Commun.ACM33:30. Li,Sing.2000.ProfessionalJini.WroxPress,Ltd. Linton,Ralph.1936.TheStudyofMan:AnIntroduction.Appleton-Century-Crofts. Lisetti,ChristineLaetitia,SarahM.Brown,KayeAlvarez,andAndreasH.Marpaung.2004.ASocialInformaticsApproachtoHumanRobotInteractionWithaServiceSocialRobot.IEEETransactionsonSystems,ManandCyberneticsPartC:ApplicationsandReviews34:195. Loebe,Frank.2005.Abstractvs.SocialRolesARenedTop-LevelOntologicalAnalysis.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Long,M.T.,R.R.Murphy,andL.E.Parker.2003.DistributedMulti-AgentDiagnosisandRecoveryfromSensorFailures.IEEE/RSJInternationalConferenceonIntelligentRobotsandSystems. Long,Matt,AaronGage,RobinR.Murphy,andKimonValavanis.2005.ApplicationoftheDistributedFieldRobotArchitecturetoaSimulatedDeminingTask.IEEEInternationalConferenceonRoboticsandAutomation. Long,Matt,andRobinMurphy.2005.AProposedApproachtoSemanticIntegrationbetweenRobotandAgentSystems.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Long,MatthewT.2004,November.Creatingadistributedeldrobotarchitectureformultiplerobots.Master'sthesis,UniversityofSouthFlorida. Long,MatthewT.,RobinR.Murphy,andJimHicinbothom.2006,September.UsingRolesandPer-sonatoAdaptRobotstoNewDomains.submittedtotheInternationalConferenceonRoboticsandAutomation. Malone,ThomasW.,andKevinCrowston.1994.Theinterdisciplinarystudyofcoordination.ACMComput.Surv.26:87. 114

PAGE 126

Martinson,Eric,andRonaldC.Arkin.2003.Learningtorole-switchinmulti-robotsystems.ICRA.2727. Masolo,Claudio,GiancarloGuizzardi,LaureVieu,EmanueleBottazzi,andRobertaFerrario.2005.Re-lationalRolesandQua-individuals.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Masolo,Claudio,LaureVieu,EmanueleBottazzi,CarolaCatenacci,RobertaFerrario,AldoGangemi,andNicolaGuarino.2004.SocialRolesandtheirDescriptions.KR.267. Mead,G.H.1934.Mind,Self,andSociety.EditedbyC.W.Morris.UniversityofChicagoPress. Micire,Mark.2002.AnalysisoftheRobotic-AssistedSeachandRescueResponsetotheWorldTradeCenterDisaster.Master'sthesis,UniversityofSouthFlorida. Moreno,J.L.,HelenH.Jennings,JoanH.Criswell,LeoKatz,RobertR.Blake,JaneS.Mouton,MerlE.Bonney,MaryL.Northway,CharlesP.Loomis,CharlesProctor,RenatoTagiuri,andJiriNehneva-jsa,eds.1960.TheSociometryReader.TheFreePressofGlencoe,Illinois. Moshkina,L.,andR.C.Arkin.2005,July.HumanPerspectiveonAffectiveRoboticBehavior:ALongitudinalStudy.Proc.IEEEInternationalConferenceonIntelligentRobotsandSystems,IROS.14441451. Murphy,RobinR.2000.IntroductiontoAIRobotics.TheMITPress. Murphy,RobinR.,K.Hughes,andE.Noll.1996,April.AnExplicitPathPlannertoFacilitateRe-activeControlandTerrainPreferences.RoboticsandAutomation,1996.Proceedings.,1996IEEEInternationalConferenceon,Volume3.2067. NationalResearchCouncilU.S..2005.AutonomousVehiclesinSupportofNavalOperations.Washing-ton,D.C.:NationalAcademiesPress. Nise,NormanS.1992.ControlSystemsEngineering.TheBenjamin/CummingsPublishingCompany,Inc. Noy,N.F.2004.SemanticIntegration:ASurveyOfOntology-BasedApproaches.SIGMODRecord,SpecialIssueonSemanticIntegration33,no.4December. OfceoftheSecretaryofDefense.2005.UnmannedAircraftSystemsRoadmap,20052030.Washing-ton,D.C.:NationalTechnicalInformationService. Paesschen,EllenVan,WolfgangDeMeuter,andMajaD'Hondt.2005.RoleModellinginSelfSyncwithWarpedHierarchies.AAAIFallSymposiumonRoles-aninterdisciplinaryperspective,2005. Parker,LynneE.2000.LifelongAdaptationinHeterogeneousMulti-RobotTeams:ResponsetoContin-ualVariationinIndividualRobotPerformance.AutonomousRobots8:239. .2004.Personalcommunication. Parsons,Talcott.1951.TheSocialSystem.Collier-MacmillanCanada,Ltd. Pickett,JosephP.etal.,ed.2000.TheAmericanHeritageDictionaryoftheEnglishLanguage.Fourthedition.Boston:HoughtonMifinCompany. Quinn,Matt,LincolnSmith,GilesMayley,andPhilHusbands.2003.Evolvingteamworkandrole-allocationwithrealrobots.ICAL2003:ProceedingsoftheeighthinternationalconferenceonArti-ciallife.Cambridge,MA,USA:MITPress,302. 115

PAGE 127

Sandhu,RaviS.,EdwardJ.Coyne,HalL.Feinstein,andCharlesE.Youman.1996.Role-BasedAccessControlModels.IEEEComputer29:38. Sattereld,Brian,HeetenChoxi,andDrewHousten.2005,March.RoleBasedOperations.ThirdInternationalMulti-RobotSystemsWorkshop. Sattereld,Brian,SteveJameson,HeetenChoxi,andJerryFranke.2005,June.ARole-BasedApproachtoUnmannedTeamOperations.AssociationforUnmannedVehicleSystemsInternationalandFlightInternational,UnmannedSystemsNorthAmerica2005Conference. Schank,RogerC,andRobertPAbelson.1977.ScriptsPlansGoalsandUnderstanding.LawrenceErlbaumAssociates,Inc. Shoham,Yoav,andMosheTennenholtz.1995.Onsociallawsforarticialagentsocieties:off-linedesign.Artif.Intell.73-2:231. Sierhuis,Maarten,JeffreyM.Bradshaw,AlessandroAcquisti,RonvanHoof,ReniaJeffers,andUszokAndrzej.2003.Human-AgentTeamworkandAdjustableAutonomyinPractice.ProceedingsofThe7thInternationalSymposiumonArticialIntelligence,RoboticsandAutomationinSpace. Sims,Mark,DanielCorkill,andVictorLesser.2004a.SeparatingApplication-SpecicandOrganiza-tionalCoordinationIssuesduringMulti-AgentOrganizationalDesignandInstantiation.TechnicalReport,UniversityofMassachusetts. .2004b.SeparatingDomainandCoordinationKnowledgeinMulti-AgentOrganizationalDesignandInstantiation.AAAI-04WorkshoponAgentOrganizations:TheoryandPractice. Smith,ReidG.1980.TheContractNetProtocol:HighLevelCommunicationandControlinaDis-tributedProblemSolver.IEEETransactionsonComputersC-29:1104December. Sowa,JohnF.2000.KnowledgeRepresentation:Logical,Philosophical,andComputationalFoundations.Brooks/Cole. Steimann,F.,andP.Mayer.2005.Patternsofinterface-basedprogramming.JournalofObjectTechnol-ogy. Steimann,Friedrich.2000.Ontherepresentationofrolesinobject-orientedandconceptualmodelling.DataKnowl.Eng.35:83. Stone,Peter,andManuelaVeloso.1999.Taskdecompositionanddynamicroleassignmentforreal-timestrategicteamwork.InIntelligentAgentsVProceedingsoftheFifthInternationalWorkshoponAgentTheories,Architectures,andLanguagesATAL-98,editedbyJ.P.Muller,M.P.Singh,andA.S.Rao,Volume1555ofLectureNotesinArticialIntelligence,293.Springer-Verlag,Heidelberg. Sty,Kasper,Wei-MinShen,andPeterWill.2002.UsingRoleBasedControltoProduceLocomotioninChain-TypeSelf-RecongurableRobots.IEEETransactionsonMechatronics7:410. Stroustrup,Bjarne.1994.TheDesignandEvolutionofC++.Addison-WesleyPublishingCompany. Stuckenschmidt,Heiner,andFrankvanHarmelen.2005.InformationSharingontheSemanticWeb.AdvancedInformationandKnowledgeProcessing.Springer. Tanenbaum,AndrewS.1995.DistributedOperatingSystems.PrenticeHall. Tate,Austin,JeffreyM.Bradshaw,andMichalPechoucek.2002.KnowledgeSystemsforCoalitionOperations. 116

PAGE 128

Thomas,EdwinJ.,andBruceJ.Biddle.1966.ChapterBasicConceptsforClassifyingthePhenomenaofRoleofRoleTheory:ConceptsandResearch,editedbyBruceJ.BiddleandEdwinJ.Thomas,2345.JohnWiley&Sons. Uszok,Andrzej,JeffreyM.Bradshaw,andReniaJeffers.2004.KAoS:APolicyandDomainServicesFrameworkfoGridComputingandSemanticWebServices.ProceedingsoftheSecondInternationalConferenceonTrustManagement. Vaughan,RichardT.,KasperSty,GauravS.Sukhatme,andMajaMataric.2000.Chapter4ofBlazingatrail:Insect-inspiredresourcetransportationbyarobotteam,111120.Springer-VerlagTokyo. Weiss,Gerhard,ed.1999.MultiagentSystems:AModernApproachtoDistributedArticialIntelligence.TheMITPress. Werger,BarryBrian.2000.Chapter2ofAyllu:DistributedPort-ArbitratedBehavior-BasedControl,2534.Springer-VerlagTokyo. Wooldridge,Michael,NicholasR.Jennings,andDavidKinny.2000.TheGaiaMethodologyforAgent-OrientedAnalysisandDesign.AutonomousAgentsandMulti-AgentSystems3:285. Wright,J.,F.Hartman,B.Cooper,S.Maxwell,J.Yen,andJ.Morrison.2006.DrivingonMarswithRSVP.Robotics&AutomationMagazine,IEEE13:37. Zambonelli,Franco,NicholasR.Jennings,andMichaelWooldridge.2003.Developingmultiagentsystems:TheGaiamethodology.ACMTrans.Softw.Eng.Methodol.12:317. Zhu,H.2003.SomeIssuesinRole-BasedCollaboration.ProceedingsofIEEECanadaConferenceonElectricalandComputerEngineeringCCECE'03,. Zimmel,BrianC.,MatthewT.Long,JenniferCarlson,andRobinR.Murphy.2004.DistributedErrorHandlingandHRI.Proceedingsofthe2004InternationalConferenceonRoboticsandAutomationICRA. 117

PAGE 129

Appendices 118

PAGE 130

AppendixA:Acronyms CoABS ControlofAgent-BasedSystems DARPA DefenseAdvancedResearchProjectsAgency DFRA DistributedFieldRobotArchitecture FSM Finite-StateMachine GPS GlobalPositioningSystem GUI GraphicalUserInterface JAUS JointArchitectureforUnmannedSystems JRMP JavaRemoteMethodProtocol KAoS KnowledgeableAgent-orientedSystem LW LittoralWarfare OCU OperatorControlUnit RAM Random-AccessMemory RBAC Role-BasedAccessControl SFX SensorFusionEffects UAV UnmannedAerialVehicle UGV UnmannedGroundVehicle UML UniedModelingLanguage UO UrbanOperations US UnitedStates USF theUniversityofSouthFlorida USV UnmannedSurfaceVehicle 119

PAGE 131

AppendixA:Continued UT UncertainTerrain MC MissionCommander SARGE Search-and-RescueGameEnvironment YARS YetAnotherRobotSimulator 120

PAGE 132

AppendixB:RoleDetails LittoralWarfare Table1:DescriptionoftheDRONEroleintheLittoralWarfarecontext.ThetabledescribestheDRONEroleandliststherolefunctions,positionandexpectations. DRONE Position HIGH Functions issrt.contexts.littoral.USVDroneScript Expectations UnmannedSurfaceVehicle Name Value Default script.cartographerEnabled TRUE script.caution LOW User drone.waypoint.0.x 30.345702 drone.waypoint.0.y -87.264215 base.waypoint.0.x 30.347936 base.waypoint.0.y -87.264331 Table2:DescriptionoftheCOURIERroleintheLittoralWarfarecontext.ThetabledescribestheCOURIERroleandliststheroledimensions:rolefunctions,positionandexpectations. COURIER Position LOW Functions issrt.contexts.littoral.USVCourierScript Expectations UnmannedSurfaceVehicle Name Value Default script.cartographerEnabled TRUE script.caution NORMAL script.releaseUAV TRUE User courier.waypoint.0.x 30.346314 courier.waypoint.0.y -87.263753 121

PAGE 133

AppendixB:Continued Table3:DescriptionoftheSCOUTroleintheLittoralWarfarecontext.ThetabledescribestheSCOUTroleandliststherolefunctions,positionandexpectations. SCOUT Position NORMAL Functions USV issrt.contexts.littoral.USVScoutScript Operator issrt.contexts.littoral.OCUScoutScript Functionsalsoperformedbyhumanoperator Expectations UnmannedSurfaceVehicle Name Value Default script.cartographerEnabled TRUE script.caution HIGH script.haltOnAnomaly TRUE User scout.waypoint.0.x 30.345702 scout.waypoint.0.y -87.264215 scout.waypoint.1.x 30.346984 scout.waypoint.1.y -87.265629 Table4:DescriptionoftheOPERATORroleintheLittoralWarfarecontext.ThetabledescribestheOPER-ATORroleandliststheroledimensions:rolefunctions,positionandexpectations. OPERATOR Position HIGH Functions issrt.contexts.littoral.OCUOperatorScript Functionsalsoperformedbyhumanoperator Expectations MissionCommander Name Value Default operator.type SHARED Table5:DescriptionoftheTELEFACTORroleintheLittoralWarfarecontext.ThetabledescribestheTELEFACTORroleandliststheroledimensions:rolefunctions,positionandexpectations. TELEFACTOR Position HIGH Functions issrt.contexts.littoral.USVTelefactorScript Expectations UnmannedSurfaceVehicle Name Value User operator.type SHARED 122

PAGE 134

AppendixB:Continued Table6:DescriptionoftheRELAYroleintheLittoralWarfarecontext.ThetabledescribestheRELAYroleandliststherolefunctions,positionandexpectations. RELAY Position NORMAL Functions issrt.contexts.littoral.UAVRelayScript Expectations Forthisdemonstration,theRELAYissimplyamarkerrole,sincetheUAVisnotcapableofautonomousight.Thus,ithasnospecicexpectationsforthisdemonstration. Table7:DescriptionoftheCOMMANDERroleintheLittoralWarfarecontext.ThetabledescribestheCOMMANDERroleandliststheroledimensions:rolefunctions,positionandexpectations. COMMANDER Position HIGHEST Functions issrt.contexts.littoral.MCCommanderScript Expectations MissionCommander Name Value Default searai.scout.waypoint.0.x 30.345702 searai.scout.waypoint.0.y -87.264215 searai.scout.waypoint.1.x 30.346984 searai.scout.waypoint.1.y -87.265629 searai.drone.waypoint.0.x 30.345702 searai.drone.waypoint.0.y -87.264215 searai.courier.waypoint.0.x 30.346314 searai.courier.waypoint.0.y -87.263753 searai.base.waypoint.0.x 30.347936 searai.base.waypoint.0.y -87.264331 Table8:DescriptionoftheSICKroleintheLittoralWarfarecontext.ThetabledescribestheSICKroleandliststheroledimensions:rolefunctions,positionandexpectations. SICK Position HIGHEST Functions UnmannedSurfaceVehicle issrt.contexts.littoral.SickScript UnmannedAerialVehicle issrt.contexts.littoral.SickScript Operator issrt.contexts.littoral.SickScript Expectations Attempttorecoverfromafault.Relinquishotherrolesandreturntobaseifthisisnotpossible. 123

PAGE 135

AppendixB:Continued UncertainTerrain Table9:DetailsoftheDRONEroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. DRONE Position HIGH Functions issrt.contexts.terrain.UGVDroneScript Expectations UnmannedSurfaceVehicle Name Value Default script.cartographerEnabled TRUE script.caution LOW User drone.waypoint.0.x 28.05725 drone.waypoint.0.y -82.41466 base.waypoint.0.x 28.05725 base.waypoint.0.y -82.41466 Table10:DetailsoftheSCOUTroleintheUncertainTerraincontext.Thetableliststherolefunctions,positionandexpectations. SCOUT Position NORMAL Functions issrt.contexts.terrain.UGVPassiveObstacleMappingScript Expectations UnmannedSurfaceVehicle Name Value Default script.cartographerEnabled TRUE script.caution HIGH script.haltOnAnomaly TRUE User scout.waypoint.0.x 28.05693 scout.waypoint.0.y -82.41434 scout.waypoint.1.x 28.05725 scout.waypoint.1.y -82.41434 scout.waypoint.2.x 28.05693 scout.waypoint.2.y -82.41466 scout.waypoint.3.x 28.05693 scout.waypoint.3.y -82.41434 124

PAGE 136

AppendixB:Continued Table11:DetailsoftheOPERATORroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. OPERATOR Position HIGH Functions issrt.contexts.terrain.OCUOperatorScript Expectations MissionCommander Name Value Default operator.type SHARED sheldon.scout.waypoint.0.x 28.05693 sheldon.scout.waypoint.0.y -82.41434 sheldon.scout.waypoint.1.x 28.05725 sheldon.scout.waypoint.1.y -82.41434 sheldon.scout.waypoint.2.x 28.05693 sheldon.scout.waypoint.2.y -82.41466 sheldon.scout.waypoint.3.x 28.05693 sheldon.scout.waypoint.3.y -82.41434 sheldon.drone.waypoint.0.x 28.05725 sheldon.drone.waypoint.0.y -82.41466 sheldon.base.waypoint.0.x" 28.05725 sheldon.base.waypoint.0.y" -82.41466 Table12:DetailsoftheTELEFACTORroleintheUncertainTerraincontext.Thetableliststhefollowingroledimensions:rolefunctions,positionandexpectations. TELEFACTOR Position HIGH Functions issrt.contexts.terrain.UGVTelefactorScript Expectations UnmannedSurfaceVehicle Name Value User operator.type ACTIVE 125

PAGE 137

AppendixC:ImplementationClasses ThischapterdescribestheimplementationoftheCONTEXTADAPTERfollowingtheapproachdelin-eatedinChapterThree.Figure46showsablockdiagramofthecomponentsthatcomprisetheCONTEXT-ADAPTERservice.Thisdiagramisahigh-levelmaptotheimplementationdescribedinthefollowingsec-tions.SectionCdescribesthelanguageandenvironmentusedfortheimplementation.SectionCdescribesthedesignandimplementationoftheCONTEXTADAPTERserviceitself.SectionCandSectionCdescribehowrolesandexpectationsinteractinthesystemandlisttheclassesthatcomprisetheROLEMANAGERandEXPECTATIONMANAGER.Finally,theSCRIPTMANAGERservicedesignandimplementationisexaminedinSectionC.ImplementationEnvironmentThesoftwaretoimplementtheCONTEXTADAPTER,ROLEMANAGER,EXPECTATIONMANAGER,CONTEXT-PLUGIN,SCRIPTMANAGERandrelatedutilitieswaswrittenintheJavaprogramminglanguage,usingthereferencevirtualmachinefromSunMicrosystems.Table13summarizeskeymetricsrelatedtotheimple-mentation. Table13:Summaryofsourcecodemetricsfortheimplementation.Codelinesarethenumberoflinesofcodeinallsourceles,includingcommentlines.Thesizeisthesizeinkilobytesofthebyte-compiledcode.Theremaybesomeerrorduetorounding. SourceArea CodeLines Sizebytes Types Files Methods Fields Statements Core249675K171515955550 Roles160342K976638381 Expectations93627K13104327187 Script160543K17149739289 LWContext3633113K201812290932 UTContext3639108K129831121190 GUI11011373K70244313613475 Other260581K28229944669 Total 27528 872K 186 119 1100 766 7673 Table13listslinesofcode,compiledcodesizeinbytes,andthenumberofdatatype,les,methods,eldsandstatementsimplemented.Cautionmustbeusedwhendrawinginferencesfromthenumbers,andthesemetricsshouldnotbeusedtodrawanyconclusionswithotherprogramsorprojects.Thelinesofcodeincludethecommentsinthesourceles,andthusdonotdirectlymeasuretheactualcodeimplemented. 126

PAGE 138

AppendixC:Continued Figure46:High-levelblockdiagramfortheCONTEXTADAPTERservice. 127

PAGE 139

AppendixC:Continued Thenumberofsourcestatementsisabetterinternalmetricofcodesize,andprovidesthethreeinterestingobservations: Thecoreoftheimplementation,theCONTEXTADAPTER,ROLEMANAGER,EXPECTATIONMANAGERandSCRIPTMANAGER,comprisesarelativelysmallpartoftheoverallcode,at1407statements. ThecodetoimplementthetwosocialcontextsUTandLWis2122statements. Thegraphicalinterfacesuse3475statements.Thesemetricsarenotsurprising;thecoreimplementationisgenericanddesignedtoimplementtherolemechanismsandproperlyinterprettheroledimensionswhichapplytoanyroleinanycontext.Thesocialcontexts,mustbemorespecicastheymustimplementactualroledimensionsthescriptstoimplementthefunctionalityofalltheroles,aswellaslogictohandletheeffectofsocialexpectationsontheroles.TheCONTEXTADAPTERServiceThissectiondescribestheimplementationoftheCONTEXTADAPTER.FigureCshowsaUMLdiagramthatrelatestheclassesintheCONTEXTADAPTERwiththeEXPECTATIONMANAGERandROLEMANAGER.TheserviceentriesareshowninFigure48;theseserviceentriespostinformationabouttheCONTEXT-ADAPTERintotheJinidistributedsystem.Figure49showsaUMLdiagramshowingtherelationoftheCONTEXTADAPTER-relatedeventsthatarepotentiallysentbytheserviceimplementation,andFigure50showstheexceptionsthatcanbethrownduringerrorsorotherexceptionalconditionsintheservice.issrt.contexts.*PackageDescriptionTheissrt.contextspackageistheprimarypackagegroupingtheserviceclassesfortheContextAdapterservice.Thefollowingarekeyclassesincludedintheissrt.contextspackage: CONTEXTADAPTER: PublicinterfacefortheCONTEXTADAPTER.Thecontextadapterprovidesaninter-faceforexternalclientstoassign,activate,deactivateandreliqushroles,aswellasrequesttherobotenterorleavecontexts.Additionally,thisinterfaceprovidesthecapabilitytoprovideuserexpecta-tionsandlinkthemtoaspecicrole.Theinterfacealsoallowsforqueryingforrole,scriptorcontextinformation. 128

PAGE 140

AppendixC:Continued Figure47:UMLdiagramshowingrelationshipsofkeyclassesintheCONTEXTADAPTERservice. Figure48:UMLdiagramshowingtherelationshipoftheJiniserviceentriesfortheCONTEXTADAPTER,SCRIPTMANAGER,ROLE,andSCRIPTclasses. 129

PAGE 141

AppendixC:Continued Figure49:UMLfortheevent-relatedclasses.Theclassesrepresentrole-,expectation-,andscript-relatedevents. Figure50:UMLfortheexception-relatedclasses.Theclassesrepresentrole-,expectation-,andscript-relatedexceptions. CONTEXTPLUGIN: TheCONTEXTPLUGINinterfacepresentsasetofmethodsforinteractingwithaCON-TEXTPLUGINIMPL.Thekeymethodsofthisinterfaceallowtheadditionofrolestothecontextandmethodstoenterorleavethecontext,makingtherolesavailableorhiddenfromtheCONTEXT-ADAPTER.AdditionallytheCONTEXTPLUGINinterfaceallowsqueryingforinformationaboutthecontextandmanipulatingdefaultroleandcontextexpectations. REMOTECONTEXTADAPTER: TheREMOTECONTEXTADAPTERextendstheCONTEXTADAPTERinter-face,turningthemethoddenitionsintoremotemethodsandallowingdistributedclientstoclassserivcemethods. CONTEXTADAPTERENTRY: ACONTEXTADAPTERENTRYistheentryforthecontextadapterintheJiniphonebook.Thatis,itisawrapperfortheinformationpostedtotheregistrarthatrepresentsthecapabilitesoftheCONTEXTADAPTER.Thisinformationincludestheactive,assignedandknownroles,aswellasinformationaboutknownsocialcontexts. 130

PAGE 142

AppendixC:Continued CONTEXTADAPTERIMPL: TheCONTEXTADAPTERIMPListheactualimplementationoftheCONTEXT-ADAPTERservice.Theprimarydutiesoftheimplementationistoconstructandmaintaintheservice,andtoroutemessagesappropriatelyfromoutsidecallersorcontextstotheROLEMANAGERorlinkEXPECTATIONMANAGER. CONTEXTADAPTERPROXY: TheCONTEXTADAPTERPROXYisthesmartproxyfortheCONTEXT-ADAPTERservice.ThedefaultimplementationsimplydelegatescallstotheunderlyingCONTEX-TADAPTERSERVER. CONTEXTADAPTERSERVER: TheCONTEXTADAPTERSERVERisthedefaultserverimplementationoftheCONTEXTADAPTERservice.Thisimplementationisbasic;itsimplydelegatescallsthroughtotheunderlyingimplementation. CONTEXTCHANGEEVENT: TheCONTEXTCHANGEEVENTisredbytheCONTEXTADAPTERservicewhenacontextisenteredorleft. CONTEXTINFO: TheCONTEXTINFOisalightweight,serializablewrapperaroundinformationrelatingtoasocialcontext.ACONTEXTINFOinstanceuniquelyidentiesacontexttoaCONTEXTADAPTER. CONTEXTPLUGINIMPL: TheCONTEXTPLUGINIMPListheimplementationoftheCONTEXTPLUGINin-terface,andimplementsasetofmethodsforinteractingwithasocialcontext.Thekeymethodsofthisclassallowtheadditionofrolestothecontextandmethodstoenterorleavethecontext,makingtherolesavailableorhiddenfromtheCONTEXTADAPTER.AdditionallytheCONTEXT-PLUGINinterfaceallowsqueryingforinformationaboutthecontextandmanipulatingdefaultroleandcontextexpectations.Thereisoneinstanceoftheisclasspersocialcontext. CONTEXTPLUGINPARSER: TheCONTEXTPLUGINPARSERisahelperclassusedinconstructingtheCONTEXT-ADAPTER.TheCONTEXTPLUGINPARSERknowshowtoextractrelevantinformationfromtheXMLcongurationleforthecontextandcreatetheRoleinstancesforthecontext ROLESTATECHANGEEVENT: TheROLESTATECHANGEEVENTisredbytheCONTEXTADAPTERser-vicewhenarole'sstatechanges. 131

PAGE 143

AppendixC:Continued Theissrt.contexts.littoralpackagecontainsthescriptsandcontextinformationusedtoenableoperationintheLittoralsocialcontext.Thefollowingarekeyclassesincludedintheissrt.contexts.littoralpackage: MCCOMMANDERSCRIPT: TheMCCOMMANDERSCRIPTimplementsthefunctionsoftheMissionCon-trolagent'sCOMMANDERrole. OCUCONTEXTCONTROLLER: TheOCUCONTEXTCONTROLLERistheimplementationoftheCONTEXT-PLUGINfortheLittoralcontext. OCUOPERATORSCRIPT: TheOCUOPERATORSCRIPTistheimplementationoftheOperatorControlUnit'sOPERATORroleintheLittoralcontext. OCUSCOUTSCRIPT: TheOCUSCOUTSCRIPTistheimplementationoftheOperatorControlUnit'sSCOUTroleintheLittoralcontext. SICKSCRIPT: TheSICKSCRIPTclassistheimplementationofthefunctionsoftheSICKrole. UAVCONTEXTCONTROLLER: TheUAVCONTEXTCONTROLLERistheimplementationoftheCONTEXT-PLUGINfortheLittoralcontext. UAVRELAYSCRIPT: TheUAVRELAYSCRIPTistheimplementationoftheUAV'sRELAYroleintheLit-toralcontext. USVCONTEXTCONTROLLER: TheUSVCONTEXTCONTROLLERistheimplementationoftheCONTEXT-PLUGINfortheLittoralcontext. USVCOURIERSCRIPT: TheUSVCOURIERSCRIPTistheimplementationoftheUSV'sCOURIERroleintheLittoralcontext. USVDRONESCRIPT: TheUSVDRONESCRIPTistheimplementationoftheUSV'sDRONEroleintheLittoralcontext. USVSCOUTSCRIPT: TheUSVSCOUTSCRIPTistheimplementationoftheUSV'sSCOUTroleintheLittoralcontext. USVTELEFACTORSCRIPT: TheUSVTELEFACTORSCRIPTistheimplementationoftheUSV'sTelefactorroleintheLittoralcontext. 132

PAGE 144

AppendixC:Continued Theissrt.contexts.terrainpackagecontainsthescriptsandcontextinformationusedtoenableoperationintheTerrainUTsocialcontext.Thefollowingarekeyclassesincludedintheissrt.contexts.littoralpackage: OCUCONTEXTCONTROLLER: TheOCUCONTEXTCONTROLLERistheimplementationoftheCONTEXT-PLUGINforthehumanoperatorintheTerraincontext. OCUOPERATORSCRIPT: TheOCUOPERATORSCRIPTistheimplementationoftheOperatorControlUnit'sOPERATORroleintheLittoralcontext. UGVCONTEXTCONTROLLER: TheUGVCONTEXTCONTROLLERistheimplementationoftheCONTEXT-PLUGINfortheUGVintheTerraincontext. UGVDRONESCRIPT: TheUGVDRONESCRIPTistheimplementationoftheUGV'sDRONEroleintheTerraincontext. UGVPASSIVEOBSTACLEMAPPINGSCRIPT: TheUGVPASSIVEOBSTACLEMAPPINGSCRIPTistheim-plementationoftheUGV'sSCOUTroleintheLittoralcontext. UGVTELEFACTORSCRIPT: TheUGVTELEFACTORSCRIPTistheimplementationoftheUGV'sTELE-FACTORroleintheLittoralcontext.RolesandtheROLEMANAGERThissectiondescribestheimplementationoftheROLEMANAGER.Figure51showsaUMLdiagramthatrelatestheclassesintheROLEMANAGERwiththeEXPECTATIONMANAGERandCONTEXTADAPTER.TherolelifecycleisshowsasaFSMinFigure10andFigure11showstheFSM,highlightingthepor-tionsofthelifecyclethatchecktheEXPECTATIONMANAGERforexpectationsinthesocialcontextwhentransitioningbetweenstates.issrt.roles.*PackageDescriptionThefollowingarekeyclassesincludedintheissrt.rolespackage: 133

PAGE 145

AppendixC:Continued Figure51:UMLdiagramshowingrelationshipsofimportantclasseswithintheROLEMANAGERintheCONTEXTADAPTERservice. 134

PAGE 146

AppendixC:Continued ROLE: ARoleistheprimaryabstractionoftheCONTEXTADAPTER.InstancesofthisclassencapsulatethestaticdescriptionofaROLEandalsocontainrun-timeinformationaboutthestateoftherole.Thestaticdescriptionofaroleincludestherole'spositioninasocialhierarchyhererepresentedbyaROLEPRIORITY,functionsandtaskshereimplementedbyaSCRIPT,andsocialexpectationsrepresentedasalistofEXPECTATIONs.RolesarecontrolledandmanagedbytheROLEMANAGER. ROLEENTRY: AROLEENTRYistheentryfortheROLEintheJiniphonebook.Thatis,itisawrapperfortheinformationpostedtotheregistrarthatrepresentsthecapabilitesandstateoftheRole. ROLEINFO: TheROLEINFOisalightweight,serializablewrapperaroundinformationrelatingtoaROLE.AROLEINFOinstanceuniquelyidentiesaroleonaparticularagent. ROLESTATECHANGELISTENER: TheROLESTATECHANGELISTENERinterfacemustbeimplementedbyanyobjectinterestedinreceivingnoticationwhenroleschangestate.Thefollowingarekeyclassesincludedintheissrt.roles.eventspackage: ROLEEVENT: TheROLEEVENTisredwheninterestingeventsoccurtoorbecauseofaROLE.Thefollowingarekeyclassesincludedintheissrt.roles.managerpackage: ROLEEXCEPTION: AnROLEEXCEPTIONisthrownwhenanerrorisencounteredwithanoperationrelatedtovariousroleactivities,suchasassignment,activation,deactivationorrelinquisingarole.Thefollowingarekeyclassesincludedintheissrt.roles.managerpackage: ROLEMANAGER: TheROLEMANAGERistheentitythatisprimarilyresponisbleforhandlingdynamicactivitiesrelatingtoRoles.Inparticular,theROLEMANAGERisresponsibleforactivating,deacti-vating,assigningandrelinquisingroles,aswellasmonitoringthestateofrolesandhandlingandresolvingcertaintypesofconictssurroundingtherolelifecycle.ExpectationsandtheEXPECTATIONMANAGERThissectiondescribestheimplementationoftheEXPECTATIONMANAGER.Figure52showsaUMLdia-gramthatrelatestheclassesintheEXPECTATIONMANAGER. 135

PAGE 147

AppendixC:Continued Figure52:UMLdiagramshowingrelationshipsofimportantclasseswithintheEXPECTATIONMANAGERintheCONTEXTADAPTERservice. 136

PAGE 148

AppendixC:Continued issrt.expectations.*PackageDescriptionThefollowingarekeyclassesincludedintheissrt.expectationspackage: EXPECTATION: InstancesoftheEXPECTATIONclassrepresentsocialexpectationsthatcaninuencethebehaviorofroles.TheconstructionofanexpectationisbasedonKAoS.Anexpectationhasatype,expectedactions,atriggercondition,andcontextandrolethattheexpectationappliesto.Theexpec-tationtypeincludespositiveandnegativeauthorizationsandobligations,withthepositiveobligationstheonlycurrentlysupportedtype.Theexpectedactionsarethepropertiesthatgointoeffectwhentheexpectationistriggered.Thetriggerstatusisaconditionthataffectswhenexpectationsareapplied. EXPECTATIONCHANGELISTENER: TheEXPECTATIONCHANGELISTENERinterfacemustbeimplementedbyanyobjectinterestedinreceivingEXPECTATIONEVENTs. EXPECTATIONCOGNIZANT: AnobjectthatisEXPECTATIONCOGNIZANTsignalsthatitknowshowtoprocessEXPECTATIONupdates.Thefollowingarekeyclassesincludedintheissrt.expectations.eventspackage: EXPECTATIONEVENT: TheEXPECTATIONEVENTisredwhennewexpectationsareappliedtoarole.Thefollowingarekeyclassesincludedintheissrt.expectations.exceptionspackage: EXPECTATIONEXCEPTION: AnEXPECTATIONEXCEPTIONisthrownwhenanerrorisencounteredwithanoperationrelatedtoEXPECTATIONs.ThismaybetheincompatibleassignmentofconictingEXPECTATIONorsomeothererror.Thefollowingarekeyclassesincludedintheissrt.expectations.managerpackage: EMFACTORY: EMFACTORYisafactoryclassthatconstructsandreturnstheappropriatetypeofEXPECTATION-MANAGERbasedonthegivenargument.BydefaultthisreturnsaSIMPLEEXPECTATIONMANAGER,butitalsoknowshowtoconstructaKAOSEXPECTATIONMANAGER. EXPECTATIONMANAGER: EXPECTATIONMANAGERistheabstractsuperclassofallexpectationman-agers.AnEXPECTATIONMANAGERisresponsibleformaintainingandprocessingexpectations.TheinterfacetothisclassallowsqueryingforEXPECTATIONsforaRole. 137

PAGE 149

AppendixC:Continued Thefollowingarekeyclassesincludedintheissrt.expectations.manager.kaospackage: KAOSEXPECTATIONMANAGER: AKAOSEXPECTATIONMANAGERisanEXPECTATIONMANAGERthatusestheKAoSagentsystemtohandleandtrackEXPECTATIONsintheformofpolicies.Thefollowingarekeyclassesincludedintheissrt.expectations.manager.simplepackage: SIMPLEEXPECTATIONMANAGER: ASIMPLEEXPECTATIONMANAGERisanEXPECTATIONMANAGERthatisprimarilyadevelopmentandtestingimplementationofanEXPECTATIONMANAGERthatpro-videsdefaultEXPECTATIONs.ScriptsandtheSCRIPTMANAGERServiceThissectiondescribestheimplementationoftheSCRIPTMANAGER.Figure53showsaUMLdiagramthatrelatestheclassesintheSCRIPTMANAGER.ThelifecycleofaSCRIPTinstanceisshownasaFSMinFigure54.issrt.scripts.*PackageDescriptionTheissrt.scriptspackagecontainsdenitionsforhigh-levelscriptclassesandsupportingclasses.Thefollowingarekeyclassesincludedintheissrt.scriptspackage: SCRIPT: ASCRIPTisahigh-levelmechanismforcoordinatingcomplexaction.InDFRA,theyareusedtosequenceandcontrollower-levelrobotbehaviorsandotherscripts. SCRIPTENTRY: ASCRIPTENTRYistheentryforaSCRIPTintheJiniphonebook.Thatis,itisawrapperfortheinformationpostedtotheregistrarthatrepresentsthecapabilitesandstateoftheSCRIPT. SCRIPTINFO: TheSCRIPTINFOisalightweight,serializablewrapperaroundinformationrelatingtoaSCRIPT.ASCRIPTINFOinstanceuniquelyidentiesascripttoaSCRIPTMANAGER.Theissrt.scripts.eventspackagecontainsdenitionsforeventsthataregeneratedbyscriptsandthescriptmanager.Thefollowingarekeyclassesincludedintheissrt.scripts.eventspackage: 138

PAGE 150

AppendixC:Continued Figure53:UMLdiagramshowingrelationshipsofimportantclassesintheCONTEXTADAPTERservice. 139

PAGE 151

AppendixC:Continued Figure54:FSMshowingtheobjectlifecycleforaSCRIPT. ANOMALYDETECTEDEVENT: TheANOMALYDETECTEDEVENTisredwhenascriptthattriedtodetectanomaliesdoesso.Registeredlistenersarenotiedoftheoccurence,aswellasthelocationofthedetectedanomalyifavailable. SCRIPTEVENT: TheSCRIPTEVENTisredwheninterestingeventsoccurtoorbecauseofachangetoaSCRIPT. SCRIPTSTATEEVENT: SCRIPTSTATEEVENTisredwhenascriptchangesstate. WAYPOINTGOALSUPDATEDEVENT: TheWAYPOINTGOALSUPDATEDEVENTisredwhenascript'sway-pointgoalsareupdated.Theeventcontainsthenewlistofgoals.Theissrt.scripts.exceptionspackagecontainsdenitionsforexceptionsthataregeneratedbyscriptsandtheSCRIPTMANAGER.Thefollowingarekeyclassesincludedintheissrt.scripts.exceptionspackage: SCRIPTEXCEPTION: AnSCRIPTEXCEPTIONisthrownwhenanerrorisencounteredwithanoperationrelatedtovariousscriptactivities. 140

PAGE 152

AppendixC:Continued Theissrt.scripts.managerpackagecontainsdenitionsfortheSCRIPTMANAGERserviceandfortheimplementationthatcontrolsandcoordinatesexecutingscripts.Thefollowingarekeyclassesincludedintheissrt.scripts.managerpackage: REMOTESCRIPTMANAGER: TheREMOTESCRIPTMANAGERextendstheSCRIPTMANAGERinterface,turningthemethoddenitionsintoremotemethodsandallowingdistributedclientstoclassservicemethods. SCRIPTMANAGER: PublicinterfacefortheSCRIPTMANAGER.TheSCRIPTMANAGERisresponsiblefordirectingandcontrollingregisteredscripts.Theinterfaceallowsclientstoregisterandunregisterscripts,aswellasreqestingtheinitialization,executionandshutdownofregisteredscripts. SCRIPTMANAGERENTRY: ASCRIPTMANAGERENTRYistheentryfortheSCRIPTMANAGERserviceintheJiniphonebook.Thatis,itisawrapperfortheinformationpostedtotheregistrarthatrepresentsthecapabilitesoftheSCRIPTMANAGER.Thisinformationincludestheactiveandknownscripts. SCRIPTMANAGERIMPL: TheSCRIPTMANAGERIMPListheactualimplementationoftheSCRIPTMAN-AGERservice.Theprimarydutiesoftheimplementationistoconstructandmaintaintheservice,andtomanagetheexecutionandstateofregisteredScriptinstances. SCRIPTMANAGERPROXY: TheSCRIPTMANAGERPROXYisthesmartproxyfortheSCRIPTMANAGERservice.ThedefaultimplementationsimplydelegatescallstotheunderlyingSCRIPTMANAGERSERVER. SCRIPTMANAGERSERVER: TheSCRIPTMANAGERSERVERisthedefaultserverimplementationoftheSCRIPTMANAGERservice.Thisimplementationisbasic;itsimplydelegatescallsthroughtotheunderlyingimplementation. 141

PAGE 153

AbouttheAuthor MatthewLongisaPh.D.candidateinComputerScienceandEngineeringattheUniversityofSouthFlorida.HeholdsaBSinMathematicalandComputerSciencesfromtheColoradoSchoolofMinesandaMSinComputerScienceandEngineeringfromtheUniversityofSouthFlorida.Mr.Long'sresearchinterestsincludemobilerobots,articialintelligence,distributedsystems,softwareagentsandsoftwareengineering.HehastaughtacourseinarticialintelligenceattheUniversityofSouthFloridaandhasgiveninvitedtalksfortheInstituteforElectricalandElectronicsEngineersandtheObjectModelingGroup.