附錄2AnIntegratedSystemforCommandandControlofCooperativeRoboticSystemsChristopherM.Clark,EricW.Frew,HenryL.Jones,&StephenM.RockAerospaceRoboticsLabDepartmentofAeronautics&AstronauticsStanfordUniversitychrisc,ewf,hlj,Abstract:Presentedisanexperimentalinvestigationintothreeissuesthatenableincreasedautonomousfunctionalitywhenusingmobilerobots.Theseissuesare(1)interfacingasingleuserwithmultiplerobots,(2)motionplanningformultiplerobots,and(3)robottrajectorygenerationfortargettracking.Forthisresearch,theMicroAutonomousRoverS(MARS)testplatformwasdevelopedthatprovidesameansforimplementingthistechnologyonlaboratoryrobotstocarryouttasksincludingcollision-freewhichasingleuserisabletocommandagroupofrobots.Experimentalresultsarepresentedinmotionandtargettracking.Keyword:multi-robotsroverMARStestplatformtargettrackingtrajectorymonitorMaintext:1.IntroductionCurrently,remoteroboticsystemsrequiremanyhumanstooperateasinglerobot.Thegoalforfuturesystemsistorequireonlyoneoperatorformanyrobots.Forexample,futurespacestructureconstructionwouldbenefitfromtheavailabilityofalargegroupofrobotsthatcanbeoperatedbyaFigure1:RoversfromtheMARStestplatform.Smallgroupsofhumans.Whiletherehasbeenasignificantamountofresearchtowardstheoperationofsingleremoterobots,moreworkisstillrequiredtowardstheoperationofgroupsofrobots.Inparticular,anincreaseddegreeofautonomymustbegiventotherobots.Torealizethisautonomy,avarietyoffundamentalcapabilitiesmustbeenabledthatinclude:(1)Providinganinterfacefromwhichadequateinformationfordecision-makingisavailabletothehuman,andcommandstooneorseveraloftherobotscanbeexecuted.(2)Providingautonomous,real-timeconstructionofcollision-freetrajectoriesforallrobotsinthegroup.(3)Providingrobottrajectorygenerationthatenablestrackingofmovingobjects.Inpreviouswork,weaddressedeachoftheseissues:Jones6developedaninterfacethatallowsasinglehumantooperatemanyrobots；Clark2designedamulti-robotmotionplanner；andFrew4designedatrajectorygeneratorthatprovidesnear-optimalsolutionsforobjectmotionestimation.Figure1:RoversfromtheMARStestplantformInthispaper,wepresentanintegratedsystemdemonstrationofthesetechnologies.Thepaperisorganizedasfollows.Section2providesabriefdescriptionofeachofthethreeresearchprojects.InSection3,theMicroAutonomousRoverS(MARS)testplatformanditsapplicationtothisresearchisdescribed.Section4detailsafinaldemonstrationthatsummarizesthepreviousresearchasanintegratedsystem.ConclusionsarepresentedinSection5.2.TechnologicalIssues2.1.HumanInterfaceForasinglehumantooperateamulti-robotsystem,thehumanmusthaveaccesstoallrelevantinformationabouttheremoteenvironmentsothatappropriatecommandscanbeexecuted.Also,thehumanmustbeprovidedwithameansofexecutingthesecommandstooneorseveraloftherobots.In6,Jonesdevelopedaninterfacebasedondialoguesbetweenthehumanandtherobotsasaneffectivemethodforoperatingmultiplerobots.Inparticular,Jonesaddressedthefollowingissues:-Establishingthestructureandscopeofthedialogue-Creatingarobotinfrastructurecapableofconductinganeffectivedialogue-Determiningmethodsfordealingwiththesocialconventionsofdialogues-Developinganinterfacethatallowstheoperatortocarryoutthedialoguewiththeroboticsystem.Hisresultisanimplementationofadialogueinteractionpatternedafterthetask-orientedDialoguescommoninhumanteams.Thehypothesisisthatsimilardialoguescanplayausefulrolewithinhuman-robotteams.TheinterfacewasimplementedusingOpenGLtoprovideathree-dimensionalviewoftherobotenvironment.Anexamplescreen-shotisshowninFigure2.Dialoguethroughtheinterfacetakesplaceelectronicallyratherthanthroughvoice.Thedialoguebeginswhenobjectsareselectedbyclickingonthemonthescreen.Theinterfacethenresolvestheidentityoftheobjectandtherobotthatsensedtheobject.Theinterfacewaitsuntilaresponsefromthecorrectrobothasbeenreturnedintheformofalistoftasksthattherobotcanaccomplishonthatobject.Thislististhendisplayedinadialogthatpopsupnexttotheobject.Theusercanselectfromthislistoftasks,andthecompletecommandofrobot/task/objectissenttotherobotforexecution.Figure2:Screenshotofthehuman-robotinterface.Thescreen-shotinFigure2providesanexampleoftheinterfaceinuse.Tworobots(denotedbywhitecylinders)and3objectsarelocatedonthetestbedworkspace.Theoperatorhasqueriedtherobotagentstodeterminewhattaskscanbeperformedontheobjectattheleftsideofthescreen.Apop-upmenuhasappearedwithalistofthequeryresults.Jonesshowedthatitispossibletobuildadialogue-basedinteractionthatenablesthecontrolofmultiplerobots.Thisinteraction,asimplementedinavirtualthree-dimensionalworld,providedanintuitivepoint-and-clickmethodfordeterminingthecapabilitiesoftherobotintheappropriatecontext,andenabledtheoperatortoparticipateintheresourcemanagementandtaskplanningfortherobots.2.2MotionPlanningWhenlargegroupsofrobotsandmovingobstaclesareworkingtogetherwithinadesignatedarea,high-levelmotionplanningisrequiredtoavoidcollisions.Continuouscommunicationbetweenallrobotsmaynotbefeasible,andnosystemofsensorscanprovideglobalknowledge.Also,tofunctioninadynamicenvironmentwithmovingobstacles,thesystemmustbeabletoreactquickly.Forthistypeofmulti-robotsystem,amotionplannerthatdoesnotneedglobalknowledgeorhighbandwidthcommunication,butthatcanstillplaninreal-time,isrequired.AmotionplanningsystemthatmeetsthisrequirementwasdevelopedbyClark2.ThealgorithmpresentedwasbasedontheplannerdevelopedbyHsuandKindel5.TheirworkdemonstratestheuseofaKinodynamicRandomizedMotionPlannerforasinglerobotmaneuveringaroundstationaryandmovingobstacles.Tohandlemorethanonerobot,theKinodynamicRandomizedMotionPlannerwasextended.Intheextendedplanner,whenrobotsdetectoneanotherusinglocalsensors,theycommunicatewitheachOther,usingaprioritysystem,therobotscoordinatetheirmotionplanstoavoidcollisions.Eachrobotcreatesaplanwithknowledgeofonlythefewobstaclessurroundingit.Byplanningaroundonlythoseobjectswithintherobotslocalarea,themotionplanningproblemisgreatlysimplifiedleadingtodecreasedplanningtimes.Whennewobjectsentertherobotsfieldofview,are-planiscalledfortoensurethattherobotstrajectoryiscollision-free.Anexampleofasimulationinvolving10rovers,and5stationaryobstaclesisprovidedifFigure3.Smallercirclesrepresentthemicro-roversasviewedfromabove,whilecrossesrepresentgoallocationsandlargercirclesrepresentobstacles.Trajectoriesconstructedbyeachrobotsmotionplannerareindicatedwithlinesthatleadtogoallocations.Themotionplannerdemonstrateditseffectivenessinplanningforalargenumberofrobotswithinaboundedworkspace.Itplannedwithahighprobabilityofsuccess,eveninclutteredenvironmentsinvolving5to15robots,Stationaryobstaclesandmovingobstacles.Planningtimesontheorderof0.1sallowedtherobotstore-planinreal-timeandreactquicklytochangesintheenvironment.Althoughthemotionplannerwasappliedtoa2Dworkspace,itshouldbenotedthattheplannerisextendibleto3Dworkspaces.Figure3:MotionplanningsimulationexampleInvolvingin10robotsand5obstacles.Figurea)illustratesrovers,theirgoals,andobstaclesbeforethesimulation.Inb),thesimulationhasjustbegunandallrovershaveconstructedtheirfirstplan.Examplesofreal-timeplanningontheflyareshowninc)andd).(Notethatsomeinitialtrajectoriespassthroughobstacles,butastherobotscomecloseenoughtosensethem,theyreplanttoavoidthem.)Finally,ine)therovershavecon