英语原文.docx
《英语原文.docx》由会员分享,可在线阅读,更多相关《英语原文.docx(11页珍藏版)》请在冰豆网上搜索。
英语原文
FuzzyLogicBasedAutonomousSkidSteeringVehicleNavigation
L.Doitsidis,K.P.Valavanis,N.C.Tsourveloudis
TechnicalUniversityofCrete
DepartmentofProductionEngineeringandManagement
Chania,Crete,GreeceGR-73100
{Idoitsidis,kimonv,nikost}@dpem.tuc.gr
Abstract-Atwo-layerfuzzylogiccontrollerhasbeendesignedfor2-DautonomousNavigationofaskidsteeringvehicleinanobstaclefilledenvironment.ThefirstlayeroftheFuzzycontrollerprovidesamodelformultiplesonarsensorinputfusionanditiscomposedoffourindividualcontrollers,eachcalculatingacollisionpossibilityinfront,back,leftandrightdirectionsofmovement.Thesecondlayerconsistsofthemaincontrollerthatperformsreal-timecollisionavoidancewhilecalculatingtheupdatedcoursetobeapplicabilityandimplementationisdemonstratedthroughexperimentalresultsandcasestudiesperformedoarealmobilerobot.
Keywords-Skidsteering,mobilerobots,fuzzynavigation.
Ⅰ.INTRODUCTION
Theexistseveralproposedsolutionstotheproblemofautonomousmobilerobotnavigationin2-Duncertainenvironmentsthatarebasedonfuzzylogic[1],[2],evolutionaryalgorithms[3],aswellasmethodscombiningfuzzylogicwithgeneticalgorithms[4]andfuzzylogicwithelectrostaticpotentialfields[5].
Thepaperistheoutgrowthofrecentlypublishedresults[9],[10],butitstudies2-Denvironmentsnavigationandcollisionavoidanceofaskidsteeringvehicle.Skidsteeringvehiclesarecompact,light,requirefewpartstoassembleandexhibitagilityfrompointturningtolinedrivingusingonlythemotions,components,andsweptvolumeneededforstraightlinedriving.
Skidsteeringvehiclemotiondiffersfromexplicitsteeringvehiclemotioninthewaytheskidsteeringvehicleturns.Thewheelsrotationislimitedaroundoneaxisandthebackofsteeringwheelresultsinnavigationdeterminedbythespeedchangeineithersideoftheskidsteeringvehicle.Samespeedineithersideresultsinastraight-linemotion.Explicitsteeringvehiclesturndifferentlysincethewheelsaremovingaroundtwoaxes.ThegeometricconfigurationofaskidsteeringvehicleintheX-YplaneisshowninFig1,whileatistheheadingangle,Wistherobotwidth,θthesenseofrotationandS1,S2arethespeedsintheeithersideoftherobot.
Thederivedandimplementedplanneratwo-layerfuzzylogicbasedcontrollerthatprovidespurely”reactivebehavior”ofthevehiclemovingina2-Dobstaclefilledenvironment,withinputsreadingsfromaringof24sonarsensorsandangleerrors,andoutputstheupdatedrotationalandtranslationalvelocitiesofthevehicle.
Ⅱ.DESIGNOFTHEFUZZYLOGICCONTROLSYSTEM
Theordertothevehiclemovement,atwo-layerMadman-typecontrollerhasbeendesignedandimplemented.Inthefirstlayer,therearefourfuzzylogiccontrollersrepondibleforobstacledetectionandcalculationofthecollisionpossibleilitiesinthefourmaindirections,front,back,leftandright.Thepossibilitiescalculatedinthefirstlayeraretheinputtothesecondlayeralongwiththeangleerror(thedifferencebetweentherobotheadingangleandthedesiredtargetangle),andtheoutputistheupdatedvehicle’stranslationalandtherotationalspeed.
Fig.1.GeometricconfigurationoftherobotintheX-Yplane.
A.firstlayerofthefuzzylogiccontroller
TheATRV-miniisequippedwithanarrayof24ultrasonicsensorsthatarevehiclesasshowninFig.2.TheultrasonicsensorsthatareusedaremanufacturedbyPolaroid.
Afterexperimentwith,andtestingseveralmethodsconcerningsonarsensordategroupingandmanagement,itwasfirstdecidedtofollowthesensorgroupinginpairsasproposedin[8](consideringtheATRV–minitwelvesonargroupAis=1,…..,12,havebeenenumeratedasshowninFig.2)andthendividethesunoftheprovidedpairsensordatabytwotodeterminethedistancefromthe(potential)obstacle.However,thismethodgaveunsatisfactoryresultsduetotheATRV–minisspecificsensorunreliability.Evenincaseswithobstaclespresentinthevicinityofthevehicle,thesensorsweredetectinga“freepath”.Toovercomethisproblem,amodified,simpler,sensorgroupinganddatamanagementmethodwastestedthatreturnmuchbetterandaccurateresults:
ThesensorswereagaingroupedinpairsaccordingtoFig.2,buttheminimumofthe(potential)obstacle.EachATRV–minisonarreturnsfromobstaclesatamaximumdistanceof4metres(experimentallyverifiedasopposedtodifferentvalueprovidedbythesonarsensorsmanufacturer
Fig.2.GroupingoftheSensors.
Theformofeachfirstlayerindividualfuzzycontroller,includingtheobstacledetectionmodule,isshowninFig.3.ObservingFig.3,datafromgroupsensorsA1,A2,….,A5(5inputs)andgroupsensorsA7,A8,…,A11(5inputs)serveasinputstotheindividualcontrollersresponsibleforthecalculationofthefrontandbackcollisionpossibilities,respectively.DatafromgroupsensorsA5,A6,A7(3inputs)serveasinputtocalculatetheleftandrightpossibilities,respectively.Theindividualfuzzycontrollersutilizethesamemembershipfunctionstocalculatethecollisionpossibilities.Thelinguisticvaluesofthevariabledistance_from_obstancearedefinedtobethree,near,meium_distance,awaywithmembershipfunctionsasshowninFig.4reflectingthemaximumdistanceof4metersasonarreturnsaccurateinformationaboutpotentialobstacles.
Fig.3.Obstacledetectionmodule.
Fig.4.InputVariableDistance_From_Obstacle.
Thefirstlayeroutputisacollisionpossibilityineachdirectiontakingvaluesfrom0to1.Thelinguisticvariablesdescribingeachdirectionoutputvariablecollisionpossibility(withempirically
Derivedforbestperformance)membershipfunctionsasshownInFig.5.ApartoftherulesbaseforleftcollisionispresentedinTableⅠ.
Anexampleoftherulesusedtoextractfrontcollisionpossibilitiesis:
IFA1isnearANDA2isnearANDA3isNearANDA4ismedium_distanceANDA5isnearTHENcollision_possibilityishigh.Similarforthebackcollisionpossibility.Forleft(equivalentlyforrightcollision)possibilitiestheruleisoftheform:
IfA5isnearAndA6isnearAndA7isnearTHENcollision_possibilityishigh.
Fig.5.OutputVariablecollision_possibility
TABLEⅠ
PARTOFTHERULESBASEFORLEFTCOLLISION
InputVariables
Output
Variables
A5
A6
A7
Near
Near
Near
High_Possibility
Away
Away
Away
Not_possible
Near
Away
Medium_Distance
Possible
Near
Away
Near
High_Possibility
B.Secondlayerofthefuzzylogiccontroller
Thesecondlayerfuzzycontrollerrecivesasinputsthefourcollisionpossibilitiesinthefourdirectionsandtheangleerror,andoutputsthetranslationalvelocity,whichisresponsibleformovingthevehiclebackwardorforwardandtherotationalspeed,whichisresponsibleforthevehiclerotationasshowninFig.6.
Theangleerrorrepresentsthedifferencebetweentherobot-headingangleandthedesiredangletherobotshouldhaveinordertoreachitstarget.Theangleerrortakesvaluesrangingfrom-1800to1800.Thelinguisticvariablesthatrepresenttheangleerrorare:
Backwards_1,Hard_Left,Left,right,Hard_right,Backwards_2with(empiricallyderivedfromtests)membershipfunctionsasshowninFig.7.
Thetranslationalvelocity(m/sec),whichisoneoftheoutputsofthesecondlayercontroller,isdescribedwiththefollowinglinguisticvariables:
back_full,back,back_solw,stop,front_slow,front,frontfull,withmembershipfunctionsasinFig.8.
Fig6.Blockdiagramofthe2ndlayerofthefuzzylogiccontroller
Fig7.InputVariableAngleError.
Fig8.OutputVariableTranslational_Velocity.
Therotational_speed(rad/sc)isdescribedwiththefollowinglinguisticvariables:
Right_full,right,no_ratation,left,let_fullwithmembershipfunctionsasinFig.9.
Anexampleoftherulesthatcontrolthevehicleisdemonstrated:
Iffront_collisionisNot_PossibleANDBack_CollisionisNot_possibleAndLeft_CollisionisNot_possibleAndRight_CollisionisNot_possibleAndAngleErrorisAheadTHENTranslational_velocityisFront_FullANDRotational_VelocityisNo_Ratation.
Fig.9.OutputVariableRotational_velocity.
Ⅲ.RUSULTS
ThefuzzylogiccontrollerhasbeendesignedandimplementedusingC++inanATRV-minimanufacturedbyRealWorldInterface(RWI).Inallexperimentstherobotisconsideredtohavereacheditstargetwhenstoppinginsideacirclewithradiusof30cm.Thisassumptionhasbeendictatedbecauseallcalculationshavebeenmaderelativetothecenteroftherobot.Soiftherobotstopsinsidethatcircleitisassumedthatithasreacheditstarget.
Severalscenariosinanindoor2-Dobstaclefilledenvironmenthavebeentestedtostudytherobotbehaviorandthecontroller’sapplicability.
ThearrowinFig.10,Fig.15,Fig.20isshowingtheinitialdirectionofthevehicle.
Intestcase1weexaminethebehaviorofthevehicleinanenvironmentwiththreeobstacles.Thetestcase1ispresentedinFig.10.Fig.11showsthetranslationalvelocity,whiletherotationalvelocityisgiveninFig.12.Fig.13presentsthefrontcollisionpossibility.InFig.14,thesolidlineindicatestheleftcollisionpossibilitywhilethedotedtherightcollisionpossibility.Thebehaviorofthevehicleisdefinedfromthesurroundingobstacles.
Inthebeginningtheleftcollisionpossibilityishighduetotheobstacleintheleft.Therobotmovesforwardsandit’ssteeringrightinordertoavoidtheobstacle.Thenitsteersleftandmovestowardsitstarget.
InthesecondtestcasepresentedinFig.15,amorecomplicatedenvironmentwiththreeobstacleshasbeentested.Fig.16showsthetranslationalvelocity,whiletherotationalvelocityisgiveninFi