Personalnavigationsystembasedondualshoemount

PersonalNavigationSystembasedonDualShoe‐MountedIMUsandIntershoeRangingAlonzoKelly*NationalRoboticsEngineeringCenterCarnegieMellonUniversityPrecisionPersonnelLocatorWorkshop1August2011(ApprovedforPublicRelease,DistributionUnlimited)TheviewsexpressedarethoseoftheauthoranddonotreflecttheofficialpolicyorpositionoftheDepartmentofDefenseortheU.S.Government.2MotivationTechnicalApproachI:InertialSensorsmountedonbootsII:AidingwithZeroVelocityUpdatesIII:AidingwithRangeUpdatesResultsSummaryOutline3Military:LocalizingdismountsinGPSdeniedenvironmentslikecaves,tunnels,canyons,forests,urbancanyonsDualUse:Localizingfirstresponders,mineworkers,civiliansindoorsDARPAGoals:MotivationPhaseMissionErrorI30min10mII4hours5mIII10hours1m4PersonalNavFormFactorMEMSInertialSensorsLongMissionsLowPowerMEMSIMUBut,...TechnicalApproachI:InertialSensorsforGPSChallengedNavigation10008006004002000-1000-800-600-400-200Toemountedpitchgyro(deg/s)Time(s)557.325557.35557.375557.4557.425557.45557.475557.5557.525557.55-50050100150200Time(s)ZAccelerationBodyFrame(m/s2)IntersenseIC3HoneywellHG1930Z-accelBodyFrame(m/s2)200150100500-50Time(s)MEMSensenanoIMUinheel5HighDynamicRangeUseTacticalGradeMEMSIMUBut,…In30min,locationerroris1,131km(1/6earthradius)WithConsumerMEMS1xearthradiusTechnicalApproachIcont’d:MEMSIMU6TechApproachII:ZeroVelocityUpdatesVelocityerrorisconstantintimewhilewalkingPositionerrorduetoaccelsbecomeslinearintimeCrosstrackpositionerrorduetogyroerrorremainsuncorrectedRollandpitcherrorsareconstantintimetvvVaidedtxx,yyfreeinertialfreeinertialVaidedy7ZUPTDetectionHumanLimbKinematicsAccel.-baseddetectorTerrainRelativeVel.SensorSingletoneCWradarDistance≈phaseshiftofgroundreflectedwave‘Zero’velocityifdistancevarianceislessthanagiventhresholdResolution:190μm/szv=false;xtest=var(ax(t-k:t));ytest=var(ay(t-k:t));ztest=var(az(t-k:t));if(xtest&&ytest&&ztestn)zv=true;endVxVy=0VzAssumption:Nearconstantacceleration=zerovelocity1S.Kwakkelet.al.GNSSAidedInSituHumanLowerLimbKinematicsDuringRunning.IONGNSS2008.PowerSupplyinandIFOutputLNAMixerchipSplitterVCOComponentsideAntennasidePotentiometer8NoopeningsinheelRubberheeltransparenttoRFsignalTRV+IMUgenerateZUPTTRVfittedintoheelw/PVCbracketTRVBoardMountingbracketPower&DataconnectorInnersoleTopviewBottomviewFasteningscrewsFasteningscrews1cmSwingHeelstrikeStanceHeelLift9‘Complementary’inertialnavfilterDuringZUPT,modelisnominallyLTIObservability:Rank8/15statesHeadinggyrobiasisthelargestunobservablecomponentSingleIMUwithZUPTstepInertialNavigationR,V,ΨComplementaryKalmanFilterf,ωz=δVδR,δV,δΨδf,δωgKeyR-PositionV–VelocityΨ–Orientation(Euler)f–Non-gravitationalaccelerationg–Gravityω–AngularrateδR–PositionerrorδV–VelocityerrorδΨ–Orientationerrorδf–Accelerometerbiasδω–Gyrobiasx-Statez–KalmanmeasurementF–StatetransitionmatrixG–NoiseinputmatrixH–Measurementmatrixn,p–whitenoise10DuringZUPT,modelisnominallyLTIObservability:Rank8/15statesUnobservablesubspaceδR(x,y,z)–positionerror(3)δΨ(z)–headingerror(1)fmxδΨ–gyrobiasaboutspecificforcevector(3)SpecificforcevectorwhileZUPTingisgravityreactionforcewhichmeansheadinggyrobiasisthelargestunobservablecomponentIssues–ModelObservability11TechnicalApproachIII:ShoeRangerUserangebetweenshoestoaidinertialnavfilterErrorobservedintherangemeasuredisdependentontheheadingerrorateachfootRangeaidcancorrectheadingerrorT.Brand&Phillips,Foot-to-FootRangeMeasurementsasanAidtoPersonalNavigation.ION59thAnnualMeeting.2003.IMUSonarIMU CableSonar CableBoot Mount12SeparateINSsgeneratesX,V,ΦforeachfeetIndirectfeed-backFilterconcatenatestwosingleshoeversionsZuptonlyfallbackifrangefailsConfigurable(perfoot):9(X,V,Φ),15(δa,δω)or27states(δωgS,δg)DualShoeKalmanFilterArchitectureRightInertialNavigationKalmanFilteraR,ωRZuptsδaR/L,δωR/L,δωgR/LSR/L,δgKeyX-PositionV–VelocityΦ–Orientation(Euler)a–Measuredaccelerationω–Angularrateδa–Accelerometerbiasδω–Gyrobiasδωg–gInducedGyrobiasS–Scalefactorsδg–GravitymodelerrorsLeftInertialNavigationaL,ωLX,V,ΦRangeX,V,Φ13Shoe-to-ShoeRangeSimulationsSimulated30minwalkover900mTwoshoesfollowidenticalpath,withphaseoffset,independentbiasesandnoiseSingleKalmanFilterwithinitialblockdiagonalstructure(P2=[P1O;OP1])Simulatednoiseof1inchMeasurementaccuracy=3inRangemeasurementsat~20Hz25cmFinalerrors(30minsim)NoRangeAidingSimulatedRangeAidingPositionX1250m1.7mPositionY400m4mPositionZ10m10mRoll2deg2degPitch2deg2degYaw180deg4deg14010020030040050060070080005101520253035404550XYError(m)1ScalarRange2ScalarRanges1VectorRangeRelativeHeadingTime(s)Simulatedpotentialconfigurationsover30min:1scalarrange2scalarrangesSRSSimulationResultsVectorrangeerror~1cmSimulatedAccuracies:Range–0.1inchstd.dev.Vector–0.1inchx,y,zs.d.Heading–0.5ºstd.dev.1vectorrange1Relativeheading15BootIMUvsshockisolatedTorsoIMUGyroIntegrationonly,nonavfilter,noZUPTsLargeheadingerrorfromgyroresponsetofootdynamicsHowever:Can’tZUPTsattorso,sonopositioningTorsogoodforgyros,footgoodforaccelerometersRecallIMUShockateachfootstepHeadingErrorafter30minRectangularWal