2024年4月29日发(作者:)
Apr.2019TransactionsofNanjingUniversityofAeronauticsandAstronauticsVol.36No.2
AGenericPlug‑and‑PlayNavigationFusionStrategyfor
LandVehiclesinGNSS‑DeniedEnvironment
LAIJizhou
*
,BAIShiyu,XUXiaowei,LÜPin
KeyLaboratoryofNavigation,ControlandHealth‑ManagementTechnologiesofAdvancedAerocraft,MinistryofIndustryand
InformationTechnology,CollegeofAutomationEngineering,NanjingUniversityofAeronauticsandAstronautics,Nanjing
211106,
(Received15March2019;revised3April2019;accepted4April2019)
Abstract:Achievingaccuratenavigationinformationbyintegratingmultiplesensorsiskeytothesafeoperationof
landvehiclesinglobalnavigationsatellitesystem(GNSS)‑r,currentmulti‑sensor
fusionmethodsarebasedonstovepipearchitecture,whichisoptimizedwithcustomfusionstrategyforspecific
gtodevelopadaptablenavigationthatallowsrapidintegrationofanycombinationofsensorstoobtain
robustandhigh‑precisionnavigationsolutionsinGNSS‑deniedenvironment,weproposeagenericplug‑and‑play
posedstrategycanhandledifferentsensorsinaplug‑and‑play
mannerassensorsareabstractedandrepresentedbygenericmodels,whichallowsrapidreconfigurationwhenevera
veestimationsarefusedwithabsolutesensorsbasedon
improvedfactorgraph,whichincludessensors’errorparametersinthenon‑linearoptimizationprocesstoconduct
uatetheperformanceofourapproachusingalandvehicleequippedwithaglobal
positioningsystem(GPS)receiveraswellasinertialmeasurementunit(IMU),camera,wirelesssensorand
sarecomparedwiththemost
commonfiltering‑sthatourstrategycanprocesslow‑qualityinputsourcesina
plug‑and‑playandrobustmanneranditsperformanceoutperformsfiltering‑basedmethodinGNSS‑denied
environment.
Keywords:GNSS‑denied;multi‑sensorfusion;plug‑and‑play;factorgraph;landvehicles
CLCnumber:V249.32Documentcode:AArticleID:1005‑1120(2019)02‑0197‑08
0Introduction
Oneoftheessentialtechnologiesthatensurere‑
thefieldoflandvehicles
[2‑4]
.However,thesenaviga‑
tionsystemsarebasedonstovepipearchitecture
[5]
,
whichiscustomizedforspecificsensorsandmea‑
gsabouthugecostswhen‑
everthenavigationsystemrequireschangesorup‑
geexistingfusionarchitectures,De‑
fenseAdvancedResearchProjectsAgency(DAR‑
PA),USAlaunchedAllSourcePositioningand
Navigation(ASPN)projectin2010
[6]
.ASPNproj‑
ectaimstodevelopadaptablenavigationthatallows
rapidintegrationofanycombinationofsensorstoen‑
ablelowcost,andseamlessnavigationsolutionsfor
militaryusersonanyoperationalplatformandinany
‑
rentlandvehiclesheavilyrelyonglobalnavigation
satellitesystem(GNSS).However,whenlandvehi‑
clesruninthedenseorevenGNSS‑deniedenviron‑
ment,GNSSsignaldegradesorevenfailstolocate
landvehicles.
[1]
WhenGNSSsignalisunavailable,accurate
navigationsolutionscanbeobtainedthroughinte‑
‑sensorfusionmeth‑
odshavebeendeeplystudiedandwidelyappliedin
*Correspondingauthor,E‑mailaddress:laijz@.
Howtocitethisarticle:LAIJizhou,BAIShiyu,XUXiaowei,icPlug‑and‑PlayNavigationFusionStrategy
forLandVehiclesinGNSS‑DeniedEnvironment[J].TransactionsofNanjingUniversityofAeronauticsandAstronautics,
2019,36(2):197‑204.
http:///10.16356/j.1005‑1120.2019.02.002
198
TransactionsofNanjingUniversityofAeronauticsandAstronauticsVol.36
searchershaveperformedre‑
searchonASPN.
Forthesoftwaresystems,ElsnerandJuangde‑
signedtheplug‑and‑playmultisensoryfusion
schemesbasedonrobotoperatingsystem
(ROS)
[7‑8]
.Forthefusionarchitecturesandalgo‑
rithms,filtering‑basedestimationmethodsaremost‑
edreconfigurableinte‑
grationfilteringEngine(RIFE).InRIFE,various
classisdefinedbythetypeofsensormeasurement
andthefiltercanbereconfiguredbyinstantiatinga
sensorobjectwheneveranewsensorisconnectedto
system
[9]
.edmulti‑sensor‑fusion
extendkalmanfilter(MSF‑EKF)toprocess
time‑delayed,relativeandabsolutemeasurements
fromatheoreticallyunlimitednumberofdifferent
ulardesignallowsseamlesshan‑
dlingofadditional/lostsensorsignals
[10]
.Groves
proposedsensorfusionmodularintegratedarchitec‑
ture,wheredifferentsubsystemsareconstructedto
processandintegratedifferentsources
[11]
.Zhuetal.
presentedagoal‑
time,power,andweightarecombinedtoreconfig‑
uresensorsuiteandallchosenmeasurementsarein‑
tegratedusingEKF
[12]
.Althoughaboveresearch
hasachievedsatisfactoryresults,thefiltering‑based
methodshaveincommonthattheyrestrictthestate
vectortothemostrecentstateandmarginalizeout
alloldinformation,whichbringsoutsuboptimalper‑
formance
[13‑14]
.Incontrasttofiltering‑basedmeth‑
ods,agraphicalmodelknownasfactorgraphrepre‑
sentsinformationfusionproblemasagraph‑based
desthe
connectivitybetweentheunknownvariablenodes
ensoryfusion
methodsviafactorgraphcanhandledelayedand
asynchronoussourcesinaflexiblewaybecausepast
statesarekeptduringtheglobaloptimizationpro‑
cess
[15]
.AnditoutperformsEKFbecauseofthe
re‑linearizationprocess
[16]
.eda
constrainedoptimalselectionforsensorsbasedon
factorgraphandtheoptimalsubsetsofsensorsare
selectedwithavailableresources,navigationaccura‑
cyandobservabilityindex
[17]
.Consideringthere‑
al‑timeapplication,edaslid‑
ing‑windowfactorgraphmethodforautonomousve‑
hicles
[18]
.tedtheeffectiveness
ofrobustoptimizationtechniquesusingthefactor
sthatthefactorgraphal‑
gorithminconjunctionwithrobustoptimizationcan
achievereasonableperformanceintheGNSS‑de‑
gradedenvironment
[19]
.However,aboveresearchis
stilloptimizedwithcustomfusionsolutions,which
isinadequatefortheflexibleandextensibleneedsof
landvehiclesnavigationsystem.
Seekingtodevelopadaptablenavigationthatal‑
lowsrapidintegrationofanycombinationofsensors
toenableseamless,robustandaccuratenavigation
solutionsinGNSS‑deniedenvironment,wepro‑
poseagenericplug‑and‑playfusionstrategybased
ategyisde‑
sabstract
sensormodelsaredesignedbythetypeofsensors,
ensoris
connectedintothenavigationsystem,thespecific
sensormodelisbuiltfromtheabstractmodelandits
posed
strategyallowsrapidreconfigurationofanycombina‑
,itsmodularityenablesthefu‑
sionarchitecturetobeflexibleandextensibletonew
tion,time‑de‑
layedsensordata,whichpresentslow‑qualitychar‑
acteristics,canbeprocessedinanaturalwaybased
ontheimprovedfactorgraph,inwhicherrorparam‑
etersofsensorsarealsoaddedintothegraphmodel
uate
performanceoftheproposedstrategyusingaland
showsthatourstrategycanprocesslow‑qualitydata
inaplug‑and‑playandrobustmanneranditsperfor‑
manceoutperformsthemostcommonfilter‑based
method.
1GenericSensorFusionStrategy
TheproposedstrategyisshowninFig.1,whi‑
chconsistsofthreeparts,preprocessinglayer,ab‑
stractinglayerandfusinglayer.
No.2LAIJizhou,icPlug‑and‑PlayNavigationFusionStrategyforLandVehicles…
199
Fig.1Genericmulti‑sensorfusionstrategy
1.1Preprocessinglayer
Inthepreprocessinglayer,rawmeasurement
sourcesareprocessedintousablenavigationinfor‑
ensorisconnectedintothesys‑
tem,itisrecognizedandcorrespondingIDisat‑
,dataconversionis
‑
ample,imagesofcameraareconvertedintoposees‑
eringthatsensorsareplacedindif‑
ferentlocationsofavehicle,spatialparameters
amongdifferentsensorsobtainedfromanofflinecal‑
ibrationareoffsetinspace‑,
ve
andabsolutemeasurementsarealsoalignedbytrans‑
formationbetweendifferentframes.
1.2Abstractinglayer
Intheabstractinglayer,variousabstractsensor
modelsaredesignedaccordingtothetypeofsen‑
yerconsistsoffourabstractmodels,
thatis,deadreckoningmodel,positionmodel,ve‑
locitymodel,cificmod‑
elofasensorcanbeinstantiatedusingitstemplates
byidentifyinginformation’,sensorerror
mple,asensor’s
specificnoiseanderrorparametersareaddedinto
thebuiltmodel.
Deadreckoningmodelrepresentsrecursivesen‑
sors,suchasinertialorotherdeadreckoningsen‑
tractmodelcanbeconceptuallyde‑
scribedbyfollowingcontinuousnonlineardifferen‑
tialequation
x=f
DR
(
x,α,Δ
)
(1)
where
x
isthenavigationstate,representingtheve‑
hicle’sposition,attitudeandvelocity;
Δ
theincre‑
mentofthevehiclemeasuredbysensorsand
α
the
odels
representsensorsthatprovidewithothermeasure‑
mentinformation,thatis,position,velocityandat‑
bstractmodelscanbedescribedina
unifiedway
z=h
M
(
x
)
+n
(2)
where
x
isnavigationstate,representingthevehi‑
cle’sposition,attitudeandvelocity;
z
theinforma‑
tionmeasuredbysensorsand
n
ameasurement
noise,whichisassumedtobezeromeanGaussian
noise.
h
M
isthemeasurementfunction,relatingbe‑
tweenthemeasurementandnavigationstate.
1.3Fusinglayer
Inthefusinglayer,non‑linearoptimization
‑
torgraphisabipartitegraph
G=
(
F,X,E
)
with
twotypesofnodes:Factornodes
f
i
∈F
andvari‑
ablenodes
x
i
∈X
.Edges
e
ij
∈E
canexistonlybe‑
tweenfactornodesandvariablenodes,andarepres‑
entifandonlyifthefactor
f
i
involvesavariable
x
i
.
Thefactorgraph
G
definesonefactorizationofthe
function
f
(
X
)
as
f
(
X
)
=f
i
(X
i
)
(3)
where
X
i
isthesetofallvariables
x
i
connectedby
anedgetofactor
f
[
i
20]
.
Afactordescribesanerrorbetweenthepredict‑
ngaGaussian
noisemodel,ameasurementfactorcanbewrittenas
f
i
(X
i
)=d[h
i
(X
i
)-z
i
]
(4)
where
h
i
(X
i
)
isthemeasurementmodelasafunc‑
tionofthestatevariables
X
i
;
z
i
theactualmeasure‑
mentand
d(⋅)
acostfunction,whichisthesquared
Mahalanobisdistance,definedas
d
(
e
)
≜e
T
Σ
-1
e
,
with
Σ
s
modelscanberepresentedusingfactorsinasimilar
manner.
Eq.(3)shouldbeminimizedbyadjustingthe
estimatesofthevariables
X
.Theoptimalestimateis
200
TransactionsofNanjingUniversityofAeronauticsandAstronauticsVol.36
theonethatminimizestheerroroftheentire
graph
[21]
X
=argmin
X
(
∏
f
i
(X
i
i
))
(5)
Differentsensorinformationisaddedintothe
time‑delayedandasynchronousmeasurementscan
beincorporatedintothefactorgraphinanatural
way,leadingtobetterestimatesforcurrentstates.
2AnImprovedSensorFusion
MethodforLandVehiclesBased
onFactorGraph
Thestructureoftheimprovedmultisensoryfu‑
nfactorgra‑
phframework,sensorerrorsareaddedintothe
optimizederrorparametersareutilizedtocalibrate
osensorerroronline
calibration,betterestimatesforthewholetrajectory
canbeobtained.
Fig.2Structureoftheimprovedfusionmethod
Consideringthatthemostcommonsensorsin
typicalnavigationapplicationsoflandvehicles,im‑
provedfactorgraphforlandvehiclesisbuiltin
sideredsensorsareIMU,GPS,od‑
ometer,visualsensors,
thispaper,GPSfactorisbuiltinthegraphmodelto
r,GPS
signalisnotfusedwithothersensorsbutusedas
groundtruthinthefieldteststoprovetheperfor‑
manceoftheproposedalgorithminGNSS‑denied
environment.
Sensors’errorparametersareaddedintograph
Fig.3Improvedfactorgraphforlandvehicles
ollowcir‑
clesmeannavigationstatesand
f
IMU
meansIMU
hollowcirclesmeanIMUbias,which
isintroducedatalowerfrequencythannavigation
solidcirclesmeanodometerfactorwhilegreyhol‑
lowcirclesrepresentscalefactorerrorofodometer.
Red,yellowandpurplesolidcirclesmeanvisual
odometry,wirelesssensor,andGPSfactor,respec‑
ollowcirclesrepresentscaleerrorof
tionstatesoflandvehiclesanderror
parametersofsensorsareoptimizedtogethertoim‑
arametersare
‑
sorfactorsarebuiltasfollows
2.1IMUfactor
IMUfactorisbuilttoconnectnavigationstates
eringtime
k
andtime
k+1
,IMUfactorisderivedas
f
IMU
(x
k+1
,x
k
,α
k
)≜d(x
k+1
-h(x
k
,α
k
,z
k
))
(6)
where
x
k+1
and
x
k
arenavigationstatesattime
k+1
and
k
,respectively;
z
k
=
[
α
k
ω
k
]
isthegiv‑
enIMUmeasurements,thatis,accelerationandan‑
gularrate;
α
k
thebiasofinertialsensor,whichises‑
er
integrationpredictionfunctionwithanoiseisadopt‑
edtorepresent
h(⋅)
.Inthesameway,biasfactor
canbedescribedas
f
bias
(α
k+1
,α
k
)≜d(α
k+1
-g(α
k
))
(7)
where
α
k+1
and
α
k
arethebiasesattime
k+1
and
k
,modelledasconstanterror.
2.2Odometerfactor
Odometerprovideswithvelocityinformation
anditsfactorcanberepresentedas
f
ODO
(x
k
,β
k
)≜d(z
ODO
k
-h
ODO
(x
k
,β
k
))
(8)
No.2LAIJizhou,icPlug‑and‑PlayNavigationFusionStrategyforLandVehicles…
201
where
z
ODO
k
and
x
k
arethevelocitiesofodometerand
navigationstateattime
k
;
β
k
isthescalefactorer‑
ror,
thesameway,scalefactorerrorcanbederivedas
f
scale
(β
k+1
,β
k
)≜d(β
k+1
-g(β
k
))
(9)
where
β
k+1
and
β
k
arethescalefactorerrorsattime
k+1
and
k
,actorerrorismod‑
elledasconstanterror.
2.3GPSfactor
GPSfactorisbuilttoprovidewithabsolutepo‑
sitionanditsfactorcanbemodelledas
f
GPS
(x
k
)≜d(z
GPS
k
-h
GPS
(x
k
))
(10)
where
z
GPS
k
and
x
k
arethepositionsofGPSandnavi‑
gationstateattime
k
.
2.4Wirelesssensorfactor
Wirelesssensorprovidesranginginformation
relesssensorcanreceive
atleastthreeranginginformationtobasestations
whosepositionsareobtainedinadvance,itcanpro‑
videwithpositioninthegivenframesanditsfactor
canbemodelledas
f
WS
(x
k
)≜d(z
WS
k
-h
WS
(x
k
))
(11)
where
z
WS
k
and
x
k
arethepositionsofwirelesssensor
andnavigationstateattime
k
.
2.5Visualsensorfactor
Visualsensorprovideswithrelativeposition
he
relativeandabsolutemeasurementsarealigned,it
provideswithposeinformationintheglobalframe.
Itsfactorcanberepresentedas
f
VOP
(x
k
)≜d(z
VOP
k
-h
VOP
(x
k
,λ
k
))
(12)
f
VOH
(x
k
)≜d(z
VOH
k
-h
VOH
(x
k
))
(13)
where
z
VOP
k
and
x
k
arethepositionofvisualsensor
andnavigationstateattime
k
;
z
VOH
k
istheyawofvi‑
sualsensorattime
k
;
λ
k
thescaleerroranditismod‑
torcanberepresented
as
f
scale
(λ
k+1
,λ
k
)≜d(λ
k+1
-g(λ
k
))
(14)
where
λ
k+1
and
λ
k
arethescaleerrorsattime
k+1
and
k
,respectively.
3Experiment
Inthefieldtests,weusealandvehicle
equippedwithaGPSreceiveraswellasIMU,ste‑
reocamera,UWB(akindofwirelesssensor)and
receiverprovideswithprecisepositioningofcenti‑
meter‑levelsolutionswhenitoperatesinreal‑time
kinematic(RTK)mode,whichistreatedasground
otintegratedintothenavigationsys‑
tem,whichonlytoevaluatetheperformanceofthe
proposedstrategyinGNSS‑deniedenvironment.
DataacquisitionmoduleisdesignedbasedonROS.
Fig.4Landvehicleusedinthefieldtest
Thetrajectoryofthefieldtestisshownin
rtingpointismar‑
kedwithastarandarrowsshowthedrivingdirec‑
incolorofthetrajectorymeansthecor‑
respondingsectionwhereacertaincombinationof
sensorsisintegratedintothenavigationsystem,be‑
causesomesensorsareavailableinspecificcircum‑
mple,redlineissurroundedby
basestations,andtheUWBisavailableonlyinthis
,theroadwayinbluepartisthearea
wherefeatureissparse,whichleavesthecamerain
anunusablestateandnotbeintegratedintothenavi‑
est,differentinformation
sourcesareintegratedtothesystemwheneverthey
areavailable.
Whenasensorisconnectedintosystem,spe‑
cificmodelsareconstructedandcorrespondingfac‑
e‑de‑
layedandasynchronousmeasurementscanbefused
inthefactorgraphinatrulyplug‑and‑playmanner
sincepaststatesarekepttoperformglobaloptimiza‑
tion.
Wecompareourresultswiththemostcommon
filtering‑basedmethod,wbackofa
202
TransactionsofNanjingUniversityofAeronauticsandAstronauticsVol.36
Fig.5FieldtestingeneralroadcampusofNUAA
basicEKFisthatlinearizationhappensonlyonce,
,EKF
issensitivetotime‑delayedmeasurementswhich
presentslow‑qualitycharacteristics,asstatescannot
uateimpacts
oflow‑qualityinformationonEKF,weaddtimede‑
laysandnoiseintosensordataatdifferenttimes,
delayissettobe1s,whichislargeenoughtoim‑
plementerrorexcitation.
ThetrajectorycomparisonisshowninFig.6.
Theeastandthenorthpositionerrorcomparisons
areplottedinFigs.7,8,inwhichthetimeperiods
withfaultedVOmeasurementsaremarkedwith
eethatbothfusionmethods
presentslowdriftasthereisnoabsoluteposition
,positionerrorsof
EKFarehighlyincreasedduringperiodswhenfault‑
ontrary,sincepaststates
arekeptinglobaloptimizationprocess,delayedin‑
formationcanbeaddedtographmodelbasedon
theirtimestampinaplug‑and‑playway,leadingto
‑mean‑square
Fig.6Trajectorycomparison
Fig.7Eastpositionerrorcomparison
Fig.8Northpositionerrorcomparison
(RMS)errorsofthepositionerrorisillustratedin
Table1.
Table1RMScomparisonm
RMSEastNorth
EKF10.7813.80
Proposedmethod5.999.44
4Conclusions
Weproposeagenericplug‑and‑playmulti‑sen‑
sorsfusionstrategyforlandvehiclesinGNSS‑de‑
ategyhandlesdifferent
sensorsinaflexiblewayassensorsarerepresented
veestimationsare
fusedwithabsolutesensorsbasedonimprovedfac‑
torgraph,inwhichsensors’errorparametercanbe
addedintographoptimizationtoperformsensoron‑
nstratetheperformanceof
sthattradi‑
tionalfilteringmethodisheavilyinfluencedby
low‑ategycanprocess
time‑delayedinputsourcesinaplug‑and‑playandro‑
bustmanneranditsperformanceoutperformsEKF
inGNSS‑deniedenvironment.
Inourfuturework,theintegratedqualityof
themeasurements,notjustrestrictedtosensoraccu‑
No.2LAIJizhou,icPlug‑and‑PlayNavigationFusionStrategyforLandVehicles…
203
racy,willbeconsideredtomeasuresensor’sconfi‑
dencelevelinthefusionprocess,thusfurtherim‑
provingrobustnessandaccuracyofthesystem.
References
[1]MAH,SMARTE,AHMEDA,m‑
age‑basedpositioningforusvunderGPSdenialenvi‑
ronment[J].IEEETransactionsonIntelligentTrans‑
portationSystems,2017,19(1):72‑80.
[2]LIX,blefusionpositioningstrategyfor
landvehiclesinGPS‑deniedenvironmentsbasedon
low‑costsensors[J].IEEETransactionsonIndustrial
Electronics,2017,64(4):3205‑3215.
[3]SUHRJK,JANGJ,MIND,fu‑
sion‑basedlow‑costvehiclelocalizationsystemfor
complexurbanenvironments[J].IEEETransactions
onIntelligentTransportationSystems,2017,18(5):
1078‑1086.
[4]LINF,WANGH,WANGW,estate
andparameterestimationbasedondualunscentedpar‑
ticlefilteralgorithm[J].TransactionsofNanjingUni‑
versityofAeronauticsandAstronautics,2014,31
(5):568‑575.
[5]emonstrationofplugand
playnavigationsystemusingscorpionandsmartsen‑
sors/cables[C]//‑
ton:ION,2017.
[6]‑BAA‑11‑14allsource
positioningandnavigation[EB/OL].(2010‑1‑1)
[2019‑3‑15].https:///index?s=
opportunity&mode=form&id=cef82a45ab7d64a5445
eadf277f13dbe&tab=core&_cview=1.
[7]salplug‑n‑playsensorintegra‑
tionforadvancednavigation[D].Dayton,USA:Air
ForceInstituteofTechnology,2012.
[8]JUANGJC,HSIEHWL,p‑
mentofaplug‑and‑playROS‑basedlandvehicularnav‑
igationsuite[C]//30thInternationalTechnicalMeet‑
ingoftheSatelliteDivisionoftheInstituteofNaviga‑
nd:ION,2017:1959‑1963.
[9]SOLOVIEVA,igurableintegra‑
tionfilteringengine(RIFE)forplug‑and‑play(PnP)
navigation[C]//26thInternationalTechnicalMeeting
ofTheSatelliteDivisionoftheInstituteofNavigation.
Nashville:ION,2013:16‑20.
[10]LYNENS,ACHTELIKMW,WEISSS,etal.A
robustandmodularmulti‑sensorfusionapproachap‑
pliedtomavnavigation[C]//2013IEEE/RSJInterna‑
tionalConferenceonIntelligentRobotsandSystems
(IROS).Tokyo:IEEE,2013:3923‑3929.
[11]plexityprobleminfuturemul‑
tisensornavigationandpositioningsystems:Amodu‑
larsolution[J].TheJournalofNavigation,2014,67
(2):311‑326.
[12]ZHUZ,ADAMSK,VENABLED,‑driven
sensorconfigurationinanavigationsystem[C]//2016
IEEE/IONPosition,LocationandNavigationSympo‑
sium(PLANS).Savannah:ION,2016:527‑534.
[13]STRASDATH,MONTIELJM,DAVISONA.
VisualSLAM:Whyfilter?[J].ImageandVisionCo‑
mputing,2012,30(2):65‑77.
[14]MASCAROR,TEIXEIRAL,HINZMANNT,et
:Graph‑optimizationbasedmulti‑sensor
fusionforrobustUAVposeestimation[C]//2018IE‑
EEInternationalConferenceonRoboticsandAutoma‑
tion(ICRA).Karlsruhe:IEEE,2018:1421‑1428.
[15]QINT,CAOS,PANJ,aloptimiza‑
tion‑basedframeworkforglobalposeestimationwith
multiplesensors[EB/OL].(2019‑1‑11)[2019‑3‑10].
https:///abs/1901.03638?context=cs.
[16]INDELMANV,WILLIAMSS,KAESSM,etal.
Informationfusioninnavigationsystemsviafactor
graphbasedincrementalsmoothing[J].Roboticsand
AutonomousSystems,2013,61(8):721‑738.
[17]CHIUHP,ZHOUXS,CARLONEL,‑
strainedoptimalselectionformulti‑sensorrobotnaviga‑
tionusingplug‑and‑playfactorgraphs[C]//2014IEEE
InternationalConferenceonRoboticsandAut‑omation
(ICRA).HongKong:IEEE,2014:663‑670.
[18]MERFELSC,sionwith
chainposegraphsforautomateddriving[C]//2016IE‑
EE/RSJInternationalConferenceonIntelligentRo‑
botsandSystems(IROS).Deajeon:IEEE,2016:
3116‑3123.
[19]WATSONRM,navigationin
GNSSdegradedenvironmentusinggraphoptimization
[EB/OL].(2018‑6‑23)[2019‑3‑10].https://arxiv.
org/abs/1806.08899.
[20]KSCHISCHANGFR,FREYBJ,LOELIGER
H‑graphsandthesum‑productalgorithm
[J].IEEETransactionsoninformationtheory,2001,
47(2):498‑519.
204
TransactionsofNanjingUniversityofAeronauticsandAstronauticsVol.36
[21]DELLAERTF,rootSAM:Si‑
multaneouslocalizationandmappingviasquarerootin‑
formationsmoothing[J].TheInternationalJournalof
RoboticsResearch,2006,25(12):1181‑1203.
Acknowledgements
Thisworkwaspartiallysupported
bytheNationalNaturalScienceFoundationofChina(No.
61703207),theJiangsuProvincialNaturalScienceFounda‑
tionofChina(20170801),theAeronauticalScience
FoundationofChina(No.2017ZC52017),andtheJiangsu
ProvincialSixTalentPeaks(No.2015‑XXRJ‑005),andthe
JiangsuProvinceQingLanProject.
innavi‑
gation,guidanceandcontrolfromNanjingUniversityof
AeronauticsandAstronautics,Nanjing,China,
iscurrentlyafullprofessoratAutomationEngineeringCol‑
lege,earchhasfocusedoninertialand
GNSSnavigation,multi‑sensorfusionandfault‑tolerantnav‑
igation,autonomousnavigationinGNSS‑deniedenviron‑
ment,visualnavigation,deeplearningandsensing,andcol‑
laborativenavigation.
ateatNanjing
earchis
focusedonmulti‑sensorfusion.
ateinNanjing
earchis
focusedondownholeautonomousnavigation.
Dr.LÜinnavigation,guid‑
anceandcontrolfromNanjingUniversityofAeronauticsand
Astronautics,Nanjing,China,rrentlyalec‑
turerinNanjingUniversityofAeronauticsandAstronautics.
Hisresearchisfocusedonrotationalinertialnavigationand
dynamicmodelassistednavigation.
houdesignedthestudy
yuconductedthe
weipartici‑
.LÜPincontributedtothedis‑
horscommented
onthedraftandapprovedthesubmission.
CompetinginterestsTheauthorsdeclarenocompetingin‑
terests.
(ProductionEditor:ZhangBei)
发布评论