ECASS: Edge computing based auxiliary sensing system for self-driving vehicles

(1)

ContentslistsavailableatScienceDirect

Journal of Systems Architecture

journalhomepage:www.elsevier.com/locate/sysarc

ECASS: Edge computing based auxiliary sensing system for self-driving vehicles

Xiong Wang

^a

, Tianpeng Wei

^a

, Linghe Kong

^a^,^∗

, Liang He

^b

, Fan Wu

^a

, Guihai Chen

^a

a Shanghai Jiao Tong University, China

b University of Colorado Denver, USA

a r t i c le i n f o

Keywords:

Autopilot systems Self-driving vehicle Trajectory GPS

a b s t r a ct

Self-drivingvehicles,combiningautomobileswithautopilotsystems,enableintelligentandsafedriving.Self- drivingvehiclescanachieveaccurateautomaticnavigation,trajectorytracking,andautomaticovertakingby usingGPS,radars,andinertialmeasurementunit(IMU).Amongthem,overtakingisessentialinordertoavoid excessivewaitingtimeandimprovethetrafficefficiency.Whenfollowingalargetruckorbus,theself-driving vehiclecannotensurethesafeovertakingbecausetheline-of-sight(LOS)rangedetectedbytheradarandcamera isblocked,thusunabletoperceivethesurroundingenvironmentaccurately.Acommonlyadoptedmitigation istofollowthetruckorbusatareducedspeed,atthecostofreducedtrafficefficiencyandmoretrafficjams.

Tomitigate thisdeﬁciency,thispaperdevelopsanauxiliarysensingsystemusingedgecomputingtolocate nearbyvehiclesforself-drivingvehicles,calledECASS.Specially,infrastructuredeployedalongtheroadlike serversareutilizedtoaccuratelylocatevehiclesaccordingtoGPSandwirelessinformationsuchasWiFiorDSRC.

Subsequently,theserverwilltransmitthelocalizationinformationofnearbyvehiclestotheself-drivingvehicle, basedonwhichitcandeterminethedrivingstateforthenextmomentdespiteoftheobstruction.Extensive simulationsverifythatECASSbasedtrajectoryismuchclosertotherealtrajectorythanGPS.Especiallywhen GPSerrorissetwithin10m,ECASScanreducethemeanabsolutelocalizationerrorfrommorethan7mtoabout 3m.

1. Introduction

Inrecentyears,self-drivingvehicleshaveattractedextensiveatten- tionfrombothindustryandacademiabecauseofits safetyandtrav- ellingefficiency[1–3].Inparticular,withtheincreaseinthenumber ofvehicles,theinjuredpersonsanddeathsintrafficaccidentsarealso increasing.Asacriticaltechniqueinfuturevehicles,autopilotcansub- stantiallydecreasethetrafficaccidentsinducedbyhumanfactorsand thusenhancethetravellingsafety.Thetrafficsharingbasedontheau- tomaticdrivingtechniquecaneffectivelyalleviatethetrafficcongestion andpollutionproblems[4,5].

Researchershavedesignedvariousdetectionalgorithmsbased on theimageandsignalprocessing,toaccuratelysensetheenvironments aroundtheself-drivingvehicle[6–8].Theself-drivingvehiclewillthen determinethedrivingstateforthenextmoment,suchasovertaking, acceleration/deceleration,ortravellingattheoriginalspeedbasedon theacquiredenvironmentalinformation.

Compared to acceleration/deceleration, or travel at the original speed, theexecution of overtaking for self-drivingvehicles becomes muchmorecomplicated.Itmainlyconsistsofthreesteps.Firstly,the

∗Correspondingauthor.

E-mailaddress:[email protected](L.Kong).

self-drivingvehiclechangesthelaneaccordingtotheplannedtrajec- tory.Secondly,itdrivesalongtheovertakenvehicleataprescribedlat- eraldistance.Finally,itwillreturntotheoriginallaneinfrontofthe overtakenvehicle[9,10].Majorityof researchworkon thisproblem hasfocusedontheplanningorpredictionoftheovertakingtrajectory [9,11,12].

Availableliteraturehasplannedtheirovertakingtrajectoriesunder thepremisethattheradarorcameraintegratedintheself-drivingvehi- clecandetectandtrackobstacleswithoutbusesortrucksaroundwhen overtaking.However,undersomespecialcircumstances,thecameraand radarmaybeblockedbythetruckorbusinfrontorbehind,rendering theself-drivingvehiclepartiallyorevencompletelyunknownaboutthe surroundings.Therefore,ithastofollowthetruckorbusatareduced speed[13].Obviously,thisapproachincursincreasedtimeconsumption andtraﬃccongestion.

However,GPSbasedlocalizationerrorscanreachupto10m,and evenlocalizationerrors inmap-matching basedGPS suﬀerfrom5m [14,15].Suchalargelocalizationerrormaycausethewrongdriving stateadoptionforself-drivingvehicles,riskingthetraﬃcsafety.Inre- sult,theauxiliarysensingsystemnamedECASSisdevelopedtoprovide accuratevehiclelocalizationinformationintheproximityoftheself-

https://doi.org/10.1016/j.sysarc.2019.02.014

Received5October2018;Receivedinrevisedform21January2019;Accepted13February2019 Availableonline13February2019

(2)

drivingvehiclebasedon edgecomputing,whenitcannotaccurately locatenearbyvehiclesonlyusingthecameraandradar.

Inparticular,whentheself-drivingvehiclecannotaccuratelysense thesurroundingenvironmentsduetoobstruction,itwillsendarequest tonearby servers,soastoinformtheseserversofobtainingthecur- rentvehiclelocalizationneartheself-drivingvehicle.Intheseservers, aninformationfusionalgorithmwithregardtothewirelesssignaland GPSisdesignedtoestimatethepositionofnearbyvehicles.Finally,the locationofeachvehicleneartheself-drivingvehiclewillbetransmit- tedtotheself-drivingvehicle.Theself-drivingvehicledeterminesthe drivingstateforthenextmomentbasedontheacquiredlocalization information,despiteoftheobstruction.Inthispaper,thedrivingstate includesovertaking,changinglanes,acceleration,deceleration,braking, andtravellingattheoriginalspeed.

Thecontributionsofthispaperarelistedasbelow:

• We propose an edge computing based framework toassist self- drivingvehiclestoachieveaccuratenearbyvehiclelocalizationand trackingwhenself-drivingvehiclescannotaccuratelysensethesur- roundingenvironment.

• Inroadsideservers,anfusionalgorithmrelatedtoGPSandwireless signalinformationisdevelopedtomeasurethelocationofvehicles neartheself-drivingvehicle.Accordingtotheseacquiredlocaliza- tioninformation,theself-driving vehiclecandeterminethe driv- ingstateforthenextmomentevenwhenpartiallyor completely blocked.

• ExtensivesimulationsarecarriedouttodemonstrateECASS’shigh eﬃciency. Compared toGPS, ECASS based localization is much closertotherealvehicleposition,especiallywhentheGPSbased localizationerrorbecomeslarger.

Thepaperis organized asfollows: Section2presents therelated literatureonself-drivingvehicles.Thepreliminariesareintroducedin Section3,includingthemotivationandproblemstatement.Then,the systemoverview,algorithmdesign,andtheselectionstrategyofservers arepresentedinSection4,andwepresenttheperformanceevaluation inSection5.Finally,weconcludethispaperinSection6.

2. Relatedwork

Uptonow,thereexistssubstantialresearchworkrelatedtotheau- topilottechnique,includingthehardwaredesignsuchasdetectionradar andcamera,algorithmdesignintermsofinformationfusionfromthese hardware,andthetrajectoryplaninthetravellingprocess[16–18].

HardwareHardware isthefundamentalpartforautopilot.High- qualityhardwareisabletoperceivenearbyinformationmoreaccurately [19,20].Forexample,Mercedes-BenzequipS-ClassS500INTELLIGENT DRIVEwithclose-to-productionsensorhardware.Inparticular,vision andradarsensorscombinedwithdigitalmapsareemployedtosense nearbytraﬃcconditions[21],andthissystemhasbeentestedinanau- tonomousmannerfromMannheimtoPforzheim,Germany.Asapromis- ingtechnique,lidarcanalsoperceivetheenvironmentinthesameway asradar.Duetomuchshorterwavelength,thehighresolutionandrelia- bilityrenderitanecessityfordriverlesscarsinthefuture[2].However, thesize,complexity,andcostofthecurrent generationof lidarsen- sorshinderitscommercialization. Therefore,extensiveacademicand industryresearchhasattemptedtomakelidarsensorssmaller,easierto manufacture,andcheaper[6,22].

Information fusion algorithm After obtaining the information aboutnearbyenvironments,howtodealwiththesemassiveinformation becomescrucialforself-drivingvehicles[23,24].Basedonthestereo camerasystem,Frankeetal.[25]presentthevisionalgorithmsforob- jectrecognitionandtracking,free-spaceanalysis,traﬃclightrecogni- tion,lanerecognition,aswellasself-localization.Further,inorderto realizeanaccuratevisualunderstandingofcomplexurbanstreetscenes, abenchmarksuiteandlarge-scaledatasetnamedCityscapesisintro- ducedtotrainandtestpixel-levelandinstance-levelsemanticlabeling.

Bothdetailedanalysisandperformanceevaluationhavebeencarried outbasedontheproposedbenchmark.Unlikecamerabaseddetection, asignalprocessingalgorithmbasedonradarisdesignedtoestimatethe speedandsizeofvehiclesin[26].Finally,someresearchworkproposes tobuildavehicledetectionsystemfusingradarandvisiondata[27].

TrajectoryplanalgorithmAccordingtotheaccurateunderstanding of surroundingenvironments,self-drivingvehiclescan determinethe drivingstateforthenextmoment[28,29],suchasbraking,acceleration, changingthelane,orovertaking.Amongthem,overtakingisthemost complicatedprocess.Toachievesafeovertaking,Milanesetal.[30]de- velopafuzzy-logicbasedcontrollertocontrolthelateralmovementand longitudinalmovementofvehicles.Meanwhile,astereovisionsystem isappliedtodetectanyprecedingvehicleandtriggertheautonomous overtakingmanoeuvre.Furthermore,amathematicalmodelandadap- tivecontrollerforautonomousovertakingmaneuverispresentedin[9]. Especially, anadaptive controlschemeis designedtoallowtracking thedesiredtrajectorieswithunknownvelocityoftheovertakenvehicle comparedtopreviouswork.Theauthorsof[31]proposedanpathplan- ningschemefortheself-drivingcarunderthecomplexenvironments.

Itmainlyconsistsofthreeparts,respectivelyasthenovelpathrepre- sentation,thecollisiondetectionandthepathmodiﬁcation.Finally,a multiple-goalreinforcementlearningframeworkisconstructedtotackle multiplecriteriaforovertakingin[12].Simulationresultsdemonstrate thehigheﬃciencyoftheproposedstrategy.

3. Preliminaries

Inthis section,we presentthemotivationbehindECASSandthe problemstatement.

3.1. Motivation

Nowadays,thetraﬃcconditionbecomesincreasinglycomplicatedin metropolises.Therefore,thereexistseveralcriticalchallengesforself- drivingvehiclestodealwithvariouskindsoftraﬃcscenarios.Forexam- ple,theself-drivingvehiclecanbeeasilyblockedbythetruckorbusin frontorbehind,asshowninFig.1,inwhichcasetheself-drivingvehi- cleistravellingbehindthetruck.Then,theradarorcameramountedon thetopoftheself-drivingvehiclecannotdetectandlocateanyobstacle inregion1,duetobeblockedbythetruckinfront.

Fig.1. Thescenariowhentheself-drivingvehicleisblocked.

(3)

Consequently,theself-drivingvehiclecannotdeterminethetravel- lingstateforthenextmomentbecauseitcannotaccuratelyperceivethe surroundingenvironment.Thetraditionalstrategythatself-drivingve- hiclesadoptistofollowthetruckorbusinfrontatareducedspeed.

Therefore,thismechanismwillincurmuchmoretimeconsumptionand traﬃccongestion,thusresultinginmorepollution.

3.2. Problemstatement

Inrecentyears,edgecomputinghasbeenproposedtoprovidefaster networkresponseandmoresafetyguaranteeusingtheopenplatform integratedwithnetworking,computing,storageandapplicationclose toobjectsordatasources,comparedtodatacenterbasedcomputing [32].Inthispaper,wecombineself-drivingvehicleswithedgecomput- ingtoaccuratelylocatevehiclesneartheself-drivingvehicleincaseof obstruction.

Inself-drivingvehicles,onesecondisdividedintontimeslots.For eachslot,theself-drivingvehiclewillexecuteinstructionsfromvehi- clecontrollers,toensurethetravellingsafety.Assumingattimet,the self-drivingvehiclewaitsforthedrivingstateinstructionforthenext moment–𝑡+1∕𝑛.Then,thevehiclecontrollerwilldeterminethetrav- ellingstateforthenextmomentaccordingtotheacquiredlocalization information.

Asin Fig.1, theself-driving vehicle, referred toas vehiclea, is blockedbythetruckinfront.Althoughequippedwiththecameraand radar,theself-drivingvehiclestillcannotdetectandlocatevehiclesb andcontheleftsideofthetruck.Theroadsideinfrastructurelikeservers areproposed toassistvehicleatodetectandlocatevehicles bandc throughinformationinteraction.Accordingtotheseacquiredinforma- tion,theself-drivingvehicleacandeterminethedrivingstateforthe nextmoment.

Toaccuratelylocatevehiclesneartheself-drivingvehicle,GPSlo- calizationinformationandwirelesssignalsfromvehicleswillbedeliv- eredtonearbyservers. Accordingtothewireless signalinformation, theservercanobtaintheangleofarrival(AOA)relativetothevehicle [33].Inthepresenceofobstruction,anaccuratevehiclelocalizational- gorithmforself-drivingvehiclescanbedevelopedbasedonthefusionof GPSlocalizationandAOAinformation.Assumingthetruelocationfor vehiclebisb_t,xandb_t,y,themeasuredpositionb_m,xandb_m,ybasedon thedesignedvehiclelocalizationalgorithmshouldsatisfytheformula asbelow.

min (√

(𝑏_𝑡,𝑥−𝑏_𝑚,𝑥)²+(𝑏_𝑡,𝑦−𝑏_𝑚,𝑦)² )

. (1)

4. Systemdesign

TheframeworkofECASSisshowninFig.2.Inthissection,thework- ﬂowofECASSispresented,followedbytheintroductionofvehiclelo- calizationalgorithm.Finally,wepresenttheselectionschemeofservers neartheself-drivingvehicle.

4.1. Systemoverview

Intheproposedsystem,everyself-drivingvehicleisintegratedwith GPSandonewirelessantennasuchasWiFiantennaorDSRCantenna.

The GPS localization informationof vehicles is delivered to nearby serversthroughwirelesscommunication.Thewirelessantenna inve- hicleaisreferredtoasAn_a.An_awillselecttwonearbyserverstocom- municatewith,soastoaccuratelylocatevehiclea. Theworkﬂow of ECASSispresentedasfollows,asshowninFig.3.

Iftheself-drivingvehiclecanaccuratelysensethesurroundingen- vironmentaccordingtothetraﬃcconditionobtainedfromthecamera, radar,andIMU,thenitplansthedrivingstatebasedonthedecision algorithm.Otherwise,itwillsendarequesttoroadsideseversforob- tainingnearbyvehiclelocation.Subsequently,thesenearbyserverswill estimatethevehicle’slocationnear theself-drivingvehicleusingthe

Fig.2. TheframeworkofECASS.

Fig.3.TheworkﬂowofECASS.

designedinformationfusionalgorithm.Theself-drivingvehiclewillre- ceivenearbyvehicles’locationinformationfromtheseservers.Finally, itcandeterminewhethertoovertake,decelerate,oracceleratebasedon theobtainedlocalizationinformation.

Wefocusonthelocationmeasurementofvehiclesontheserverside, which isbasedon thefusionof GPSandwireless signalinformation transmittedfromvehicles.Therefore,thetimeoverheadcanalsobecut down sincethevehiclelocationestimationisexecutedontheserver side.Inthispaper,wecallitthevehiclelocalizationalgorithm.

4.2. Designofthevehicledetectionalgorithm

Toaccuratelylocatevehiclesintheblockedareafortheself-driving vehicle,edgecomputingbasedontheroadsideinfrastructureisutilized.

AsshowninFig.4(a),thewirelessantennaAn_bonvehiclebcommuni- cateswithtwonearbyserversBandCsimultaneously.Vehiclebdelivers itsownGPSlocalizationinformationtonearbyserversBandC.Specif- ically,serverscanharnesstheincomingwirelesssignalstoderivethe

(4)

(a) (b)

Fig.4. Communicationbetweenthevehicleandservers.

AOA[34].Ontheserverside,thelocationandAOAinformationcanbe fusedtoimprovethelocalizationaccuracyofvehiclebaccordingtothe designedlocalizationalgorithm.Meanwhile,serverswilltransmittheir ownlocationinformationandIDstovehicleb,basedonwhichitcan realizetheselectionofnearbyservers.

Foreaseofunderstanding,Fig.4(a)canbeabstractedintoFig.4(b).

InFig.4(b),pointsBandCrepresentserverBandC,respectively.The wirelessantennaAn_bisplacedatpointD.Theconnectionlinebetween pointsBandCissetastheYaxis.Andthelineperpendiculartotheline BCissetastheXaxis.b_xandb_yrepresenttheabscissaandordinateof theantennaAn_b.Theselocalizationinformationcanbeobtainedfrom GPSembeddedinthevehicle.Meanwhile,thecoordinatesofserverB aredenotedasb_B,xandb_B,y,whileb_C,xandb_C,yarethecoordinatesof serverC.Thedistancebetweentwoserversissettod.

Weconsiderthescenariothatthewireless antennacommunicates withtwonearbyservers.Speciﬁcally,An_btransmitstheGPSlocalization informationb_xandb_ytoserversBandC.Therefore,ageometrictriangle

△BCDcanbeestablished,asshowninFig.4(b).AssumingthattheAOA fromDtoBis𝛼 andtheAOAfromDtoCis𝜃 accordingtothereceived wirelesssignals,thenthefollowingequationcanbeestablished.

𝑑_𝐵𝐷× sin(𝜋 −𝛼)=𝑑_𝐶𝐷× sin(𝜃), (2)

whered_BDrepresentthelengthoflinesegmentBD,andd_CDdenotethe lengthoflinesegmentCD.Thisequationcanbeformulatedasbelow.

√

(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²× sin(𝜋 −𝛼)

=

√

(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²× sin(𝜃). (3) Inthemeantime,thefollowingequationcanalsobeestablishedac- cordingtothelawofcosines.

2× cos(𝛼 −𝜃)×

√

(𝑏𝑥−𝑏𝐵,𝑥)²+(𝑏𝑦−𝑏𝐵,𝑦)²×

√

(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²=(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²+ (𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²−𝑑².

(4)

Finally,wecanalsogetanotherequation:Thesumofthelengthof linesegmentBEandthelengthoflinesegmentCEisequaltothelength oflinesegmentBC,whichcanbeformulatedas:

√(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝑏_𝐶,𝑦)²× cos(𝜃)+

√

(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²× cos(𝜋 −𝛼)=𝑑. (5) Theseequations holdwhentheGPS basedlocalizationandAOAs areaccurate.However,asmentionedabove,evenmap-matchingbased GPSlocalizationsuﬀersfrommetersoferrors.Theerrorofmeasured AOAsareat alevelof severaldegrees. Therefore,thereexists agap betweenthelengthoflinesegmentBCandthesumofthelengthofline segmentBEandthelengthoflinesegmentCE.Theaimofthedesigned localizationalgorithminECASSistominimizethegapbetweenthese twovalues,whichcanbeformulatedas:

min (√

(𝑏_𝑥−𝑏_𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²× cos(𝜃) +√

(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²× cos(𝜋 −𝛼)−𝑑)

. (6)

Thereexistsomeconstraintsontheoptimizationproblem.Firstly, Eqs.3and4shouldbesatisﬁedatthesametime.Secondly,𝜃 shouldbe largerthan0,yetsmallerthan𝜋/2.Inconclusion,theoptimizationisre- latedtotheantennaAn_bmountedonthevehicleb,andtheoptimization problemcanbeconvertedinto:

min (√

(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²×𝑐𝑜𝑠(𝜋 −𝛼)+

√

(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²×𝑐𝑜𝑠(𝜃)−𝑑) , 𝑠.𝑡.

2×𝑐𝑜𝑠(𝛼 −𝜃)×

√

(𝑏_𝑥−𝑏_𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²×

√

(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²=(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²+ (𝑏𝑥−𝑏𝐶,𝑥)²+(𝑏𝑦−𝑏𝐶,𝑦)²−𝑑²,

√

(𝑏_𝑥−𝑏_𝐵,𝑥)²+(𝑏_𝑦−𝑏_𝐵,𝑦)²×𝑠𝑖𝑛(𝜋 −𝛼)

=

√

(𝑏_𝑥−𝑏_𝐶,𝑥)²+(𝑏_𝑦−𝑏_𝐶,𝑦)²×𝑠𝑖𝑛(𝜃),

0≤𝛼 −𝜃 ≤𝜋, 𝜋∕2≤𝛼 ≤𝜋, 0≤𝜃 ≤𝜋∕2,

𝑑𝐿,𝑆≤𝑏𝑥≤𝑑𝐿,𝑠+𝐵𝑤, 0≤𝑏_𝑦,

whereB_w representsthewidthof theroad,d_L,S denotesthedistance betweentheserverandtheroad.Foreachrequest,theoptimizationop- erationisexecutedontheserversideonce,andthensendtheestimated vehiclelocationtotheself-drivingvehicle.Throughcollectingtheselo- calizationinformationfromnearbyservers,theself-drivingvehiclecan determineitsdrivingstatebasedonthedecisionstrategy.

4.3. Theselectionofnearbyservers

Theproposedsystemassumesthatnearbyserversareemployedto assisttheself-drivingvehiclerealizevehicledetectionandlocalization, andtheycancommunicatewitheachother.AsdesignedinSection4.2, twonearestserversonthesamesideareselectedastheassistantinfras- tructuretolocatethevehicleaccurately.Thisisbecausetheclosertothe vehicle,thestrongerwirelesssignalfromthevehiclecanbereceivedby servers.Further,itcontributestomoreaccurateAOAestimation.

Asdemonstratedintheframework,servers,withinthecommunica- tioncoverageofself-drivingvehicles,willsendtheirinformationinclud- ingtheirIDandlocalizationtotheself-drivingvehicle.Subsequently, theself-drivingvehiclewillselecttwonearestserversonthesameside toassistaccuratevehiclelocalization.AsshowninFig.5,theantenna An_awillselectserversBandCsincetheyarethetwonearestservers whentheself-drivingvehicleisatposition1.Afteraperiodoftime, whenatposition2,serversCandDareselected.Detailedalgorithmfor theserverselectionaredemonstratedinAlgorithm1.

(5)

Fig.5. Thetravellingprocesswhenpassingthroughoneserver.

Algorithm1:Theserverselectionalgorithm.

input :(𝑥_𝑣,𝑎,𝑦_𝑣,𝑎):thepositionofvehicle𝑎;𝑆:thesetofservers thatareinthecommunicationrangeofthevehicle𝑎;𝑁:

Thenumberofserversinset𝑆;𝑆_𝑖:server𝑖;(𝑥_𝑠,𝑖,𝑦_𝑠,𝑖):the positionofserver𝑖

output:𝑆_𝑝𝑟𝑒:thesetofcommunicationbasestations

1 𝑑min←√

(𝑥𝑣,𝑎−𝑥(𝑠,1))²+(𝑦𝑣,𝑎−𝑦(𝑠,1))²;

2 𝑆𝑝𝑟𝑒←∅;

3 𝑚←0;

4 𝑘←0;

5 while(𝑗<𝑁)do

6 𝑑←√

(𝑥_𝑣,𝑎−𝑥(𝑠,𝑗))²+(𝑦_𝑣,𝑎−𝑦(𝑠,𝑗))²;

7 if(𝑑<𝑑min)then

8 𝑑min←𝑑;

9 𝑘←𝑚;

10 𝑚←𝑗;

11 𝑗←𝑗+1;

12 𝑆_𝑝𝑟𝑒←𝑆_𝑝𝑟𝑒∪𝑆_𝑚;

13 𝑆_𝑝𝑟𝑒←𝑆_𝑝𝑟𝑒∪𝑆_𝑘;

14 return𝑆𝑝𝑟𝑒;

5. Performanceevaluation

Extensivesimulationsareconductedtoverifythehigheﬃciencyof ECASS.

5.1. Simulationsettings

ThesimulationsaredevelopedbyMATLAB.AsshowninFig.6,there aretwolanesinthesamedirection.Eachlanewidthissetas3m.The distancebetweentwoadjacentserversalongtheroadissetas200m.

Thedistancefromtheservertothenearestlaneissetas10m.Vehicles travelalongtheXaxis.Therefore,thecoordinatesoftheﬁrstservercan besetas(0,0),andthecoordinatesofthesecondservercanbesetas (200,0).

Fig.6.Thesettingsinthesimulations.

5.2. Simulationresults

Inthesimulationpart,weevaluateECASSwithvariousvelocities, diﬀerentGPSerrors,anddiﬀerentnumberofvehicles.

5.2.1. Simulationwithvariousvelocities

Firstofall,wemimicthescenarioofonlyonevehicletravellingat aconstantspeedof10m/s.Thevehiclemovesinastraightlineoflane 1.TheGPSlocalizationerrorissetwithin2m,andtheangleestimation errorbasedonwirelesssignalsissetwithin5°.Accordingtothevelocity, realAOAvaluesoftheantennarelatedtotwonearbyserversareequalto arctan(11.5/10t)and𝜋 −𝑎𝑟𝑐𝑡𝑎𝑛(11.5∕(200−10𝑡)),asshowninFig.7(a) and(b),respectively.

Thesimulationresults aboutthetrajectorytracking areshownin Fig.8(a).Obviously,ECASSbasedtrajectory trackingisclosertothe vehiclerealtrajectorythanGPSbasedtrajectory.

The absolute error between ECASS based positionandreal position is equal to √

(𝑏_𝑡,𝑥−𝑏_𝑚,𝑥)²+(𝑏_𝑡,𝑦−𝑏_𝑚,𝑦)² for vehicle b, and the absolute error between GPS position and real position is equal to

√

(𝑏_𝑡,𝑥−𝑏_𝑥)²+(𝑏_𝑡,𝑦−𝑏_𝑦)².Thesetwokindofabsoluteerrorsarereferred toas𝐸_{𝐸𝐶𝐴𝑆}_𝑆−𝑅𝑒𝑎𝑙and𝐸_{𝐺𝑃 𝑆−}_{𝑅𝑒𝑎𝑙},respectively.

TheCDFof𝐸_{𝐸𝐶𝐴𝑆}_𝑆−𝑅𝑒𝑎𝑙and𝐸_{𝐺𝑃 𝑆−}_{𝑅𝑒𝑎𝑙}alongthetrajectoryisplot- tedinFig.8(b).Fromthisﬁgure,itcanbeobservedthattheabsolute localizationerrorsbasedonECASSaresmallerthanthatbasedonGPS.

WhentheGPSlocalizationerrors aresetwithin2m,themeanabso- luteerrorbetweenECASSbasedpositionandrealpositionisabout1m, whichis0.4mlessthanthatbetweenGPSbasedpositionandrealpo- sition.Consequently,itveriﬁesthehighlocalizationaccuracyof the proposedalgorithm.

Withregardtotheangleestimation,thesimulationresultsareplot- tedinFig.9(a)and(b),fortheAOAbetweenthevehiclewiththeserver behind,andtheAOAbetweenthevehicleandthefrontserver,respectively.BothﬁguresverifythatECASSbasedanglesaremuchcloserto therealanglesandremainmuchstablerthantheestimatedanglesbased onwirelesssignals.

In the meantime, we also investigate the angle values based on the serverselectionscheme whenpassingthrough oneserver in the travellingprocess.ThesimulationresultsareshowninFig.10(a)and (b).ItclearlydemonstratesthatthemeasuredanglesbasedonECASS can tracktherealanglemuchmore accuratelycomparedtowireless signalbasedanglesalongthetrajectory.Evenwhenthevehiclepasses throughoneserver,ECASSbasedangleestimationisstillmuchmore accurate.

Abovesimulationsarebasedontheconstantvelocity,whichcannot besatisﬁedinmostactualscenarios.Therefore,weinvestigatethesce- nariothatthevehiclevelocitychangesovertime.Thesetvelocitiesare showninFig.11(a)and(b),respectively,inwhichthehighestspeedof thevehicleislimitedto20m/s.Asshownintheleftﬁgure,thevehicle

(6)

0 5 10 15 20 Time(s)

0 0.5 1 1.5 2

Angle(rad)

(a)

0 5 10 15 20

Time(s) 1.5

2 2.5 3 3.5

Angle(rad)

(b)

Fig.7. Theanglevalueswhenthevehiclefollowsthe uniformmotioninastraightline.

0 50 100 150 200

X axis(m) 10

12 14 16 18

Y axis(m)

Real trajectory GPS based trajectory ECASS based trajectory

(a)

0 0.5 1 1.5 2

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

EGPS-Real E_ECASS-Real

(b)

Fig.8.Thetrajectorytrackingresultswhenthevehiclefol- lowstheuniformmotioninastraightline.

0 2 4 6 8 10 12 14 16 18 20 Time(s)

0 1 2

Angle(rad)

Real angle Wireless angle ECASS based angle

(a)

0 2 4 6 8 10 12 14 16 18 20 Time(s)

1 2 3 4

Angle(rad)

(b)

Fig.9. Theanglevalueswhenthevehiclefol- lowstheuniformmotioninastraightline.

0 4 8 12 16 20 24 28 32 36 40 Time(s)

0 1 2

Angle(rad)

(a)

0 4 8 12 16 20 24 28 32 36 40 Time(s)

1 2 3 4

Angle(rad)

(b)

Fig. 10. The angle values when passing throughoneserver.

(7)

0 2 4 6 8 10 12 14 16 18 Time(s)

9 10 11 12 13 14

Velocity(m/s)

(a)

0 2 4 6 8 10 12 14 16 18 Time(s)

0 20

Velocity(m/s)

(b)

Fig.11. Thesetvelocity.

0 50 100 150 200

X axis(m) 10

12 14 16

Y axis(m)

(a)

0 0.5 1 1.5 2

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

E_GPS-Real E_ECASS-Real

(b)

Fig.12. Thetrajectorytrackingwhentravellingwiththeuni- formaccelerationmotion.

0 2 4 6 8 10 12 14 16 18 Time(s)

0 1 2

Angle(rad)

(a)

0 2 4 6 8 10 12 14 16 18 Time(s)

1 2 3 4

Angle(rad)

(b)

Fig. 13. The angle values when travelling withtheuniformaccelerationmotion.

followstheuniformaccelerationmotion,andthevelocityintheright ﬁgureisirregular,whichisclosertorealisticscenarios.

ThesimulationresultsareplottedinFig.12(a)and(b).ECASSbased vehicletrajectory showsasuperiorperformancein termsofboththe trajectorytrackingandabsoluteerrorcomparedtothosebasedonGPS information.Inthemeantime,themeanabsoluteerrorbasedonECASS remainsabout0.4mlessthanthatbasedonGPS.

TheangleestimationisshowninFig.13(a)and(b).Wecanobserve thatECASSbasedanglemeasurementsnearlymatchwiththerealan- gles,whileanglesestimatedbasedonwirelesssignalsﬂuctuateinalarge range.

Finally, we carry out simulations based on the velocity set in Fig.11(b).Thesimulationresults shownin Fig.14(a)and(b)verify thatECASSbasedtrajectoryismoreaccuratethanGPSbasedtrajectory evenwhenthevehiclevelocityisirregular.

Therefore,itcanbeconcludedthatwhentheGPSlocalizationerror issetwithin2m,ECASSbasedtrajectorytrackingshowsasuperiorper-

formancecomparedtoGPSbasedtrajectorydespiteofvariousvehicle speeds.

5.2.2. SimulationwithdiﬀerentGPSerrors

AbovesimulationsarecarriedoutwiththeGPSlocalizationerrorset within2m.Inthissubsection,wewillexplorethesimulationscenario withdifferentGPSlocalizationerrorssincetheGPSlocalizationaccu- racycanbeinfluencedbydifferentfactors,suchastherefractioneffect causedbyionosphereandmultipatheffect.

Firstly,wesettheGPSlocalizationerrorwithin6m.Fig.15(a)and (b)demonstratethesimulationresultsintermsofthelocalizationaccu- racy.ECASSbasedtrajectoryismuchclosertotherealtrajectorycom- paredtothatbasedonGPS.Themeanabsoluteerroris2mlessthan thatbasedonGPS.

TheanglemeasurementsareshowninFig.16(a)and(b).Bothtwo AOAsbasedonECASSshowanaccurateestimationandstableperfor-

(8)

0 50 100 150 200 X axis(m)

10 12 14 16

Y axis(m)

(a)

0 0.5 1 1.5 2

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

(b)

Fig.14. Thetrajectorytrackingwhentravellingwithirregular velocities.

0 50 100 150 200

X axis(m) 5

10 15 20 25

Y axis(m)

(a)

0 2 4 6

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

(b)

Fig.15. ThetrajectorytrackingwhentheGPSerrorisset within6m.

0 2 4 6 8 10 12 14 16 18 20 Time(s)

0 1 2

Angle(rad)

(a)

0 2 4 6 8 10 12 14 16 18 20 Time(s)

1 2 3 4

Angle(rad)

(b)

Fig.16. TheanglevalueswhentheGPSerror issetwithin6m.

mancecomparedtotheestimatedAOAsbasedonwirelesssignalsduring thesimulationtime.

Finally,Fig.17(a)and(b)plotthetrackedtrajectorywhentheGPS localizationerror issetwithin10m.Thevehicletrajectorybased on ECASSshowsamuchbetterperformancethanthetrajectorybasedon GPS.Themeanabsoluteerrorbasedonthedesignedalgorithmisabout 3m,whichis4.1mlessthanthatbasedonGPS.

Finally, it is observed that the accuracy improvement based on ECASScanbeenhancedwithlargerGPSlocalizationerrors.Although themeanabsoluteerrorbasedonthedesignedalgorithmisabout3m whentheGPSerrorissetwithin10m.Wecanseethatthisrelatively largeerrorismainlycausedbythelocalizationerrorinXaxis.Incon- vention,thereshouldexistalargesafetydistancebetweentwonearby vehicles.Therefore,thesemuchsmallererrorsincomparisonwiththe safetydistancehaveasmallinﬂuenceonthedrivingstatedetermina- tion.Yet,thelargeGPSlocalizationerrorwillleadtowrongdetermina- tions.WecanconcludethatECASScanworkmoreeﬃcientlywhenthe GPSerrorsbecomelarger.

5.2.3. Simulationwithtwovehicles

Inthesubsection,wewilldeploytwovehiclestodemonstratethe higheﬃciencyandrobustnessofECASS. Thevelocitiessetforthese twovehiclesareshowninFig.11(a)and(b).Inordertosimulatemore complexscenarios,vehicle1willchangethelaneinthe10secondwith theinitiallocationsetinlane2,whilevehicle2willchangethelane inthe4second.Theinitialpositionofvehicle1issetas(0,0),andthe initialpositionofvehicle2issetas(20,0).

Thesimulationresults forvehicle1areplotted inFig.18(a)and (b).ItclearlydemonstratesthatthetrajectorybasedonECASSperforms muchbetterthanGPSbasedtrajectory.Itcanbeseenthatthetrajec- torycanalsoaccuratelytracktherealtrajectoryevenwhenthevehicle changesitslaneinthetravellingprocess.

Thesimulationresultsoftheestimatedanglevaluesunderthesce- nariooftwovehiclesareshowninFig.19(a)and(b).Itveriﬁesthat ECASScanachieveahighaccuracyofAOAestimationevenwhenthe vehiclechangesitslane.

(9)

0 50 100 150 200 X axis(m)

0 10 20 30

Y axis(m)

(a)

0 2 4 6 8 10

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

(b)

Fig.17. ThetrajectorytrackingwhentheGPSerrorisset within10m.

0 50 100 150 200

X axis(m) 0

10 20 30

Y axis(m)

(a)

0 2 4 6 8 10

Absolute error(m) 0

0.2 0.4 0.6 0.8 1

(b)

Fig.18. Thetrajectorytrackingwhensettingtwovehicles.

0 2 4 6 8 10 12 14 16 18 Time(s)

0 1 2

Angle(rad)

(a)

0 2 4 6 8 10 12 14 16 18 Time(s)

1 2 3 4

Angle(rad) Real angle

Wireless angle ECASS based angle

(b)

Fig.19. Theanglevalueswhendeployingtwo vehicles.

6. Conclusionandfuturework

Withtherapiddevelopmentoftheautopilottechnique,self-driving vehicles has attracted substantial attention from both industry and academia.Majorityofavailableliteraturehasfocusedonthehardware designsuchasvehicleborneradarandcamera,algorithmdesignre- latedtotheinformationfusion,andvehicletrajectoryplanning.How- ever,mosttrajectoryplanningalgorithmsarebasedontheassumption thatradar,camera,andIMUcanperceivetheenvironmentaroundthe self-drivingvehicle.Nevertheless,inrealtraﬃcscenesthevehiclemay beblockedbythetruckorbusaheadorbehind,theradarorcamerain- tegratedwithinthevehiclecannotsensethesurroundingenvironments accurately.Inaddition,GPSbasedlocalizationaccuracycannotensure thesafetyforautomaticdriving.

Therefore,wecombineself-drivingvehicleswithedgecomputingto realizenearbyvehicledetectionandlocalizationwhenself-drivingvehi- clesarepartiallyorevencompletelyblockedbytrucksorbuses.Lever- agingtheroadsideinfrastructure,accuratevehiclelocalizationnearthe

self-drivingvehiclecanbeachieved.Theseinformationwillbedelivered totheself-drivingvehicle,basedonwhichitcandeterminethedriving stateforthenextmoment.Extensivesimulationresultshaveveriﬁedthe higheﬃciencyoftheproposedsystem.

Althoughwehaveproposedanovelframeworkusingtheroadside infrastructuretolocatevehiclesintheproximityoftheself-drivingvehicle,thereexistsomedeﬁciencies,whicharelistedasbelow.

• Eﬃcient communicationstrategiesarenecessarytocut down the transmissiontimeconsumptionsincethetimecostplaysavitalrole inthereactionofautopilotsystem.

• Inthefuturework,thevehiclelocalizationalgorithmshouldbede- signedincombinationwiththetrajectoryplan,inordertodealwith variouskindsoftraﬃcconditions.

• Insteadofsimulations,realtraceexperimentsshouldbecarriedout totestthedesignedschemeinactualscenarios.

(10)

Acknowledgments

ThisworkispartlysupportedbyNationalKeyResearchandDevelop- mentProgramgrant2016YFE0100600andNSFC61672349,61672353, 61472252.

Supplementarymaterial

Supplementarymaterialassociatedwiththisarticlecanbefound,in theonlineversion,atdoi:10.1016/j.sysarc.2019.02.014.

References

[1] G.H. Lee , F. Faundorfer , M. Pollefeys , Motion estimation for self-driving cars with a generalized camera, in: in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2013, pp. 2746–2753 .

[2] E. Ackerman , Lidar that will make self-driving cars aﬀordable [news], IEEE Spectr.

53 (10) (2016) 14 .

[3] J. Carlson, Mapping large, in: Urban Environments with GPS-aided Slam.

[4] H. Xiong , L.N. Boyle , Drivers’ adaptation to adaptive cruise control: examination of automatic and manual braking, IEEE Trans. Intell. Transp. Syst. 13 (3) (2012) 1468–1473 .

[5] J.V. Mierlo , G. Maggetto , E.V. de Burgwal , R. Gense , Driving style and traﬃc mea- sures-inﬂuence on vehicle emissions and fuel consumption, Proc. Inst. Mech. Eng.

Part D 218 (1) (2004) 43–50 .

[6] R.W. Wolcott , R.M. Eustice , Visual localization within LIDAR maps for automated urban driving, in: in: International Conference on Intelligent Robots and Systems, IEEE, 2014, pp. 176–183 .

[7] M.A. Brubaker , A. Geiger , R. Urtasun , Lost! leveraging the crowd for probabilistic visual self-localization, in: in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2013, pp. 3057–3064 .

[8] J. Xing , H. Dai , Z. Yu , A distributed multi-level model with dynamic replacement for the storage of smart edge computing, J. Syst. Archit. 83 (2018) 1–11 . [9] P. Petrov , F. Nashashibi , Modeling and nonlinear adaptive control for autonomous

vehicle overtaking, IEEE Trans. Intell. Transp. Syst. 15 (4) (2014) 1643–1656 . [10] C.M. Sosa-Reyna , E. Tello-Leal , D. Lara-Alabazares , Methodology for the model–

driven development of service oriented IoT applications, J. Syst. Archit. 90 (2018) 15–22 .

[11] A. Houenou , P. Bonnifait , V. Cherfaoui , W. Yao , Vehicle trajectory prediction based on motion model and maneuver recognition, in: in: International Conference on Intelligent Robots and Systems, IEEE, 2013, pp. 4363–4369 .

[12] D.C.K. Ngai , N.H.C. Yung , A multiple-goal reinforcement learning method for complex vehicle overtaking maneuvers, IEEE Trans. Intell. Transp. Syst. 12 (2) (2011) 509–522 .

[13] E. Coelingh, S. Solyom, All aboard the robotic road train, IEEE Spectr. 49(11).

[14] M.A. Quddus , W.Y. Ochieng , R.B. Noland , Current map-matching algorithms for transport applications: state-of-the art and future research directions, Transp. Res.

Part C 15 (5) (2007) 312–328 .

[15] N.M. Drawil , H.M. Amar , O.A. Basir , et al. , GPS localization accuracy classiﬁcation:

a context-based approach, IEEE Trans. Intell. Transp. Syst. 14 (1) (2013) 262–273 . [16] A.K.S. Rajan , A. Feucht , L. Gamer , I. Smaili , et al. , Hypervisor for consolidating

real-time automotive control units: its procedure, implications and hidden pitfalls, J. Syst. Archit. 82 (2018) 37–48 .

[17] T. Fanni , L. Li , T. Viitanen , C. Sau , R. Xie , F. Palumbo , L. Raﬀo , H. Huttunen , J. Takala , S.S. Bhattacharyya , Hardware design methodology using lightweight dataﬂow and its integration with low power techniques, J. Syst. Archit. 78 (2017) 15–29 .

[18] P. Pelliccione , E. Knauss , R. Heldal , S.M. Ågren , P. Mallozzi , A. Alminger , D. Bor- gentun , Automotive architecture framework: the experience of volvo cars, J. Syst.

Archit. 77 (2017) 83–100 .

[19] P. Falcone , F. Borrelli , J. Asgari , H.E. Tseng , D. Hrovat , Predictive active steering control for autonomous vehicle systems, IEEE Trans. Control Syst. Technol. 15 (3) (2007) 566–580 .

[20] Q. Li , N. Zheng , H. Cheng , Springrobot: a prototype autonomous vehicle and its algorithms for lane detection, IEEE Trans. Intell. Transp. Syst. 5 (4) (2004) 300–308 . [21] J. Ziegler , P. Bender , M. Schreiber , H. Lategahn , T. Strauss , C. Stiller , T. Dang , U. Franke , N. Appenrodt , C.G. Keller , et al. , Making bertha drive-an autonomous journey on a historic route, IEEE Intell. Transp. Syst. Mag. 6 (2) (2014) 8–20 . [22] N. Jeong , H. Hwang , E.T. Matson , Evaluation of low-cost LiDAR sensor for applica-

tion in indoor UAV navigation, in: in: Sensors Applications Symposium, IEEE, 2018, pp. 1–5 .

[23] L. Meier , P. Tanskanen , L. Heng , G.H. Lee , F. Fraundorfer , M. Pollefeys , PIXHAWK:

a micro aerial vehicle design for autonomous ﬂight using onboard computer vision, Auton. Robots 33 (1–2) (2012) 21–39 .

[24] J. Wei , J.M. Snider , J. Kim , J.M. Dolan , R. Rajkumar , B. Litkouhi , Towards a viable autonomous driving research platform, in: in: Intelligent Vehicles Symposium, IEEE, 2013, pp. 763–770 .

[25] U. Franke , D. Pfeiﬀer , C. Rabe , C. Knoeppel , M. Enzweiler , F. Stein , R. Herrtwich , Making bertha see, in: in: Proceedings of the IEEE International Conference on Com- puter Vision Workshops, 2013, pp. 214–221 .

[26] S.J. Park , T.Y. Kim , S.M. Kang , K.H. Koo , A novel signal processing technique for vehicle detection radar, in: in: MTT-S International Microwave Symposium Digest, Vol. 1, IEEE, 2003, pp. 607–610 .

[27] G. Alessandretti , A. Broggi , P. Cerri , Vehicle and guard rail detection using radar and vision data fusion, IEEE Trans. Intell. Transp. Syst. 8 (1) (2007) 95–105 . [28] S.J. Anderson , S.C. Peters , T.E. Pilutti , K. Iagnemma , An optimal-control-based

framework for trajectory planning, threat assessment, and semi-autonomous control of passenger vehicles in hazard avoidance scenarios, Int. J. Veh. Auton. Syst. 8 (2–4) (2010) 190–216 .

[29] S. Glaser , B. Vanholme , S. Mammar , D. Gruyer , L. Nouveliere , Maneuver-based trajectory planning for highly autonomous vehicles on real road with traﬃc and driver interaction, IEEE Trans. Intell. Transp. Syst. 11 (3) (2010) 589–606 .

[30] V. Milanés , D.F. Llorca , J. Villagrá, J. Pérez , C. Fernández , I. Parra , C. González , M.A. Sotelo , Intelligent automatic overtaking system using vision for vehicle detection, Expert Syst. Appl. 39 (3) (2012) 3362–3373 .

[31] U. Lee , S. Yoon , H. Shim , P. Vasseur , C. Demonceaux , Local path planning in a complex environment for self-driving car, in: in: IEEE 4th Annual International Con- ference on Cyber Technology in Automation, Control, and Intelligent Systems, IEEE, 2014, pp. 445–450 .

[32] S. Sardellitti , G. Scutari , S. Barbarossa , Joint optimization of radio and computa- tional resources for multicell mobile-edge computing, IEEE Trans. Signal Inf. Pro- cess. Netw. 1 (2) (2015) 89–103 .

[33] C. Yang , H.R. Shao , WiFi-based indoor positioning, IEEE Commun. Mag. 53 (3) (2015) 150–157 .

[34] C. Chen , Y. Chen , Y. Han , H.Q. Lai , K.R. Liu , Achieving centimeter-accuracy indoor localization on wiﬁ platforms: a frequency hopping approach, IEEE Internet Things J. 4 (1) (2017) 111–121 .

Xiong Wang received the B.S. degree in electronic informa- tion engineering from the Wuhan University of Science and Technology in 2013 and the master’s degree in information and communication engineering from the Huazhong Univer- sity of Science and Technology in 2016. He is currently pur- suing the Ph.D. degree with the Department of Computer Sci- ence and Engineering, Shanghai Jiao Tong University, China.

His research interests include wireless networks and mobile computing.

Tianpeng Wei is currently studying in Shanghai Jiao Tong University as a junior student. He majors in computer science.

Linghe Kong received the B.E. degree from Xidian Univer- sity in 2005, the Dipl.-Ing. degree from TELECOM SudParis in 2007, and the Ph.D. degree from Shanghai Jiao Tong Univer- sity in 2012. He was a Post-Doctoral Fellow with Columbia University and McGill University. He is currently a Research Professor with Shanghai Jiao Tong University, China. His research interests include wireless communications and mobile computing.

Liang He is an assistant professor at University of Colorado Denver. His research mainly focuses on cyber-physical systems, IoTs, and mobile computing. Before joining UCD, he worked as a research fellow at The University of Michigan at Ann Arbor, MI, USA, with Prof. Kang G. Shin, as a Research Scientist at Singapore University of Technology and Design, Singapore, with Dr. Yu (Jason) Gu, and as a research assistant at University of Victoria, Canada, with Prof. Jianping Pan. He is a senior member of IEEE and a member of ACM.

(11)

Fan Wu is a professor with the Department of Computer Sci- ence and Engineering, Shanghai Jiao Tong University. He received the BS degree in Computer Science from Nanjing Uni- versity, in 2004, and the PhD degree in Computer Science and Engineering from the State University of New York at Buﬀalo, in 2009. He has visited the University of Illinois at Urbana-Champaign (UIUC) as a Post Doc Research Associate.

His research interests include wireless networking and mobile computing, algorithmic game theory and its applications, and privacy preservation. He has published more than 100 peer- reviewed papers in technical journals and conference proceedings.

Guihai Chen received the B.S. degree from Nanjing University in 1984, the M.E. degree from Southeast University in 1987, and the Ph.D. degree from the University of Hong Kong in 1997. He is a Distinguished Professor with Shanghai Jiaotong University, China. He had been invited as a Visiting Professor for many universities, including the Kyushu Institute of Tech- nology, Japan, in 1998, the University of Queensland, Aus- tralia, in 2000, and Wayne State University, USA, from 2001 to 2003. He has a wide range of research interests with focus on sensor network, peer-to-peer computing, and high perfor- mance computer architecture.