/
Distributed Control Applications Within Sensor Networks BRUNO SINOPOLI  STUDENT MEMBER Distributed Control Applications Within Sensor Networks BRUNO SINOPOLI  STUDENT MEMBER

Distributed Control Applications Within Sensor Networks BRUNO SINOPOLI STUDENT MEMBER - PDF document

mitsue-stanley
mitsue-stanley . @mitsue-stanley
Follow
586 views
Uploaded On 2014-12-12

Distributed Control Applications Within Sensor Networks BRUNO SINOPOLI STUDENT MEMBER - PPT Presentation

SHANKAR SASTRY FELLOW IEEE Invited Paper Sensor networks are gaining a central role in the research community This paper addresses some of the issues arising from the use of sensor networks in control applications Classical control theory proves to ID: 22990

SHANKAR SASTRY FELLOW

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "Distributed Control Applications Within ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

DistributedControlApplicationsWithinSensorNetworks BRUNOSINOPOLI,STUDENTMEMBER,IEEE,COURTNEYSHARP,LUCASCHENATO,STUDENTMEMBER,IEEE,SHAWNSCHAFFERT,STUDENTMEMBER,IEEES.SHANKARSASTRY,FELLOW,IEEEInvitedPaperSensornetworksaregainingacentralroleintheresearch keysissuespresentingthemselvestimeandtimeagain:locationdetermination,timesynchronization,reliablecom-munication,powerconsumptionmanagement,cooperationandcoordination,andsecurity.Thegoalofourresearchistodesignrobustcontrollersfordistributedsystemsthatviolatetypicalcontrolassumptions.Designedcontrollerswillbeevaluatedonadistributedcon-trolapplicationtestbed.Amongthewealthofavailableap-plications,wehaveselectedapursuit–evasiongame(PEG)application.Inourparticularapplication,theSNisdeployedintheenvironmentwherethegameisplayedandcooperateswiththepursuers’team.Thisapplicationincludesmanyinterestingresearchproblemsintheareasoftracking,controldesign,security,androbustness.ForaPEG,theSNmustbecapableofmultiple-vehicletrackingthatcandistinguishpursuersfromevaders.Furthermore,thenetworkneedstohaveadynamicroutingstructuretodeliverinformationtopursuersinminimaltime.Sincethegamewillbeplayedinadistributedfashion,distributedsensing,control,andactuationneedtobeaccountedforduringcontrollerdesign.Topreventtheevader’steamfrominterceptingsensitiveinformation,thenetworkmustprovidesecurityfeatures.Finally,sinceanyonenodeofanSNcanfail,controlalgorithmsshouldshowgracefulperformancedegradation.II.PEGTheframeworkofPEGscapturesfundamentalfeaturesformodelingmultiagentsincooperativeroboticsandhasbeenanactiveareaofresearchinthepastdecades.Inthissection,wegiveabriefoverviewoftheresearchhistoryonPEGs,describetheadvantagesofaddingSNstostandardPEGs,andenumerateadditionalissuesthatarisewhenusingSNs.A.PEGOverviewPEGsareamathematicalabstractionarisingfromnumeroussituationswhichaddressestheproblemofcon-trollingaswarmofautonomousagentsinthepursuitofoneormoreevaders.Typicalexamplesaresearchandrescueoperations,surveillance,localizationandtrackingofmovingpartsinawarehouse,andsearchandcapturemissions.Insomecases,theevadersareactivelyavoidingdetection,asincapturemissions,whereasinothercasestheirmotionisapproximatelyrandom,asinrescueoperations.DifferentversionsofPEGshavebeenanalyzedaccordingtodifferentframeworksandassumptions.DeterministicPEGsonfinitegraphshavebeenextensivelystudied[14],[15].Inthesegames,theplayingfieldisabstractedtobeafinitesetofnodes,andtheallowedmotionsforthepursuersandevadersarerepresentedbyedgesconnectingnodes.Anevaderiscapturedifboththeevaderandoneofthepursuersoccupythesamenode.Oneofthemostimportantproblemsarisingfromthisframeworkisthecomputationofthesearch,i.e.,thesmallestnumberofpursuersnecessarytocaptureasingleevaderinafinitetime,regardlessoftheescapingpolicyadoptedbytheevader.IthasbeenshownthatthisproblemisNP-hard[15],[16].Thisapproachislimitedonlytoworstcasemotionsoftheevaders,anditisingeneraloverlypessimistic.Agreatdealofresearchhasfocusedonhowtoreduceacontinuousspaceintoadiscretenumberofregions,eachtobemappedintoanodeofthegraph,sothatthegameonthereducedgraphisequivalenttotheoriginalgameinthecontinuousspace.Forexample,LaValleetal.proposedamethodofdecomposingthecontinuousspaceintoafinitenumberofregionsforpolygonalenvironments[17]andsimplyconnected,smooth-curved,two-dimensional(2-D)environment[18].AnotheractiveareaofresearchdealswithPEGswheretheenvironmentis.Inthisframework,anaddi-phaseisrequiredtoprecedethepursuitphase.Themap-learningphaseis,byitself,time-consumingandcomputationallyintensiveevenforsimple2-Drectilinearenvironments[19].Moreover,inaccuratesensorscomplicatethisprocessandaprobabilisticapproachisoftenrequiredrequiredFinally,arecentapproachtoPEGshasdealtwithcom-biningmaplearningandpursuitintoasingleproblem.Thisisdoneinaprobabilisticframeworktoavoidtheconserva-tivenessinherentinworstcaseassumptionsonthemotionoftheevader.Aprobabilisticframeworkalsonaturallytakesintoaccountinaccuratesensorreadings,uncertainapriorimapofterrain,andevadersmotionpolicies[21],[22].B.SNsinPEGsTheuseofanSNcangreatlyimprovetheoverallperfor-manceofaPEG.Pursuershavearelativelysmalldetectionrange.Theyusuallyemploycomputervisionorultrasonicsensors,providingonlylocalobservabilityovertheareaofinterest.Thisconstraintmakesdesigningacooperativepur-suitalgorithmharderbecauselackofcompleteobservabilityonlyallowsforsuboptimalpursuitpolicies(seeFig.1).Fur-thermore,asmartevaderisdifficulttocatchunlessappropri-atelydetected.Communicationamongpursuersmaybedifficultoveralargearea.Lackofcommunication,evenpartially,amongpursuersisamajordisruptionforanypursuitpolicy.Becauseoftheexpenseofunmannedvehicles,itisunrealistictode-ployalargenumberofthemtocontinuouslymonitoralargeregion.WithSNs,completevisibilityofthefieldandcommunica-tionoveralongradiusispossible(seeFig.2).Globalpursuitpoliciescanthenbeusedtoefficientlyfindtheoptimalsolu-tionregardlessofthelevelofintelligenceoftheevader.Also,withanSN,thenumberofpursuersneededislikelyafunc-tionexclusivelyofthenumberofevadersandnottothesizeofthefield.ThisdistributedPEG(DPEG)scenarioexposesanumberofissuesfundamentaltoanySN.Resolvingtheseissuesiscomplicatedbythedesiretomakethesolutionsrobusteveninadynamicadhocnetwork.Time—Thenotionoftimepresentstwodistinctproblems.First,coordinatingsensingandactuatinginthephysicalworldrequireseitherasenseofglobaltimeortheabilitytoresolvedifferenttimemeasurementstoameaningfulrepresentation.Second,manyexistingdesigntechniquesPROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003 Fig.1.PEG:whatpursuersreallysee. Fig.2.PEG:SNincreasesvisibility.etal.:DISTRIBUTEDCONTROLAPPLICATIONSWITHINSENSORNETWORKS1237 assumethatthecomputationofcontrolandtheprocessingofsensingandactuationoccurwithinanegligibleamountoftime,thusrequiringnewdesignandanalysistechniquesforSNs.—Itisexpectedthatanetworkofmoteswillspanaspatialareasignificantlygreaterthanasinglemote’smaximumcommunicationarea.Foramotetosendamessagetoanother,distantmote,intermediatemotesmustbeabletorelaythemessage.Additionally,becausemotescangoofflinewithoutwarning,theunderlyingcommunica-tionprotocolmustberobusttonetworkchanges.—Sensingandactuatingeventsinthephysicalworldmustbepairedwiththerelativeorabsolutelocationofthemotetobeusefultocontrolalgorithms.Thatlocationmustbeassumed,provided,ordeduced.Cooperation—Tasksthatrequirethecombinedeffortoftwoormoremotes,suchasanyformofdistributedsensingordistributedcomputing,requireprotocolsandstructuresthatprovidehandshaking,coordination,andpossiblyhierarchy.Power—EnergyisavaluedresourceinanSN.ServiceandperformanceguaranteesprovidedbyanSNmustbebalancedagainstoverallpowerconsumption.—TopreventnumerouspotentialabusesofanSN,acommunicationsecuritylayermustprovideknownguaran-teesforaccesscontrol,messageintegrity,andconfidentiality.WhendevelopingcontrolapplicationsonanSNplatform,weareparticularlyinterestedwithissuesrelatedtotime,communication,andlocation.WewillfocusontheseissuesC.DPEGsTostartourDPEGscenario,themotescomposingtheSNaredeployedontotheplayingfieldinasleepstate.ThemoteSNthengoesthroughaninitializationandcalibrationstageforbootstrappingtheirprovidedservices.Thepursuersandevadersthenentertheplayingfieldandremainwithinthefieldforthedurationofthegame.TheSNprovidesavarietyofservicestobothpursuersandothersensormotes:timesynchronization,localization,movingentity(pursuerorevader)estimation,etc.Forthepurposeofthegame,thesolegoaloftheseservicesistopro-duceestimatesonthepositions,velocity,andidentityofen-titiesintheplayingfield.Thisinformationistime-stampedandroutedtoallpursuersintheplayingfield.Thepursuershaveonboardcomputationfacilitiescomparabletoalaptopcomputer.Wemaychoosetohavethepursuerscommunicatethroughaseparaterobustchanneltocoordinatetocapturetheevaderwhenandifthatchannelisavailable.Whenallevadersarecaptured(acaptureoccurswhenapursueris“closeenough”toit),thegameends.Abasesta-tionisoutsidetheplayingareaandprovidesloggingandvi-sualizationservices.III.IMPLEMENTATIONOurimplementationsspanhardware,software,andvar-iousapplicationscenariostoexploreanddemonstratedis-tributedcontrolviaSNs.Inthehardwaresection,wediscussourcurrentembeddednetworkdevices.Then,inthesoftwaresection,wereviewournewprogramminglanguage,oper-atingsystem(OS),andsystemservicearchitecture.Finally,wesurveyourcurrentandfuturetestbedsforinteractingandlearningatthewhole-systemlevel.A.HardwareThehardwareplatformdevelopedbytheTinyOSgroupatBerkeleyconsistsofnumerous,small,extendableembeddednetworkdevices.Eachdevicehaslimitedpower,computa-tion,andstorageresources—significantlylimitedwhencom-paredtomoderndesktopcomputersystems.Thegoalofeachhardwareplatformistoprovidecomputation,sensing,actu-ation,andcommunicationresourcesembeddedinminiaturepackaging.Bymakingtheconsciousdesigndecisiontosig-nificantlylimittheresourcesavailablepermote,weleavethedooropenforreachingthegoalofdust-sizeddevices.Thecurrentplatformsaredesignedtobebothmodularandflexible,providingeaseinretargetingmotestonewandunanticipatedapplicationswhileallowingforsignificantcodereuse.Fig.3showstheevolutionofthebasecomputa-tionmodules.Inparticular,themostrecenttransitionfromtheRene2/DottoMica(seeFig.4)gaveatleastafourfoldincreaseinprogrammemory,RAM,andradiotransmissionrate.Allmoteshaveotherwisehadsomeformofa4-MHz,8-bAtmelmicrocontrollerandanRFMTR1000radio.Add-onboardsforthemotesmaybedesignedforgeneral-purposesensingortargetedtowardaparticularapplication.Forinstance,theweathersensingboardhashumidity,baro-metricpressure,infrared,temperature,andlightsensors,andisusedforexperimentsonGreatDuckIsland[23]inMaine.Andamotor/servoandwhisker/accelerometerboardweredevelopedforCOTS-BOTS[24](seeFig.5)forcontrollingoff-the-shelfminiaturecars.Wealsohavevariousgeneral-purposesensorboardsthathavesomecombinationofpho-todiodes,temperaturesensors,magnetometers,accelerome-ters,microphones,andsounders.Theoverallmodularityofthesedevicescomesatthecostofsize.Adevicetargetedatlarge-scaledeploymentcandoawaywiththeadd-onconnectorandsupportingcircuitry.Theresultingspacesavingsinthecurrentplatformseasilyallowsforafinalform-factorwithdiametersmallerthanaquarter.Alltogether,thehardwareplatformshavebeensufficienttomeettheneedsofbothresearchandexperimentation.B.SystemServicesWebuildourembeddedsoftwarewithNesC[25],anew,open-sourceprogramminglanguagedevelopedatBerkeley.NesCextendsthestandardClanguagewithsemanticsandsyntaxforcomponent-basedarchitectures.Componentbe-haviorsaredescribedwithbidirectionalinterfacesthateitherprovidecommandsorrequirethedependenttohandleevents.Componentsarestaticallywiredtogethertoformawholeprogramorsystem,which,whencompiledwithawhole-programcompiler,allowsforgreateroptimizationsandef-ficiency.Wholeprogramsalsomatchwellwithformalanal-ysistoolsforverifyingsystemfunctionality.PROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003 Fig.3.EvolutionofmotesfromtheBerkeleyTinyOSgroup. Fig.4.Micamote 1  withattachedweatherboardmodule.Berkeley’sopen-sourceembeddedOS,TinyOS[26],providesbasicsystemservices,suchascommunicationandsimpleprocessscheduling,andaccesstohardwarecomponents,suchassensorsandactuators.Itisspecificallydesignedforextremelyresource-limiteddevicesthathaveonlyafewkilobytesofmemory.TinyOSiswritteninNesCusingacomponent-basedarchitecturewithlayeredaccesstohardwareresources,whichprovidesrobustness,flexibility,andextensibility.UsingNesCandTinyOSasbuildingblocks,wehavebeenworkingwithanumberofothergroupsontheNESTprojectfundedbytheDefenseAdvancedResearchProjectsAgency(DARPA)todevelopacoherentarchitectureofsystemser-vicestohelpsolvefundamentalSNstumblingissues.The Fig.5.COTS-BOTSdevelopedbyS.BergbreiterandK.Pister.crucialserviceswehavecurrentlyidentifiedareestimation,grouping,localization,powermanagement,routing,servicecoordination,andtimesynchronization.WefeelthatthesecomponentswillfacilitatealargesetofrichandadaptiveToaddresstimeissueswithinanSN,weproposeatimesynchronizationapplicationprogramminginterface(API)thatsupportstwotimemanagementprotocols:aglobalNetworkTimeProtocol(NTP)-likesynchronizationpro-tocol,andalocaltimeprotocolwiththemeanstotransformtimereadingsbetweenindividualmotes.ItisexpectedthatNTP-likeglobalsynchronizationwillofferlowerprecisionetal.:DISTRIBUTEDCONTROLAPPLICATIONSWITHINSENSORNETWORKS1239 timemeasurements,butotherwiseprovideanimmediatelyavailableglobaltimeonthemote.Localtransformationsbetweenindividualmote“timezones”havetheadvantagesofhigherprecisionbetweenpairsofmotes,beingabletoback-calculatesynchronizedtimesforpastevents,andguar-anteesmonotonicityinlocaltimebynotdirectlymodifyingthelocalclock[8].Variousapplicationscanhavevastlydifferenttimesynchronizationrequirements,andwefeelthesetwomethodologiestogethercanmoreadequatelyserveabroadsetofapplications.ToaddresscommunicationissueswithinanSN,wepro-poseageneralroutingframeworkthatsupportsanumberofroutingmethodologies.First,becauseSNsprimarilysenseandinteractwithphenomenainthephysicalworld,routingtogeographicregionsisexpectedtobethecommoncase.Second,toassistinroutingpacketsaroundphysicalobsta-cles,routingbasedongeographicdirectionisexpectedtobeuseful.Third,themoreobviouscaseofroutingtosym-bolicnetworkidentifiersisreservedfordynamicallyroutingtophysicallymovingdestinationswithinthenetwork.Fi-nally,thegeneralcaseofconstraint-basedroutingprovidesmeanstoroutebasedonarbitrarycriteria,suchaspowerlevel,sensorvalues,andsoon.ToresolvethephysicallocationofmotesinanSN[9],weproposeatop-to-bottomlocalizationframework.Alo-calizationservicerequiresabroadsetofcoordinationandprocessingstagesbetweenmotes:coordinatedsensorsandactuators,groupdatamanagement,andcomputation.Sepa-ratinglocalizationintoanumberofdistinctcomponentsthatworktogetherallowsforanamountofheterogeneityintheSNthatmaybenecessarygiventhelimitedresourcesoftheTheissueofcoordinationbetweenmotesneedstobead-dressed.Weproposebothapplication-specificgroupingal-gorithmsandgeneral-purposegroupingservices.Agroupmanagementservicemustprovidemeanstosendandre-ceivedatafromagroup,theabilitytojoinandleaveagroup,andleaderelection.FortrackingamovingevaderinaPEGscenario,decisionstojoinandleavegroupscanbetiedtosensorreadings.Thissimplifiesthehandshakinganddeci-sionprocess,allowingforoverallloweroverhead.Thereareconcernsthattheseservicesinthegeneralcaseimposesig-nificantoverheadonanSN.Issuesofpowermanagementareontheagendabutarecur-rentlyunaddressedinthearchitecture.IssuesofsecurityarebeingsolvedattheOSlevel,providingtransparentauthenti-cation,encryption,andconcealment.Ourmethodologyforcreatinganinfrastructurefortheseservicesistofirstspecifyasetofprototypesthatdefineab-stractprogramminginterfacesforclassesofcomponentsandservices.Developersthencreatethatinstantiateusingoneormoreprototypes.Somecompo-nentsmaybehaveaswhoseexecutionandbehavioraremanagedbyacentralcoordinator.Finally,interactionbe-tweencomponentsmustbeformalizedbyspecifyingproto-.Fig.6illustratesthismethodologyinasamplearchitecturethatshowstheinteractionsandprotocolsbe-tweencomponents,services,aservicecoordinator,sensors, Fig.6.Samplecomponentarchitecturedemonstratingthedesignmethodology.andradiochannel.Fig.7furthershowstherelationshipbe-tweentheseservices,components,TinyOS,andaDPEGap-plicationlayer.C.TestbedsOurcurrentexperimentalplatformisfunctionalbutlim-itedwhencomparedtothescopeofafullDPEGscenario.Itistheresultofafocusedefforttoproduceasolutionforasetofparticulargoalsratherthantoprovideageneralframe-work.Tothatend,itexistsmoreasaproofthatahighlyconstrainedDPEGsolutionisachievableandthatNesCandTinyOSprovideasuitableplatformfordevelopment.Fig.8showsthesetupforthatplatform.Ahumanre-motelycontrolsaminiaturecar,andtheSNremotelycontrolsapan–tilt–zoomcameratotrackthecar.Becausewehavenotyetintegratedaself-localizationserviceonthemotes,theSNisauniformgridof25motes,whereeachmotepresumesitslocationgivenitsnetworkaddress.Eachmotesharesitslo-cationwithitslocalneighborhood,whichisnecessarybothforpositionestimationandgeographic-basedrouting.Whenamotedetectschangeinitslocalmagneticfield,itbroad-castsitsreadingstoitslocalneighborsandrecordssimilarbroadcastsfromother,nearbymotes.Inthisway,localbe-haviorsareexpected;currently,wearenotattemptingtoag-gregatereadingsfromtheentirenetworktoproduceasingle,globalestimate.Themotewiththehighestreadingisim-plicitlyelectedtheleader,whocalculatesapositionestimatefromitscachedneighborhoodreadings.Thatestimateissentviareliablegeographic-basedmultihoproutingtoabasesta-tionmote,whichrelaysittoacameramote.Thecameramoteperformstheactuationnecessarytopointtowardtheesti-matedlocation.WhatwewouldliketodoistousetheSNsoftwarearchi-tecturetoimplementthisscenarioinamoreversatile,gen-eralframework.Wearelookingforwardtoamorecomplete,outdoorPEGscenario,showninFig.9.Beyondthatsce-nario,welookforwardtoexpandingourunderstandingofwhole-systembehaviorthroughformalismandparameteri-zationofdistributedSNs.IV.MOurinitialDPEGimplementationhasprovidedvaluableinsightintothepressingissuesthatacontroldesignmethod-PROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003 Fig.7.Relationshipbetweenproposedservicesandcomponents. Fig.8.Indoorsensor-basedtrackingtestbed. Fig.9.OutdoorDPEGtestbed.ologymustaddress,andwewillusetheseideastoinformourproposeddesignmethodology.Inthissection,wewillfirstreviewexistingdesignapproachesthataddresstheissuesofscalabilityanddistribution.Duringthisdiscussion,wewillbeinterestedinextractingtheessenceofexistingalgorithmswhileabstractingawaytheparticularchoiceofmodel.Fol-lowingthis,modelsofcomputation(MOCs)willbeexploredthatareusefulfordescribingsuchsystems.Finally,wewilldiscussourproposeddesignmethodologythatwillbeap-pliedtothenextDPEGimplementation.A.ScalabilityandDistributedControlDistributedcontrolsystemsareanintegralpartofourworldandhavebeenstudiedinmanydifferentcontexts,rangingfrombiologytoartificialintelligencetocontrolsystems.Naturallyoccurringdistributedsystemssuchasantssearchingforfood,bacteriaforaging,andtheflightformationsofsomebirdshavebeenwellstudiedbybiolo-gistsandarebeginningtoreceivemoreattentionfromothercommunitiesinterestedindistributedalgorithms.Indeed,theartificialintelligencecommunityhasconsideredsuchsystemsinmoreabstracttermsforseveralyears.Addition-ally,thecontinuoustimecontrolcommunityhasaddressedmanyofthefeaturesthatdistinguishdistributedcontrolsystemsfromclassicalcentralizedcontrolsystems.Natureprovidesuswithseveralgoodexamplesofdis-tributedcontrolinaction.Forexample,schoolinginfish[27]andcooperationininsectsocieties[28]exhibitcomplexcollectivepatternsarisingfromrathersimpleindividualbe-havior.Thesesocialbehaviorshavebeenarguedtoimprovefoodsearch,predatoravoidance,andcolonysurvivalforthespeciesasawholeratherthanfortheindividual.Someresearchershavebeenturningtosuchexamplestogaininsightintothesenaturallyoptimizeddistributedalgorithms.InvestigatingbacteriaforagingofE.coli,Passino[29]hasdevelopedadistributedoptimizationalgorithm.ThealgorithmmodelshowE.colibacteriamoveinasolutionastheycollectivelysearchfornutrientsandavoidtoxinstoreachanoptimalstatewherethecollectionofbacteriaissatisfiedwiththeirsurroundings.Theartificialintelligencecommunityhasaddressedsuchsystemsunderthetitleofdistributedagentsforseveralyears[30].Someresearchersinthiscommunityhavedevelopedapproachessuchasfreemarketsystems[31]thatmimicourowntradesystem.Inthisarchitecture,eachagent,whichcouldbearobotwithaspecializedability,bidsonapar-ticulartaskbasedonitscostfunctionwhichcombinestherobot’srewardandeffort.Itisevenpossibleforrobotstobe-etal.:DISTRIBUTEDCONTROLAPPLICATIONSWITHINSENSORNETWORKS1241 Fig.10.SectionofadistributedcontinuouscontrolMOCwithsensing,actuation,andcommunicationjitter.Shadedblocksrepresentatimedelay.comeleaderswhobidontasksandthensubcontractthetaskouttoseveralotherrobots.Thecontinuouscontrolcommunityhaswrestledwithdis-tributedsystemsformanyyearsintherealmofprocesscon-trolandhasindependentlyaddressedmanyofthecaveatsofdistributedsystems,suchasjittercompensationandsched-uling.Martíetal.[32]haveidentifiedthetypesofjitterthatcanoccurindistributedsystemsandinvestigatedcompensa-tiontechniques.Theirmethodfirstanalyzeswhetheronlineorofflinecompensationisneeded.Ifonlinecompensationisfeasible,thentheparametersofthecontrollawaredynam-icallyupdatedaccordingtothenexttimethecontrollerwillbeexecuted.Otherresearchershavereformulatedthetypicalschedulingproblemasadynamicsystemsothatthetech-niquesofcontroltheorymaybeapplied[33].In[34],acen-tralizedschedulingruleisreplacedwithlocalinstantiationsofintegralcontrollersthatareshowntodrivethestatetoaviablesolution.B.ModelsofComputationTheimpossibilityofcharacterizingthesesystemswithintheclassicalcontrolframeworkraisestheneedtoselectoneormoreMOCsinordertoaccuratelyanalyzedis-tributedcontrolproblemsinSNs.Ourhopewouldbethatsuchacombinationcapturesthecontinuouslychangingdynamicsoftheenvironment,thedistributionofresources,andthediscretenatureofthehardware.Toaddressthisissuemorespecifically,weinvestigateseveralcommonMOCs,includingdiscreteevents,continuousdynamicalsystems,discrete-timedynamicalsystems,hybridautomata,synchronousreactivelanguages,anddataflowmodels.Foreachofthese,weconsideritsadvantagesanditsdrawbackswithrespecttocontrolapplicationswithinSNs.Continuoustimedynamicalsystems[35],[36]areawell-studiedformalmodel.Keypropertiessuchasstabilityandreachabilitycanbedeductedusingavailableanalyticalandnumericalmethods.Controllerscanbedesignedtomeetde-siredspecifications.Additionally,theyarefamiliartothecontrolcommunityand,hence,preferredforcontrolappli-cations.However,fordistributedcontrolapplicationsinSNs,thistheoryisnotabletocapturecommunicationdelays,timeskewbetweenclocks,ordiscretedecisionmaking.Sinceallthevariablesarecontinuous,itisdifficulttomodelsuchdis-cretephenomena.Additionally,controllersmustbeimple-mentedonmicroprocessors,andcontrolmustbepiecewiseTodescribethecontroller’spiecewiseconstantnature,weturntodiscretetimedynamicalsystems[35],[36].However,weareagainlimitedtocharacterizingsystemswithoutmodechanges.Additionally,thisMOCassumesperiodicactivationofthecontrollerwithinstantaneouscomputationofthecon-trollawwhichisnotpreservedbytheunderlyingplatform.Thismodeldoesnotdirectlyaddresssensingandactuationjitter,butitcanbetakenintoaccountbyaugmentingwithtimedelaysbetweentheplantandthecontroller.Thisap-proachassumesthatthecontrollawiscomputedsynchro-nouslyoneachnodeevery seconds,butdifferentsensingandactuationjittersareallowedforeachnode.Thismodelisusefulwhenweassumethattheprocessschedulerrunningoneachnodecanensuresynchronousoperation.Addition-ally,thesystemcanbemodifiedtodistributecontrolcompu-tationacrossnodeswithstatecommunicationbetweenthem,asshowninFig.10.Themultimodalnatureofsuchsystemscanbedescribedbyahybridautomaton[37].Thesesystemsnicelyaccountforboththe“continuousflow”anddiscretejumpsofsuchPROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003 systems.Notethat“continuousflow,”orjustflow,inahybridautomatonmaybemodeledbyeitherdifferentialequationsordifferenceequations.Theyallowthesystemtoevolveaccordingtotheflowwithoccasionaldiscretetransitions.Additionally,witheachdiscretetransition,theequationsgoverningtheflowareallowedtochange.Differ-enceequationsallowsuchamodeltocapturethepiecewiseconstantnatureofthecontroller.Modechangescanthenbecharacterizedbythediscretedynamics,whereallthediscretepropertiesofourapplicationmustbeencoded.Thediscretedynamicsaresimilartofinite-statemachinesinthatencodingmanydiscretevariablesleadstoadiscretestateexplosionproblemandquicklybecomesunmanageableforToconsiderMOCsmoreappealingforalgorithms,wecandiscreteeventsystems[38].Suchamodelworkswellformodechangesortaskschedulingandcharacterizesthehardwareplatformnicely,aswell.Italsoallowsforthesystemtobeevent-triggered,whichisoftenthecaseinSNs.However,itdoesnotsupportcontinuousvariables,andgiventhediscretenatureofvariablesweagainrunupagainstastateexplosionproblemwhenmodelingalargenumberofnodes.Finally,suchsystemsgenerallydonotcorrelatetimestepsofthemodelwithrealtime.[39]MOCsareintendedtodescribedatatrans-formations.Inparticular,theyareusefulforcharacterizingseveralcommunicatingprocesses.However,thisparadigmisawkwardforcontrol,sinceitgenerallyconsiderstherelation-shipbetweensequencesofinputsandsequencesofoutputs,ratherthantheevolutionoftheoutputforeachinputsignalinturn.Ingeneral,whencomposingseveraldataflowmodelsinafeedbackloop,theresultmaynotbedeterministic[40].Anothersetofcommonmodelingparadigmsarechronousreactivelanguages,suchasSignal[41],Lustre[42],andEsterel[43].Theselanguagessupportabroadrangeofformalverificationtoolstoaidindebugging.Additionally,itispossibletogeneratecodefortheplatformdirectlyfromthesynchronousreactivelanguage.However,weagainfindthatthereisnorelationbetweentimestepsofthelanguageandrealtime.Furthermore,synchronousreactivesystemspresumetheexistenceofaglobalclockandthattimesteps—and,hence,thecomputationoffixedpoints—happeninstantaneously.ThisMOCisnotappro-priatebecauseitisnotcongruentwiththeeventtriggerednatureofSNs.Finally,suchamodelcanbecounterintuitive,sinceitsearchesforafixedpointateverystep.C.DesignApproachesIntheprevioustwosections,wedescribeddifferentapproachestoaddressscalabilityandsynchronus/asyn-chronoussystems.OurapproachtoscalabilityforDPEGswillrelyheavilyondistributedprocessingofsensorsread-ingsinordertogetgoodestimatesofpositionsandvelocitiesofbothevadersandpursuers.Thecontrolofeachpursuerdynamicsisperformedwithinthepursueritselfbasedonnetworkreadings,buthigherlevelcoordinationwillbedistributedbetweenallthepursuerstomaximizerobustnesstoadversarialattack.InordertoaddresstheissuesarisingfromthefactthatDPEGsincludesynchronousandasyn-chronousdynamics,severaladhocsolutionsareavailable.Tocompensatefornonuniformtimedelays,oneapproachistobuffertheincomingdataforacertainamountoftimesuchthatmostofthedatahasarrived.Withthisapproach,theproblemhasbeenreducedtotheclassicalcontrolproblemofdrivingasystemwithafixedtimedelay.However,thisresultcomesatthepriceofsuboptimalperformance.Asformissingdata,themostcommonsolutionsareeitherusingthemostrecentdataregardlessofitsexacttimeofarrival,orestimatingthemostprobablemeasurementthatisconsistentwithpreviousmeasurementsandthedynamicsofthesystem.Someissuesrelatedtotheevent-triggerednatureofdis-tributedcontrolhavebeenaddressedbythehybridsystemcontrolcommunity.Here,theideaistodevelopaformalismthatcombinesthebestofcontroltheoryandstatemachinetheory[44]–[46].Althoughfewanalyticalresultsareavail-abletoday,thisratherintuitiveandpromisingapproachisanactiveareaofresearch.TimesynchronizationresearchforSNshasbeenintense,yieldingpromisingresults[8].Inourmodel,weconfidentlyassumethatsensorreadingscomewithanaccuratetimestamp.Also,weassumethatsensorsknowtheirlocationinspace.Alocalizationserviceensuresthatthenodesinadeployednetworkcancomputetheirlocationrelativetoeachother[9].Withthesetwoassumptions,weusethestandardcontrolformalismwithSNs.Achoiceofamodeliscriticaltothedesignofcontrollersforsuchsystems.IndealingwithcomplexapplicationssuchasDPEGs,controlmustbeexercisedatseverallevels,andahierarchicalsystemseemstobethenaturalmodelingchoice.AgraphicalrepresentationisshowninFig.11.Atthelowlevel,thecontinuoustimedynamicsofthesystemneedtobecaptured.Sincetheimplementedcon-trollersaredigital,themodelisdiscretizedtoyieldadiscretetimecontrolsystem.Atthislevel,thesystemistimebased,inthesensethattimetriggerseachtransition.Ateachtimestep,anobservation,generatedfromasensorreading,needstobeprovidedtothecontroller,whichwillinturnproduceaninputtothedynamicsofthesystemviaanactuator.Instandardcontrolproblems,thesensorsarephysicallyattachedtotheplant;therefore,itisassuredtoreceiveasensorreadingateachtimestep.InthecaseofSNs,thesensingisdistributed.Thismeansthatitmaytakesometimefortheobservationtoreachitsdestination,sincepacketsoverthenetworkaresubjecttodelayandloss.Additionally,thecontrollawneedssomeinformationabouttheplanttocomputethenextinput,whichwillheavilyrelyonstateestimation,prediction,andsmoothing.Intheabsenceofanobservation,wewillmakeuseofthemodelalonetoprovidestateestimationforcontrol.Inthisway,latepacketscanbeusedtoimprovecurrentestimate.Severalmethodscanbeusedforestimation,fromKalmantoparticlefiltering.Agraphicalrepresentationofthelow-levelcontrollerisshowninFig.12.etal.:DISTRIBUTEDCONTROLAPPLICATIONSWITHINSENSORNETWORKS1243 Fig.11.Ahierarchicalsystemrepresentation. Fig.12.Low-levelcontroller. Fig.13.Theproposeddesignmethodology.Atthehigherlevel,thesystemiseventbased.Inthisdo-main,thecontrolreactstooneormoreevents,sequencesofwhicharecalledbehaviors.EventsaredetectedbytheSNandtransmittedtoadiscretecontrollerthatgeneratestheap-propriatereaction.Eachreactionisthentransmittedtothelowerlevelbychangingthecontrolobjectivetoagreewiththenewspecifications.Onceagain,eventsoccurinanasyn-chronousfashion,makingformalanalysisdifficult.Toworkwithsuchevents,weimplementthesystemusingasyn-chronousreactivelanguage,wherebehaviorscanbeverifiedandmappedtoourasynchronousplatform,makingsuretheverifiedpropertiesarepreserved.Theproblemofmappingbehaviorsfromdifferentdomainshasbeentackledinsev-eraldifferentways.WefollowtheapproachofBenveniste[47]bydesigningcontrollersinasynchronousfashion,ver-ifyingthebehavior,andthendesynchronizingthealgorithmtobeimplementedontheasynchronoustargetarchitecture.Theadvantageofthisapproachalsoincludesthepossibilityofautomaticallygeneratingembeddedcodedirectlyfromahigh-levelspecificationlanguage,thusenormouslyspeedingupthedevelopmentphase.AgraphicalrepresentationofthedesignflowisshowninFig.13.V.CInthispaper,wepresentedanoverviewofresearchactiv-itiesdealingwithdistributedcontrolinSNs.WeintroducedSNsandrelatedresearchissues.Wethenpresentedourhard-wareandsoftwareplatformswhileproposinganopenar-chitecturetohelpdeveloprichdistributedapplications.Wepresentedanoverviewofthetheoreticalissuesfacingre-searchersinterestedinusingSNsfordistributedcontrolap-plications.Weidentifiedkeypropertiesthatcauseclassicalcontroltheorytofail.Wesuggestedageneralapproachtocontroldesignusingahierarchicalmodelcomposedofcon-tinuoustime-triggeredcomponentsatthelowlevelanddis-creteevent-triggeredcomponentsatthehighlevel.Inthefuture,wewillfocusonimplementation,verification,andtestingofourmethodologiesindistributedcontrolsystemsonourproposedDPEGtestbed.CKNOWLEDGMENTTheauthorswouldliketothanktheTinyOSteamforpro-vidingagreathardwareandsoftwareplatform.Additionally,theauthorswouldliketothanktheentireNESTteamatBerkeleyandIntelResearch-Berkeley.Finally,theauthorswouldliketothankthefollowingindividuals:S.Bergbreiter,E.Brewer,D.Culler,D.Gay,J.Hill,B.Hohlt,C.Karlof,P.Levis,S.Madden,K.Pister,J.Polastre,N.Sastry,R.Szewczyk,R.vonBehren,D.Wagner,M.Welsh,K.Whitehouse,andA.Woo.PROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003 [1]Crossbow.[Online]Available:http://www.xbow.com/[2]Millennial.[Online]Available:http://www.millennial.net[3]Sensoria.[Online]Available:http://www.sensoria.com[4]B.Warneke,M.Last,andK.S.J.Pister,“SmartDust:Communi-catingwithacubic-millimetercomputer,”IEEEComputer,vol.34,pp.44–51,Jan.2001.[5]Smartbuildingsadmittheirfaults.LabNotes[Online]Available:http://www.coe.berkeley.edu/labnotes/1101smartbuildings.html[6]A.Mainwaring,J.Polastre,R.Szewczyk,D.Culler,andJ.Anderson,“Wirelesssensornetworksforhabitatmonitoring,”presentedatthe1stACMInt.WorkshopWirelessSensorNetworksandApplications,Atlanta,GA,2002.[7]A.N.Knaian,“Awirelesssensornetworkforsmartroadbedsandintelligenttransportationsystems,”Ph.D.dissertation,Mass.Inst.Technol.,Cambridge,1999.[8]J.Elson,L.Girod,andD.Estrin,“Fine-grainednetworktimesynchronizationusingreferencebroadcasts,”presentedatthe5thSymp.OperatingSystemsDesignandImplementation(OSDI),Boston,MA,2002.[9]S.Klemmer,S.Waterson,andK.Whitehouse.(2000,Dec.)Towardalocation-basedcontext-awaresensorinfrastructure.[Online]Avail-able:http://guir.berkeley.edu/projects/location/Location.pdf[10]D.Estrin,R.Govindan,J.S.Heidemann,andS.Kumar,“Nextcen-turychallenges:scalablecoordinationinsensornetworks,”inProc.5thAnnu.ACM/IEEEInt.Conf.MobileComputingandNetworking1999,pp.263–270.[11]E.M.RoyerandC.-K.Toh,“Areviewofcurrentroutingprotocolsforad-hocmobilewirelessnetworks,”IEEEPers.Commun.,vol.6,pp.46–55,Apr.1999.[12]F.Zhao,J.Shin,andJ.Reich,“Information-drivendynamicsensorIEEESignalProcessingMag.,vol.19,pp.61–72,Mar.2002.[13]L.Guibas,“Sensing,trackingandreasoningwithrelations,”SignalProcessingMag.,vol.19,no.2,pp.73–85,Mar.2002.[14]T.Parsons,“Pursuit-evasioninagraph,”inLectureNotesinMath-matics,TheoryandApplicationofGraphs,Y.AlaniandD.Lick,Eds:Springer-Verlag,1976,vol.642,pp.426–441.[15]M.Megiddo,S.Hakimi,M.Garey,S.Johnson,andC.Papadim-itriou,“Thecomplexityofsearchingagraph,”J.ACM,vol.35,no.2,pp.18–44,1988.[16]A.S.Lapaugh,“Recontaminationdoesnothelptosearchagraph,”J.ACM,vol.40,no.2,pp.224–245,Apr.1993.[17]S.LaValle,D.Lin,L.Guibas,J.Latombe,andR.Motwani,“Findinganunpredictabletargetinaworkspacewithobstacles,”inIEEEInt.Conf.RoboticsandAutomation,vol.1,1997,pp.737–742.[18]S.LaValleandJ.Hinrichsen,“Visibility-basedpursuit-evasion:thecaseofcurvedenvironments,”inIEEEInt.Conf.RoboticsandAu-,vol.3,1999,pp.1677–1682.[19]X.Deng,T.Kameda,andC.Papadimitriou,“Howtolearnanun-knownenvironmentI:therectilinearcase,”J.ACM,vol.45,no.2,pp.215–245,1998.[20]S.Thrun,W.Burgard,andD.Fox,“Aprobabilisticapproachtoconcurrentmappingandlocalizationformobilerobots,”Mach.Learn./AutonomousRobots(jointissue),vol.31,no.5,pp.1–25,[21]J.Hespanha,H.J.Kim,andS.Sastry,“Multiple-agentprobabilisticpursuit-evasiongames,”inProc.38thConf.DecisionandControl1999,pp.2432–2437.[22]J.Kim,R.Vidal,H.Shim,O.Shakernia,andS.Sastry,“Ahier-archicalapproachtoprobabilisticpursuitevasiongameswithun-mannedgroundandaerialvehicles,”Proc.40thIEEEConf.Deci-sionandControl,pp.634–639,2001.[23]GreatDuckIsland.[Online]Available:http://www.greatduckisland.[24]S.BergbreiterandK.Pister.COTS-BOTS.[Online]Available:http://www-bsac.eecs.berkeley.edu/~sbergbre/CotsBots/cotsbots.[25]NesC.[Online]Available:http://nescc.sourceforge.net/[26]TinyOS.[Online]Available:http://webs.cs.berkeley.edu/tos/[27]J.Parrish,S.Viscido,andD.Grunbaum,“Self-organizedfishschools:anexaminationofemergentproperties,”Biol.Bull.,vol.202,pp.296–305,June2002.[28]J.Deneubourg,A.Lioni,andC.Detrain,“Dynamicsofaggregationandemergenceofcooperation,”Biol.Bull.,vol.202,pp.262–267,June2002.[29]K.M.Passino,“Biomimicryofbacterialforagingfordistributedop-timizationandcontrol,”IEEEControlSyst.Mag.,vol.22,pp.52–67,June2002.[30]G.Weiss,Ed.,MultiagentSystems:AModernApproachtoDis-tributedArtificialIntelligence,2nded.Cambridge,MA:MITPress,2000.[31]M.DiasandA.Stentz,“Opportunisticoptimizationformarket-basedmultirobotcontrol,”presentedattheIEEE/RSJInt.Conf.IntelligentRobotsandSystems,Lausanne,Switzerland,[32]P.Martí,G.Fohler,K.Ramamritham,andJ.M.Fuertes,“Jittercom-pensationinreal-timecontrolsystems,”inReal-TimeSystemsSymp.London,UK,2001.[33]M.CaramanisandA.Sharifnia,“Nearoptimalmanufacturingflowcontroldesign,”Int.J.FlexibleManuf.Syst.,vol.3,no.3–4,pp.321–336,June1991.[34]V.V.PrabhuandN.A.Duffie,“Nonlineardynamicsindistributedarrivaltimecontrol,”IEEETrans.ControlSyst.Technol.,vol.7,pp.724–730,Nov.1999.[35]F.M.CallierandC.A.Desoer,Eds.,LinearSystemTheory,1sted.NewYork:Springer-Verlag,1991.[36]C.-T.Chen,Ed.,LinearSystemTheoryandDesign,3rded.London,UK:OxfordUniv.Press,1999.[37]C.Tomlin,J.Lygeros,andS.Sastry,“Computingcontrollersfornon-linearhybridsystems,”inLectureNotesinComputerScience,Hy-bridSystems:ComputationandControl,F.W.VaandragerandJ.H.vanSchuppen,Eds.Heidelberg,Germany:Springer-Verlag,1999,vol.1569,pp.238–255.[38]E.A.Lee,“Modelingconcurrentreal-timeprocessesusingdiscreteevents,”AnnalsSoftw.Eng.(SpecialVolumeonReal-TimeSoftware,vol.7,pp.25–45,1999.[39]W.B.Ackerman,“Dataflowlanguages,”IEEEComputer,vol.15,pp.15–25,Feb.1982.[40]E.A.Lee,“ADenotationalsemanticsfordataflowwithfiring,”Elec-tron.Res.Lab.,Univ.ofCal.,Berkeley,Tech.Rep.No.UCB/ERLM97/3,1997.[41]A.BenvenisteandP.L.Guernic,“Hybriddynamicalsystemstheoryandthesignallanguage,”IEEETrans.Automat.Contr.,vol.35,pp.525–546,May1990.[42]P.Caspi,D.Pilaud,N.Halbwachs,andJ.A.Plaice,“LUSTRE:Adeclarativelanguageforprogrammingsynchronoussystems,”inConf.Rec.14thAnnu.ACMSymp.PrinciplesofProgrammingLan-guages,1987,pp.178–188.[43]G.BerryandG.Gonthier,“TheEsterelsynchronousprogramminglanguage:design,semantics,implementation,”Sci.Comput.Pro-gram.,vol.19,no.2,pp.87–152,Nov.1992.[44]R.AlurandT.Henzinger,“Modularityfortimedandhybridsys-tems,”inLectureNotesinComputerScience,CONCUR97:Con-currencyTheory,8thInternationalConference.Berlin,Germany:Springer-Verlag,1997,vol.1243.[45]M.S.Branicky,V.S.Borker,andS.K.Mitter,“Aunifiedframeworkforhybridcontrol:modelandoptimalcontroltheory,”IEEETrans.Automat.Contr.,vol.43,pp.31–45,Jan.1998.[46]J.Lygeros,C.Tomlin,andS.Sastry,“Controllersforreachabilityspecificationsforhybridsystems,”Automatica,vol.35,no.3,pp.349–370,Mar.1999.[47]A.Benveniste,“Somesynchronizationissueswhendesigningem-beddedsystemsfromcomponents,”inLectureNotesinComputerScience,ProceedingsoftheFirstInternationalWorkshoponEm-beddedSoftware.Berlin,Germany:Springer-Verlag,2001,vol. BrunoSinopoli(StudentMember,IEEE)receivedtheLaureadegreeinelectricalengi-neeringfromtheUniversityofPadova,Padova,Italyin1998.HeiscurrentlyworkingtowardthePh.D.degreeinelectricalengineeringfromtheUniversityofCalifornia,Berkeley,underthesupervisionofProf.S.Sastry.Hisresearchinterestsincludesensornetworks,designofembeddedsystemsfromcomponents,distributedcontrol,andhybridsystems.etal.:DISTRIBUTEDCONTROLAPPLICATIONSWITHINSENSORNETWORKS1245 CourtneySharpreceivedtheB.SdegreefromtheUniversityofOklahoma,Norman,incom-puterengineeringin1997andtheM.S.degreefromtheUniversityofCalifornia,Berkeley,inelectricalengineeringin2000.HeiscurrentlyaResearchSpecialistwiththeIntelligentMachinesandRoboticsLaboratory,UniversityofCalifornia,Berkeley.Hisresearchinterestsincludedesignofdistributedandnetworkedembeddedsystems,composablear-chitectures,distributedprogramminglanguages,andcomputervision. LucaSchenato(StudentMember,IEEE)wasborninTreviso,Italy,in1974.HereceivedtheDr.Eng.degreeinelectricalengineeringfromtheUniversityofPadova,Padova,Italy,in1999.HeiscurrentlyworkingtowardthePh.D.degreefromtheDepartmentofElectricalEngineering,UniversityofCalifornia,Berkeley.Hisresearchinterestsincludemodelingofbiologicalnetworks,insectlocomotion,millirobotics,andavionics. ShawnSchaffert(StudentMember,IEEE)receivedtheB.S.degreeinelectricalengineeringfromtheUniversityofNebraska,Lincoln,in1998,andtheM.S.degreeinelectricalengineeringfromtheUniversityofCalifornia,Berkeley,in2001.HeiscurrentlyworkingtowardthePh.D.degreeattheUniversityofCalifornia,Berkeley,underthesupervisionofhisresearchadviser,Prof.S.Sastry.In2001,hewasanInternatXeroxPARC,PaloAlto,CA,investigatingthecomplexityofcon-strainedoptimizationproblems.Hisresearchinterestsincludeembeddedsystems,distributedcontrol,hybridsystems,androbustcontrolinsensornetworks.Mr.SchaffertisamemberofTauBetaPiandEtaKappaNu. S.ShankarSastry(Fellow,IEEE)receivedtheM.S.degree(honoriscausa)fromHarvardUniversity,Cambridge,MA,in1994,andthePh.D.degreefromtheUniversityofCalifornia,Berkeley,in1981.From1980to1982,hewasanAssistantProfessoratMassachusettsInstituteofTech-nology,Cambridge.In2000,hewasDirectoroftheInformationTechnologyOfficeattheDefenseAdvancedResearchProjectsAgency,Arlington,VA.HeiscurrentlytheNECDis-tinguishedProfessorofElectricalEngineeringandComputerSciencesandBioengineeringandtheChairmanoftheDepartmentofElectricalEngineeringandComputerSciences,UniversityofCalifornia,Berkeley.Hisresearchinterestsareembeddedandautonomoussoftware,computervision,computationinnovelsubstratessuchasDNA,nonlinearandadaptivecontrol,robotictelesurgery,controlofhybridsystems,embeddedsystems,sensornetworks,andbiologicalmotorcontrol.Dr.SastrywaselectedintotheNationalAcademyofEngineeringin2001“forpioneeringcontributionstothedesignofhybridandembeddedsystems.”HehasservedasAssociateEditorforIEEETRANSACTIONSONUTOMATICONTROL,IEEECONTROLAGAZINE,andIEEERANSACTIONSONIRCUITSANDPROCEEDINGSOFTHEIEEE,VOL.91,NO.8,AUGUST2003