ArchitecturalApplicationsofComplexAdaptiveSystems
CoryClarkeandPhillipAnzalone
XO(eXtendedOfce)
Abstract
Thispaperpresentsmethodsandcasestudiesofapproachingarchitecturaldesign
andfabricationutilizingComplexAdaptiveSystems(CASs).Thesecasestudies
andobservationsdescribedherearendingsfromacontinuingbodyofresearch
investigatingapplicationsofcomputationalsystemstoarchitecturalpractice.CASs
arecomputationalmechanismsfromthecomputerscienceeldofArticialLife
thatprovideframeworksformanaginglargenumbersofelementsandtheirinter-
relationships.TheabilityoftheCAStohandlecomplexityatascaleunavailable
throughnon-digitalmeansprovidesnewwaysofapproachingarchitecturaldesign,
fabricationandpractice.
1INTRODUCTION
ThispaperdocumentsndingsfrominvestigationsoftheapplicationofComplex
AdaptiveSystems(CASs)tothepracticeofarchitecture.CASsarecomputational
mechanismsfromthecomputerscienceeldofArticialLifethatprovideframeworks
formanaginglargenumbersofelementsandtheirinter-relationships.Some
algorithmsthatareclassiedasCASsareCellularAutomata,LindemayerSystems,
TuringMachinesandFlockingalgorithms.Therearemanypotentialapplicationsfor
thesesystemsinthearchitecturaldesignprocess:
•Toolsfortheautomateddesignoflargescalebuildingsandurbanprojects;
•Techniquesforthedesignofserializedbuildings(e.g.masshousing,franchise
buildings,pre-fabricatedconstruction)thatarecontextually-sensitiveand
differentiated;
•Platformsforroboticizedself-assemblingandself-adjustingbuildings;
•Methodsfordesigningadaptablebuildingswithredundantstructureswith
theabilitytomoresuccessfullywithstanddamageandcatastrophicevents.
CriticaltoourinvestigationisthemarriageofCASswithsuitablegeometricand
structuralsystems.Complexcomputationalsystemshavefoundlittleapplicationin
contemporaryarchitecturalpracticebecausetheireffectiverealizationisoftennot
possiblewithtraditionalgeometries,suchastheCartesiancoordinatesystem,or
traditionalstructuralmethods,suchastrabeatedconstruction.Inparallelwithour
investigationofapplicationsofCASstoarchitecturaldesignhasbeenthedevelopment
offormalandstructuralsystemswithgeometriestomatchtheCASmorphology.
Thispaperdescribes:
•TheconceptofCASsandtheirpresentationinarchitecturalterms,addressing
ideasofcontext,site,programandform;
•Developmentofstructuralandconstructionsystemsthatprovideameansof
constructingresultsfromCASbaseddesignsystems;
•CasestudiesoftheapplicationofCASsinarchitecturaldesign.
2CASCONCEPTSINARCHITECTURALDESIGN
CASsareaclassofdynamicsystemsfromtheeldofArticialLife.Thesesystems
typicallyinvolvesetsofdiscreteelementsthatchangestate(typicallyvisualized
bychangesincolor),basedontheiterationofasimplesetofdeterministicrules
(Wuensche,Lesser1992).Theelements’stateschangeasafunctionoftheircurrent
stateandtherules,producinganunpredictable,complexglobalbehaviorpattern.
ACASchangesitsstatebasedontherulesofthesystem.Theserulesareappliedto
thecurrentstateofeveryelementinthesystemtocalculateeachelement’snewstate.
Despitethefactthatthesesystems’behaviorsaredenedbycompletelydeterministic
rules,thebehaviorsofthesesystems(withsufcientelementsinplay)aresocomplex
thattheycannotbepredicted.Sincethenewstateofthesystemisdeterminedbythe
currentstate,differentinitialstatesofaCAScanfosterentirelydifferentemergent
behavior.TochangethebehaviorofthesystemaCAScanmerelybestartedwithnew
initialstate.
DifferentinitialstatesinCASswillleadtodifferentglobalbehaviorandpatterns.In
thiswaytheinteractionoftheelementswithinaCAScanbethoughtofasa‘calculator’
(Neumann,1966),computingoutputsbasedoninputs.Thiscapacitytoderiveresults
fromacomplexmatrixofinputsisanobviousanalogtothedesignprocess–which
musttakethecomplexmatrixofrequirements–program,siteandstructure–and
translatethemintoanarchitecturaloutcome.
ThereareseveralpropertiessharedbyCASsuponwhichwehavefocusedourresearch.
Thepropertieslistedbelowhaveprovideduswithameansoflinkingtheprocessesand
behaviorsofCASstothedesignandproductionofarchitecture.Whiletheseproperties
maynotbelongtoallCASs,theyprovidealistoftraitsthatcanbecombinedand
exploitedintheproductionofarchitecturaldesigntools.Thepropertiesarediscrete
composition,algorithmicrelationships,exogenouscontrolandscalability.
2.1DiscreteComposition
TheunderlyingsubstructuresofCASsarediscretematricesofelements,treatedas
independententities,operatinginparallel.Despitethediscretenatureofthesesystems
atthelocalscale,theyareabletoexhibitglobalbehaviorswithvaryingdegreesof
continuity.Thetraitofhavingadiscontinuoussubstructure,butvaryingcontinuity
atthelevelofsuperstructure,isoneofthebehaviorsthatmakeCASssuitablefor
translationintoarchitecture.Thistraitprovidesananalogtothewayinwhich
architectureisproduced;assembledfromdiscreteelements–eitherprogrammatic
(roomsandzones)orstructural(beamsandcolumns)–thatdenespaceswith
varyingdegreesofcontinuity(e.g.spatial,programmatic,acoustical,environmental).
2.2AlgorithmicRelationships
ThebehaviorofaCASisprescribedbysetsofrulesandalgorithmsthatdictate
therelationshipsbetweenelementswithinthesystem.Therelationshipsbetween
elementsinaCASareiterativelyre-evaluatedbasedontheserules,providingthe
meansforthesystemtoadapttointernalandexternalchanges.Thecleardenitionof
relationshipsbetweenelements,andthereiterationofthoserelationalrulesovertime,
ndsananaloginthedesignprocess–wheredesignintentionsaredescribedinterms
ofrelationships(privacy,lighting,proximity),andthenalproductemergesfromthe
continualreiterationandrenementoftheserelationships.
Aspecicexampleofthiscanbeseeninthecomputationalalgorithmdescribedas
ocking.Basicockingsystems(Figure1)deneasetofdesiredrelationshipsbetween
elementsinaock(suchasidealdistancebetweenelementsandpreferredposition
relativetootherelements)andattempttomaintaintheserelationshipsbycontinually
readjustingthepositionsofeachelement(Reynolds,1987).Whentherulesofproximity
andpositionareappliedtoagroupofelementstheyexhibitcomplexadaptivebehavior
similartoaockofbirds.Thissimplesetofrulesofproximityandpositioncannd
applicationinarchitecture,forexample,asrulesforpositioningofprogrammaticor
structuralelements.
2.3ExogenousControl(ContextualAwareness)
Alongwiththeideasofinternalconstraints,manyCASshavethecapacitytoreactto
exogenouscontrol,whichinarchitecturaltermscanprovidemechanismsforcontextual
Figure 1 - Paths described by the motion
of ocking particles .
andenvironmentalawareness,andinuencingthesystemtoincorporateadditional
designintentions.CASshavetwoprimarymechanismsforreactingtoenvironment
andcontext–initialstatedenitionandavoidancebehavior.
SinceallCASsoperateonadeterministicsetofrules,theinitialstateofthesystem
ultimatelydeterminesitsoutcome.Thoughtofinthisway,theoutcomeofaCAScan
beseenasaregistrationoftherulesuponaparticularinitialstate.Withasimple
CAS,suchasCellularAutomata,theentirerangeofpossibleoutcomesforeveryrule
fromasingleinitialstatecanbemapped(Wuench,Lesser1992).Inanarchitectural
applicationtheinitialstatecanbedenedintermsoftheenvironment,andthe
behavioroftheCASwouldbeadirectexpressionoftherelationalrulesofthesystem
withinthatenvironment.
Alongwiththeabilitytoregistertheinitialstateoftheirenvironment,CASscanreact
toexternalchangeviamechanismssuchasavoidancebehavior.Thisbehavioris
comprisedofbasicmechanismsthatmonitortheincrementalgrowthofaCASfor
collisionswithobjectsoreldsintheenvironmentandprovideforadditionalreactions.
Ifelementsinthesystemarecurrentlyonacollisioncoursethemonitoringmechanism
signalsfornewbehavioralrulestobeinvokedthatreacttothepotentialcollision.
Theobviousarchitecturalimplicationofthecollisiondetectionandavoidancelogic
istoinstallthesystemwithanawarenessofsurroundingbuildingsandlandscape.
However,thebehaviorcanalsobeappliedtomoreabstractlyzonedvolumeswithinthe
systemssurroundingcontextaswell.Thegrowthofasystemcouldrespondtozoning
envelopes,acousticalenvelopes,sunandshadevolumes,viewcorridors,program
areas,andanyotherabstractconstraintsthatcanberepresentedvolumetrically.
Collisiondetectioncanbefurtherextendedasamechanismforcontrolbyinvertingthe
logicofavoidancebehaviorsothatthecollisiondetectionmechanismcanbeusedas
atoolforsearchingoutattractiveareasofanarchitecturalcontext.Iftheavoidance
behaviorweretobechangedtotendtowardspotentialcollisioninsteadofaway,the
morphologicaltendenciesoftheCASwillgrowtowardsspecicareasinasitethatare
designatedasdesirable.Furthermorethevolumesofdesirability,andundesirability,
canbeweightedininuencetoproduceamultivalentmapofcontext.
2.4Scalability
ManyCASsexhibitafractalbehavior-thebehaviorofasmallnumberofelementsis
congruenttothesamesystemwithexponentiallymorecomponents.Thisscalability
ofbehaviorhastwobenetstothestudyandimplementationwithinarchitectural
practice.Therstisoneofstudyandtesting;sincethebehaviorofalargesystem
issimilartothatofasmallsystem,methodsfordesigncanbemodeledinalimited
capacityandstilltranslatetothelargerscaleofabuildingorurbanproject.The
secondbenetofthescalabilityofCASsistheirpotentialtobenestedfractally.Their
scalabilitymakesthemessentiallyscale-less;aparticularsetofrulesthatfunctions
wellasaglobalorganizationsystemcouldbeusedatthescaleofanurbanproject,
whileeach‘cell’couldinturnbelledwithasmallerversionofaCASwithadetailed
behavioratthescaleofbuildingordwellingunit.
3STRUCTURALSYSTEMS
Thecharacteristicsoftheorganizationalstrategyoutlinedabove,andthediverserange
ofpossiblecomplexformsthatmayresult,requireastructuralsolutionwithsimilar
geometricandalgorithmictraits.Ourresearchfocuseduponthethree-dimensional
differentialspace-truss.
Thetraditionalspace-trussisalatticestructureofstandardelements,typicallyleading
toarchitectureofregulargeometricalforms,asinthegeodeticdomesofBuckminster
FullerandprojectssuchasI.M.Pei’sJavitsConventionCenterinNewYork.The
traditionalspace-trussemploysstandardelementsbecauseofconstraintsofdesign,
analysisandfabrication–constraintsnowsurmountablethroughcomputer-based
techniques.Thedifferentialspace-trussusesnon-standardelements;byallowing
eachelementtobeuniqueitcantakeoncomplexthree-dimensionalcurvilinearform
aswellasbasiclineargeometry.
Duringourresearchwebuiltandtestedseveralscalemodelsofthree-dimensional
differentialspace-trusseswithnon-uniformelements(Figure2).Themanufacturing
ofthestructuralelementsrequirestheuseofComputerNumericallyControlled(CNC)
fabricationtechniques.Thespecicbiasesandlimitationsoftheseconstruction
methodscanbebuiltintotherelationalrulesoftheCASsasfurthermeansofcontrol.
Thedifferentialspace-trusshasthreemajortraitsthatmatchmanyofthebehaviors
andpropertiesofCASs:discretecomposition,latticegeometryandscalability.
3.1DiscreteElements
Space-trusssystemsarecomprisedoftwobasiccomponents,linearstrutsand
connectingnodes;throughthemanipulationofthesetwotypesofelementsthesystem
canyieldadiverserangeofcomplexthree-dimensionalforms.Theuidnaturein
whichaspace-trusscantransitionbetweencurvilinearformandlineargeometry
issimilartothatofmonolithicstructuralsystemssuchascastconcrete,yetitis
constructedfromrepetitivediscreteelements.Itsrangeofformalexpression,coupled
Figure 2 - Fabricated prototype of
differential space-truss.