Strategies of Orientation in Environmental Spaces

Tobias Meilinger

Strategies of Orientationin Environmental Spaces

MPI Series in

Biological Cybernetics

No. 22, July 2008

www.ebook3000.com

Strategies of Orientation in Environmental Spaces

Inaugural-Dissertation zur

Erlangung der Doktorwrde der Wirtschafts- und Verhaltenswissenschaftlichen Fakultt

der Albert-Ludwigs-Universitt Freiburg. i. Br.

vorgelegt von

Tobias Meilinger aus Ebersberg

SS 2007

www.ebook3000.com

2

Dekan: Prof. Dr. Dr. Jrgen Bengel 1. Gutachter: Prof. Dr. Gerhard Strube 2. Gutachter: Prof. Dr. Markus Knauff Datum des Promotionsbeschlusses: 7.12.2007

www.ebook3000.com

3

Abstract

Orientation in space is fundamental for all humans and most other animals.Accomplishinggoalsoftenrequiresmovingthroughenvironmentalspacessuchasforests,houses,orcities.Severalmechanismswereevolvedinordertosolvetheseorientation problems, including spatial updating, route navigation, andreorientationby landmarks andgeometry.Humanorientation capabilitiesbuildupon these and other fundamental mechanisms. Compared to nonhumananimals,humansdemonstrateagreater flexibilityduringorientation tasks.Theyare able to apply various strategies to fulfil one orientation goal, such asnavigating to a known location. In thiswork,we examinedwhich orientationstrategies human navigators apply and how efficient these strategies are. Wefocusedonmemory strategies for encoding spatialknowledge andonplanningstrategiesespeciallyduringwayfinding.

InStudy1,weexaminedmemorystrategiesused toencodearoute.Participantslearned two routes inanovelphotorealisticvirtualenvironmentdisplayedona220screen,whiletheyweredisruptedbyeitheravisual,aspatial,averbalorinthe case of the control group no secondary task. In a subsequentwayfindingphase, the participants were required to navigate the routes again. Theinterferences between verbal and spatial secondary task and the encoding ofwayfindingknowledgeweregreaterthanthatbetweenthevisualsecondarytaskand theencodingofwayfindingknowledge.This suggests thatparticipants relyon a verbal and spatialmemory strategy. In Study 2 and Study 3,we tried todeterminemorepreciselythekindofspaceinvolvedinthisspatialstrategy:eitheramaplike space (called figural space), or the visible area in the surroundingenvironment(calledvistaspace).

InStudy2,wetestedthehypothesisthatthespatialmemorystrategyreliesonamaplikespace.Ifnavigatorsuseamapencodingstrategy,specifictransformationcostsshouldoccurfortasksperformedinrouteperspective(e.g.,wayfinding),butnotfortasksperformedinthesamebirdseyeviewastheencodedmap(e.g.,mapdrawing). To test these predictions, participants learned two routes. In onecondition,theroutewaslearnedfrommaps,inthecontrolcondition,participantslearnedtherouteonlyfromverbalinstructionsconstructedfromthesemaps.Bothgroupsthentriedtofindtheseroutesandperformedrouteknowledge,direction,and distance estimation tasks from varying perspectives. When tested in thebirdseyeperspectiveofthemaps,thegroupsdiffered.Participants,therefore,didremember knowledge from themaps.However, for tests in route perspective,especiallywayfinding,nodifferencescouldbeobserved.Thiswasalsoconfirmed

www.ebook3000.com

4

inapoweranalysis.AswithStudy1, the resultssuggest thatparticipantsuseaverbal strategy forwayfinding. Thiswas supported by participants subjectivereports,aswellasbytheirsuperiorperformanceingivingdirectionscomparedtodrawingroutemaps.Theseresultsspeakagainstastrong involvementoffiguralspaceinhumanspatialmemorystrategy.

InStudy3,weexaminedwhetherthespatialmemorystrategyofwayfindersrelieson the geometry of vista spaces.Data from Study 1was analysed in order tocompareperformanceofallparticipantsatdifferent intersections.We formalisedthegeometryofanintersection,applyinganewdirectionspecificisovistanalysis.Suchanisovistanalysiscomputesparametersfromtheviewshedpolygonofthevisiblearea.Using theseparameters,we could cluster the intersections into twogeometrically dissimilar groups, i.e., tintersections and nontintersections.Participants exhibited better wayfinding performance at nontintersections aswell as more thorough landmark and route knowledge. Therefore, it seemsplausiblethatthegeometriclayoutofvistaspacesplaysaroleinwayfindingandthatthememorystrategynavigatorsuseforspatialorientationreliesatleastpartlyon thegeometryof thevistaspace.Additionalresults fromStudy3showed thatparticipants seemed to apply awhenindoubtfollowyournose strategy. Theyencodedprimarily intersectionswhich required a turn; theydid not recallwellintersections they walked straight through. Better route knowledge andwayfinding performance were traded for a decreased ability to recognisestraighton intersections. Thewhenindoubtfollowyournose strategy showshowplanningcaninteractwithmemory.

Study 4 andStudy 5were concernedwithplanning strategiesmoredirectly. InStudy 4, we examined wayfinding strategies as a function of familiarity.Participants who were familiar and unfamiliar with a complex multilevelbuilding performed six wayfinding and several survey knowledge tasks. Wemeasured strategy by route choice and by thinking aloudprotocols. Familiarparticipantspreferredaregionalplanningstrategy that involved firstheading tothe flooronwhich thegoalwas located.Overall, thisstrategywas tied tobetterwayfinding performance compared to a least angle strategy or a strategy ofrelyingonwellknownpartsof thebuilding.The regionalplanningstrategycanreducememoryworkloadduringplanningandnavigationwhile stillprovidingrather short routes. Route knowledge showed a greater impact onwayfindingperformance compared tometric surveyknowledge.Thiswas also indicated inStudy1andStudy2andvariedactivelyinStudy5.

InStudy5,weexaminedmetricandnonmetricstrategies.Firsttimevisitorstoacomplex building solved twowayfinding and two selflocalisation tasks. Theyeither used a standardmap,which depictedmetric relations correctly, or theyusedahighlyschematicmaponlydepictingroute information, i.e., the topology(which decision point is connected towhich other decision point) and turninginformation at decision points (straight on, left or right).No differenceswere

www.ebook3000.com

5

found forselflocalisation, indicating thatallparticipants focusedmoreon routeinformation.Participantswitha schematicmapwereevenbetteratwayfinding.Metricinformation,therefore,didnotseemtocontributemuchtowayfindingandselflocalisation.

Asindicatedinthestudies,humanstrategicchoicecanoftenbedescribedwithacostefficiency criterion. A spatial memory strategy relying on vista spacesrequireslesstransformationcoststhanaspatialstrategyrelyingonfiguralspaces.Thewhenindoubtfollowyournosestrategyrequiresfewerdecisionpointstobeencoded in memory, without the risk of getting lost. The efficient regionalwayfinding strategy also reduces memory workload during planning andnavigation,whilestillprovidingrathershortroutes.Althoughtheyfacilitatemoreprecisenavigation,metricstrategiesrequirehighermemoryand/orcomputationalloads,leadingtoworseperformanceoverall.

Based on our resultswe formulated a theoretical framework for orientation inenvironmental spaces. The strong involvement of a verbalmemory strategy inStudy 1 and Study 2 led to the dual coding theory of spatial orientation. Thistheoryassumesthathumannavigatorsencodeenvironmentsnotonlyinavisualor spatial format (asprobably alsononhuman animalsdo),but also in a verbalformat.This theory can explainbiases in spatialmemory,help interpret resultsfrom wayfinding and reorientation, and provide a basis for more elaboratestrategies.Thecharacteristicsofmerespatialmemoryandplanningstrategiesaredescribed in the network of reference frames theory. This theory assumes thatspatialmemory for environmental spaces consists of a network of vista spacereference frames. It proposes a commonmemory structure forwayfinding andreorientation, provides a common framework for route and survey navigation,and highlights similarities and differences between human and nonhumannavigators.Itcanexplainresultsfromvariousareasofspatialorientation,suchasorientation specificity, changes due to familiarity, and asymmetries in spatialmemory.Italsoprovides ideasforfutureresearch includingtestablepredictions.Theresultsofthisthesis,aswellastheproposedtheoreticalframework,aremeanttobeastepforwardsinapproachingafunctionaltheoryoforientationinspace.

www.ebook3000.com

6

Zusammenfassung

Die Fhigkeit sich imRaum zu orientieren ist frMenschen und andere Tiereessenziell wichtig. Um Ziele zu erreichen ist es oft notwendig, sich durchNavigationsrume (environmental spaces), wie z.B. Wlder, Gebude oderStdte, zu bewegen. Zur Lsung solcher Orientierungsaufgaben evolviertenverschiedeneOrientierungsmechanismen,z.B.Pfadintegration,Routennavigationoder die Reorientierung an Landmarken bzw. der Umgebungsgeometrie.MenschlicheOrientierungsfhigkeitenbauenaufdiesenMechanismenauf.DabeilegenMenscheneinevielgrereFlexibilittandenTagalsandereTiere.SosindMenschen in der Lage, unterschiedliche Strategien zu verwenden, um ein unddasselbeZielzuerreichen.

DievorliegendeArbeituntersucht,welcheStrategienMenschenzurOrientierungim Raum einsetzten undwie erfolgreich diese Strategien sind.Dabei liegt derFokusaufGedchtnisstrategienzurEnkodierung rumlichenWissens sowieaufPlanungsstrategien,vorallembeimWegfinden.

Studie 1 untersuchte Strategien zur Enkodierung von Wegen. DieVersuchspersonen beobachteten ein Video, dass zwei Wege durch einephotorealistische virtuelle Stadt darstellte, die auf eine 220 groe Leinwandprojiziert wurden. Whrend sie sich die Wege einprgten, bearbeiteten dieVersuchspersonen verschiedene Nebenaufgaben entweder eine visuelle, einerumliche, eine verbale oder keine Nebenaufgabe. In der anschlieendenWegfindungsphase sollten die Versuchspersonen die gelerntenWegemitHilfeeines Joysticksablaufen.Zubeobachtenwar,dassdieverbaleunddierumlicheNebenaufgabestrkermitderEnkodierungvonWissenberWegeinterferierten,alsdievisuelleNebenaufgabe.DiesesErgebnisdeutetdaraufhin,dassmanzurWegfindungeineverbaleundeinerumlicheGedchtnisstrategieeinsetzt.

ZielderStudien2und3wares,genauerzubestimmenaufwelcheArtvonRaumsich dieGedchtnisstrategie bezieht: den kartenhnlichenObjektraum (figuralspace)oderdensichtbarenRaum,derdieMenschenunmittelbarumgibt(vistaspace)?

Studie2berprftedieHypothese,dassdierumlicheGedchtnisstrategiensichaufkartenhnlicheRume(figuralspaces)beziehen.DieserHypothesefolgendist zu erwarten, dass das Enkodieren einer Karte bestimmte Transformationskostennachsichzieht,wenndiesesWissenauseinerRoutenperspektive,z.B.beimWegfinden,abgerufenwird.SolcheTransformationskostensolltenjedochnichtbeiAufgabenauftreten,die inderVogelperspektivebearbeitetwerden,wiez.B.das

www.ebook3000.com

7

ZeichneneinerKarte.DieVersuchspersonenlerntenzweiWege,entwederanhandvonKartenoder alsKontrollbedingung anhandvonWegbeschreibungen,dieaufBasisdieserKartengeneriertwurden.AnschlieendsolltenbeideGruppendieWege laufen sowie Aufgaben zum Routenwissen und zur Schtzung vonRichtungen und Entfernungen aus unterschiedlichen Perspektiven bearbeiten.FanddieAufgabeausderVogelperspektiveherausstatt,sounterschiedensichdieGruppenvoneinander.DieVersuchspersonenhattenalso etwasvondenKartengelernt.FanddieAufgabeallerdingsausderRoutenperspektiveheraus statt, soergaben sich, z.B. in der Wegfindungsaufgabe, keine Unterschiede. DieserNulleffekt wurde auch durch eine Analyse der Teststrke untersttzt. Wie inStudie 1 legt das Ergebnis nahe, dass die Versuchspersonen eine verbaleGedchtnisstrategie zumEnkodieren vonWegen einsetzen.Diese InterpretationwirdzudemdurchBefragungsdatengesttztsowiedurchdiebessereLeistungderVersuchspersonen bei der Beschreibung als beim Aufzeichnen der Wege.Insgesamt deuten die Ergebnisse nicht auf eine starke Beteiligung vonkartenhnlichen Rumen (figural spaces) beim rumlichen Enkodieren vonWegenhin.

Studie3untersuchte,obdierumlicheGedchtnisstrategieaufderGeometriedersichtbaren Umgebung (vista spaces) beruht. Dazu wurden die in Studie 1generiertenDaten so ausgewertet,dassdieLeistung allerVersuchspersonen anunterschiedlichen Kreuzungen miteinander verglichen werden konnten. Wirparametrisierten die Kreuzungsgeometrie mit Hilfe einer neuen,richtungsabhngigen IsovistAnalyse, die das eingeschrnkte menschlicheSichtfeld bercksichtigt. Eine IsovistAnalyse berechnet unterschiedlicheParameter anhand des Polygons, das der sichtbaren Bodenflche entspricht.Aufgrund dieser Parameter konnten wir die Kreuzungen in zwei, einandergeometrischmglichstunhnliche,Gruppeneinteilen:TKreuzungenundnichtTKreuzungen.InWegfindungsaufgabensowieinAufgabenzumLandmarkenundRoutenwissen zeigten die Versuchspersonen an nichtTKreuzungen bessereLeistungen. Es erscheint daher plausibel anzunehmen, dass dieGeometrie dessichtbarenUmgebungsraumesinderWegfindungeineRollespielt.EbensoscheintdierumlicheGedchtnisstrategie,wenigstenszumTeil,aufeinemsolchenRaumaufgebaut zu sein. Weitere Ergebnisse aus Studie 3 zeigen, dass dieVersuchspersonen eine imZweifelgeradeaus Strategie verwenden. Sieerkennen zwar vor allem Kreuzungenwieder, an denen sie abbiegenmssen,allerdings zeigen sie bessere Leistungen sowohl beimWegfinden als auch beiAufgaben zum Routenwissen, wenn sie an einer Kreuzung geradeaus laufenmussten.SolcheineStrategie isteinschnesBeispiel frdieVerschrnkungvonGedchtnisundPlanungsstrategien.

DieStudien4und5beschftigtensichkonkretermitPlanungsstrategien.Studie4untersuchte Wegfindungsstrategien und deren Vernderung aufgrund vonErfahrung.Versuchspersonen,diemiteinemkomplexenmehrstckigenGebude

www.ebook3000.com

8

vertraut waren, wurden in sechs Wegfindungs und verschiedenenberblicksaufgabenmitVersuchspersonenverglichen,denen dasselbeGebudenurwenigvertrautwar.DieErhebungdereingesetztenStrategienerfolgteanhandder Wegentscheidungen sowie anhand von Protokollen des lauten Denkens.Versuchspersonen, die mit dem Gebude vertraut waren, bevorzugten eineregionalePlanungsstrategie, indemsie immerversuchten,zuerst indas richtigeStockwerk zu laufen. Diese Strategiewar ehermit guterWegfindungsleistungverknpft,alseineberblicksstrategieodereineStrategie,diedaraufbasierteinenWeg zuwhlen der soweitwiemglich durch gut bekannteGebudeteilefhrt. Die regionale Planungsstrategie reduziert die notwendigerweisegespeichertenEntscheidungspunkteund fhrt trotzdem zu eherkurzenWegen.Insgesamt hatte Routenwissen einen greren Einfluss auf dieWegfindungsleistungalsberblickswissen.DieszeigtesichauchindenStudien1und2wurdeinStudie5aktivvariiert.

Studie 5 untersuchtemetrische und nichtmetrische Strategien. In einem ihnenunbekannten Gebude lsten Versuchspersonen zwei Wegfindungs und zweiSelbstLokalisationsaufgaben.DazuverwendetensieentwedereineStandardkartemit korrekten metrischen Relationen oder eine stark schematisierteRoutenwissenKarte,dienurdieTopologie (welcheEntscheidungspunktesindmitwelchenverbunden)unddieAbbiegeinformationanEntscheidungspunkten(rechts, links, geradeaus) richtig darstellte. In der SelbstLokalisationsaufgabekonnten keine Unterschiede zwischen den Gruppen gefunden werden. Diesdeutetdaraufhin,dassalleVersuchspersonensicheherander, inbeidenKartenkorrektdargestellten,Routeninformationorientierten.InderWegfindungsaufgabezeigtedieGruppemitder schematisiertenKarte sogar bessereLeistung.DiesesErgebnis lsst vermuten, dass metrische Informationen nicht wesentlich zurWegfindungundSelbstLokalisationbeitragen.

Insgesamt knnen die Studien so interpretiert werden, dass menschlicheStrategiewahl oft einem Kosteneffizienzkriterium folgt. Eine rumlicheGedchtnisstrategie, die auf die sichtbaren Umgebungsrume (vista spaces)zurckgreift, fhrt zu geringeren Transformationskosten als eine rumlicheStrategie,dieaufkartenhnlichenRume(figuralspaces)basiert.MiteinerimZweifelgeradeaus Strategie mssen weniger Entscheidungspunkte enkodiertwerden, ohne Gefahr zu laufen, sich hinterher zu verlaufen. Die effizienteregionale Planungsstrategie bentigt weniger Gedchtniskapazitt sowohlwhrend der Planung als auch whrend des Laufens der Wege und lieferttrotzdem eher kurze Wege. Metrische Strategien ermglichen zwar przisereNavigation, stellen aber hhere Anforderungen an das Gedchtnis, bzw. dieVerarbeitung.DaherknnensieschlechtereLeistungenzurFolgehaben.

Die beobachteten Ergebnisse fhrten zur Formulierung einiger theoretischerPositionen bezglich der Orientierung in Navigationsrumen (environmentalspaces).DiestarkeBeteiligungderverbalenGedchtnisstrategieinStudie1und2

www.ebook3000.com

9

resultierteinderZweifachkodierungstheoriederRaumorientierung.LautdieserTheorieenkodierenMenschenbeiOrientierungsaufgaben ihreUmweltnichtnurineinemvisuellenoderrumlichenFormat,wiedasandereTierevermutlichauchtun, sondern verwenden zudem ein verbales Format.DieZweifachkodierungstheorie erklrt systematischeVerzerrungen imRaumgedchtnis, siehilftbeiderInterpretation von Ergebnissen aus der Forschung zur Wegfindung undReorientierungundsiestellteineBasisfrelaboriertereStrategienbereit.Diereinrumlichen Aspekte von Gedchtnis und Planungsstrategien sind in der hierformulierten Referenzrahmennetztheorie beschrieben. Diese noch spekulativeTheorie nimmt an, dass das Raumgedchtnis frNavigationsrume aus einemNetzwerkeinzelnerReferenzrahmenbesteht,diesichjeweilsaufeinensichtbarenUmgebungsraum beziehen. Im Gegensatz zu anderen Theorien schlgt dieReferenzrahmennetztheorie eine einheitliche Gedchtnisstruktur fr WegfindungundReorientierungsowiefrRoutenundberblickswissenvor.Zudemzeigt sieGemeinsamkeitenundUnterschiede zwischenmenschlicherundnichtmenschlicherRaumorientierungauf.SieerklrtunterschiedlicheErgebnisse,unteranderem zur Orientierungsspezifitt, dem Einfluss von Erfahrung und derAsymmetrie desRaumgedchtnisses.Weiterhinwirft sie Fragen fr zuknftigeExperimente auf und macht dabei verschiedene empirisch berprfbareVorhersagen.Die hier vorgestellten empirischenErgebnisse sowieder indieserArbeit vorgeschlagene theoretische Rahmen stellen einen Betrag dar, zurEntwicklungeinerfunktionellenTheoriederOrientierungimRaum.

10

Acknowledgements

Without the help of a number of people who supported me throughout thevariousstagesofmythesisthisworkwouldnothavebeenpossible.

Iwouldliketothank:

GerhardStrubeandMarkusKnauffforsupervisingthisthesis.

Heinrich H. Blthoff and the MaxPlanck Society for providing a fantasticenvironmentfordoingresearch.

The Cognitive Science Department in Freiburg and the TransregionalCollaborative Research Center Spatial Cognition for providing input oncognitioningeneralandspatialcognitioninspecificfromdiverseperspectives.

TheGermanResearchFoundation(DFG)andtheEUfortheirfinancialsupportintheprojectsMapSpace(SFB/TR8)andWayfinding(6thFPNEST).

Christoph Hlscher, Simon Bchner, Martin Brsamle, Hanspeter Mallot,Bernhard Riecke,Manuel Vidal, Jrg SchultePelkum, Gerald Franz, JanMalteWiener, Jack Loomis,Gottfried Vosgerau, Lars Konienczny,Daniel Berger andmanyothersfordiscussingideas.

DanMontelloforhisdistinctionbetweenfigural,vistaandenvironmentalspacesandforsuggestionsforliterature.

Michael Weyel, HansGnter Nusseck, Thomas Fangmeier, Harald Teufel,BenjaminTurski,FranckCaniardforintensiveandsustainedsupportwithvariousproblemsregardingthetechnicalsetupandprogramming.

AnnaWidiger,NaimaLaharnar,TanjaWaltherandGottfriedVosgerau forhelpwithdatacollectionandanalysis.

Alltheparticipantswhogotlostduringtheexperiments(notalwaysaccordingtomyinitialhypotheses)andalsotothosewhodidnotgetlost.

Betty Mohler, Jenny Campos, Simon Bchner, Christoph Hlscher, MartinBrsamle,KarinPilz,JoshSiegle,JohnButler,GregorHardie,AlexanderSchnee,BernhardRiecke,andDanielBergerforreadingpartsofmythesis,givinghelpfulsuggestions, and transforming the sentences Iwrote into something resemblingEnglish.

CoraKrner,MartinBreidtandStefanStreuberforhelpwithlayouting.

11

Manyotherfriendsandcolleagues(whicharenotseparategroups)fromFreiburg,Bremen,andTbingenfortheirinspirationandtheirsupport.

Cordula,Maxandmyparentsforalwaysbeingthereforme.

Notes on the printed version

More recentversions of the studiespresented in this thesis arepublished or inpressinthefollowing:

Study1isnowpublishedinCognitiveScience,32,755770.

Study2iscurrentlyinpressintheJournalofSpatialScience.

Study3iscurrentlyinpressinEnvironmentandPlanningB.

Study5isnowpublishedinT.Barkowsky,M.Knauff,G.Ligozat&D.R.Montello(Eds.)SpatialCognitionV(pp.381400).Berlin:Springer.

12

frCordula

13

Contents

1Introduction.............................................................................................................................................. 17

2Theory........................................................................................................................................................ 20 2.1Orientation,spaces,andmaps ........................................................................................................ 20 2.2Goals ............................................................................................................................................. 21 2.3Orientationmechanisms .................................................................................................................. 23

2.3.1Orientationprocesses ............................................................................................................ 23 2.3.1.1Classificationsoforientationprocesses ................................................................ 23 2.3.1.2Reorientation ............................................................................................................ 26 2.3.1.3Updatingbypathintegration................................................................................. 32 2.3.1.4Routenavigation ...................................................................................................... 35 2.3.1.5Therelationbetweenreorientation,pathintegrationandroutenavigation ... 39

2.3.2Strategies ................................................................................................................................. 41 2.3.2.1Thedifferencebetweenprocessesandstrategies,humanandnonhuman

navigators ................................................................................................................. 41 2.3.2.2Reorientationstrategies........................................................................................... 43 2.3.2.3Wayfindingstrategies.............................................................................................. 44

2.4Spatialknowledge ............................................................................................................................ 47 2.4.1Familiarity............................................................................................................................... 48

2.4.1.1Effectsoffamiliarity................................................................................................. 48 2.4.1.2Orientationdependency ......................................................................................... 48 2.4.1.3Summary................................................................................................................... 50

2.4.2Landmark,routeandsurveyknowledge ........................................................................... 50 2.4.2.1Developmentalsequence ........................................................................................ 51 2.4.2.2Relationtowayfindingmechanisms ..................................................................... 51 2.4.2.3Formingdiscreteknowledgefromacontinuousinput ...................................... 52

14

2.4.3Cognitivemap.........................................................................................................................52 2.4.3.1Whatisacognitvemap?..........................................................................................52 2.4.3.2Vistaspacesandenvironmentalspaces.................................................................53 2.4.3.3Nospontaneousintegrationofseparatelylearnedvistaspaces

intoacognitivemap ................................................................................................54 2.4.3.4Shortcutting ...............................................................................................................54 2.4.3.5Multiplecognitivemapsarenecessary .................................................................56 2.4.3.6Summary....................................................................................................................57

2.4.4Biasesinspatialknowledge ..................................................................................................57 2.4.4.1Straighteningedges,aligningedgesandsquaringobliqueintersections.........57 2.4.4.2Biasesduetoregionalisingspaces..........................................................................59 2.4.4.3Biasesduetoconnectednessandavailablelandmarks .......................................60 2.4.4.4Summary....................................................................................................................60

2.4.5Framesofreference ................................................................................................................61 2.4.5.1Theexistenceofegocentricandallocentricreferenceframes.............................61 2.4.5.2Thenatureofallocentricreferenceframes ............................................................65 2.4.5.3Summary....................................................................................................................72

2.4.6Sourceofknowledge..............................................................................................................72 2.4.6.1Knowledgeacquiredfromenvironmentalspace .................................................73 2.4.6.2Knowledgeacquiredfromvistaspace...................................................................74 2.4.6.3Knowledgeacquiredfromfiguralspace ...............................................................75 2.4.6.4Knowledgeacquiredfromverbaldescriptions ....................................................76 2.4.6.5Knowledgeacquiredfromvirtualenvironments.................................................77 2.4.6.6Summary....................................................................................................................80

2.5Representations .................................................................................................................................81 2.5.1Definitionofrepresentations ................................................................................................81 2.5.2Descriptiveversusdepictiverepresentations:theimagerydebate .................................82

2.5.2.1Argumentsfordepictiverepresentations..............................................................82 2.5.2.2Argumentsagainstdepictiverepresentations ......................................................83 2.5.2.3Conclusion.................................................................................................................84

2.5.3Workingmemory ...................................................................................................................85 2.5.3.1Thephonologicalloop .............................................................................................86 2.5.3.2Thevisuospatialsketchpad.....................................................................................86 2.5.3.3Thecentralexecutive................................................................................................88 2.5.3.4Workingmemoryinspatialorientation ................................................................89 2.5.3.5Summary....................................................................................................................90

15

2.6Strategiesfororientationinenvironmentalspaces ...................................................................... 90 2.6.1Study1:Memorystrategiesappliedforwayfinding ........................................................ 93 2.6.2Study2:Doesthespatialmemorystrategyrelyonfiguralspaces?................................ 93 2.6.3Study3:Doesthespatialmemorystrategyrelyonvistaspaces? ................................... 94 2.6.4Study4:Familiarityandtheefficiencyofwayfindingstrategies .................................... 94 2.6.5Study5:Metricandnonmetricstrategiesinwayfindingandreorientation ................ 95 2.6.6Notesonmethodology.......................................................................................................... 96

2.7References .......................................................................................................................................... 96

3Experiments ............................................................................................................................................ 120 3.1Workingmemoryinwayfinding.................................................................................................. 120 3.2Askfordirectionsoruseamap .................................................................................................... 138 3.3Fromisovistsviamentalrepresentationstobehaviour............................................................. 154 3.4Upthedownstaircase.................................................................................................................... 171 3.5Schematicmapsinwayfindingandselflocalisation ................................................................. 201

4Discussion............................................................................................................................................... 224 4.1Summaryanddiscussionoftheindividualstudies................................................................... 224

4.1.1Study1:Memorystrategiesappliedforwayfinding ...................................................... 224 4.1.2Study2:Doesthespatialmemorystrategyrelyonfiguralspaces?.............................. 225 4.1.3Study3:Doesthespatialmemorystrategyrelyonvistaspaces? ................................. 225 4.1.4Study4:Familiarityandtheefficiencyofwayfindingstrategies .................................. 227 4.1.5Study5:Metricandnonmetricstrategiesinwayfindingandreorientation .............. 228 4.1.6Summary............................................................................................................................... 229

4.2Dualcodinginspatialorientation ................................................................................................ 230 4.2.1Dualcodingoffigural,vista,andenvironmentalspaces ............................................... 230

4.2.1.1Figuralspaces ......................................................................................................... 230 4.2.1.2Vistaspaces............................................................................................................. 231 4.2.1.3Environmentalspaces............................................................................................ 231

4.2.2Dualcodingcomparedtoothertheoriesofspatialmemory ......................................... 232 4.3Thenetworkofreferenceframestheory...................................................................................... 236

4.3.1Structureandprocessesassumed...................................................................................... 236 4.3.1.1Structure .................................................................................................................. 236 4.3.1.2Encoding ................................................................................................................. 237 4.3.1.3Reorientationbyrecognition................................................................................ 242 4.3.1.4Routenavigationbyactivationspread ............................................................... 242 4.3.1.5Surveynavigationbyimagination ...................................................................... 243

16

4.3.2Thenetworkofreferenceframestheoryinthetheoreticalandempiricalcontext ......244 4.3.2.1Vistaspacereferenceframesasthebasicunitinknowledge

ofenvironmentalspaces........................................................................................244 4.3.2.2Therelationbetweenvistaspacereferenceframes:

Networkvs.hierarchicstructureofenvironmentalspaces..............................245 4.3.2.3Thesubstructureofvistaspaces...........................................................................246 4.3.2.4Allocentricandegocentricreferenceframes.......................................................249 4.3.2.5Asymmetryinspatialmemory.............................................................................253 4.3.2.6Imaginingdistantlocationsasadifferencebetweenhuman

andnonhumannavigators...................................................................................255 4.3.3Summaryandopenquestions ............................................................................................256

4.4Strategiesoforientationinenvironmentalspacesconcludingremarks................................258 4.5References.........................................................................................................................................262

17

1 Introduction

Orientation in space is fundamental for all humans and other animals.Accomplishing our goalsmost often requiresmoving through complex spacessuch as houses or cities also called environmental spaces.Wemove from ourhometowork,toouroffice,tothesupermarketortothenewpub.Forwellknownroutes this is basically effortless. We do not even notice the tremendousperformanceweaccomplisheveryday.Thischangeswhenwearenolongerabletouse thememory of our environment, e.g.,Alzheimerpatients sometimes areunable to find even theirway back home (e.g.,Monacelli,Cushman,Kavcic&Duffy,2003).Inroboticsitisconsideredadifficultproblemtobuildrobotsthatareabletonavigatethroughtheworldandatthesametimekeeptrackofwheretheyare (e.g.Stachniss,2006).We realizewhatorientationproblems canoccurwhentryingtoreachacompletelyunknownlocation.Fornewlocationswecannotrelyonourmemoryand just follow learned routes,butwehave to relyon externalsourcesofinformationwhichtelluswhereourgoalislocatedandhowwecangetthere.Someof theseexternalsourcesaregestures,verbalwayfindingdirections,signsandmaps.

Navigating to unknown locations is not a specific human capability. Somemigratory birdsgenetically know inwhichdirection to fly towhenwinter isapproaching(e.g.,Berthold,Helbig,Mohr&Querner,1992).Otherspeciesareableto communicate locations in order to reach unknown locations: Ants followolfactorytrailslaidoutbytheirconspecifics(Hlldobler,1971)andbeeslearntheangleanddistanceofa food sourcebyobservinganotherbeesdance (e.g.,vonFrisch, 1967).However,none of these communications of spatial locations is asflexibleashumansuseofverbaldirectionsormaps.

Theexamplesmentionedshowthatseveralbiologicalandtechnicalsolutionsarepossible to one specific orientation problem. The present work is bound tounderstand the information processing underlying orientation in such a broadperspective seen from different point of views such as psychology, biology orcomputer science. Themost relevant questions are What strategies enable toachieve goals, e.g., to knowwhere one is or to find a specific location in theenvironment? and How is the environment represented? Our approach toanswerthesequestionsfollowsseveralimplicitassumptions:

In the lastdecades, in the fieldofpsychology,aswellas inartificial intelligenceandcognitivescience,explanationsforvariousproblemschangedfromassumingone general representation format together with one general problem solvingmechanism towards domainspecific knowledge andmultiple, problem specific

18

formsofrepresentationsaswellasmechanismsoperatingonthem(e.g.,Tye,1991;Sloman, 1984; Strube, 1996). This change brought these fields closer to biologywheremultiplemechanisms guiding behaviourwere assumedmore often (e.g.,McFarland,1999).Thepresentworkwillfollowthisdevelopmentintheory.Itwillfocusonmultiple formsofrepresentationsaswellasmultiplemechanismsusedfororientation.1

A second assumption of this work is a rather continuous phylogenetic andontogenetic development. Orientationmechanisms can evolve alongwith newsensory inputs (e.g.Wehner, 1994) or strategies developwith higher cognitivecapacitiesinchildren(e.g.,Cornell,Heth&Alberts,1994).Othermechanismssuchashumanpathintegrationmightnotbeusedsointensivelyanymore(cf.,Loomisetal.,1993).Newabilities,sucas language,enablenewstrategies fororientationproblems, e.g., giving verbal directions to somebody. Such new strategies,however, are not assumed to replace all other existingmechanisms. It is ratherunlikely that humans orient completelydifferently than nonhuman animals orthat human adults orient completely differently than children. Taking thisperspective of a rather continuous development implies also thatwe can learnabout orientationmechanisms in adults by looking at how children and nonhumananimalsorient.

This research is within the area of cognitive psychology. It is not indevelopmental,biological andpersonalitypsychology,nor inbiology, cognitivescience, neuroscience or artificial intelligence. It is strictly empirical andbehavioural.This reseachdoesnot implyany formalmodellingofexperimentaldata. However, this work tries to take into account findings from all thementioned research areas. Psychology in general and cognitive psychology inparticularisintouchwithallthementioneddisciplines(Prinz&Msseler,2002).Results from thesedisciplinesarebound toplaceconstraintsandsuggestionsonpossible theories used to explain adult orientation in space. In that sense thepresent work will consider findings from various sources. However, it is farbeyondthescopeofthisworktodiscussallrelatedareas,eveninsucharelativelyyoung interdisciplinary fieldsuchasspatialcognition.Due to theempiricalandbehaviouralorientationof thiswork, itwillmainly considerbehavioural resultsfromorientation inanimalsaswellasdevelopmental,biologicalandpersonalitypsychology. However, when highly relevant, findings from neuroscience,cognitivescienceandartificialintelligencewillalsobediscussed.

1 Due to Occams razor, explanations requiring fewer theoretic assumptions are preferable toexplanations requiringmore assumptions.This is not necessarily an argument againstmultiplerepresentations or mechanisms. Explaining empirical results with one general representationformatusuallypropositional/symboliccanrequiremoreadditionalassumptionsthanexplainingthemwithmultiplerepresentations.

19

We will, first, review the literature concerning orientation in environmentalspaces, second,wewillpresentanddiscuss the resultsof five studies regardingopenproblemsinstrategiesoforientationandthird,wewilldiscusstheseresultsinthebroadperspectiveofspatialorientationresearch.

20

2 Theory

Inorder todiscussorientation in environmental spaces,onehas toknow,whatorientationandwhatanenvironmental space is.Therefore, the first sectionwilldefine the terms space and orientation. Spatial orientation is about achievinggoals,e.g.,tolocateourcurrentpositionorreachacertainlocation.Insectiontwowewill look at thesegoals.To reach thesegoalsweapplydifferentorientationmechanisms e.g., the strategy of trying to minimise the deviation from thedirectionof thegoal locationduring locomotion.Section threewill regard thesemechanisms. Inorder toapplyastrategyoranothermechanisms fororientationknowledge is required, e.g., the location of the goal relative to our currentposition. In section four we will discuss types and organisation of spatialknowledge. We obtain this knowledge either directly by experiencing ourenvironmentorweobtainitindirectlyfromothersources,suchasmapsorverbaldirections.Wenotonlyobtaintheknowledgefromdifferentsourceswealsousedifferentmemorysystemstorepresentit,e.g.visual,spatialorverbalmemory.Insectionfivewewilllooktherepresentationalformatofthisknowledge.

2.1 Orientation, spaces, and maps

This thesis isaboutorientation inenvironmental spaces.Byorientation in spacewemeanorientation inaphysicalspace,notorientation in lifeor inmetaphoricalspacessuchasmathematicalspaceorthehyperspaceoftheinternet.Wealsodonot consider allpossiblephysical spaces toorientwithin,but limitourselves tospaceswhich surround us andwhichwe apprehend by locomotion.Montello(1993)distinguishestheseenvironmentalspacesfromgeographicalspaceswhicharetoobigtoapprehendbylocomotionandthereforehavetobelearnedviamaps.IngeographicalspacesonecouldtellwhetherNapoliorNewYork isfurthertothenorthorwhetherRenoorSanDiego is further to thewest (cf.Stevens&Coupe,1978).Inenvironmentalspacesonecouldwalktoalocation.Mapsalsooccupyaspace.Montello (1993) calls this figural space. It is projectively smaller than thebodyandnolocomotionisneededtoperceiveitsproperties.Hesubdividesfiguralspaceintopictorialandobjectspaces,theformerreferringtosmallflatspacessuchasmapsandpicturesand the latter tosmall3Dspacessuchassmallobjectsordistant landmarks. Pictorial spaces can be used to depict environmental orgeographicalspaces.The lastspace in thisdistinction iscalledvistaspace.It isaslargeorlargerthenthebody,butcanvisuallybeapprehendedfromasingleplacewithout locomotion. It is the spaceof single rooms, town squares, smallvalleys

2.2 GOALS

21

and horizons.2 Thisworkwill consider orientation in environmental spaces. Inordertodososeveralvistaspaceshavetobecrossedandparticipantsoftenrefertothefiguralspaceormorespecifictothepictorialspaceofamap.Spaceswhichareexclusivelylearnedbymaps,suchasverylargegeographicspaces,willnotbeconsidered. However, as far as maps display environmental spaces they arerelevantforthiswork.

Graphic representations such asmaps imply abstraction from the environment,however, sincemaps often distil and highlight important information, scale issometimes only roughlypreserved (Tversky, 2000).Therefore, for ourpurposeswe can define maps as twodimensional graphic representations of spatialrelationsinanenvironment.Thisincludes,e.g.,acrosssectionofabuildingoraroutedrawninthesand.Itexcludespicturesandthreedimensionalmodels.

We defined what spaces we examine, but what exactly do we mean byorientation?Orientation involves goaldirected interactionwith an environment(cf.Montello,2005).Thesegoalscancomprisereachingaknownlocation,findingan unknown location, reorienting oneself after getting lost, or exploring theenvironment.Orientationinvolvescognitiveaspectssuchasplanning,recognisinglandmarks,orupdating the locationofanobjectduringmovement.Orientationalso involves the execution ofbehaviour, e.g.,motor control for locomotion (cf.Passini, 1992).Wewill, however, focus on the cognitive rather than themotoraspects of orientation.3Definitions for the individual cognitive aspects such asstrategies or representations will be given in the respective sections.Following thementioneddefinitionswewill examinedifferent strategies in thegoaldirectedinteractionwithenvironmentalspaces.Thenextsectionwilldiscussthevarioustypesofgoalsinthisgoaldirectedinteraction.

2.2 Goals

Several goals can be pursued in an environmental space.Wewill distinguishwayfinding,reorientation,andexploration.

Wayfindingoccurseverydaywhenwewanttogetsomewhere:fromourbedtothebathroom, from our flat to ourwork, or from the train station to a conferencecentre.Whenfindingourwayweknowwhereweareandwherewewanttogo.We,e.g.,visitedourgoal locationbeforeand, therefore,know it, i.e.,werelyon

2Manyauthorsusethedistinctionbetweensmallscaleandlargescalespaces(e.g.Fields&Shelton,2006).Asthisdistinctionisnotdefinedverywellwewillnotuseithere.Bothfiguralandavistaspace can be considered as small scale spaces and bothvista and environmental spaces can beconsideredaslargescalespaces.ForadistinctionofspacessimilartoMontello(1993)seeTversky(2005).3Asthisworkdoesnotfocusonmotorcontrol,wedonotfurthersubdividevistaspaces,e.g.,intoactionspace,i.e.,thespacewithinoneperformsmotoractions(cf.Cutting&Vishton,1995).

CHAPTER 2 THEORY

22

ourmemoryoftheenvironment inordertogetthere.Ifwehavenotbeentothegoallocationbefore,wecannotrelyontheinternalrepresentationofourmemory,butmustuseanexternalrepresentation,e.g.,amap,verbaldirection,orgestureswhichtelluswherethegoalistobefound.Ifwedonotknowthelocationofthegoalandnoexternalrepresentationsareavailable,wemustsearchforthegoal.Fororientation indaily lifesuchanuninformedsearch isratheruncommon.Usuallyweknowatleastsomethingaboutagoallocationanddonotsearch,e.g.,awholetown.Duetothehigherrelevanceindailylifeandforthiswork,wewillfocusonwayfindingwhereinformationaboutthegoallocationisavailable.

Wayfinding,bydefinition,alsoinvolvesaselectioninwhichdirectiontogoinanenvironmentalspace.Aconferencecentrecould,e.g.,alsobereachedbyhiringataxiandtellingthedriverwheretogo.Thisdoesnotinvolveselectingadirection,asthedriverisdoingthisforus.Wayfindingisdelegatedtothedriver.Thesameargumentation holds true for taking the train or the plane. In these cases theproblem of reaching a goal is solved by planning processes which do notnecessarilyinvolvetherepresentationofspace.

Weconsiderwayfindingasinvolvingsituationsinwhichweknowwhereourgoalistobefoundandwhichinvolvesinteractionwithspaceinasensethatwedecideourselveswheretogobasedonthisknowledge.

Askingwheresomethingisinanenvironmentandhowitcanbereacheditisonequestion,anotherquestion isaskingwhereone is in theenvironment (cf.Allen,2004).Thecorrespondinggoalistolocaliseoneselfwhennotknowingwhereoneis.Wewillcall this taskof findingonesposition inanenvironmentreorientation,butitisalsoreferredtoasselflocalisation.Forconvenience,thetermpositionwillbe used to refer to the combination of a particular location and a particularheading in space (cf. Rump & McNamara, in press). Reorientation is onlynecessarywhenone isdisorientedwith respect to theenvironment.Contrary tothat,wedefinedbeingorientedasaprerequisiteforwayfinding,i.e.,wayfindingimpliesthatnavigatorsknowwhereareandwherethegoalis.Sometimestheygetlostduringwayfinding.Due to ourdefinitionsnavigators firsthave to reorientbeforebeingabletopursuetheirgoalagain.

The last goalwewant to describe is simply to learn something about a newenvironment. For this purposewe look at amap orwalk around, e.g, duringwindowshoppinginanewtown.Traditionally,thisbehaviouriscalledexploration(at least for walking around). The explorer is oriented with respect to theenvironmentasinwayfindingandsearch,butcontrarytowayfindingandsearchno specific location is soughtafter. Exploration is an important goal whenorienting in space. This work, however, will focus on wayfinding andreorientation.

2.3 ORIENTATION MECHANISMS

23

2.3 Orientation mechanisms

Thegoalstoknowwhereoneis(reorientation)andhowonecangetfromheretoacertain location (wayfinding) are very common in humans and the rest of theanimalkingdom.Asachievingthesegoalsiscrucialforsurvivinginmanyspecies,variousmechanismshaveevolvedtosolvetheseproblemsoverthepastmillionsofyears.Thesemechanismscanbeclassifiedinseveralways(e.g.,Franz&Mallot,2000;Gillner&Mallot,1998;Mallot1999;Trullier,Wiener,Berthoz&Meyer,1997;Wang&Spelke,2002).Oneverybasicdistinctionisbetweenprocesseswhichcanbe observerd inhumans and other animals and strategieswhich are specific tohumans.

2.3.1 Orientation processes

2.3.1.1 Classifications of orientation processes

Several authors have provided classifications of orientation processes.Wewilllook at the approaches provided by Wang and Spelke and by the groups ofTrullierandMallot.

Wang and Spelke.WangandSpelke(2002)distinguishthreeprocessesrelevantfororientation: path integration, viewpointdependant place recognition andreorientation.Pathintegrationisaprocessbywhichtherelationofahumanoranonhuman animal to one or more significant places in the environment isupdated continuously during movement. The viewpointdependant placerecognition operates by template matching of viewpointdependentrepresentationsof landmarks. Itallowsnavigating fromone location toanother.The reorientation system looks for congruences between representations of theshapeofthesurfacelayout.Itfocusesonthegeometryofthesurroundingsurfacelayoutasacuefororientationafterbeingdisorientated.

Wang and Spelke regard encapsulated representations of the environment asbuilding blocks for these three proposedmechanisms in animals and humans.Specifically human symbolic capacities enable to construct new spatialrepresentations and strategies to overcome the limits of the more primitivenavigationalsystems.

Trullier and colleagues. Trullierand colleagues (1997)propose severalprocessesbasedonareviewofbiologicallyinspiredcomputationalmodelsofnavigationinanimals: guidance (move in relation toperceptions),place recognitiontriggeredresponse (orient relative to specific places), topological navigation (move alongknownpaths) andmetricnavigation (move in relation to an overview of thewholeenvironment/movenewpaths).Asaprerequisiteforallotherprocessesanavigatorhas tobe able to approach a location.This couldbe achieved e.g.by

CHAPTER 2 THEORY

24

aligningthebodywiththedirectiontowardsthegoalandthenmovingforward.For visual stimuli this is also called beaconing. In guidance a navigator actsimmediatelyonthesensorinputandastoredsensoryinput,e.g.followsawall,ortries tominimise thediscrepancyof thecurrentview toastoredviewe.g.at thegoal.Withthismechanism,e.g.,aspecific location inbetweenseveral landmarkscanbe reached inavista space. Inplace recognitiontriggered responseanavigatorrecognisesaplacevisitedbeforeandselectsadirectiontomoveonsoastoreachitscurrentgoal.Thegoalisnotvisiblefromthecurrentlocationsoitisaprocessfor reaching goals in environmental spaces. A place in this sense is a set ofcontinuous locationswhere anavigator selects the same action.The situation isperceived as identical or very similar to a learned one.Withplace recognitiontriggered response a navigator is able to reach a goal, but has no internalrepresentation of the relations between the current place and other places.Noplanningisinvolved,i.e.itisnotpossibletorepresentthecompleteroutefromthecurrent place to the goal.With this process only, a barrier on the routewouldmakeitimpossibletoreachthegoal.Intopologicalnavigationnoplacegoalactionassociationsarestored,butplaceactionplaceassociationsarebuiltandstoredinmemory.Anavigatorknowshow toget fromone rememberedplace toanotherandviceverse.Thiscanbethoughtofasagraphwherethenodesrepresentplacesand the edges represent actions to reach from one place to another. With arepresentationlikethis,planningispossible,i.e.searchingforasequenceofplacestovisitinordertoreachthegoal.Iftheselectedroutewasblocked,analternativeroutecouldbeplannedifcontainedintherepresentation.Gettingasequencefromthecurrentpositiontothegoalmustnotbeasymbolicsearchalgorithm,butcanbeachievedbyactivationspreadfromthenoderepresentingthecurrentpositionand/orfromthenoderepresentingthegoal.Amechanismselectingthesequenceofnodeswiththehighestactivationwouldgivesimilarresults.Suchamechanismwouldalwaysselecttheroutewiththefewestnumberofnodes.Itcan,therefore,explain identical behaviour without assuming symbolic planning processes.Contrary to the three already described processes metric navigation allows fornovel trajectories,especially shortcuts.Forexampleanavigatormightknow theroutearoundabig forestandnow tries todirectly cross the forest to reach theothersideonamuchshorterpath.Inordertodothismetricpropertieshavetobestored, i.e. angles and distances between locations. To identify a shortcut novisible landmarks indicating the goal position are permitted or the behaviourcouldbeexplainedbyguidance.

Mallot and colleagues. Similar to Trullier and colleagues (1997) Mallot andcolleagues(Franz&Mallot,2000;Gillner&Mallot,1998;Mallot1999)distinguishseveral processes, e.g., search, direction following, path integration or routenavigation. In addition, Mallot (1999) proposed a complexity hierarchydistinguishingtheprocessesbasedonthetypeofmemoryrequired(seeFigure1).Onlevelone,adirectmappingbetweensensorsandeffectorsoccurs.Thisenables


25

e.g.a femalecricket toreachherchirpingpartnerduringmating (e.g.,Pollack&Plourde,1982),orabraitenbergroboticvehicletofollowamovinglightsourceorto hide in darkness (Braitenberg, 1984).With a simpleworkingmemory (leveltwo)sensoryinputscanbeintegratedovertimeandspacewhichallowsforpathintegration.Anantcanstoreavectorpointing from itscurrentposition towardsthenest.Duringwalkingaroundthisvectorisupdatedcontinuouslyenablingtheanttoreturnto itsnestonastraightpath(e.g.Wehner&Wehner,1986).Onthethird level long termmemoryallows for learning.Landmarkbasedmechanismssuchasguidanceandrecognition triggeredresponsesarepossiblewherecertainbehaviour is associated with a recognised stimulus or is derived from thediscrepancybetweenthecurrentandamemorisedstimulussituation.Declarativememory(levelfour)isrequiredtoplanandtotraveldifferentroutescomposedofpiecesandstepsstored inmemory.Movementdecisionsdependnotonlyonthecurrent landmark information,butalsoonthegoalthenavigator ispursuing.Atthesamelocationonecanturnrightinordertoreachhomeorturnleftinordertogetsomefood.Declarativememorydoesnotnecessarilyimplymetricinformation.

Figure 1: Four levels of complexity of behaviour according to Mallot (1999). Level 1 allows reflex-like behaviour based on the wiring of effectors and the current sensory input. Level 2 includes spatio-temporal processing of inputs arriving at different sensors and at different times. Learning is introduced at Level 3, by allowing for plasticity of the spatio-temporal processing. Except for this plasticity, behaviour is still determined completely by the sensory input. At level 4, one sensory input may elicit different behaviours depending on the current goal of the agent. The figure is taken from Mallot (1999).

Summary. Several orientation processes have been proposed. Most authorsdiscusspath integrationandwhatwewould like tocall routenavigationwhichwouldencompassviewpointdependentplacerecognition,recognitiontriggered

CHAPTER 2 THEORY

26

response and topological navigation. Questions relating to metric or surveynavigationwillbediscussedinSection2.4.Reorientationisnotconsideredbyallmentionedauthors.However,researchinthisareaisalsodirectlyrelevantforthecurrentwork.Inaddition,manyprocessesrequireanavigatortobeorientedwithrespect to the environment, which is accomplished by reorientation. We will,therefore, discuss reorientation, path integration and route navigation inmoredetail.

2.3.1.2 Reorientation

Asdescribedinthegoalsection,wedefinereorientationastryingtoregainonesposition, i.e., location and heading, with respect to an internal or externalrepresentation of an environment.We use the same name reorientation for thegoal and for the process enabling to fulfil it. Finding specific locations in anenvironment can be seen asmeasures of reorientation. Under the label placelearning thishasbeen the subjectofmany studies inhumansandespecially inotheranimals.Inthissectionwewillfocusontheprocessofreorientationandthecuesused,ratherthanonthepropertiesoftheunderlyingknowledge.ThesewillbediscussedinSections2.4and2.5.Thecuesusedforreorientationcanbedividedinto the geometric layout of an environment (e.g., the shape of a room) andspecificlandmarks(e.g.,objectsinanenvironment).4Wewillfirstregardgeometryandrelatedaspectsofthisresearchandthendealwithreorientationonlandmarks.As research on hippocampal place cells is tightly related to thiswork,wewillintroducethisrelatedworkafterwards.

Reorientation on geometry, colour and texture: the rectangular room paradigm.Intherectangular room paradigm, subjects see an object hidden in a corner of arectangular room and are then disoriented (for a recent review see Cheng &Newcombe, 2005).Reorientation is partially specified by the rooms shape andfullyspecifiedbyboth theroomsshapeandnongeometric information,e.g. thecolourofawall(seeFigure2),apatternonthewallorsometimesalsoalandmark.Subjectsdemonstrate their ability to reorient themselvesby locating thehiddenobject.Rats(e.g.,Cheng,1986)andchildreninasmallroom(HermerundSpelke,1994, 1996; Learmonth, Newcombe & Huttenlocher, 2001) orient only on thegeometry,i.e.,whentheobjectishiddenintheupperleftcornerofFigure2theysearchequallyofteninthetwocornersmarkedbyadot,butmoreofteninthesecorners than in the other two corners.An encapsulated (cf. Fodor, 1983) shapebasedreorientationspecificmechanismwasproposedasanexplanation(Hermer& Spelke, 1994; 1996;Cheng, 1986) and called geometricmodule (e.g.Cheng,1986;Gallistel,1990;seealsoCheng&Newcombe,2005).Encapsulatedmeansthat

4Many authors use the term landmark also forwalls and corners, i.e., features of a geometriclayout.However,werestrictouruseofthetermlandmarkstoobjects,e.g.,trees,polesorhousesintheenvironment.


27

nootherkindsof informationother thangeometry are involved in theprocess,eventhoughgeometriccuesmightbeavailableforotherprocesses.

Adultsgenerallyusebothgeometricandnongeometric informationunless theyaredisturbedbyaverbalshadowingtaskwheretheyhavetoimmediatelyrepeatwords from a text presented via headphones. This interference does not occurduring clapping a rhythm or repeating syllables (HermerVasquez, Spelke &Katnelson,1999).HermerVasquezetal.(1999)assumedlanguagetobenecessarytocombinegeometricandnongeometricinformation.Thisassumptionaswellastheproposalofanencapsulatedgeometricmodule,however,isquestionedbythefinding thatprimates,birdsandeven fishcanusegeometricandnongeometricinformation for reorientation (Gouteux, ThinusBlanc & Vauclair, 2001; Kelly,Spetch&Heth,1998;Sovrano,Bisazza&Vallortigara,2002).Also,1724montholdchildrenareable tousebothkindsof information ina larger room (Learmonth,Newcombe&Huttenlocher,2001)and theshadowingeffectsof languagedonotoccur when the adults receive a training trial and more explicit instructions(Ratkliff & Newcombe, 2005). Although language processes do not seemnecessary, they are stillhelpful as there is aboost in reorientationperformancewithin children around the ages of five and sixyears regarding their emergingspatial language abilities e.g. verbal expressions involving the terms left andright (HermerVazquez, Moffett & Munkholm, 2001; Learmonth, Nadel &Newcombe,2002).Anotherrecentexplanationfocusesonhemisphericcrosstalkasa prerequisite for combining geometric and nongeometric information(Newcombe,2005).

When only geometric cues are available, rats and chicken seem tomatch localgeometrysuchasthesizeofwallsandtheanglebetweenwallsinordertofindafoodsourceinaroomwithadifferentgeometry,e.g.arhombus,aparallelogramorakiteshapedroom(McGregor,Jones,Good&Pearce,2006;Pearce,Good,Jones&McGregor,2004;Tommasi&Polli,2004),ratherthanorientonthemainaxisofaroomasproposedbyCheng&Gallistel(2005).They,therefore,seemtofocusonlocalgeometriccuesratherthanonglobalgeometriccues.

Therectangularroomparadigmwasused toexamine the influenceofgeometry,colourandlanguageonreorientation.Thenextsectionwillconsiderreorientationonobjectsaslandmarksandtheirconfiguration.

Figure 2: The rectangular room with one wall painted in a different colour, as used in many experiments. Opposite corners of the rectangular room are geometrically identical. To disentangle this ambiguity the colour of the walls has to be taken into account.

CHAPTER 2 THEORY

28

Reorientation on visible landmarks. When reorienting on visible landmarks, asubject sees objects4 from a location in avista space.Afterbeingdisoriented inordertoexcludeupdatingbypathintegrationthatvistaspaceisenteredagainandthesamelocationhastobereachedbasedonthememoryofthelandmarks.Thistaskoftenisreferredtoasplacelearning.Inordertofindaplaceagainthesubjecthastoreorient,i.e.,toidentifyonespositioninrelationtotheenvironment,usingvisiblelandmarks.AtypicalsetupforratsistheMorriswatermazetask(Morris,1981;Morris,Garrud, Rawlins&OKeefe, 1982). In this task a rat swims in acircular basin surrounded by landmarks and filled with milky water until itreachesa smallplatformhiddenunder thewater surfacewhere itcan rest.Thisplatform isnotvisibleasthewater isopaque.Inatesttrialafterwards,therat isputagainintothepoolatarandompositionandtriestofindtheplatformasfastas possible. To do so the rat has to rely on memory of distal landmarkssurroundingthebasin.

Forhumans,theseexperimentsmostoftenweredoneusingvirtualenvironments,adesktopsetupwhereparticipantsuseamouseorkeyboardtomovearoundoranimmersivevirtualenvironmentwhereparticipantscanwalkthroughavirtualenvironmentdisplayedonvideoglassescalledheadmounteddisplays.Insuchanimmersive setup participants can receive inertial cues such as proprioceptivefeedback, efference copies or vestibular information which is not possible indesktopvirtualrealities.InadesktopversionoftheMorriswatermaze,humanslearn to take straight trajectories to theplatform in thepresenceof conspicuousdistalcuesunlesstheysufferfromhippocampallesions(Astur,Taylor,Mamelak,Philpott & Sutherland, 2002).Men perform better thanwomen in such a task(Astur,Ortiz,&Sutherland,1998).Ingeneral,theseexperimentshaveshownthatplace learning inhumanscanoccurreadily incomputersimulatedenvironmentsandthatsuchlearningfollowsmanyoftheprinciplesofplacelearninginanimals(Hamilton and Sutherland 1999; Jacobs, Laurance & Thomas, 1997; Jacobs,Thomas,Laurance&Nadel,1998;Sandstrom,Kaufman&Huettel,1998). In thefollowingwewill focus on the cues used for orientation in humans and nonhumans.

Awidevarietyofspeciesareabletoreorientonvisible landmarks.Beesseemtomatchstoredvisualsnapshotsoflandmarkswiththeiractualview:whenthesizeofthelandmarkwaschangedbetweentrainingandtesting,theareainwhichbeessearchedwasdisplaced toonewhere the landmarkappeared roughly the samesizeasthetraininglandmarkwhenviewedfromthetraininglocation(Cartwright&Collett,1982;1983).Contrarytobees,mice(Collett,Cartwright&Smith,1986),pigeons (Cheng, 1988) and humans (Spetch et al., 1997) take the distance intoaccount:e.g.,theysearchinaconstantdistancefromthelandmarkevenwhenthesize of the landmark is changed.When several landmarks arepresent,humansand other animals consider the configuration of landmarks in order to find alocation (e.g.Collett et al., 1986;Cheng, 1988). For example if one landmark is


29

shifted,thesearchpositionusuallyliessomewherebetweenthelocationindicatedbythemoved landmarkandthe location indicatedbythestatic landmarks.Bothhumans and nonhumans have been shown to rely on landmarks in closeproximitytoagoallocation(Bennett,1993;Cheng,1989;Cheng,Collett,Pickard&Wehner,1987;Spetch,1995;Spetch&Wilkie,1994).Whilerodentsareabletouseinformation about landmark identity, theymay principally rely on informationabout the geometric properties of the landmark arrangement (Benhamou &Poucet, 1998; Collett et al., 1986; Greene & Cook, 1997; Maurer & Derivaz,2001).Birdscanusebothdistanceanddirectionalinformationtofindaplaceagain,however, they seem to focus more strongly on the direction towards thelandmarks thanon thedistance (Cheng,1994;Kamil& Jones,2000).Contrary tothat, humans in an immersive virtual environment seem to focus more on(relative) distance than on the angle between different landmarks, unless theanglesconsistofrightorstraightangles(Waller,Loomis,Golledge&Beall,2000).This isdespite the fact that fewerdistance cuesareavailable in suchanvirtualenvironment than in a real environment. Theoretically, imprecise distanceinformationresultsinasmallerareaofpossiblelocationsthanimprecisedirectioninformation (Walleretal.,2000).As imprecision indistanceestimation increases(roughlylinearly)withthedistancetobeestimated(forareview,seeWiest&Bell,1985), focusing on closer landmarks should lead to better performance. Fordirectioninformationthisisnotthecase.

Contrary to absolute distances, only adult humans seem to consider relativedistances.They search in themiddle (oranother constant ratio)between twoormore landmarks when the landmarks are moved further apart. Children,monkeys, rodentsandpigeons typically search in two locationsabout the samedistancetoeachlandmark(Collettetal.,1986;MacDonald,Spetch,Kelly&Cheng,2004; Spetch et al., 1997).However, some birds are able to learn to search at alocation that is identified by an equal distance or an equal angle towards twolandmarksatleastaslongasthetestdistancesbetweenthelandmarksarewithinthedistancesexperiencedintraining(Kamil&Jones,1997;Spetch,Rust,Kamil&Jones,2003).

Pigeons also seem to consider only one or two landmarks when multiplelandmarksareavailable(Spetch&Mondloch,1993),whereashumansinadesktopvirtualenvironmentseem to focuson theconfigurationofmorethan just twoorthreelandmarks:contrarytoremovingsomelandmarks,removingalllandmarks(Jacobsetal.,1998),removingallbutonelandmark(Chamizo,AznarCasanova&Artigas, 2003), or changing their configuration disrupts human performance(Jacobs, et al., 1998).The ability to orient ondistal landmarks and on relationsbetween landmarksdevelopsduringmaturation (Laurance,Learmonth,Nadel&Jacobs,2004).

Bothbirdsandhumansinimmersivevirtualenvironmentsrelyheavilyonthelinedefined by two landmarks, i.e., they aremore precisewhen finding a location

CHAPTER 2 THEORY

30

locatedonsuchaline(Walleretal.,2000;Spetchetal.,2003).Thisinformationcanconstrainthepossiblelocationsoftheviewpointtolieononesideofthelinethatconnects the landmarks (Levitt&Lawton1990;Thompson,Valiquette,Bennet&Sutherland, 2000). In an immersive virtual environment this qualitativeinformationcanalsooverridedistanceinformationusuallyused.Alocationwithinthreelandmarksisnotsearchedforoutsideoftheselandmarksinadifferenttestenvironment,evenwhenthelocationcorrespondingtothesamerelativedistancesasduringtrainingliesoutsidethisenclosure(Walleretal.,2000).Hereorientationonqualitativemightoverridequantitativeinformation.Suchadditionalcategoricalencoding isalso indicated inspecificbiases towards landmarks (Fitting,Allen&Wedell, inpress) aswell as in recalling locations ofdots on a computer screen(e.g.,Huttenlocher,HedgesandDuncan,1991).

Both in nonhuman animal and human desktop virtual reality studies, thelandmarks had to be visible during learning the goal location. Landmarksoccludedduring learning,butvisiblewhilenotnavigating to thegoalwere lesshelpful (Hamilton, Driscoll & Sutherland, 2002; Sutherland, Chew, Baker &Linggard, 1987). This speaks against a spontaneous integration of landmarkssuccessivelyvisibleintwovistaspacesintoonecoherentvistaspace(cf.alsoSturz,Bodily&Katz,2006).

To sum up, the available data suggest that humans and animals usemultiplemechanisms for reorientingand findingplaces inavistaspaces.Ratshavebeenfound to switch between two mechanisms even within one trial (Hamilton,Rosenfelt&Whitshaw,2004).

Reorientation on geometry and landmarks. When both geometric cues andlandmarkscanbeused,reorientationfocusesongeometryratherthanlandmarksbothinrats(Benhamou&Poucet,1998;Pearce,WardRobinson,Good,Fussell&Aydin,2001;Weisendetal.,1995)andhumanchildren(Gouteux&Spelke,2001;Hermer&Spelke,1996).WhenaMorriswatermazeisshiftedwithinaroomwithlandmarksonthewalls,ratssearchatthesamelocationinthepoolnotatthesamelocationdefinedbythelandmarks(Weisendetal.,1995).Childrenreorientinginarectangular room do not use landmarks to disambiguate between the twogeometrically identical corners of the room (Hermer& Spelke, 1996).Childrenreorientinaroomwithadistinctivegeometry,butnotonan identicalgeometricfigurebuiltbyidenticallandmarks(Gouteux&Spelke,2001).

Geometric cues are not overshadowed. The term overshadowing refers to thefindingthatthepresenceofasecondrelevantcuewillcauseanimalstolearnlessabout a first than theywould havedone if trained on the first cue in isolation(Kamin, 1969; Pavlov, 1927). In animals landmarks overshadow landmarklearning,butnotgeometry(Brown,Yang,&DiGian,2003;Hayward,McGregor,Good,&Pearce,2003;Pearceetal.,2003;Pearceetal.,2001;Wall,Botly,Black,&Shettleworth,2004).Insomecaseslandmarkovershadowingmaybeexplainedby


31

animalsonlyencoding someof the landmarkswhich results ina loweraverageperformanceonalllandmarks(cf.Spetch&Mondloch,1993).

Hippocampal place cells. Evidence for place learning is not only found inbehavioural experiments, but also in electrophysiological single cell recordingsespeciallyinrodents(e.g.,OKeefe&Nadel,1978),butalsoinhumans(Ekstrometal., 2003). In these studiesvarious electrodeshavebeenplaced inbrain regionssuch as thehippocampus or theparahippocampus.These electrodes record theactivity of a single or a few neurons while e.g. the rat is actively navigatingthroughanenvironment.Intheratshippocampus,socalledplacecellshavebeenfound(e.g.OKeefe&Dostrovsky,1971).Placecellsshowincreasedactivitywhenthe rat is located at a specific area of the experimental space nomatterwhichdirection it is facing.Alsoanother typeofneuronssocalledheaddirectioncellshave been identified in several areas of the brains limbic system including thehippocampus (e.g.Wiener& Taube, 2005; Taube,Muller&Rank, 1990). Theseheaddirectioncellsshow increasedactivitywhenever the rat is facingaspecificdirectionnomatterwhereintheexperimentalareaitislocated.

The activity of place cells directly represents the location with respect to theimmediateenvironmentdefinede.g.bygeometryand landmarks.Placecells fireat a constant distance to the nearestwallswhen the shape and the size of arectangular room is changed (OKeefe & Burgess, 1996). When a landmark isrotatedaround inacircularsquareby90aplacecell isactiveat thesamearearelative to the new position of the landmark, i.e., the active area is rotatedtogetherwith the landmark by 90 (Muller&Kubie, 1987).Hippocampalplacecells have been regarded as a neural correlate of ametric cognitivemap (e.g.OKeefe & Nadel, 1978). However, this only holds true for vista spaces. Inenvironmental spaces the sameplacecellscanbeactive indifferentpartsof theenvironmentwhichisthecaseforabout30%ofallplacecells(Thompson&Best,1989).Whenplace cells identify a specific area in an environmental space, thendifferent cell populations should be used for different locations (Trullier et al.,1997).Intwoidenticalroomsconnectedviaanalleyaplacecellcancodethesamerelative area, e.g. the northeast corner, but also different areas (Skaggs &McNaughton, 1998). If a place cellwould represent locations in environmentalspaces, cells should be found with firing areas separated by a wall. In thementionedexperimentnosuchcellshavebeenfound.Thesetwoobservationsareevidence against the hypothesis that place cells code a specific area in anenvironmental space and therefore function as a cognitive (metric) map forenvironmentalspaces.

However,hippocampalplacecellsdorepresentmetricrelationsinacertainvistaspace.They could, therefore,beused toencodean important location inavistaspacesuchastheplatformintheMorriswatertask.Inprincipalthisinformationcouldbeusedtoplanapathfromoneplacecellareatothenextuntilreachingthegoal.

CHAPTER 2 THEORY

32

Summary. Many studies have examined reorientation, e.g., in place learning.Humans and other animals are able to profit from geometric layout aswell aslandmarks in order to find a location again. Geometry is considered moreimportant than landmarks.Humans andmost other animals use distances andangles to find locationsagain.Languageeases reorientationongeometry,but isnotnecessary.Hippocampalplacecellscanbeseenasaneuralrepresentationofalocation inavista space.Fornonhumananimals reorientation seems to relyonfeatures of specific vista spaces. Contrary to that, humans seem to be able toreorientonmapsandongeometricstructureofanenvironmentalspace.Thiswillbediscussedin2.3.3.1.

2.3.1.3 Updating by path integration

Cues used for updating.Wecanupdateourpositioninspacebypathintegrationorbyreorientation.Pathintegrationisanorientationprocessinwhichselfmotionisintegrated over time to obtain an estimate of ones current position (Loomis,Klatzky&Golledge,2001).Incontrasttoreorientation,path integrationdoesnotinvolve the recognition of external features such as geometry or landmarks(Montello, 2005). No internal or external long term representation of anenvironment is needed. Instead, in path integration sensory inputs indicatinglocomotionareintegratedovertimetokeeptrackofoneormorelocationsintheenvironment.Generally,workingmemoryisseenassufficienttodothat,withoutthe need for longterm memory (e.g., Mallot, 1999). Sensory inputs used forupdating includeexternalcuessuchasoptic flow from theeye (e.g.,Riecke,vanVeen&Blthoff,2002)andaudition(Loomis,Klatzky,Philbeck&Golledge,1998),butmainlyinertialcuesarereferredtowhentalkingaboutpathintegration(e.g.,Loomis,etal.,1993).Inertialcuescomprisevelocityandaccelerationsignalsfromthevestibular system,proprioceptive feedback from skin, joints,andmusclesaswellasefferencecopiestothe limbs centrally initiatedneuralcommandstothemusculature.There is,however,noevidence thatefferencecopiesplaya role inwholebody locomotion (Montello, 2005). In nonhuman animals also compassinformationbasedonskylight(polarization)patterns(Wehner,1994)andbasedonamagnetic sense areused todetermine turning angles (e.g.Kimichi,Etienne&Terkel, 2004). In humans magnetic sensing has been proposed as a source ofinformation, too (Baker, 1980). However, this claim lacks any clear evidence(Montello, 2005). In humans distant landmarks and slant can also providecompassinformation(cf.Restat,Steck,Mochnatzki&Mallot,2004).

An impressive example forwhat can be accomplishedwith path integration isgivenby thedesertant (Cataglyphis).Theseants leave theirnestand take longandcircuitousexplorationsforfood.Whentheyfindfood,theydirectlywalkbacktotheirnest(e.g.Wehner&Wehner,1990).Antsarethoughttocomputetheirnetdistance and direction from the nest throughout their outward and return


33

journeysandsocanalwaysreturndirectlyhomefromtheircurrentlocation.Thiscanbedescribedbyahomingvectorwhichisupdatedovertheentirejourney.Ifahomewardboundantispassivelycarriedindarknesstoanewlocation,itmoveson aparallelpath for the appropriatedistance (Wehner& Srinivasan, 1981). Inordertocomputethisvectorantsuseaskylightcompasstoestimatetheirturningangles(forareviewseeWehner,1997)andcounttheirstepstoestimatedistances(Wittlinger,Wehner&Wolf,2006).Whenhomewardboundantsareplaced inajar and allowed to continue their journey after a variabledelay, their ability tofollowtheappropriatehomingvectorcoursevanishesafterafewdays(Ziegler&Wehner,1997).

Likeother animals,humans can return towards theoriginof apathusingonlyinertial (vestibularandsomatosersory)cues.However,evenovershortdistancesbelow 30meters this path integration is quite inaccurate (Loomis et al., 1993).Physical turning isrequired forcorrectpath integration. Imagined turning,opticfloworwatchinganotherpersonturningisnotsufficient(Klatzky,Loomis,Beall,Chance & Golledge, 1998, Riecke & Wiener, in press). Nevertheless, pathintegrationcanbedonebasedonopticalflowonly(Rieckeetal.,2002).However,whenbothopticalflowandinertialcuesareavailable,inertialcuesandespeciallyproprioception seem todominate (Bakker,Erkhoven&Passenier, 1999;Kearns,Warren,Duchon&Tarr,2002).However,activeversuspassivemovementdoesnotmatter inmanycircumstances(Klatzky,etal.,1998;Wraga,CreemRegehr&Proffitt,2004).

Updating processes. As mentioned, path integration is more difficult duringimagined movement compared to physical movement especially for rotations(Klatzky et al., 1998; Rieser, Guth & Hill, 1986; May, 2004). This fact can beexplainedbydifferentprocesses.Thenecessarytransformationcouldbefacilitatedbyphysicalmotion(cf.Farell&Robertson,1998).Alternatively,interferencecouldoccur from a conflict between the awareness of ones physical position in anenvironment and thediscrepantpositiononehas to adopt in imagination (May1996; 2004). This interference theory is supported by results showingdisorientation to improve the performance in imagined rotations (May, 1996).Merefacilitationofthenecessarytransformationbyphysicallocomotioncouldnotexplain this. The interference theory can also be applied to navigating virtualenvironments.Heretherealandvirtualworldcaninterferetostrongerorsmallerextentsdependingonthequalityofthevirtualrealitysetup,e.g.,desktopversusimmersive setups, the field of view, etc. (e.g.,Riecke,Cunningham& Blthoff,2006;SchultePelkum&Riecke,inpress).

Updatingbypath integrationwhenmovingphysicallyoftenseems tohappen inanautomaticmanner:Participantscanupdateveryaccuratelyandeasily,withoutanyawarenessofhavingtothinkaboutthetask(Rieseretal.1986;Rieser,1989).Participantsareevenunable tovoluntarily refrain fromupdatingwhenmovingphysically (Farell&Robertson, 1998; Farell& Thomson, 1998;May&Klatzky,

CHAPTER 2 THEORY

34

2000; but see Waller, Montello, Richardson & Hegarty, 2002). However, forimagined environmentswe are able to refrain fromupdating, even ifwemovephysically(Wang,2004).

What is updated in path integration. Single locations can be updated by pathintegration,e.g., thestartofaroute inahoming task (Loomis,etal.,1993).Alsomultiplelocationsareupdatedsuchasthecornersofaroomoranarrayofobjects(e.g.,Wang&Spelke,2000;Holmes&Scholl,2005).Notonlyenvironmentsseenbefore can be updated by path integration, but also unknown environmentsdescribed by language (Avraamides, 2003; Loomis, Lippa,Klatzky&Golledge,2002),oranobjectarrayona tableexploredhaptically (Pasqualotto,Finucane&Newell,2005).Familiarenvironmentalspacessuchasauniversitycampuscanalsobeupdatedduringphysicalrotationsand imagined translations (Easton&Sholl,1995;Holmes&Sholl,2005).

Updating on longer routes and storing routes in memory. Due to its integrativenature,pathintegrationisnecessarilypronetotheaccumulationofrandomerrors(Wehner,1999),especiallywhenonlyrelyingoninertialcues(Benhamou,Sauv&Bovet,1990).This isempiricallyshown inantswhodeterminetheoriginoftheirjourney increasingly lessaccurately the farther theyhaveventuredout from thestart (Wehner&Wehner, 1986).Most studies on human path integration onlyconsidered routes shorter than about 30meters (e.g., Loomis et al., 1993). Forexploring environmental spaces such as houses or cities usually much largerdistanceshave tobecovered.Forsuchspaces,path integrationbasedon inertialcuesprobablyplaysonlyaminorrole,ifatall:correctinertialcuesasduringacarride do not enhance spatial knowledge compared to no orwrong inertial cueswhilewatchingavideoofthatride.However,fullfieldofviewandtheabilitytoturnonesheaddoenhanceonesspatialknowledge insuchsituations(Goldin&Thorndyke,1982;Waller,Loomis&Steck,2003).Estimatesof the time travelledmaybemoreimportantfortravellinglongerdistancesthanpathintegration.

Intheprevioustext,updatingreferredtokeepingtrackofonespositionrelativeto a location or an environment. This can be explained by working memory.However, also the velocity profile of a translation can be stored in memory(Berthoz, Israel, GeorgesFranois, Grasso & Tsuzuku, 1995). In trianglecompletionexperimentsparticipantsalsoseemtomaintainahistoryoftheroutestravelled: latencies increased formore complex trajectorieswhich shouldnotbefoundwhenonlystoringahomingvector(Loomisetal.,1993).Evenantsandbeesstore information aboutwhere a food source is to be found (Collett,Collett&Wehner,1999;Srinivasan,Zhang,Lehrer&Collett,1996).Forlongtermstorageofpathslearnedbyinertialcuestheencodingerrormodel(Fujita,Klatzky,Loomis&Golledge,1993)mightbeappropriate.Thismodelassumesthatpeopleencodethedistancesand turnsofa route travelled.Pointingorhomingerrorserrors reflectsystematicinaccuraciesintheencodingprocess,asparticipantscancomputeandexecute pointing movements and walking trajectories without any systematic


35

error.Itwasoriginallydevelopedtoexplainonlineupdatingbypathintegration,butwasnot supported in its original form (Klatzky,Beall,Loomis,Golledge&Philbeck,1999,Rieckeetal.,2002).Itmight,however,explainpointingtounseenlocations frommemory. To imagine standing at a location and use longtermmemoryofaroute tocompute thedirectionofanother location isanalternativesolution for many imagined updating studies (May & Klatzky, 2000; see alsoAmonrim, Glasauer, Corprinot & Berthoz, 1997; Avraamides, 2003). Twoalternative processes can, therefore, explain results from updating by pathintegration: First, online updating of one ormore locations in the environmentrelying only on working memory which happens obligatory during physicalmotion and, second, accessing trajectories stored in long term memory andimaginingthegoaltopointortowalkto.Inanabstractformthisknowledgecanbe described by a chain of vectors, which means not preserving the velocityprofile, but simply the angles anddistances from aposition.The latterprocessprobablyislimitedtohumanorientation,asisimaginedupdatingingeneral.

Summary. Path integration is an orientation process in which selfmotion isintegratedovertimetoobtainanestimateofonescurrentpositioninspace.Whenmoving physically, updating happens automatically and leads to betterperformance especiallywhen turns occur. Interference can explain the drop inperformance for imagined updating compared to physical updating aswell asproblemswith virtual environments.Onlineupdating bypath integration is anerrorproneworkingmemoryprocess.Inhumansaccurateupdating is limitedtorather short distances. Movement trajectories can also be stored in long termmemory.

2.3.1.4 Route navigation

Routenavigationisawayfindingprocess,aprocessenablingustoreachaknowngoal in an environmental space, by navigating a known route. The knowledgenecessary for route navigation is called route knowledge (see 2.4.2). It can belearned directly by navigating a route or indirectly via, e.g., maps or verbaldirections.Wewillproposeatheoreticalframeworkforroutenavigationincludingelementsandsubprocessesnecessaryforexecutingroutenavigation.

Route navigation as we understand it involves two parts: first, identifying alocation,andsecond,movingtowardsthegoal(althoughnotnecessarilydirectly).The latter involves selecting one among several possible directions to movetowards.Thesetwopartsformthebasicelement inroutenavigation.Inordertoreach a goal several of these basic elements have to be combined, i.e., severallocationshavetobeidentified,andadirectionhastobeselectedateachlocation.

Identifying a location. To reach a goal by route navigation a navigator has toidentify the start, the goal and several locations inbetweenwhere correct routedecisionshave tobechoosen.These locationscanbe identified invariousways.

CHAPTER 2 THEORY

36

For exampleone could identify the specific formof an intersection, recognise afamiliar landmark, e.g., a yellow house, or one could identify a location bycounting,e.g.thethirdintersectionasdescribedinroutedirectionsordisplayedinamap.Astheexamplesdemonstrate,forhumannavigatorssuchlocationsdonothavetobevisitedbefore,butcanbelearnedfromothersources.Ifalocationwasvisitedbeforeandisrecognisedduringroutenavigation,probablythesamecuesdescribed for reorientationplaya role, too (cf.2.3.1.2).These cues can compriseproximal anddistal cues (Steck&Mallot, 2000).Cueusagewilldependon cuesaliency.Acuewhichiscommonlyencounteredinanenvironmentalspaceisnotveryspecificforalocationalsoreferredtoasasnotverysalient.Suchacueisusedlessoftenthanasalientcuewhichencounteredlessoften(cf.Stankiewicz&Kalia,inpress).Fromatheoreticalpointofviewonlylocationshavetoberememberedwherealternativeroutechoicescouldoccur.Whenarouteonlygoesstraightonwithout any route alternatives, one does not have to think aboutwhere to go.Consequentlyparticipants arebetter in remembering information fromdecisionpoints: in familiar environments, on routes in virtual environments, routesdisplayed on amap, and routes presented via slides landmarks arementionedmore frequently and are recognised better when located at decision points(Aginsky, Harris, Rensink & Beusmans, 1997; Appleyard, 1969; Cohen &Schuepfer,1980;Janzen,2006;Lee,Tappe&Klippel,2002).

Routenavigationreliesondiscretelocations,notonacontinuousrepresentationofthe environment (Mallot, 1999;Trullier, et al., 1997).How thediscrete locationsemerge from a continuous perceptual input when navigating through anenvironment is largelyanopenquestion.We canhowever say somethingabouttheextensionofsuchalocation.Wewanttodefinealocationreferredtoinroutenavigationasanareathatanavigatorisalsoabletoreorientoneself.Definingsuchanareaastheareawhereanavigatorselectsthesameaction(Trullier,etal.,1997)is problematic, because it can lead to circular explanations. For example anavigatorturnsleftexactlyatthatareawhichisdefinedbythenavigatorturningleft.

When identifying a location during route navigation, a navigator is oriented,knowing where the last location visited is situated. This distinguishes routenavigation fromreorientationwhereonedoesnotknowones locationonamapor with respect to familiar locations. If a navigator gets lost during routenavigation he or she has to reorient before being able to navigate the routetowardsthegoalagain.Forshortdistances,recognizingafamiliarlocationmightbe substituted by path integration, e.g.,walking in darkness to the next room.However,thiswillbeanexception.

Directional information in route navigation.Afteridentifyingalocationthenavigatorhastoselectadirectiontomovefromthecurrentlocation,inordertoreachhisorher goal. This might include approaching a visible landmark, also calledbeaconing.OfferedastheonlypossibilitybyWangandSpelke(2002)tonavigate


37

from one location to another, beaconing might be as simple as following anextendedlandmarksuchasawall,butmostofteninvolveslocomotinginacertaindirection in relation to one or to several landmarks (guidance). Selecting adirection to move towards always involves moving away from the currentlocation.Thisdistinguishesroutenavigation fromreorientation,where findingalocation in vista space is usually themeasure for knowing ones positionwithrespect to theenvironment.Contrary to that,routenavigationpointsaway fromthe current location along a route leading towards a known goal that is notdirectlyaccessible.Thisroute,and thereforealso theselecteddirection,doesnothave to lead straight to thegoal.The route can involve a loop if, e.g.,nootherrouteispossibleorknown.

Fordeterming thedirection tomove towardsTrullierandcolleaguesandMallotandcolleaguesproposeanassociatedbehaviour,a triggeredresponse.Aspecifictriggeredresponsemight,however,notexplainallobservedbehaviour.Ratsswima route learnedbywalking (MacFarlane,1930).Catswalka route learnedwhilebeingpassivelycarriedalongtheroute(Hein&Held,1962).Wecancycleapathlearnedbywalkingor learned fromamap.Thedirection information, therefore,hastobemoreabstractthanaspecificbehavioursuchaswalkingorcycling.

Thedirection tomove towardswillalsonotbe egocentric ingeneral:often ratswalkintoaspecificcorridorinordertogetfood,evenwhentheylearnedtoturnleft inorder toget the food,but in the test trialhave to turnright (Restle,1957).Thedirection information, therefore, ismore likelyadirection in relation to thecurrent environment rather than a direction in relation to ones current bodyorientation,which typically is thecase forpath integration.5 In the followingwewould like to describe this direction information by a vector pointing into thedirection indicating where to move next. Note that this applies for routenavigationinfamiliarenvironmentsexperienceddirectly,whichisthecaseforallnonhuman animals. Humans also can navigate using maps and verbal routedirections.Hereleft/rightdecisionswithinanegocentricreferenceframeprobablyplayamoreprominentrole.

Combining basic elements. Locomoting into a direction at one specific locationnormallyisnotsufficienttoreachagoal.Ithastobedoneseveraltimesthebasicelementofidentifyingalocationandselectingadirectionhastoberecombinedtogeta chainof theseelements.Hereby thenavigator learnshow toget fromonelocationtoanother,e.g.,walkingthisstreetIwillreachthecityhallnext.Suchasequenceof identifyinga locationanddecidingwhere togodoesnothave tobespecific for a certain goal. Rats can latently learn the structure of theirenvironment,e.g.,learntheroutetoaroomwithoutgettingfoodinthatroom,b

Documents

Strategies of Orientation in Environmental Spaces