212_place recognition and wayfinding

Place Recognition and Wayfinding:Making Sense of Space

Reginald G. Golledge

ReprintUCTC No. 212

The University of CaliforniaTransportation CenterUniw;rsity of CaliforniaBerkeley, CA 94720

The University of CaliforniaTransportation Center

The University of CaliforniaTransportation Center (UCTC)is one of ten regional unitsmandated by Congress andestablished in Fall 1988 tosupport research, education,and training in surface trans-portation. The UC Centerserves federal Region IX andis supported by matchinggrants from the U.S. Depart-ment of Transportation, theCalifornia Department ofTransportation (Caltrans), andthe University.

Based on the BerkeleyCampus, UCTC draws uponexisting capabilities andresources of the Institutes ofTransportation Studies atBerkeley, Davis, Irvine, andLos Angeles; the Institute ofUrban and Regional Develop-ment at Berkeley; and severalacademic departments at theBerkeley, Davis, Irvine, andLos Angeles campuses.Faculty and students on otherUniversity of Californiacampuses may participate h~

Center activities. Researchersat other universities within theregion also have opportunitiesto collaborate with UC facultyon selected studies.

UCTCs educational andresearch programs are focusedon strategic planning forimproving metropolitanaccessibility, with emphasison the special conditions inRegion IX. Particular attentionis directed to strategies forusing transportation as aninstrument of economicdevelopment, while also ac-commodating to the regionspersistent expansion andwhile maintaining and enhanc-ing the quality of life there.

The Center distributes reportson its research in worldngpapers, monographs, and inreprints of published articles.It also publishes Access, amagazine presenting sum-maries of selected studies. Fora list of publications in print,write to the address below

University of Ca~ormaTransportation Center

108 Naval Architecture Bu/Id/ngBerkeley, California 94720Tel: 510/643-7378FAX: 510/643-5456

The contents of this report reflect the views of the author who is responsiblefor the facts and accuracy of the data presented herein. The contents do notnecessarily reflect the official views or policies of the State of California or theU.S. Department of Transportation. This report does not eot~slitute a standard,s~fie.~ion, or r~galafioa.

Place Recognition and Wayfinding:Making Sense of Space

Reginald G. GolledgeDepartment of Geography

University of California at Santa BarbaraSanta Barbara, CA 93106

Reprinted fromGeoforum

1992, Vol. 23, No. 2, pp. 199-214

UCTC No. 212

The University of California Transportation CenterUniversity of California at Berkeley

(:-eofo,,um. VoL 23. No. 2, pp. ]99-2|4. 1992~) 1992 Pergamon Press Ltd

PlaceRecognition and Wayfinding:Making Sense of Space

REGINALD G. GOLLEDGE,* Santa Barbara, CA, U.S.A.

Abstract: In this paper I examine processes involved in place recognition andwayfinding in the context of spatial knowledge acquisition generally. Recognizingplaces is seen to be of vital importance in developing a declarative base: wayfinding isviewed as the most common means of acquinng place knowledge. Characteristics ofplace recognition are examined along with discussion of errors in place cognition andthe role that spatial familiarity plays in attaching importance weights to distinguishprimary nodes (anchor points) from other places. Wayfinding is characterized route knowledge acquired via procedural rules. Parameters of wayfinding arediscussed in reference to navigation in familiar and unfamiliar environments. Theexpression of wayfinding in terms of computational process models is examined, andthe future role of geographic information systems in such modelling is explored in thepenultimate section.

Ilntroduction

Unless we are blind, or completely lost in a pitch-darknight, fog or a blizzard, or for the first time swimmingunder water in an unknown area, we know that byusing vision we can begin to make sense of our,;urroundings. We look for things that stand outbecause they are different from their surrounds, orbecause they have a shape or form or structure thatwe believe we could recognize again. If nothingcatches our attention, we create something--wescratch a mark on the sidewalk or wall, or build acairn or mound of dirt, anything than can represent tous a sense of location. Once established, this anchorsother information processed by our senses. Order canbegin replacing chaos. Things we sense now haveproperties of distance, direction, orientation, prox-imity~ linkage, and association, both with respect tospatial anchors and with regard to each other. We canbegin to classify, to cluster, to regionalize, and to

"Department of Geography, University of California,Santa Barbara~ CA 93106, U.S.A.

impose hierarchies. Where information is sparse, wecan create another anchor, establish a relation be-tween this and the initial one (e.g. by establishing path or base line), and can continue the process ofordering the mass of information bombarding oursenses. With such ordering comes security~ recog-nition capability, and, even when al/things appearstrange and difficult to identify, according to our well-established perceptual norms, we can at least identifyand use the environment in which we find ourselves.

My purpose in this paper is to discuss some facets ofthe process of acquiring spatial knowledge. Twoimportant components of this process, recognizingplaces and finding ones way between places, willbe emphasized. To achieve this purpose, I beginby examining different types of spatial knowledge.This is followed by a discussion of place location,recognition, linkage, and choice. The next sectionexamines wayfinding from its elemental spatia! per-spective, followed by discussion of route learning byhumans. The penultimate section examines the inte-gration of these components into a knowledge struc-ture, and briefly points to the role that environments

199

200

[represented as geographic information systems(GISs)] play in the knowledge acquisition process.

Types of Spatial Knowledge

Geoforum/Volume 23 Number 2/1992sequence will be undertaken in this paper. Conven-tion also assumes that there are a few different typesof spatial knowledge, usually referred to as land-mark, route, and survey. This classification is ofimportance here and will be discussed further.

It is usual to distinguish between declarative andprocedural knowledge. A persons declarativeknowledge consists of the inventory of pieces ofinformation contained in long-term memory. In thespatial domain, this consists of places (such as land-marks or nodes), lines (such as routes, edges, andboundaries), and areas (such as neighborhoods, dis-triers, cities, regions, or countries). Proceduralknowledge includes the rules for linking pieces ofinformation into ordered strings. In the spatialdomain this is usually taken to include rules of pathdefinition, segment concatenation, associations andrelations between points, lines, and areas (e.g. hier-archical ordering), and rules for wayfinding and navi-gating within a sensed or experienced spatial system.

But some researchers interested in spatial knowledgeargue that there is yet another component. Usuallycalled survey or configurational knowledge, this con-sists of an awareness of configurational properties orlayout characteristics of various types of spatial fea-tures. Such features are usually not directly sensed,but are inferred from spatial primitives and theirderivatives, and are used to infuse meaning into anenvironment while experiencing it or while think-ing about an experience. Thus configurationalunderstanding is achieved by integrating piecemealinformation into a comprehensive spatial knowledgesystem. The existence of this type of knowledge isinferred because of the apparent failure to date oftheories based exclusively on declarative and pro-cedural components to account for the nature anddevelopment of spatial knowledge.

Behavioral geography and environmental psychologyhave conventionally assumed that, in the process ofacquiring spatial knowledge, individuals pass throughstages: from egocentric to allocentric frames of refer-ence and from topological to fully metric comprehen-sion of space (HART and MOORE, 1973; PIAGETand INHELDER, 1967). Critical evaluations of thishypothesis exist [e.g. LIBEN (1982), GARLING andGOLLEDGE (1989), and GARLING and EVANS(1991)], and no further elaboration of this inferred

Place knowledge, often referred to by the Lynchianterm landmark knowledge, consists of lists of fea-tures perceived to exist in both the natural and builtenvironments. Such features may be mountains,rivers, trees, beaches, buildings, roads, recreationalareas, and so on. Attached to each feature is a stringof perceived attributes, including things such as loca-tion, size or magnitude, identity, time, colour,uniqueness, function, and so on. While Lynch orig-inally used the terra landmark to differentiate thefeatures which had outstanding characteristics (suchas dominance of visible form or unique size, shape,color, or functional significance), the term hasdegraded and is now applied generically to knownplaces. New terms, including reference node andchoice point, are now applied to those places whosesignificance is such that many people use them toanchor directions or to act as loci for wayfinding orthe regionalizing of information (SADALLA et al.,1980; GOLLEDGE, 1990; G,~RLING et al., 1986).Anchors, landmarks, reference nodes, choice points,and other bits and pieces of place knowledge areabsorbed and organized as different environmentsare experienced.

Most of the information we collect about any givenenvironment is obtained by travelling through it. Wemay travet through it physically, as by following apath between an origin and a destination in objectivereality. We may also travel through an environmentby reading a book, by listening to a verbal descrip-tion, or from viewing image records such as slides,tape, or live television transmission. We may alsoacquire not just piecemeal but layout information byexamining models, maps, areal photos, satellite im-ages, or by simply looking out the window of anairplane as we pass over any given environment. Inthese latter cases, we appear to do more than simplyrecord information about a long list of differentplaces. We link information in some way, by some setof procedural rules o Such rules allow a person todevelop what KUIPERS (1977) calls a commonsense" understanding of an environment. Knowingthat A is linked to C through B allows one to under-

Geoforum/Volume 23 Number 2/1992

stand how spatial movement can take place and planfor ,;uch activity.

Recognizing Places

In the spatial domain recognizing a place meansbeing able to identify its location. In addition tolocation, occurrences found at particular places haveother characteristics, including a name or identity,physical features such as color, shape, size, and so on,a temporal life or episodic interval at which an occur-rence occupies a location, and a magnitude ormea,;urement of how much of the occurrence is foundat that place. Thus, although place is a dimensionlessspatial term, conventionally it is interpreted as amultidimensional phenomenon. Places which areeasily identified are said to be familiar, but, asGALE et al. (1990) and PERON et al. (1990) argue,familiarity itself is a complex multidimensional con-cept Some people claim familiarity with a place evenwhen they only know its name. Others claim famili-arity if they have observed, visited, or passed by theplace frequently. Yet others claim familiarity becausethey ,can identify an image of it.

Some: features stand out from their surroundings.Such places are often regarded as being familiar orwell known by a significant number of people. Im-portance accrues to the place because of this commonknowledge. Often designated as landmarks, theseplaces usually provide a significant part of both indi:viidual and common cognitive maps of the environ-rrtent in which they are found. When speaking toothers of ones knowledge of a place, these landmarksare referenced most frequently. When directing astranger to a specific location, a global or super-ordinate frame of such landmarks dominates generaldirectional and orientation information beforespecific information found in the immediate neigh-borhood of a destination is identified.

Given that places can be identified and their featureslearned, how can such information be stored andused? In answering this question I do not attempt tosummarize the neuro-bioiogical or genetic codingliterature that speculates on how the brain physicallystores information, but instead turn to the literatureof geography, geodesy, surveying, and cartographyfor guidance.

Recording Spatial Information201

Places can be located using global or local referencingsystems. Since the development of the chronometerand the acceptance of a spherical earth, the mostwidely accepted global referencing system is via lati-tude and longitude. In everyday life few use thissystem or are indeed aware of its essential properties.It is often replaced by a less exact global system that isbased on cardinal compass directions. North, south,east, and west are convenient labels to define theedges of a grid that can be superimposed over anysurface. Global latitude and longitude measurementscan be replaced by more local grid coordinate systemsthat can be as coarse or as precise as desired. Inaddition to the coordinate structure, however, adirectional component gives a finer existencemeasure. Having established a location, one coulddescribe another location as being 25 miles south-east of the first. This establishes a distance anddirection so that a vector can be constructed from thefirst point to the second [Figure l(a)]. Using odometer and a compass, the location of the secondpoint can then be determined. The position of the twoplaces are established with respect to each other andwith respect to the superordinate grid. The inter-vening distance is often described as the crow-flydistance and may depart considerably from the over-the-road distance, which is the total number ofground units covered during the journey from theorigin to the destination.

Another way of locating a place is to use an offsetmeasure. Assume a possible destination or place ofinterest is at a location not before visited. Assumefurther that two landmarks bounding the segment ofspace containing the desired destination are known.The unknown destination can be located by movingalong a path joining the two landmarks until it ispossible to take an offset from the main path; thedistance of the desired offset and the angle at which itdeparts from the main route can establish the locationof the desired place [Figure l(b)].

Triangulation is yet another method for establishingthe location of a place. A fisherman who finds aproductive reef within sight of land may take mechan-ical or perceptual bearings on three prominentobjects. The intersection of the back bearings fromthe objects can establish location. Similarly, if one

202

(a)GeoforurrffVotume 23 Number 2/1992

ANGLE - DISTANCE

A BA B

ANGLE / OPPOSITE SIDE

P

a bA

RESECTION

(B) OFFSETq

F

DISTANCE INTERSECTION

Figure 1. (A) Directional and distance techniques for locating features---direction.(B) directional and distance techniques for locating features---distance.

wishes to find the location of a destination that isknown but not seen, and one has no technical assist-ance, one can replace the precision of the compasswith a less exact but convenient method called pro-jective convergence. Here one uses a cognitive mapof the environment, by imagining being located ateach of three places in turn and pointing at the desireddestination. The triangle of error produced by theintersection of the three pointing vectors gives anapproximate location for the destination. If the point-ing error is large, and the resulting error triangle isalso large, calculating the triangles mean areal centerprovides a feasible first approximation for the poss-ible location of the destination.

If a destination is out of sight but the distance to itfrom two or more locations is known, a procedurecalled trilateration can be used to locate the place.The more accurate the distances, the more precise thelocational estimate. In objective reality lasermeasuring devices allow extremely accurate locationwith only two locations as long as they are not closelylinearly aligned with the possible destination. In thecognitive domain, encoding of both distance anddirection are subject to error and it is unwise to useless than three points to try to establish an unknownplaces location.

In each of the above cases, I have identified a meansfor locating a place. In the world of objective reality,precise instruments have been developed to ensurethat places can be located with great accuracy (e.g.global positioning systems). In the cognitive domainwe must rely on more primitive human abilities.These are the abilities to estimate or reproduce adistance, and to estimate or reproduce an angle, or touse both distance and angle within a particular frameof reference or context. Without a frame of refer-ence, distances and directions can be extremely error-prone. With a good frame of reference, such errorscan be substantially reduced. In either the physical orthe cognitive domain, places are located using funda-mental spatial concepts--location, the interval be-tween locations or distance, and the bearing of onelocation from another within the superimposed frameof reference.

Example 1: the Errors Associated with LocatingPlaces Using Distances

The data used in this study were obtained from anexperiment performed by RICHARDSON (1981).Five subjects were asked to make paired comparisonjudgments about 19 location cues in a 2~ x 1 mile

Geoforum/Volume 23 Number 2/1992Table 1. Average total familiarity, Santa Barbara study*

AverageCue: familiarity

State Street and Highway 101 8.4in,tersection

La Cumbre Plaza 7.7State Street and Cabrillo Boulevard 7.5

intersectionState Street and Mission Street 7.3

intersectionSanta Barbara Airport 7.3Mission 7.0Picadilly Square 7.0Arlington Theatre 7.0Fairview Avenue and Hollister Avenue 6.8

intersectionHollister Avenue and Storke Road 6.7

intersectionGoleta Beach 6.7Fairview Shopping Center 6.6Magic Lantern Theater 6.2Cou~xy Court House 6.2Robinsons Department Store 6.2Botanical Gardens 6.2Santa Barbara Harbor 5.9East Beach 5.8Biltmore Hotel 55Rob Gym at UCSB 5.5lsla Vista Beach 5.5Museum of Natural History 5.4Bank of America, Isla Vista 5.4Isla Vista Market 5.3Ledbetter Beach 5.2Childs Estate 5.2Administration Building at UCSB 5.1

* Source: compiled from data collected by author

neighborhood in Goleta, California. The cues were"selected on the basis of extensive questionnaireswhich were circulated throughout the neighborhood,and they included places such as banks, shops, res-taurants, theaters, and so on. The procedure usedwas the same as that previously used by GOL-LEDGE (1974). Subjects were first asked to assign scale score of 9 to the pair or pairs of cues that wereconceived to be the farthest apart, and a score of i tothe pair or pairs thought to be the closest together.The remaining pairs of cues were then scaled accord-ingly between 1 and 9. In addition, subjects wereinstructed to indicate how familiar they were witheach place. This was also done on a nine-point scale,with l representing no knowledge or familiarity, and9 indicating that the place was very well known (Table1).

Having obtained distance estimates, TOBLERs

2O3

(1978) trilateration procedure, TRILAT, was chosenas a method for obtaining configurations from thesesubjective proximities. Toblers original work re-mains an unpublished manuscript, but there exists abrief description of the algorithm in GOLLEDGEand RUSHTON (1972, ppo 14-17). In its automatedform, the trilateration procedure is a two-dimensional metric algorithm that begins with amatrix, D, of proximities, dq, between n points.Given such a matrix, the problem is to find a configur-ation of the n points defined by x, y coordinates suchthat Y.(de - d~)2 (where d3 represents the distancescalculated from these coordinates). GOLLEDGEand RUSHTON (1972) outline the iterative pro-cedure for obtaining a configuration.

Another measure of some interest concerns themetricity of the space represented by the distanceestimates. One of the properties of a metric space isthat of triangular inequality, i.e. for any triplet ofpoints i, ], k, dij + d]k

2O4

familiarity and error. Specifically, both distortion andfuzziness were less for the configurations based onbest known places This was particularly evident withdistortion, supported by all cases studied. Similarsuccess was achieved with fuzziness. These results areconsistent with the hypothesis that the locations ofbetter known cues are more accurately and moreprecisely fixed in the cognitive images of the testsubjects, thus lending empirical support to the anchorpoint theory of spatial cognition. However, thesample that was used is a small one, only one urbanarea was tested, and the method of extractingcue familiarity from the subjects may not have pro-vided the necessary detail for a complete test of thetheory.

Example 2: Problems of Locating Places UsingAngles

Establishing the location of a place by pointing pro-cedures bypasses the errors associated with distanceestimation, but is subject to the errors of angle esti-mation. In studies with blind and blindfolded sightedindividuals, WORCHEL (1951) and KLATZKY etal. (1990) have shown that, while people can learn particular turn angle and reproduce it with consider-able accuracy if both the learning and reproductionphase are restricted to rotation of the body, pointingfrom memory produced errors of up to 35 for avariety of different angles. SIEGEL (1981) andMOAR and CARLETON (1982) have shown that,when people are quite familiar with an environment,their pointing accuracy increases enough to allowprojective convergence to give an approximate loca-tion for the object being pointed to. This lattertechnique, however, requires being able to pointfrom a number of different physical locations. Whenasked to imagine that one was at a number of differentlocations and to draw a vector from each to an unseendestination, the pointing error appears to increase.


cedures into a configurational knowledge structure.The evidence in support of this contention is notstrong and has generally failed to thoroughly examinethe effect that the pointing error, or the errors in-volved in reproducing directions, can contribute toconfigurational distortion. Much of the error alreadydiscovered in cognitive maps has not been decom-posed into appropriate components. Thus, whilelocational displacement can be obtained from tasksinvolving sketch mapping, toy play, or multi-dimensional scaling analysis of paired proximities, itis not clear as to whether the error so made is a resultof incorrect distance estimation, incorrect directionestimation, or a combination of both.

Recent experiments using two partially overlappingroutes in a 1 mile 1 mile square area of SantaBarbara, California (GOLLEDGE etaI., 1991), haveproduced some interesting results as far as pointingerror is concerned. In these experiments two partiallyoverlapping routes of different configurations andapproximately the same length were learned on asequence of unidirectional and bidirectional trials.Subjects were given two pointing experiments toconclude a set of tasks [the total experiments andtasks involved are described elsewhere (GOL-LEDGE etal., 1991)]. Of importance here is theresult of a particular pointing task. In one task, aftersubjects had learned a route through a relativelyunfamiliar environment via exposure to slide se-quences, they were shown a series of slides, with onebeing described as an origin, and the other beingdescribed as a place that had to be pointed to. Thepointing task was completed on a standard sheet ofpaper which was said to be oriented in the directionshown by the origin slide. The individual then drew avector on the sheet of paper equivalent to the direc-tion of the second slide from the origin. In the case ofpointing to places on the same route as the origin,some considerable success was achieved, althougherrors were still usually in the range of 10--25.

Pointing is the externalization of cognized directions.In cognitive science theory generally, it is assumedthat, given a declarative knowledge base, and a set ofrules for linking bits of that knowledge base to form aseries of routes, the opportunity presents itself forrecovering a representation of the spatial layout of allthe places experienced by learning individual routes.This assumes an ability to integrate lists and pro-

To test the suggestion that layout information couldbe recovered by integrating knowledge learned onseparate routes, a second task was developed inwhich the subjects were again shown a pair of slidesand were asked to imagine that they were standing atthe site of and facing in the same direction as the firstslide. This time a second slide showed a point on theother route learned in the same environment. Very

Geoforum/Volume 23 Number 2/1992little success was obtained on this task. Errors rangedfrom as little as 10" to as much as 170. Certainly therewas little evidence that integration of the two separ-ately learned routes had been achieved, or, if someintegration had been achieved, that this resulted in areasonable understanding of the spatial layout of thetask environment. One inference to be drawn fromLhis might be that integration of route knowledgealone; does not necessarily produce the type of layoutknowledge that is commonly assumed to exist at thesurvey stage. It also very obviously raises the questionof what survey knowledge really is, as well as sugges-ting that our knowledge of route learning, if theproce.dures involved in route learning can be used togenerate configurational understanding, is not wellknown.

Recognizing Places: the Concept of Spatial Fam-iliarlW

What do we mean when we say we are able torecognize a place, or that we know it well? Such astate is usually taken to mean that we are quitefamiliar with a place. Familiarity, in this case, is acatch..all term. It may include both spatial and non-spatial components. For example, GALE et aL(1.990) have suggested there are at least four possibledimensions of familiarity when used in this sense. Thefirst of these indicates an ability to identify a place byrecognizing its name or label. Most Americans wouldsay they are quite familiar with the Statue of Libertyby name~ just as most French citizens would adopt thesame position with respect to the Eiffel Tower.Knowing a places name, however, carries with it nospatial identity. A second dimension of familiaritymight be an ability to recognize a place when shownart image of it. Again, using the previous examples,pictures of the Statue of Liberty and the Eiffel Towercould be readily identified. Such pictures need nolocational reference nor background information toassist in such recognition. They stand as images intheir own right, distinct from the environments inwhich they stand. Being familiar with a place byknowing where it is represents yet another dimensionof this multidimensional concept. Knowledge ofwhere a cue is can be absolute (e.g. in terms ofknowing its coordinates in a global coordinate sys-tem) or relative (e.g. knowing where it is in relation other known places).

205A fourth dimension of familiarity is interaction fre-quency. Many places are seen or visited on a regularbasis and become integrated into a local knowledgestructure. In some cases neither the name nor thefunction of a place may be known but, because offrequency of exposure when travelling a weU-knownroute, or by knowing the approximate distance ordirection of a place from other known places, somecombination of visual recognition and locationalpositioning can be achieved.

At the individual level, the dimensions of identity,visual recognition, and iocational accuracy are highlycoUinear (GALE et al., 19913). It appears that visi-tation frequency is a reasonable independent indi-cator of spatial familiarity (as distinct from the moregeneric concept of familiarity). Spatial familiaritythus implies an ability not only to recognize andlocate phenomena, but an ability to relate phenom-ena to other places contained in a spatial knowledgestructure. Locational errors are less for the mosthighly familiar places. Pointing to highly familiarplaces apparently produces less error than pointing toless familiar ones. Just as familiarity of a placeappears to influence the accuracy of spatial judg-ments with respect to that place (e.g. interpointdistance judgments, interpoint directional judg-ments, spatial sequencing, and spatial linkage) so toodoes it influence the type and amount of errorattached to a place upon its inclusion in a cognitivemap or internal representation of a layout. The con-cept is obviously a significant one to include in theor-ies of cognitive mapping and spatial cognition, but,like other concepts discussed in this paper, it issubject as yet to error which is not as yet well speci-fied.

A fundamental axiom of geography is that no twodiscrete things can occupy the same position in spaceat the same moment in time. By necessity, things arelocationally separated, and linking places often re-quires moving between them or otherwise connectingthem. Wayfinding is such a process. It consists ofprocedures for searching an environment to find apath that can link an origin and a destination. Navi-gation is the process of choosing headings, and defin-ing the set of angles, path segments, and speeds ofmovement needed to ]ocomote over a path. A part ofnavigation is the designation of choice points where

206changes of direction or speed are desirable. This needlinks place recognition and wayfinding: the former isneeded to identify origins, destinations, choicepoints, check points, or general layout information,while the latter is required to specify the path of travelfrom among all possible connecting links. Althoughone can differentiate the two processes, they arefrequently used synonymously, though navigation isoften more focused on the task of reading instrumentsfor guiding a vehicle over a route, while wayfindingfrequently refers to the individual movement of livingorganisms. Recently, research on robotic and auto-mated vehicle guidance has also been couched inwayfinding terms, as has research on computationalprocess models of movement.

Wayfinding

Information about environments usually accumulateswhile in the process of experiencing them. For themost part, experience implies travel. Regardless ofthe type of motive power used, travel invariablyinvolves following a route between a specific originand destination. Much recent effort has gone intobuilding computer models to guide robotic move-ment or to act as navigational aids for automaticvehicle guidance systems. In both cases, the claim ismade that the system being developed is similar tothat used in human wayfinding. While the desire hasbeen there, most wayfinding and navigational devicesuse complex solution algorithms which do not matchhuman wayfinding capabilities or procedures. Cri-teria such as shortest path or other network-basedoptimal routing models dominate this literature, mostof them requiring an a priori map of the existingenvironment, including the existing routes serving it.Computers examine the map and quickly and expedi-tiously find a route that satisfies particular maximiz-ing or minimizing criteria built into the relevant pathselection algorithm. Given the criteria driving thealgorithm, mistakes are rare and optimal pathsusually are defined.

But this is not how humans find their way throughknown or unknown environments. As GARLING etal. (1987) point out, wayfinders are not generally leasteffort, short path, or distance minimizers. Given thisknowledge, the problem examined in this section


concerns the processing of information by humans toallow wayfinding to take place.

The Basic Elements

Finding ones way through an environment impliespurposive behavior. In a spatial sense, this requiresdefinition of an origin, the ability to recognize adestination when confronted with it, and the ability tostring together path segments and turn angles in anappropriate order to allow the destination to bereached. Information abstracted from the environ-ment and encoded for use in wayfinding includesbeing able to estimate and reproduce path segmentlengths and to be aware of and to be able to reproducethe turn angles between consecutive path segments.While this exercise may be simple for short trips withfew segments and turns, as a journey increases incomplexity the encoding and storage of distance andturn angle combinations becomes more difficult.Losing ones way usually means either an incorrectencoding of a path segment or turn angle or incorrectdecoding of correctly stored information duringtravel. How often have we been faced with theproblem of deciding whether a particular destinationis down one turn street or the next? In such cases it isnot just the recall and recognition of path sequencesand turn angles that allows successful trip com-pletion, but more likely the identification of an en-vironmental cue, either on or off the route beingfollowed, that helps select the next path segmentunder conditions of uncertainty.

Learning a route involves not only remembering thenumber of segments and turns, but also the angle ofthe turns, the length of the segments, critical environ-mental features at choice points, and other environ-mental features that occur along or in the vicinity ofpath segments (e.g. ones that can be seen from theroute and used as orientation and locational aids).Thus, when traversing a route, even when a path isfollowed that was not part of the original learnedroute, a successful trip can be completed if recog-nition of a nearby landmark assures the traveller thatshe/he is at least heading in the right direction.Learning a route, therefore, involves being aware notonly of the route segments experienced during asuccessful trip, but also developing a strategy ofrecognizing environmental features that allows for

Geol0rum/Volume 23 Number 2/1992

compensation of wrong choices when turning or whenviewing a landmark from a different angle.

Route Learning in Familiar and Unfamiliar En-vironments

Over the years a research team at the University ofCalifi~rnia, Santa Barbara has examined theacquisition of route knowledge over time via re-peated learning trials on selected routes withinfamil!iar and unfamiliar environments (DOHERTY,1984; DOHERTY and PELLEGRINO, 1986;DOHERTY et al., 1989; GALE et al., 1990; GOL-LEDGE et al., 1991; PELLEGRINO er al., 1990).l:,ach of these studies used a battery of tasks toexamine the ability of subjects to recognize scenesthat were either found on the routes to be learned orin the: same or different neighborhoods.

Two types of scenes were considered--piots, whichwere scenes of individual cues such as houses, naturalenvironmental features, or other signs of humanoccupancy, and views, which were extended imagesof what one might see when looking straight aheaddown a route segment. Experiments were conductedboth in the field with immediate debriefing in amobile laboratory following field experience, and inthe laboratory where route learning took place byviewing slides or videotapes of a particular route.Recognition tasks were developed which requiredsubjects to be able to determine whether or not aspecific plot or view scene was on or off a route or inor out of the neighborhood through which the routeextended. In addition, sequencing tasks, again usingplots and views, helped determine if the correctordering of route segments took place, while sketchmapping procedures provided evidence of whetherthe approximate number of segments, their length,and the turn angles connecting them were appropri-ately encoded. Other tasks were defined to testw!hether cues were located cognitively in their correctsegments, and whether or not subjects were able toinfer proximities between readily identifiable cueswhen across-segment judgments were made (GALE,1985; GOLLEDGE etal., 1991).

Most recently, a series of orientation and directional

207

estimation tasks were devised using pointing fromimaged locations to other imaged locations to see ifencoded route information duplicated the real con-figuration of learned routes. To examine whether theMOAR and CARLETON (1982) and SIEGEL et alo(1984) suggestions that people were able to integrateknowledge obtained from routes into a generalknowledge of the layout of an environment can beduplicated, tasks were devised to see how accuratelysubjects could estimate the orientation and directionof a point of one learned route to a point on adifferent, but partially overlapping learned route.The lack of success in completing these tasks(GOLLEDGE et al., 1991) runs counter to the routeintegration hypothesis and appears to suggest that anadditional type of knowledge structure~configurational or spatial relational knowledge~develops in the spatial domain.

Segments and Turns

Our past research has tended to confirm the findingsof ALLEN (1981, 1985, 1987) and ALLEN andKIRASIC (1985) that route learning usually involveschunking of the routes. These chunks or segments canbe sequenced far more effectively than would be thecase if one tried to learn the correct sequence of anextremely large number of individual cues that makeup each route segment. This chunking hypothesisaccounts for the ability to order masses of infor-mation into correct sequences. Errors are sometimesmade, however, when the turn angle connectingsequences is encoded incorrectly. For example, 90turns may be encoded as 270~ turns, resulting in a 180switch in route direction. This could result in immedi-ate disorientation and a sense of being lost, eventhough the balance of information along the routecould be encoded correctly and followed accurately.Alternatively, smooth curves may be encoded ascontinuous straight lines with no change of direction,and non-right-angle turns can be encoded as rightangles. Each of these cases may pose no hindrance tothe successful completion of a trip through an en-vironment. However, if stored incorrectly in onescognitive map, attempts to guide other people or toreproduce ones knowledge structure externally foruse by other people may result in error or mis-direction.

2O8Environmental Complexity

ALLEN (1981) showed that, if the environment wassufficiently differentiated, route learning could takeplace quickly and efficiently through a segmentationor chunking process. Chunks of the environment thatwere internally similar, or that were differentiatedfrom other chunks by a significant landmark or otherenvironmental cue (eog. a change in housing styles, change in housing density or lot size, an increase inthe quantity of neighborhood vegetation, or changefrom rectilinear to curvilinear street patterns), canprovide criteria for the cognitive segmentation pro-cess. A given route under these circumstances mayproceed without terms for quite a while, but still bechunked as part of the process of classifying environ-mental information and simplifying it. Alternatively,where the environment is relatively undifferentiatedand uniform (as is often the case in some largesurburban residential tracts, or in inner-city areasdominated by row houses or sets of uniformly con-structed high-rise apartment buildings) it is a muchmore difficult task to undertake segmentation.

Route Complexity

Conventional wisdom assumes that, as routes in-crease in length and complexity, the ability to remem-ber and traverse them without some type of aid (e.g. map) decreases. Reference is frequently made to themagic number 7 + 2 as being the upper asymptote ofroute segments that can be learned with relativeefficiency and low error.

In one of the first studies combining distance percep-tion with route learning, BRIGGS (1972) showedthat cognized route distance remained a power func-tion transform of objective distance regardless ofwhether the routes were straight or curved, andwhether intermediate turns were towards or awayfrom the final destination. Recently, HAYASHI etal. (1990) provided clear evidence that, as the numberof route segments increased, the perceptual errorincreased.

The substantive problem with which Bfiggs was con-cerned was the influence of environmental structureupon the cognition of distance. The hypothesis putforward was that, for equivalent objective distances

Geoforum/Volume 23 Number 2/1992between points within an urban environment, cog-nized distance varies as a function of route complexitybetween points. In particular, Bfiggs hypothesizedthat, for equivalent objective distances, the cognizeddistance from a common origin point would begreater to locations towards downtown than to placesaway from downtown; and, secondly, that for pairs ofpoints an equivalent objective distance apart, butlinked by routes involving turns or bends, the routeswith bends or turin will be cognized as longer and theorigin and destination points cognized as fartherapart than points linked by a straight route.

Briggs results generally showed that distances todowntown locations were overestimated relative tothose away from downtown. With respect to straightroutes against those involving turns, Bfiggs inferredthat a different scale was used by subjects for routeswith bends compared to routes as straight lines. Thissuggested that, along routes with bends, interpointdistances were cognized more as crow-fly distancesthan as over-the-road distances. This is an interestingconclusion. It appears from this that routes are notencoded simply as concatenations of line segments.Rather the anchoring end-points, and their relativeinterpoint distances, are perceived more as shortcutor crow-fly distances than as over-the-road distances.Such a finding would tend to support a hypothesis thatinformation learned from routes could readily beintegrated into a configurational knowledge struc-ture. In such a knowledge structure, the relativeposition and direction of points from one anotherestablish the positional layout. Over-the-roadmeasures of connectivity are important only for esti-mating travel time. These results also suggest thattravel time and distance are not mere substitutes forone another, but reflect two quite different purposesin the construction of cognitive maps. The linking ofpoints in a configuration by straight-line distanceshelps establish the two-dimensional layout uponwhich more detailed information can be grafted atwill. Thus, when required to travel between an originand a destination, it could be hypothesized that onefirst establishes an image of the relative locations ofthe places, and then attempts to determine a routebetween them that will satisfy some type of move-ment criteria (such as minimizing time, maximizingaesthetic value of the journey, or minimizing stressand anxiety). This could help account for the manyand varied results which tend to imply that individuals

GeoforurrffVolume 23 Number 2/1992are not minimum-path travellers (when the minimumpath is interpreted in terms of route distanceraeasurements). Certainly, envisaging connectionsbetween locations as straight lines provides the direc-tional information that over-the-road travel may not.

Returning to a point made earlier in this paper, theuse of the crowofly distance to connect points allowsthe use of simple triangulation procedures to locatee, ach point--a procedure which over-the-road con-nectivities do not allow whenever there are bends orturns, in the road. It also implies that individuals whoknow only the time of travel between places may havegreat difficulty in constructing accurate configura-tional or layout information. Since it is suspected thatrnany people encode route information in temporalterm:s, one should not be surprised by results thatindicate that people who learn environments byroutes are less than precise at locating landmarks inthose; environments and knowing the approximatedirectional and distance information of each land-raark, from others.

Waflinding Errors

Wayfinding errors can occur in both encoding anddecoding information. For example, the incorrectsensing of (forward) speed can result in distancetravelled being overestimated or underestimated.Incorrectly integrating velocity to obtain distancetravelled on each leg of a multi-segment path can als6produce error, which may result in misspecification ofthe kmation of a critical choice point or destination.Such errors may also result in improper identificationof key landmarks, particularly where little perceptualvariation exists---as might be the case in a uniformresidential neighborhood.

Incorrectly sensing the role of turn also leads to anencoding error. This may result in incorrectly inte-grating turn rate when trying to define a new heading,resulting in the incorrect encoding of a turn anglegreater or less than a right angle, or encoding smoothgradual curves as straight lines.

Wayfinding errors may also result from using anerroneous or biased representation of an environ-ment as the spatial basis for decision making.Stretched, folded, or incomplete cognitive maps lend

209themselves to both navigation and wayfinding errors.Here the error may result from incorrectly decodingcorrectly sensed information (e.g. because of stress,anxiety, brain damage, etc.) or correctly decodingincorrectly sensed information (e.g. as specifiedabove).

Yet another type of wayfinding error can occur iferroneous updating of self-position occurs while navi-gating. This may result from incorrectly performingthe mental trigonometry used to estimate bearing ordistance of a target location, or when attempting tosolve a shortcutting problem. Humans are notori-ously poor path integrators~which is the reason formillennia of experimentation to produce more accu-rate ways of measuring distance, angle, and speed.The simple fact is that most human travel unassistedby technical aids is error-prone.

In recent experiments using blind and blindfoldedsighted individuals, LOOMIS et al. (1991) haveshown that an increase in route complexity producesan increase in error when subjects are required tolearn an incomplete configuration of segments, andthen return directly to home using a shortcuttingprocedure. The different configurations used in theseexperiments varied from simple right angled trianglesto squares, rectangles and quadrilaterals. The mostcomplex route was a crossover in which it was necess-ary for the traveller to cross an initial leg beforereturning correctly to the home base.

Computational Process Models of Wayfinding

A variety of computer simulation models---calledcomputational process models (CPMs)--have beendeveloped for navigational purposes [see KUIPERS(1977), GOPAL et al. (1989), and LEISER andZILBERSHATZ (1989), for examples]. LEISERand ZILBERSHATZ (1989) suggest that experiencewith a large-scale environment is at first unstructured.Although specific environmental cues are recognizedand learned, they are not necessarily spatially con-nected one to another, but simply listed in a declara-tive structure. With increasing experience, however,routes are learned which connect specific locations.As the relational characteristics between places onand off these routes become recognizable, a survey-type knowledge structure develops. The sequence is

210

similar to that argued by SIEGEL and WHITE(1975) and GOLLEDGE (1977). Once a survey-levelknowledge structure is obtained, then geometric re-lations between points such as crow-fly distances,orientation to frames of reference, and directionsbetween places can be comprehended. They alsoargue that many people are unwilling to recognize theextent of their environmental knowledge, preferringto develop repetitive travel patterns which confirmthe learned anchoring structure of their cognitivemap, but which may also inhibit exploration or thesearch for more efficient connecting routes

With respect to route knowledge, empirical evidenceoften contradicts theory. For example, THORN-DYKE (1980) and ANDERSON (1982) argue proceduralization acts on a declarative knowledgebase to allow route knowledge to emerge. Pro-ceduralization consists of sets of production rules thatcan be accessed to fire in the correct sequence to helprecall memorized sequences of movement through aparticular environment, which in turn allows success-ful trip completion. There would be no reason toexpect proceduralization to work differently depend-ing on the direction travelled between any two points.However, SADALLA et aI. (1980) showed that dis-tances were perceived differently depending onwhether they were to or from a reference node, andSAISA et al. (1986) have demonstrated that segmentand route length estimates are asymmetric. It isdifficult to see how a single set of production rules canhandle these asymmetries, just as it is difficult toimagine how production rules could account for thedifferent type of directional and orientation errorsthat are made in the learning process and that some-times are part of the final product.

The building of a computational process model basedon the simple symmetric hypothesis produces error-free travel along specific routes and is at best an idealsituation or norm from which one can measure depar-tures. Such models may also have difficulty inaccounting for distance and directional cognitionswhere crossing of regional boundaries occurs. GOL-LEDGE and SPECTOR (1978) have previouslyargued that, while travel habits may be repetitive andpredictive within a households activity space, at thefringes of this space and outside it each individualgenerally has only enough information to reach agiven destination from well-known places (i.e. major

Geoforum/Volume 23 Number 2/1992anchor points); knowledge of the surrounding en-vironment is sparse.

Other models designed to simulate learning of spatialnetworks have been produced by KUIPERS (1978),McDERMOTT and DAVIS (1983), SMITH et al.(1982), and GOPAL et al. (1989). The productionrules built into such models allow learning to takeplace on repetitive trials and allow the successfulcompletion of a task. However, the models do notsimulate any particular persons natural travel behav-iour because of things such as the inability to incor-porate asymmetric distances and directions, and the.inability to specify on any particular trip the mostappropriate criteria that might be selected by a hu-man traveller. Such criteria range from minimizingtime of travel, minimizing distance of travel, expend-ing least effort, minimizing stress or anxiety, minimiz-ing the number of turns, maximizing the aestheticvalue, minimizing the chance of be.ing lost by takinglonger routes via well-known anchor points, and soon. Clear sets of well-authenticated criteria forexplaining wayfinding do not exist.

The typical computational process model begins bysetting out from an origin without prior knowledge ofthe network through which a traveller must pass. It istherefore working in an unfamiliar environment, anddoes not have the access to ancillary tools such asmaps or photographs of the area, nor does it know thereference frame of the task environment. All thesemust be learned. The moving organism does notknow what paths or path alternatives it will laterencounter, nor whether the initial path that it selectsis an efficient or effective start to the linking of a givenorigin and destination. The basic components of themodel, therefore, consist of condition-action pairswhich usually take an if-then format. These pairsaccumulate consecutively to form the production rulethat defines the initial route. In a sense, a productionrule for a route segment provides a signpost forselection of the next segment.

Once a simulated traveller has been stimulated toleave an origin, it first proceeds by scanning theadjacent nodes to see if any of these are the destina-tion. If a particular node is the destination, a scan ismade to see how many routes can reach that destina-tion. Obviously, to avoid excessive enumeration of allpossible routes, some criteria such as least effort,

GeoforumNolume 23 Number 2/1992least number of turns, or minimizing the number ofsegments, need to be used as controlling factors. If anadjacent node is not the destination, a typical processis to select each adjacent node in turn and scan in itsvicinity for the destination. This process is repeatedand the search space expands uniformly in alldirections--if a regular network exists. Irregularsearching takes place if the network is incomplete(i.e. not a completely linked graph), or if the distanceapart of nearest neighbors increases in some areaswhile remaining constant or decreasing in others. Inthis way a computer accumulates information about|he e, ntire network.

This comprehensive learning process is quite differ-ent to that normally used by humans. The latter donot explore sequentially and successively all areas inthe vicinity of every node in the network. Humansearch space is invariably sectoral and may be guidedby even a small piece of information such as a chosenheading: thus, if one hypothesizes the destination isto the north, then eastern, western, and southernnodes are eliminated early in the search process.Once sectoralized, however, a further constrained setof production rules guide segment selection by elimi-nating turns that appear to direct the traveller awayfrom moving in a northerly direction. Later as moreinformation about the general layout of the environ-ment becomes known, this orientation rule may beviolated if the traveller finds that an effective routecan be determined by first moving in the nonprime.direction. For example, a person may make a shorttrip to the south to enter a freeway, which later turnsnorth and passes near a given destination, or providesaccess to a direct arterial on which the destinationmay be located.

An advantage of a CPM that uses the production rulesystem is that, once experienced, a route can beretraced at will, regardless of where the travellero~ginates along the route. For example, assume aroute is learned from A to B that involves eightsegments. Assume further that another route from Ct,a D is learned that crosses the AB route in the thirdsegment. This should give the CD traveller the abilityto remember and travel to either of the AB destinaotions once s/he is at the appropriate intersection.Only that sequence of production rules required tocomplete part of the AB trip need be activated.

211

This process of learning routes can fix networkproperties in memory. But nodes and segments maynot necessarily be fixed in a spatial sense, unlesssegment distance, heading, and external frame ofreference are all simultaneously stored. For example,one could remember segment sequence and segmentlength between multiple pairs of nodes, but quiteinaccurately fix the end point locations. Such inaccu-racies would inhibit shortcutting and spatial explor-ation between hitherto unconnected nodes, for themodel would have no way of estimating whether ornot the connected sequence that it had stored was themost effective way of joining the two places.

A recent attempt to build a cognitivety aware dis-aggregate model of household travel behaviour hasbeen undertaken by G,~RLING et al. (1991). Thismodel has a series of modules: (a) the objectiveenvironment consisting of a transportation networkand a set of functions that attract individuals, (b) long-term calendar that contains the time availablefor the execution of each activity in a householdsepisodic activity cycle, (c) a constraints identifier thatexamines information from the objective environ-ment and the activity pattern for assigned prioritiesand then lists possible obligatory and discretionaryactivities, (d) a sequencer which examines the list possible activities and matches it with environmentalconstraints and priorities to determine the sequenceof events that take place within a given time, and (e) mental executor which chooses the transportationmodes necessary, for the execution of the differentactivities.

Wayfinding and Geographic informationSystems

Each activity defined for a traveller can be describedby a set of productions. But these productions need tobe implemented in a network context reflecting theidiosyncrasies of a real environment. This is whereGISs become important in the modeling of wayfind-ing behavior. For example, in ARCINFO, the acti-vation of NETWORK subroutines atlows one todefine different types of routes once a network isidentified. In a U.S. city, for example, the appropri-ate citys TIGER files can provide the detailed streetnetwork required to activate production rules fordifferent trip purposes.

212

In general, a network can be specified as a list ofnodes and a list of links between pairs of nodes. Eachnode and arc (link) can be geo-referenced. In addi-tion to the basic network, the GIS would containlanduse information for each block facing a roadsegment. Superimposed on this information couldalso be the socio-demographic information containedin the most disaggregate census units. For any trippurpose the subset of landuses at which that purposecan be satisfied can be identified. The problem thenarises as to which combination of route segmentsshould be used to fulfil a trip purpose. Accessing aHousehold Scheduler (GARLING et aL, 1991) candetermine if a single- or multiple-purpose and single-or multiple-stop trip is scheduled. This schedule iscompleted according to criteria specifying the timerequired to complete the purpose, the institutionalconstraints on opportune open times for access to theplace at which a function could be accessed, prefer-ences as to the temporal sequencing of scheduledactivities, and selection of criteria (suchas minimaltime or distance covered) that allow a trip to beplanned with the scheduling constraints in mind.Using a combination of a Scheduler and a GIS wouldfurther enable the network to be visually displayed onscreen and the selected path or route to be high-lighted. Accessing other network-based models (e.g.TRANSCAD or TRANPLAN) could then allowcomparison between the chosen route and a routethat would conform to traditional planning principles(e.g. shortest-path or travelling salesman principles).

Early work by GOLLEDGE and ZANNARAS(1973) suggested that route knowledge settles earlyinto a fixed or habitual pattern, and the componentsof routes are oriented and integrated to conform tothe travellers guiding principles. According to PAIL-HOUS (1970.), networks have at least two differentlevels, primary and secondary, and any particularjourney is most likely to have at least three com-ponents: a secondary link to a primary entrance, aprimary segment, and a link from the exiting node onthis primary segment to a secondary connection thatleads to a destination. I suspect there are more thantwo levels in a learned network structure, but this stillremains to be proven. LEISER and ZILBERT-SHATZ (1989) suggest that an interesting researchdirection is to explore what happens to route selec-tion when networks develop special topologicalfeatures such as congestion, bottlenecking, or grid-


locking. This, of course, brings the academic way-finding problem into touch with hard reality, and isundoubtedly a major direction tor future researchthat seeks to combine wayfinding and a GIS.

Concluding Staternent

Recognizing and locating places and finding routesbetween them are essential part.,; of the process ofspatial knowledge acquisition. Often referred to aslandmark and route knowledge or declarative andprocedural knowledge, these represent the core of.spatial knowledge--particularly what can be referredto as common sense spatial knowledge which is theknowledge base for most people. Knowing places androutes suffices for everyday behavior. But when in-terpretation of an environment is needed, or whenspatial inferences (such as taking shortcuts throughunknown areas or deciding where an urban functionmight be found) is required, I suggest that anothertype of spatial knowledge--configurational or rela-tional knowledgewis essential.

Configurational knowledge is that which underliesmany geographical concepts--such as shape, pattern,distribution, and association Examination of theprocesses needed to gain such knowledge has notbeen well researched. In contrast, a large literatureexists on place recognition and wayfinding. As can beinferred from this paper, many active research pro-grams exist in a variety of disciplines focused onlandmarks and routes. Apart from the basic need tounderstand these concepts and their spatial outcomes(i.e. spatial behaviors), there is an emerging trend apply spatial knowledge concepts in scenariosranging from designing moon-walkers and auto-matically guided vehicles to understanding how dis-advantaged children can best be helped to effectivelylearn and use their environments.

As we find out more about places, routes, andlayouts, there will be a need to reexamine existingtheories of environmental knowing, the spatial com-ponent of theories of development, and theories ofthe cognitive structuring of information. As we learnmore about human understanding and use of space, itis apparent that spatial knowledge has a uniquecharacter that is not necessarily well described byexisting theories or models of learning and under-


standing. It is important that geographers realize thisand that they initiate work designed to explore morefully the concept base of the discipline they profess.

AcknowledgementsmThis research was partly funded byNSF grant No. SES-9023047. The final draft was completedwhile the author was a Visiting Professor of Geography atthe UJaiversity of New South Wales.

References

AJLLEN, S. C. (1981) A developmental perspective on theeffects of subdividing macrospatial experience, J. exp.Psychol.: hum. Learn. Theory, 7, 120-132.

ALLEN, S. C. (i985) Strengthening weak links in the studyof the development of macrospatial cognition, In: TheDevelopment of Spatial Cognition, R. Cohen (Ed.).Lawrence Erlbaum Associates, Hillsdale, NJ.

ALLEN, S. C. (1987) Cognitive influences on the acqui-sition of route knowledge in children and adults, In:Cognitive Processes and Spatial Orientation in Animaland Man, Vol. 2: Neurophysiological and Developmen-tal Aspects, NATO ASI Series, Series D: Behaviouraland Social Sciences, No. 37, pp. 274-283, P. Ellen andC Thimus Blanc (Eds). Martinus Nijhoff, Dordrecht.

ALLEN, G. L. and KIRASIC, K. C. (1985) Effects of thecognitive organization of route knowledge on judgmentsof macrospatial distance, Mere. Cognit., 13, 218-227.

ANDERSON, J. R. (1982) Acquisition of cognitive skill,Psychol. Rev., 89,369-406.

BRIGGS, R. (1972) Cognitive distance in urban space,PhD Dissertation, Ohio State University, Columbus,OH.

GALE, N. (1982) Exploring location error in cognitiveconfigurations of a city, In: An Examination of theSpatial Variation in the Distortion and Fuzziness ofCognitive Maps, Final Report, R. G. Gotledge andW. R. Tobler (Eds). NSF Grant no. SES81-10253.

GALE., N. (1985) Route learning by children in real andsimulated environments, PhD dissertation, Universityof California, Santa Barbara, CA.

GALE., N., HALPERIN, W., GOLLEDGE, R. G. andCOUCLELIS, H. (1990) Exploring spatial familiarity,..Prof. Geogr., 42, 229-313.

GARLING, T. and EVANS, G. (Eds) (1991) Environ-merit, Cognition, and Action: a Multidisciplinary Inte-.gralive Approach Oxford University Press, New York.

GARI,ING, T. and GOLLEDGE, R. (1989) Environmen-tal perception and cognition, In: Advances in Environ-mental Behavior and Design, 2, E. Zube and G. Moore..(Eds). Plenum Press, New York.

GARI,ING, T., LINDBERG~ E., CARREIRAS, M. andBOOK, A. (1986) Reference systems in cognitive maps.J. envir. Psychol., 6, 1-18.

GARI.,ING, T., KWAN, M.-P. and GOLLEDGE, R. G.(1991) Computational process modelling of travel de-cisions: review and conceptual analysis, Paper presentedat ,~he Congress of the International GeographicalUnion, Stockholm, Sweden, 2--4 September.

GOLLEDGE, R. G. (1974) Methods and methodologicalissues in environmental cognition research, In: On

213Determining Cognitive Configurations of a City, Vol. 1:Problem Statement, Experimental Design and Prelimi-nary Findings, pp. 225-242, R. Golledge (Ed.). NSFGrant no. GS-37969.

GOLLEDGE, R. G. (1977) Environmental cues, cognitivemapping, and spatial behavior, In: Behavior-Environment Research Methods, pp. 35-46, D. Burke etal. (Eds). Institute for Environmental Studies, Univer-sity of Wisconsin, Madison, WI.

GOLLEDGE, R. G. (1989) Environmental cognition, Ln:Handbook of Environmental Psychology, pp. 131-174,D. Stokois and I. Altman (Eds). John Wiley, New York.

GOLLEDGE, R. G. (1991) Cognition of physical and builtenvironments, In: Environment, Cognition, and Action:a Multidisciplinary Integrative Approach, T. Gilding andG. Evans (Eds). Oxford University Press, New York.

GOLLEDGE, R. G., PELLEGRINO, J. W., GALE, N.and DOHERTY, S. (1991) Acquisition and integrationof route knowledge, Natn. geogr. J. India, 37, 130-146.

GOLLEDGE, R. G. and RUSHTON, G. (1972) Multi-dimensional scaring: review and geographical appli-cations, AAG Commission on College Geography,Technical Paper No. 10, Washington, DC.

GOLLEDGE, R. G., SMITH, T. R., PELLEGRINO,J. W., DOHERTY, S. and MARSHALL, S. P. (1985)A conceptual modeI and empirical analysis of childrensacquisition of spatial knowledge, J. envir. PsychoL, S,125-152.

GOLLEDGE, R. G. and SPECTOR, A. (1978) Compre-hending the urban environment: theory and practice,Geogrl Analysis, 14, 305-325.

GOLLEDGE, R. G. and ZANNARAS (1973) Cognitiveapproaches to the analysis of human spatial behavior,In: Environment and Cognition, W. Itte|son (Ed.). Sem-inar Press, New York.

GOPAL, S., KLATZKY, R. and SMITH, T. R. (1989)NAVIGATOR: a psychologically based model of en-vironmental learning through nagivation, Jo envir. Psy-chol., 9,309-331.

HART and MOORE (1973) The development of spatialcognition: a review, In: Image and Environment: Cogni-tive Mapping and Spatial Behavior, pp. 246-288, R.Downs and D. Stea (Eds)o Aldine, Chicago, IL.

HAYASHI, T., FUJII, H. and INUI, T. (1990) Modelingthe cognitive map formation process based on psycho-logical experiments using a computer graphics system,Presented at IEEE SMC Conference, Los Angeles, CA.

KLATZKY, R. L., LOOMIS, J. M., GOLLEDGE, R. G.,CICINELLI, J. G., DOHERTY, S. and PELLE-GRINO, J. W. (1990) Acquisition of route and surveyknowledge in the absence of vision, ~r. Motor Behav., 22,19--43.

KUIPERS, B. (1977) Representing knowledge of largescale space, AI-TR-418, AI Laboratory, MIT, Harvard,MA.

KUIPERS, B. (1978) Modeling spatial knowledge, Cognit.Sci., 2, 129-153.

LEISER, D. and ZILBERSHATZ, A. (1989)o TheTRAVELLER: a computational model of spatialnetwork learning, Envir. Behav., 21,435-463.

LIBEN, L. (1982) Childrens large-scale spatial cognition:is the measure the message?, In: New Directions forChild Development: Childrens Conceptions of SpatialRelationships, pp. 51--64, R. Cohen (Ed.). Jossey-Bass,San Francisco, CA.

214LOOMIS, J., KLATZKY, R., GOLLEDGE, R., CICI-

NELLI, J and FRY, P. (1991) Navigation without sight,Manuscript under review.

McDERMOTT, D. and DAVIS, Eo (1983) Planning andexecuting routes through uncertain territory, Artif.InteR., 143--160.

MOAR and CARLETON (1982) Memory for routes, Q. Jlexp. Psychol., 34A, 381-394.

PAILHOUS, J. (1970) La Representation de rEspaceUrbain--lExemple du Chauffeur de Taxi. PressesUniversitaires de France, Paris.

PERON, E. M., BARONI, M. R., JOB, R. and SALA-MASO, P. (1990) Effects of familiarity in recallinginteriors and external places, J. Envir. Psychol., 5,325-333.

PIAGET, J. and INHELDER, B. (1967) The ChildsConception of Space. Norton, New York.

RECKER, W., McNALLY, M. and ROOT, G. (1986) model of complex travel behavior: Part I: theoreticaldevelopment, Transpn Res. A~ 20, 307-318.

RICHARDSON, G. D. (I981) Comparing two cogmtivemapping methodologies, Area, 13, 325-331.

SADALLA, E. K., BURROUGHS, W. J. and STAPLIN,L. J. (1980) Reference points in spatial cognition. J. exp.

.. PsvyhoL: hum. Learn. Memo, 6, 516--528.SAISA, J., SVENSSON-G.~RLING, A., GARLING, T.

and LINDBERG, E. (1986) Interurban cognitive dis-tance: the relationship between judgments of straight-

Geofomm/Volume 23 Number 2/1992line distances, travel distances and travel time, GeogrlAnalysis, lg, 167-174.

SIEGEL, A. W. (1981) The externalization of cognitivemaps by children and adults: in search of better ways toask better questions, In: Spatial Representation andBehavior across the Life Span: Theory and Application,pp. 167-194, L. L/ben, A. Patterson and N. Newcombe(Eds). Academic Press, New York.

SIEGEL, A. W. and WHITE, S. H.. (1975) The develop-ment of spatial representations of large-scale environ-meats, In: Advances in Child Development andBehavior, Vol. 10, H. W. Reese (Fd.). Academic Press,New York.

SMITH, T. R., PELLEGRINO, J. W. and GOLLEDGE,R. G. (1982) Computational process modelling spatial cognition and behavior, Geogrl Analysis, 14,305-325.

THORNDYKE, P. W. (1980) Pertbrmance models for"spatial and Iocational cognition, Technical Report R-2676-ONR, RAND Corporation, Santa Monica, CA.

TOBLER, Wo R. (1965) Computation of the correspon-dence of geographical patterns, Pap. Proc. reg. Sci.Ass., 15, 131-139.

TOBLER, W. R. (1978) TRILAT program. Departmentof Geography, University of California, Santa Barbara,CA.

WORCHEL, P. (I951) Space percept,ion and orientation the blind, PsychoL Monogro: geno Applo, 65, 1-27.

212_place recognition and wayfinding

Documents