Abstract
Previous research has suggested that the layout of urban spaces can have a substantial influence on how people navigate through those spaces. However, to date, few studies have directly investigated how changes in layout interact with changes in visual field to shape a person’s route choice. Across two experiments, the influence of visual field and spatial layout was manipulated using virtual reality. It was found that route choice was significantly influenced by the configuration of a space—The more consistently organized environment led to more systematic route choices. However, limiting the perception of distant visual information was found to influence route choice in a similar but completely independent way. These findings suggest that navigation in urban spaces is dependent on the interaction between topology and the visual features of the space, where greater visual field and a consistently organized spatial layout lead to maximally efficient route choices.
For well over a century, much of the research in human and animal navigation has focused on the roles of factors such as landmark learning, path integration, and cognitive mapping in solving wayfinding problems. However, emerging research has suggested that route choices in wayfinding tasks are often somewhat independent of such factors and are instead largely driven by more global aspects of the structure of a space (Penn, 2003).
Early studies of spatial preference suggested a relationship between the global environment and our perception of space. Kevin Lynch (1960) argued that individuals prefer spaces that have a high degree of legibility—the degree to which they can be easily recognized and organized by a person. When people were asked to imagine navigating the streets of Boston, Los Angeles, and Jersey City, they were found to have a higher probability of becoming confused or lost in locations that were isolated. Furthermore, in sketch maps of environments, people were found to include more nodes and paths when the surrounding area was both familiar to the person and was defined by visually distinctive features. Subsequent research by S. Kaplan and Kaplan (1982) and R. Kaplan and Kaplan (1989) provided further evidence for the influence of the spatial configuration of an environment on preference. People were found to prefer locations within an environment that maximized their ability to perceive useful visual information, and were embedded in an environment with a coherent and easily understood spatial configuration. Taken together, these two distinct approaches suggest that differences in the configuration of an environment can influence how one explores an environment, initially and in subsequent navigation.
Other evidence for the influence of the configuration of an environment on spatial cognition has been observed previously: A number of different experimental approaches have suggested that a more linear and orthogonal structure of the components of an environment, paths, and intersections can produce optimal route planning and spatial memory performance. In paper and pencil studies, people have been shown to prefer paths that consist of a straight initial segment (Bailenson, Shum, & Uttal, 2000). This preference for linearity and structure extends beyond route planning preferences into spatial memory. People have been shown to have poorer memory (Sadalla & Montello, 1989) and poorer accuracy at pointing at local landmarks (Montello, 1991) for intersections that deviate from orthogonal structures. Furthermore, in route traversal, one of two routes is perceived to be longer if more turns are made en route, despite both routes traversing the same distance (Sadalla & Staplin, 1980). As such, it appears that not only does the configuration of an environment influence preference but also spatial memory and spatial cognitive performance.
Research in a field known as Space Syntax (Hillier, 1996; Hillier & Hanson, 1984) has provided further evidence for the role of spatial configuration in the navigation process. This line of research seeks to quantify the configuration of an environment in a way that is consistent and reduces the environment to its key elements. Early observational studies of aggregate human movement data and the geometry and topology of cities have suggested that the layout of a city shows a probabilistic relationship, with respect to human movement (Hillier, Burdett, Peponis, & Penn, 1987; Hillier, Penn, Hanson, Grajewski, & Xu, 1993), and can serve as a predictor of vehicular movement (Penn, Hillier, Banister, & Xu, 1998a, 1998b). In subsequent research, the network structure of an environment, viewed as a spatial system rather than a combination of component parts (i.e., landmarks, neighborhoods, etc.), has placed particular emphasis on the predictive power of two different topological measures. Although many structural network properties can be defined for a spatial system derived from a city or neighborhood, those most commonly used include connectivity, the total number of routes that are connected to a specific path in a network, and integration, the average number of turns required to move from one path to any other path in the network. These global measures, computed independently of visual features of an environment, have shown a striking ability to predict people’s choice of path through an environment across a wide variety of experimental approaches, especially the measure of spatial integration (Penn, 2003). This significant correlation between the connectivity and integration of a spatial network and emergent navigation behavior bears a striking resemblance to the predictions raised by classical studies of spatial preference (i.e., S. Kaplan & Kaplan, 1982; Lynch, 1960) and studies of the influence of spatial configuration on route planning (Bailenson et al., 2000), and spatial perception or spatial memory (Montello, 1991; Sadalla & Montello, 1989; Sadalla & Staplin, 1980). Together, these bodies of research suggest that topological measures of an environment, viewed as an array of paths, may be useful in the navigational process above and beyond the role of other factors such as landmarks and attractive visual features.
Given these findings, it is possible to make certain predictions about how an individual will navigate by focusing on the configuration of an environment alone. Using correlations between the properties of connectivity and integration, one is able to compute a measure of the global spatial properties of an entire system rather than simply using local measure that might predict path preference (Hillier, 1996). In studies of navigation behavior of individuals, environments with high intelligibility (i.e., a high correlation between integration and connectivity) have been shown to promote more systematic and similar route choices than those environments with a low intelligibility (Conroy, 2001; Hillier, 1996). In contrast, environments with a low intelligibility have been shown to increase the idiosyncrasy of a person’s movement behavior (Conroy, 2001). The most commonly used paths through an environment tend to be those that have higher connectivity and integration; people appear to prefer routes that are highly connected and that extend through areas requiring fewer turns or offering a more simplified spatial layout. These results, combined with those found in aggregate movement studies, support the idea that the navigation process may be influenced by purely topological properties independent of other types of information.
In spite of the evidence for an apparent influence of topological properties on navigation, the reasons for such influences are not well understood. In an observational study of human movement, people were found to prefer paths through a city that were composed of line segments that consisted of the least angular change or contained the fewest number of turns, suggesting that the human movement preferences may be influenced by the simplicity and linearity of available roads and paths (Haq, 2003; Hillier & Iida, 2005). This finding is further supported by experimental investigations of route choice that have shown that individuals prefer to follow uniform and linear paths and that they seek to maximize the distance that they can see at any given time and that they may seek paths with topological properties conducive of this preference (Conroy, 2001; Conroy Dalton, 2003; Peponis, Zimring, & Choi, 1990). However, agent-based simulations (Turner, 2003; Turner & Penn, 2002) and an experimental study (Haq & Zimring, 2003) have suggested that visual affordances may be the primary influence on navigation performance, rather than topological properties and cues found within an environment. The desire to maximize the amount of visual space at a given time would also be consistent with the suggestions of S. Kaplan and Kaplan (1982) and R. Kaplan and Kaplan (1989). As such, it remains an open question whether differences in topological properties of a space are related to choices made throughout a navigation task or instead an emergent property of the manipulation of visual space corresponding to changes in the configuration of an environment alone.
A further possibility is that topology and visual distance may play a more complicated role in the navigational process. Haq and Zimring (2003) examined the performance of navigators as they became more experienced with a space and found that initial exploration was shaped by local spatial properties. As participants became more experienced with an environment, navigation was found to be better predicted by global topological measures rather than by local spatial measures. As such, it is possible that a more complex interplay between the structure of an environment and the navigation performance that it affords may be observed than that of one simply driven by visual factors.
In the present study, we sought to disentangle the contribution of topological information and the contribution of preference for maximized visual information on navigation performance. In two experiments, conducted in virtual reality, topological information was manipulated between participants by comparing performance in two environments differing in intelligibility. To determine whether differences in performance can be attributed to differences in visual information rather than topology, the size of the participant’s available visual field (and thus the available visual information) was manipulated between participants in two different ways. In Experiment 1, we manipulated the size of the visual field by either allowing a full field of view or a restricted field of view through the use of two different head-mounted displays (HMDs). In Experiment 2, we manipulated the availability of nearby and distant visual information through the use of simulated fog. This approach allowed us to determine the relative contribution of topology and visual field on the navigation process. It was hypothesized that evidence for the use of only topological information would be found if a significant main effect of topology was observed in the absence of a main effect of visual field. In contrast, it was predicted that a significant main effect of visual field manipulation in the absence of a main effect of topology would support the idea that individuals have a preference for maximizing their visual field as they move through space. Furthermore, to probe the effect of task on the relative use of topological and visual information, a two-part navigation task was used. The navigation task consisted of a primary naive search task, followed immediately by a wayfinding task.
Experiment 1
The purpose of Experiment 1 was to assess whether navigation behavior was influenced by changes in local visual field properties, independently of topology. To achieve this, peripheral visual information was reduced in half of the participants during the navigation task: Half of the participants were asked to navigate through two virtual environments while wearing a HMD, which did not offer full peripheral vision (44° field-of-view), whereas the other half of the participants were asked to navigate through an environment with an HMD, which offered nearly full peripheral vision (150° field-of-view). Topology was also manipulated by using two similar versions of a virtual environment that differed in intelligibility (i.e., the correlation between integration and connectivity). A significant main effect of topology and a significant main effect of visual field independent of topology on navigation behavior were hypothesized, where the latter effect would provide evidence for a change in navigation behavior driven largely by local visual characteristics.
Method
Participants
Forty-eight undergraduate students attending the University of Waterloo participated in the experiment in exchange for course credit. The average age of the sample was 20.71 years (SD = 1.85 years). Equal numbers of male and female participants were recruited, with an average age of 21.14 (SD = 2.23) for males, and 20.28 (SD = 1.46) for females. All participants were fluent English speakers and had normal or corrected-to-normal vision.
Virtual environments
Two virtual environments were constructed using SketchUp Pro 6.0 (Google Inc., Mountain View, California), a 3D modeling and graphics design package. Both models consisted of an urban environment measuring 1,200 m by 800 m. Within each environment, 38 buildings were placed of varying size and shape. Each building was given a consistent height of 8 m and was textured with identical textures of a traditional apartment building. Each environment is depicted in Figure 1. The placement of each building was adjusted in the high intelligibility environment to produce an optimal correlation between integration and connectivity across the majority of the space. The low intelligibility environment was created by adjusting the shape and position of buildings in the high intelligibility environment to reduce the correlation between connectivity and integration across the majority of the virtual environment. This method is similar to that used by previous investigations of space syntax (Conroy, 2001; Hillier, 1996). In both environments, a plaza was created and placed near the geographic center of the space. A target monument was placed in the center of the plaza in both environments, with a duplicate copy of the monument being placed at the participant’s start location on the edge of the environment. The border of the environment was demarcated by a wall, distinctively textured with a large red brick texture.
Overhead views of the high intelligibility and low intelligibility environments used in Experiments 1 and 2.
Visual displays
The virtual environment was rendered for the navigation task using a custom script written within Vizard (Worldviz Inc., Santa Barbara, California), a Python-based virtual reality toolbox. The environment was rendered stereoscopically on one of two HMDs: an nVisor SX HMD (nVis Inc., Reston, Virginia), which did not offer full peripheral vision, or a Wide5 HMD Fakespace Labs Inc., Mountain View, California), which offered a natural level of peripheral information. The nVisor SX HMD was capable of providing 60° diagonal field-of-view and was rendered at a 1,280 × 1,024 pixels per eye. A thick black fabric shroud prevented the participant from seeing the environment around them, allowing them to focus on the virtual scene presented to them by the HMD. The Wide5 HMD was capable of providing 150° diagonal field-of-view and was rendered at 1,600 × 1,200 pixels per eye. Both HMDs were fitted with an InertiaCube2 (InterSense Inc., Billerica, Massachusetts) tracking devices and were calibrated to update viewpoint changes in real time. This allowed changes in participant’s head orientation to be translated to the virtual environments.
Movement control
Participants were asked to navigate through the virtual environments using a combination of wireless mouse control and head position tracking. Movement was controlled using the wireless mouse, with the depressing of the left mouse resulting in forward movement at an average walking pace. Directional changes were produced through head position tracking by having the participant turn his or her head in the preferred heading direction. The participant could visually scan the environment by ceasing linear movement and using head turns. The participant’s location within the virtual environment and their head direction were recorded at sampling rate of 50 Hz throughout the navigation task.
Procedure
The experiment consisted of four experimental conditions (high or low intelligibility, and nVisor SX or Wide5 HMD) administered as a between-participants design. Each participant was randomly assigned to one of the four experimental conditions, while maintaining an equal number of each gender in each experimental group, prior to the commencement of the experiment. Participants were greeted in a centralized meeting area and escorted to the lab by the experimenter. Each participant was given a detailed explanation of the experiment and procedures before being asked to complete an informed consent letter. After receiving informed consent, the experimenter assisted the participants in putting on the HMD. The experimenter then familiarized the participants with the controls necessary to complete the experiment. After the participants indicated they were comfortable with the apparatus, the environment was presented to the participants through the HMD. Each participant started the task at the same corner of the environment beside a duplicate version of the target statue. The participants were informed that their task was to navigate through the city to find an identical statue located somewhere in the city nearby. The participants were further instructed that they would be required to return to the start position by the most direct route possible after successfully locating the target statue. After the participant indicated that they understood the instructions, the experiment began, and the participant’s position and heading data were collected until experimental completion. Each participant was allowed as much time as was necessary to complete the search and wayfinding navigation tasks.
Data analysis
The data collected for each participant were aggregated to provide a number of dependent measures. First, to assess the effect of topology and visual information on overall task performance, the total distance traveled and mean route similarity were calculated. Mean route similarity was calculated by constructing matrices containing the probability of all participants to travel in each possible direction at each intersection within the environment. Each participant’s movement was then referenced to this matrix and averaged. The resulting average probability therefore provided a general measure for how idiosyncratic or systematic each participant’s navigation was relative to that of other participants in the same experimental group. Second, to assess the influence of either experimental factor on navigation-specific cognitive processing, the total time spent paused throughout the task, total number of discrete gazes made during discrete pauses, and the mean size of gaze across all pauses were calculated. For the purpose of data analysis, a participant was considered to be pausing in a location if they were stopped in a specific location for more than 1 s. The mean size of gaze during pauses was calculated as the difference between the maximum and minimum gaze angle observed throughout each discrete pause.
Results
The data for Experiment 1 were first analyzed to determine whether topology and visual information (field-of-view) had an influence on overall navigation performance with a 2 × 2 (topology: high intelligibility vs. low intelligibility; field-of-view: peripheral vision vs. no peripheral vision) between-groups multivariate ANOVA. A second analysis was then performed to determine whether the influence of topology and field-of-view differed based on the level of experience a participant had with the environment, consisting of a 2 × 2 × 2 (topology: high intelligibility vs. low intelligibility; field-of-view: peripheral vision vs. no peripheral vision; task: search vs. wayfinding) mixed-models repeated measures ANOVA. Note that the search task was performed when participants had no experience with the environments and the wayfinding task was performed after the search task, that is, after they had experience navigating through the environments. Significant interactions were followed up with simple effects testing. Effect size estimates were computed using partial η2 (see Pierce, Block, & Aguinis, 2004, for a description of the calculation and interpretation of this measure). Representative traces of the navigation behavior for both the high intelligibility and low intelligibility environments are presented in Figure 1.
From the overall analysis, the effect of topology on navigation performance can be seen in Table 1. Differences in topology were found to have a significant effect on overall distance, F(1, 44) = 24.419, p < .001, η2 = 0.357, and mean route similarity, F(1, 44) = 31.395, p < .001, η2 = 0.416. Participants were found to travel shorter and follow more systematic routes in the high intelligibility environment than in the low intelligibility environment. Topology was also found to have an influence on gaze behavior throughout navigation. Participants were found to have a larger mean gaze range, F(1, 44) = 12.588, p < .001, η2 = 0.222, but significantly fewer number of gazes overall, F(1, 44) = 5.055, p < .03, η2 = 0.103, in the high intelligibility environment compared with the low intelligibility environment. However, no influence of topology was observed on pause behavior. In contrast to the influence of topology on the navigation and gaze behavior, no significant influence of the field-of-view manipulation was found, and no significant interactions were observed between field-of-view and topology.
Navigation Performance in High and Low Intelligibility Environments.
Note: Data are presented as mean (SD) values. ns = not significant.
The data were next analyzed to determine whether task was responsible for differences in the navigation and behavioral measures. Evidence was found for differences in performance based on the type of task the participant was engaged in. When initially searching for the target monument, participants were found to travel significantly shorter distance than when they attempted to navigate their way back to the start position, F(1, 44) = 104.13, p < .002, η2 = 0.191. However, no significant change in the mean route similarity was observed between the two tasks. Compared with search performance, wayfinding was found to result in significantly more frequent pause behavior, F(1, 44) = 8.472, p < .006, η2 = 0.161, a larger average gaze range, F(1, 44) = 59.019, p < .001, η2 = 0.573, and longer mean pause time, F(1, 44) = 95.089, p < .001, η2 = 0.684. Again, no evidence was found for an effect of field-of-view on either search or wayfinding components of the task, and no significant interactions were observed between task, field-of-view, and topology. The results for the search and wayfinding tasks are shown in Table 2.
Search Versus Wayfinding.
Note: Data are presented as mean (SD) values.
Discussion
In Experiment 1, we sought to determine whether alterations in the local visual field would influence the search and navigation process independent of topological changes by manipulating participants’ field of view during navigation.
Most evident was the finding that people were sensitive to differences in the topological measures between the two environments but not to changes in local visual field. This finding supports the argument that path choice, both in the behavior of individuals and at the aggregate level, is influenced by the topological makeup of the environment. Similarly, results indicated that when an environment has high intelligibility, gaze range increases significantly but does not result in a change in path, as no corresponding main effect of visual field was observed. Both findings reaffirm the potential importance of space syntax measures on navigation performance at the individual level.
No evidence was found for the role of local visual field changes in navigation task performance. Several possibilities exist to account for the lack of an effect of field-of-view on navigation performance. First, it is possible that participants compensated for the visual field differences by adjustment of navigation behavior. In previous investigations using environments differing in intelligibility, people preferred to minimize direction changes (Conroy, 2001; Conroy Dalton, 2003). This preference for minimized direction changes would result in preserved navigation performance if peripheral field-of-view is removed as it is likely that peripheral visual information was (mostly) ignored. Second, it is possible that the magnitude of field-of-view manipulations were insufficient to produce any change in behavior, if peripheral visual information was not required for successful completion of the navigation task. Previous research has found that differences in performance have not been observed consistently, but have most often been observed when visual field was reduced to values below 50° (e.g., Alfano & Michel, 1990; Chambers, 1982; Piantanida, Boman, Larimer, Gille, & Reed, 1992). In addition, peripheral vision has been found to be much more useful in the perception of vection, rather than other types of visual information (Warren, Kay, Zosh, Duchon, & Sahuc, 2001; Warren & Kurtz, 1992; Warren, Morris, & Kalish, 1988). Although the data do not discount either possibility, it is more parsimonious to conclude that field-of-view manipulations restricted to peripheral vision are not capable of producing a significant change in navigation behavior. It would therefore be reasonable to conclude that local visual field manipulations do not drive behavior in the navigation context but that more global visual field changes are necessary to produce a change in behavior.
Also worth noting was that, although we did find longer paths during the search task, consistent with naive search, we did not observe a larger gaze range compared with the wayfinding task. Instead, an increase in gaze range was found during the wayfinding task reflecting an increase in the use of visual information during effortful and experience-based navigation. These results are summarized in Tables 1 and 2.
Experiment 1 provided evidence that there was a lack of effect on navigation performance when field-of-view was changed from a near-natural (150°) field-of-view to a restricted (60°) field-of-view. It also provided evidence for the influence of topological variables derived from the surrounding environment (i.e., intelligibility) on navigation behavior.
Although it remains possible that local visual field differences may play a role in route choice in more familiar environments, it seems more likely that more global visual field changes are necessary to influence people’s spatial preferences in a way predicted by space syntax.
Experiment 2
In Experiment 2, we sought to investigate whether a change in the global visual field was capable of influencing navigation behavior and the use of topological information. Specifically, we investigated whether reducing the amount of distal spatial information available influenced navigation behavior in high and low intelligibility environments.
Method
Participants
Forty-eight undergraduate students attending the University of Waterloo participated in the experiment in exchange for course credit. The average age of the sample was 20.12 (SD = 1.50). Twenty-four participants were male 19.55 (SD = 1.20) and 24 were female 20.68 (SD = 1.78). All participants were fluent English speakers and had normal or corrected-to-normal vision.
Procedure
Experiment 2 used an identical procedure to Experiment 1, but with two exceptions. First, only one HMD was used for the experiment—the nVisor SX HMD. This HMD was selected as no performance differences were observed in the previous experiment between the two HMDs. Second, instead of manipulating peripheral vision, distal central vision was manipulated during the navigation task. Participants were randomly assigned to one of four conditions (high or low intelligibility, unrestricted vision or restricted vision), maintaining an equal number of each gender per experimental condition. Vision was restricted using simulated fog around the participant’s position in the virtual environment. The fog was rendered at a viewing distance of 50 m using a geometric concentration gradient. The distance was determined by the average size of an intersection found within each environment plus two standard deviations. In the restricted vision condition, participants were able to see the configuration of each intersection within the environment but could not perceive more distant visual cues.
Data analysis
The data for Experiment 2 were analyzed using an identical strategy as the one used in Experiment 1. Mean distance and mean route similarities were calculated as measures of navigation task performance. Mean route similarity was calculated by comparing each participant’s route choice to the cumulative performance of participants from the same experimental group at each intersection. In addition, amount of time spent paused in one location, number of head turns during pause times more than 1 s, and mean gaze range were assessed to determine whether different cognitive demands were placed on the participant as a result of the presence or absence of distal vision and differing topological information.
Results
The data were first analyzed using a 2 × 2 (topology: high intelligibility vs. low intelligibility; viewing distance: unrestricted vision vs. restricted vision) between-groups multivariate ANOVA to assess whether the lack of distant visual information would alter the use of topological visual information or navigation behavior in general. Examples of the movement behavior within both environments are provided in Figure 1. The data were then analyzed using a 2 × 2 × 2 (topology: high intelligibility vs. low intelligibility; viewing distance: unrestricted vision vs. restricted vision; task: search vs. wayfinding) mixed models repeated measures ANOVA to assess whether experience with the environment and task demands with the environment would alter the use of topology or the recruitment of distant visual information. Significant interactions were followed up with simple effects testing. Effect size estimates were computed using partial η2.
Overall, significant effects of topology and viewing distance were found on navigation performance (see Table 3). A decrease in the intelligibility of an environment was associated with longer travel distances, F(1, 44) = 9.602, p < .003, η2 = 0.179, and a higher likelihood to follow idiosyncratic routes, F(1, 44) = 31.626, p < .001, η2 = 0.418. Gaze range was also found to be higher in the high intelligibility environment, F(1, 44) = 9.607, p < .003, η2 = 0.179, but this difference was not reflected in the length of pause time or mean number of head turns across the navigation task. Interestingly, the influence of viewing distance showed a similar pattern of results. Restricting viewing distance resulted in an increase in the travel distance of participants, F(1, 44) = 10.699, p < .002, η2 = 0.196, and an increase in the idiosyncratic nature of their routes, F(1, 44) = 0.082, p < .001, η2 = 0.342. Restricting viewing distance also resulted in an increase in the mean number of head turns performed by each participant, F(1, 44) = 5.637, p < .02, η2 = 0.114, but was not found to influence pause time or gaze range. Despite having a similar effect on performance as the topology manipulation, topology and viewing distance were not found to interact significantly, suggesting that viewing distance and topology account for unique variance in navigation behavior. To investigate the potential additive nature of viewing distance and topology, multiple regression analysis was used such that we could quantify the unique contribution to the variance of the distance traveled and route similarity measures by each of the two factors. For viewing distance, R2 = .166, p < .004, whereas for topology, R2 = .149, p < .003, indicating that each explained a significant amount of unique variance in the distance traveled variable, explaining a total of 31.6% of the variance together. For the route similarity measure, R2 = .226, p < .001, for viewing distance and R2 = .313, p < .001, for topology, accounting for 54.0% of the variance observed in the propensity to engage in more idiosyncratic paths through the environment.
Intelligibility and Vision.
Note: Data are presented as mean (SD) values.
Separate analyses of the search and wayfinding components of the task revealed several significant findings. Pause time was higher in the search (M = 95.15 s; SD = 56.38 s) than wayfinding task (M = 19.24 s, SD = 29.17 s), F(1, 44) = 71.883, p < .001, η2 = 0.620. Gaze range and pause frequency were found to follow a similar pattern. The number of pauses was significantly greater, F(1, 44) = 24.198, p < .001, η2 = 0.355, and gaze range was wider, F(1, 44) = 19.277, p < .001, η2 = 0.305, in the search phase of the task, with an average of 12.62 head turns (SD = 10.12) and an average range of 134.68° (SD = 72.24), compared with the wayfinding task, where an average of 4.56 head turns (SD = 6.38) and a mean range of 71.10° (SD = 78.67) was observed. In contrast to Experiment 1, the mean route similarity was influenced significantly by experience and task demands. A tendency to make more systematic route choices was observed in the wayfinding (M = 81.49%; SD = 0.07%) rather than the search task (M = 70.41%; SD = 11.02%), F(1, 44) = 53.435, p < .001, η2 = 0.548. Significant two-way interactions were also observed between task and both viewing distance and topology. A significant interaction effect was found on distance, F(1, 44) = 5.661, p < .02, η2 = 0.114; however, subsequent simple effects testing did not reveal significance differences in performance between search and wayfinding at the restricted viewing distance (p = .099) or unrestricted viewing distance (p < .100) levels, preventing further interpretation of the result. A significant interaction effect was also found on mean route similarity, F(1, 44) = 10.273, p < .003, η2 = 0.189. Follow-up simple effects testing showed a significant decrease in route similarity in the wayfinding task compared with the search task when the visible distance was restricted (F = 52.80, p < .001). In contrast, simple effects testing revealed that participant’s route similarity increased in the wayfinding task compared with the search task when the visible distance was not restricted (F = 8.05, p < .007). The data for these interactions can be found in Figure 2a and 2b .
Graph of (a) the mean distance and (b) the mean route similarity data for participants from Experiment 2.
Discussion
In Experiment 2, we attempted to disentangle the influence of topology from changes in the viewing distance of participants by either allowing participants to have normal vision of distant objects in the environment, or by restricting vision to only the local environment. This approach allowed us to determine whether topological information and the global size of field-of-view (i.e., the visibility of global environment features) have independent influences on navigation performance or whether the two factors interact with each other in some way.
Results again indicated a consistent influence of topological information on the navigation process. High intelligibility was found to be associated with a decrease in the idiosyncratic path choice and an increase in the efficiency of navigation (i.e., participants spent less time wandering and reached their goal quicker). Similarly, the manipulation of viewing distance was found to produce the same pattern of effects as topology. When vision was restricted to the local environment, allowing only the perception of the shape of a single intersection, participants engaged in less efficient and more idiosyncratic route choices. This result is consistent with a navigation strategy that is driven by the use of distant visual features and is consistent with the suggestions of previous authors who have argued that distant visual features, as measured by integration, can best account for route preferences throughout navigation (Conroy, 2001; Conroy Dalton, 2003; Haq & Zimring, 2003). However, this finding alone is insufficient to account for all navigation performance given that a more complex relationship between the factors of vision and topology was observed.
In fact, perhaps the most surprising result was that a significant additive effect of topology and viewing distance was found on navigation performance throughout the experiment. This pattern of results provides further evidence, not just for the use of both types of information but for the independence of the two factors given that each factor explained a significant amount of unique variance in the dependent variables. More specifically, when vision was limited to only the local area, topology was still capable of influencing performance. This was not predicted due to the fact that the topological factors in the present experiment were computed based on the relationship between local (i.e., connectivity) and global space (i.e., integration). This finding suggests that the visual information within the intersections themselves may serve as a source of topological information. However, this may only be possible within highly intelligible environments, as it has been proposed that low intelligibility may entirely prevent the perceptions of topological information (Hillier, 1996) rather than simply obfuscating the perception of this type of information.
Navigation performance was also shown to be only mildly sensitive to the type of task being performance by the participant. Across the experiment, participants were found to pause for a greater amount of time during the initial search task of the experiment compared with the wayfinding task. After gaining experience with the environment, participants were then shown to be more likely to follow similar paths to other participants in the wayfinding portion of the task when viewing distance was not restricted. This result provides initial evidence for the role of experience within a novel environment in the use of topological information and is also consistent with previous investigations of this behavior (Haq & Zimring, 2003; Hölscher, Brösamle, & Vrachliotis, 2012).
Further evidence for the role of experience in the use of topological visual information was observed in a significant interaction for mean route similarity and a nonsignificant, but similarly shaped, interaction for the length of a participant’s path (as shown in Figure 2). The figure shows a decrease in idiosyncrasy in route choice during wayfinding compared with search when viewing distance is restricted. The converse relationship was seen when viewing distance was not restricted. In a similar vein, distance traveled decreased when viewing distance was restricted and increased when it was unrestricted, but only in the wayfinding task. Both interactions suggest that the influence of viewing distance on the use of topological information is greatest when distant visual information is restricted and wayfinding is required. This pattern of results is inconsistent with the hypothesis that experience does not influence the use of topology (Penn, 2003). There has been no previous experimental research investigating the use of topological information relative to experience within an environment and available viewing distance.
A more complete picture of the separable influence of global field-of-view, operationalized as viewing distance, and topological information on the navigation process was provided by Experiment 2. The results of Experiment 2 not only allow us to better define the relationship between the two factors, such that it is necessary to recognize that distant visual features and topological features are necessary for efficient navigation, but they have also provided evidence that experience and knowledge about the layout of the environment affects the degree to which they influence behavior.
General Discussion
Navigation through an environment, whether familiar or novel to a person, is influenced by the configuration of the surrounding environment. Although previous research and environmental modeling techniques have suggested that the configuration of an environment alone plays a large role in path selection, little controlled research has been performed to dissect the influence of changes in the topology of a city or spatial system on navigation performance at an individual level. In the current study, field-of-view (quantified as central vs. peripheral vision and local vs. distant visual field) and topology was manipulated across two experiments to disentangle the role of these two types of information on the navigation process. Several key results drawn from this study provide vital clues as to the basis for the role of configuration vis-à-vis topology throughout the navigation process.
First, the consistency and magnitude of the effect of topology on the navigation process across both experiments reaffirms the apparent large influence of these factors on the navigation process as a whole. In the current study, the environment with high intelligibility not only produced wayfinding behavior that was more consistent to that of other participants when compared with the low intelligibility environment but also an increase in the efficiency and directness of this behavior. This is most interesting given that the environments were novel to each participant and consisted of primarily configurational cues. Traditionally, accounts of environmental modeling approaches, such as Space Syntax, have argued that the accuracy of the prediction from these measures is a consequence of a nondiscursive, perhaps automatic, use of this type of information throughout navigation as a result of an understanding that urban spaces are designed to achieve the function of navigation (i.e., Hillier, 1996; Penn, 2003). Should participants be unable to perceive or use topological information found within an environment, it would have been expected that participants would perform more idiosyncratically during the search phase of the experiment, regardless of intelligibility level. Instead, despite having no prior experience with the location of the goal landmark, participants showed rather compelling evidence that they were able to use this type of information to navigate through entirely novel environments with ease supporting this fundamental presumption underlying space syntax research. Although it remains an open question whether sufficient prior experience in urban spaces is necessary to observe these efficiencies in navigation behavior resulting from topology, the importance of this result should not be understated.
Second, and perhaps unexpected, the findings derived from the visual field manipulations are also noteworthy. Most striking was the finding that visual field and topology have independent, additive, effects on the efficiency with which participants were able to navigate through the environments. This pattern suggests that although visual factors and a drive to maximize field of view may be important to the use of characteristic routes when navigating through an intelligible environment (e.g., Conroy, 2001; Conroy Dalton, 2003; Haq & Zimring, 2003; Turner & Penn, 2002), they are not exclusively the reason for these observations. To assess the topological properties of the surrounding environment, participants required more frequent head turns throughout the navigation task when vision was limited. This interaction also provides a potential explanation for the finding that a markedly reduced ability for topological measures to predict movement distance and systematicity observed within this experiment as compared with those presented by Penn (2003). Perhaps the best explanation for this is that in previous data sets, the influence of visual field (frequently quantified as an isovist field, see Benedikt, 1979) was contaminated by the influence of topology due to a lack of manipulation of visual field. Accordingly, the observed estimates may be a more pure measure of the predictive power of topology measures, independent from the influence of visual field and visual affordances.
Finally, the current data speak to the potential role of effortful cognitive processing in navigation through novel environments. Prior researchers have postulated that the identification of commonly used pathways at an aggregate level supports the idea that the use of topological information is nondiscursive, insensitive to planning and other cognitive demands (e.g., Penn, 2003). Although it should be stated that other authors have identified the potential role of experience and planning in the use of these factors at an individual level (Haq & Zimring, 2003; Hölscher et al., 2012), it is important to note that these studies took place in real-world environments and were therefore likely influenced by placing particular demands on visual search mechanisms not present in the current experiment due to the lack of rich, and potentially unrelated, visual information, such as landmarks, signs, and other people. A number of studies have suggested that the geometry of space is capable of placing demands on the visual attention system, especially when long sight lines are present (Frankenstein, Büchner, Tenbrink, & Hölscher, 2010; Wiener, Hölscher, Büchner, & Konieczny, 2009). As these sight lines may have been limited or their influence reduced by the specificity of our experimental approach, these effects may not have been as pronounced within our experiment in the same way as one would see in the real world. Despite this, the present study, especially Experiment 2, showed that the restriction of vision to the local space had a profound influence on the use of topological cues in the wayfinding task, but not in the search task, supporting the idea that this process likely does demand effortful processing within a spatial cognitive system when the task places emphasis on goal-oriented wayfinding.
By examining the data from both experiments, a complicated picture of the role of navigation performance appears to emerge whereby topological and visual features have a profound influence on navigation performance. Although innate preference for linear routes in planning (Bailenson et al., 2000) and wayfinding (Conroy, 2001; Conroy Dalton, 2003) may explain part of this effect, they do not account for all of the characteristic, systematic behavior observed across the two experiments. This is especially true of the influence of intelligibility across the whole navigation task. One might expect that a preference for linearity, resulting in systematicity in route choice, would be more likely to be observed during the search phase of an environment, regardless of the intelligibility of the overall environment. However, this was not found to be the case, and the influence was observed across the entire navigation experience. In addition, spatial memory is expected to play a sizable role in the observed effects. One possibility is that the alteration of the topological structure of environments may result in more varied intersection shapes, placing greater demand on spatial memory and spatial memory biases across the experiment that would, presumably, interact with visual field and topology manipulations. In such a case, the complexity of the features of an environment, accumulated across the navigation tasks (in the form of the feature accumulation hypothesis: Montello, 1997), would be more demanding in the low intelligibility environment, resulting in more idiosyncratic movement due to an increase in effectual task difficulty. However, both of these proposals extend beyond the current data set and are therefore important in future investigations into these phenomena.
Despite the limitations of the present study, evidence was found for understanding how topology is capable of influencing the navigation process through the interplay between topological properties themselves and the size of the discrete visual field. The first evidence for an additive interaction between the availability of distant visual information and topology was revealed, while also showing only a minimal influence of experience within an environment. This is suggestive that the findings predicted by the influence of configuration, especially those derived from environmental modeling techniques, are not an emergent property of the manipulation of configuration, but rather the result of a much more complex model of spatial navigation performance whereby topological information is used to enhance or drive navigational route choices. Although further study is needed to identify the precise factors that may be driving this pattern, this study provides vital clues as to mechanisms that are of importance in this process. Generally, these results place emphasis on the need for researchers to pay attention to the construction of their environments when studying navigation to better understand the construct of interest. The structure of an environment is capable of influencing topological measures and the availability of visual information incidentally, potentially obscuring a rich and complex model of spatial processing within any given environment.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by funding provided to Colin Ellard through the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Social Sciences and Humanities Research Council of Canada (SSHRC).