This study examined the effects of two variables on spatial learning: the objects' array (with or without a central reference point) and the learning method (virtual walk through or seeing still images). After learning the objects' locations and their positional relations in a virtual room in egocentric reference frames, participants judged the directions to a target from an imaginary position in the room. The results revealed that having a central reference point facilitated learning, especially in the virtual-walk condition with a central node. The findings are discussed in terms of the interactions between the effect of the central reference point and virtual walk and the relationship between intensive encoding in egocentric reference frames and stored representations of the layout in environmental reference frames.
In everyday life, people move about in their surrounding
space, they get out of bed, walk into the kitchen for coffee,
go into the garage, and drive to their office. Such daily
spatial behavior is based on spatial cognitive functions in
which we comprehend positional relations between our
bodies and objects. Because spatial locations are essentially
relative, reference systems to describe position and
orientation are important
When walking around a familiar town, people rely on their
spatial memory of the environment, including elements such
as routes, landmarks, and their layouts. Many studies
indicate that spatial layouts tend to be organized in the
frame of environmental reference systems rather than
egocentric reference frames
Learning spatial layouts has been a major topic of spatial
cognition, and much research has been conducted with
respect to the reference frames. However, fewer studies
have examined a layout’s own effect on spatial learning.
Some studies have focused on layouts of objects in learning
tasks set in middle-scale spaces.
In the present study, we examined whether differences in layout of objects affect spatial learning, using a freeexploring task that invokes egocentric reference frames in a middle-scale virtual space. Specifically, using post-tests, we compared the learning results of two different layouts: one consisting of four objects arranged like spots on a die and the other consisting of five objects in which one more object was added at the center.
Our previous research using a real labyrinth
Although our experiment was conducted to investigate the effects of the different types of directional inference (unidirectional versus multidirectional), in the course of examining the results, we conceived a hypothesis that along with the updating, the difference in the reference points’ layouts recognized by the participant as a consequence of their inferences might also have resulted in the superiority of the Non-Circle-Order condition. In particular, we presumed that the participants in the Non-Circle-Order condition recognized the intersection as a fifth reference point in addition to the target locations, whereas those in the Circle-Order condition recognized only four reference points in the target locations. In our experiment, the locations of the targets ought to have been recognized as important reference points, but the significance of the intersection might depend on the conditions. For those in the Non-Circle-Order condition, the intersection was more important than for those in the Circle-Order condition, because it was only in the former condition that the participants needed to pay extra attention to determine the multidirectional relation in the egocentric reference frame and choose their way. In contrast, in the Circle-Order condition, the participants could automatically turn left or right during the revisiting after perceiving the layout.
The term “reference points” was originally used in
categorization in cognitive science
If the central point acts functionally like an anchor point
that organizes other spatial information into a layout
Based on the hypothesis that the fifth central reference point added to the four reference points serves as an anchor that organizes other points into a layout and facilitate spatial learning through navigation, we conducted an experiment to examine the effect of the central reference point when learning via virtual navigation and from multiple vantage points using still images. Via virtual navigation, one might well be able to recognize the relations between the reference points in egocentric and environmental reference frames, whereas via still images, one basically see the objective relations of the reference points shown by images.
One factor was the array of objects: Square or Central Node (Figure 2). The other factor was the learning method: Virtual-Walk or Still-Image.
Figure 2: The Arrays of Objects.
A virtual circular room with a diameter of 12 m (Figure 3), generated by CAD software (shade dreamhome 2.0.3) was used in all the conditions. Four common objects (Figure 4) and curbs indicating the front of the objects were placed, and a central node object (Figure 4) was added at the center of the room only in the Central Node conditions. Six kinds of rooms that included all combinations of the placement of common objects were prepared, and each one was assigned to four participants in four groups. In the Virtual-Walk conditions, participants were seated in front of a computer display and operated a keyboard for the virtual walk through in which continuous images from the perspective of a virtual camera (height: 1500 mm, field-of-view: 80 degrees) were shown according to the key operation. The starting points in the Virtual-Walk conditions were in front of a common object that varied between the four participants assigned to each kind of room. In the Still-Image conditions, still images of the six kinds of rooms from 12 vantage points (three perspectives for each common object, see Figure 5) were shot by the virtual camera in the same terms as the Virtual-Walk conditions. Each participant watched the four sets of three images in front of the same display. For practice in all the conditions, another virtual circular room with a diameter of 5 m in which three objects and curbs were placed was used. Images of the practice room for StillImage conditions were shot similarly as for the experimental room.
Figure 3: Images of the Virtual Circular Room from Virtual Camera.
Four Common Objects (cue objects in the post-tests) and Node Object (the potted plant)
Ninety-six undergraduate and graduate students with normal vision were randomly assigned to each group. The ratio of male to female was the same in each group (12 male and 12 female). The average ages of each group were 22.5 (SD: 4.65; Square-Walk), 22.4 (SD: 4.02; Square-Still), 21.0 (SD: 3.60; Central Node-Walk), and 21.4 (SD: 2.74; Central Node-Still).
The experiment was divided into three phases: practice, exploring, and post-test. At the beginning, participants were instructed to remember the objects, their locations, and the positional relations between them.
Practice Phase The participants in Virtual-Walk conditions explored the practice room for one minute, whereas those in the Still-Image conditions saw images of the room for one minute. Then, participants in both conditions took Post-test 1 so that they would consciously remember the relative positons of all objects in the exploring phase. Throughout the practice phase in both conditions, experimenters monitored and checked whether the participants understood what they needed to judge in the post-test.
Exploring Phase At the beginning of the phase, the participants in the Virtual-Walk conditions were instructed to move to four positions where they could see each common object in front of them, and then allowed to move around the room freely. Their walking trajectories were recorded by a screen capture program. The overall time of the virtual-walk was four minutes. In contrast, those in the Still-Image conditions randomly saw the four sets of three images showing each object from three vantage points. Each set was presented four times for 15 seconds each (5 seconds per image) so that the overall viewing time would be four minutes.
Post Test Phase Two post-tests and a sketch map test were conducted for each group. In Post-test 1, participants were asked for directions to one of the common objects from imaginary locations. A cue target was presented first, followed by a fixation point, and then the target object (Figure 6). The participants selected one of the keys to indicate directions to the target objects as if they were standing and facing a cue object. One set of 12 randomized trials, including all possible combinations of two objects, was conducted twice.
In Post-test 2, participants were asked for directions to one of the common objects while assuming they were standing in the center of the room and seeing a cue object in a particular position. The cue object indicated which direction one was facing. After the first picture showing the cue object, the target object was displayed following a fixation point, and the participants selected one of the directions in the same manner as in Post-test 1. One set of 24 randomized trials with all possible combinations of two objects was conducted twice.
Finally a Sketch Map Test was conducted in which participants wrote down, on a circular piece of paper, the names of the objects and curbs in the position they experienced during the exploring phase.
Post-test 1
Post-test2 + + Cue Object 500 ms Target Cue Object 500 ms Target 2500 ms 2500 ms “if you were standing and facing an object “if you were standing in the center of the TXh,esecloercrtetchteandsiwreecdrtsi:obn, odf, aortfarget”. rpToihcoetmucroaern,resdcetsleaecnetisntwhgeearsdn:irao,ebccjte,iceot,noXorfgatsheintathreget”. c b a e f The participants selected g one of the seven directions.
The sketch maps drawn by the participants were checked to determine whether the locations of objects were recalled correctly. Eleven participants (three in Square-Walk, three in Square-Still, three in Central Node-Walk, and two in Central Node-Still) could not recall the objects and/or put them in the wrong positions. Since some of the 11 participants seemed to abandon the judgment in the post tests (e.g. selecting the same directions in any trial), we decided to exclude the participants from the analyses based on the success or failure of the map.
After the angular transformation, the mean error rates in each post-test were analyzed by a two-way ANOVA. Results for Post-test 1 revealed a main effect of the learning method [F(1,81) = 38.69, p < 0.01] and a significant interaction effect [F(1,81) = 4.98, p < 0.05] (Figure 8). Results for Post-test 2 revealed main effects of the array [F(1,81) = 11.40, p < 0.01] and learning method [F(1,81) = 5.74, p < 0.05], and a significant interaction effect [F(1,81) = 5.10, p < 0.05] (Figure 8). Another two-way ANOVA was performed with only the results of the Virtual-Walk conditions in Post-test 1, using the following factors: the array and correct directional responses (front, diagonally forward left and left-hand side). The analysis revealed only a main effect of the layout [F(1,40) = 6.15, p < 0.05].
The mean response time (Figure 9) in each post-test was also analyzed by a two-way ANOVA, and a main effect of the learning method was detected in Post-test 1 [F(1,81) = 7.73, p < 0.05] and Post-test 2 [F(1,81) = 4.95, p < 0.05]. Except for two participants’ trajectories that could not be recorded due to technical difficulties, 40 recorded trajectories (19 in the Square-Walk and 21 in the Central Node-Walk) were examined by the experimenters. In both the Virtual-Walk and Still-Image conditions, after visiting the common objects as instructed, some participants kept moving from one zone to another (see Figure 8) with short stops, while others were more likely to stay at some locations longer and overlook the room. Among the behavioral patterns common to the groups, frequencies of zone migration and staying at the center zone (turning left or right to overlook the room for more than 10 seconds) were counted (Table 1). Two-sample t-tests were performed on the mean values of the zone migration and staying, and there was a significant difference only in the frequencies of staying [t(38) = 2.19, p < 0.05]. Analysis of the correlations between each value and the mean error rate in each post-test was performed, and a significant possible negative correlation was found only between the migration frequency and the mean error rate in Post-test 2 in the Central NodeWalk condition; however, a regression analysis revealed no significant relationship between the variables.
0.80 0.70 0.60
We hypothesized that, when learning the reference points through navigation, the central point would serve as an anchor that organizes other points into a layout, which then facilitates layout learning. The results supported the hypothesis and suggested that the central reference point is more effective when learning the spatial relations of objects via virtual virtual-walk than via still images from multiple viewpoints. The number of participants who failed the sketch map test in each group indicated that, regardless of condition, nearly equal numbers of participants could remember and recall the objects’ locations. However, in Post-test 1, when participants were asked to indicate the direction to a target object, the presence of the central reference point led to a major difference in the judgments, which varied according to the learning method. When learning occurred via still images, the central point did not affect the judgments, as suggested by the absence of a main effect of array; however when learning via virtual-walk, the central point seemed to affect learning. This could be explained by an interaction between the factors, in which mean error rate in the Central Node-Walk condition was notably lower than the rates of the other groups, and a main effect of the layout in the ANOVA with only Virtual-Walk conditions.
In Post-test 1, learning method affected the judgments due to the large difference between Virtual-Walk and StillImage conditions in the error rates and the response times. The difference in the error of judgments would arise from a qualitative difference in layout representations between the learning conditions. The cognitive manipulation performed by the participants in the first test can be speculated as follows: recalling a layout stored in the environmental reference frames, reorienting a target object in the egocentric reference frames from the layout, and imagining being in front of a cue object. Of course, one could provide the directions using representations only in the egocentric reference frames by remembering 12 possible positional relations from one target to the others, but it is rather unlikely that many participants would have applied such a cumbersome and uneconomic strategy. Instead, it seems that participants remembered what they were asked in the first test and made an effort to remember the layout for the posttest. Considering that the participants in both conditions remembered the layouts equally well, it could be inferred that those in the Virtual-Walk condition could reorient a target object in egocentric reference frames from the imaginary locations more easily and faster than those in the Still-Image condition. This would mean that the representations in environmental reference frames in the
Virtual-Walk condition were more elaborated than those in the Still-Image condition. However, the most influential condition was the Central Node-Walk; participants’ representations in that condition might have been the most elaborated. Indeed, the ANOVA performed on the VirtualWalk conditions using the correct directional responses as an extra factor suggests that overall positional relations of the objects as layout were learned better in the Central Node-Walk than in the Square-Walk.
In Post-test 2, the participants did not know that they would be asked for directions to common objects from the center of the room, so they could not intentionally remember the positional relations from the center point in egocentric reference frames during the exploring phase. Therefore, when they judged directions, they had no choice but to reorient the positional relations between one’s body and the targets using the representations of the layout in environmental reference frames. The results of the second test also support the hypothesis. Although there were main effects of the two factors, the significant interaction effect and the finding of the highest percentage of correct answers in the Central Node-Walk condition showed that the two factors had positive effects on mainly the Central NodeWalk condition. In contrast, the error rates did not differ greatly between the Square-Still, Square-Walk, and Central Node-Still conditions, although the participants in the Virtual-Walk conditions, including the Square-Walk, could answer quicker than those in the Still-Image conditions.
Analyses of the participants’ trajectories suggest that the participants in Square-Walk recognized that the center zone was an important vantage point for layout learning, as they tended to stay in that zone and overlooked the room with a higher frequency than those in Central Node-Walk. It could be interpreted that, at the center zone, they paid attention to the positional relations of the four common targets in their view and tried to remember the layout in environmental reference frames, instead of thinking of the relations between the targets and themselves—standing at the center—in egocentric reference frames. In contrast, the participants in Central Node-Walk who thought they had to remember the relations, including the central node object, would have intensively encoded the relations from the central node object to the others in environmental reference frames at the center zone near the object.
Lastly, we would like to suggest the process whereby the
central reference point act like an anchor point in our
experiment. The first factor lies in the general feature of the
layout. When connecting the objects with a straight line, the
lines in both conditions overlap each other. However, the
diagonal relations in the Central Node Array are segmented
by the central node object. The segmented components
might result in elaborate layout representations and facilitate
accurate judgments because the components allow one to
describe and encode more diverse positional relations. The
second factor lies in the superiority of the virtual walk
through over learning by still images. Learning structures
via navigation and wayfinding involves the integration of
local perspectives and views that a traveler has learned
independently
The present study revealed the effect of differences in layouts with and without a central reference point under the condition of a virtual walk through. The intensive encoding of the positional relations from the central reference point by virtual navigation apparently results in elaborated layout representations. In our previous experiment using a real labyrinth, a process similar to that in this experiment would have occurred, although there was no obvious central object indicating a prominent location in the labyrinth. Future work should confirm the speculations arising from the present findings by clarifying (1) that the layout contains a distinctive reference point that serves as an anchor, and (2) the relationship between encoding a reference point in egocentric reference frames and stored representations of the layout in environmental reference frames.