
Maze Experiment #3 Method
Maze Experiment #3
Survey Representations
Maze experiment II established that VE’s can enable
effective
route representations. However, an important question remained: if people
are given enough time in a VE, will they eventually be able to form as
accurate survey representations as they do in the real world or with a
map? Experiment III was conducted to determine if there are any
differences between configurational knowledge formed from VE’s, maps, or
the real world. This experiment used what we feel are two very important
principals experimental designs for VE research: 1.) The independent
variable of primary interest was manipulated within subjects. 2.) The VE
was programmed with features specifically designed to enhance spatial
knowledge.
Participants were given as much time as they wanted to explore two maze
environments: one in either a desktop or immersive VE and one in either
the real world or on a map. Configurational knowledge was measured
primarily through a variant of Siegel’s projective
convergence technique.
The experiment also provided us an opportunity to employ
a new measure of
survey knowledge involving reaction times.
Method
Participants. 66 people (34 women) recruited from the
University
of Washington department of Psychology’s subject pool were given extra
credit in their introductory psychology class in return for participation
in the experiment.
Materials. We designed two mazes to have a similar pattern
of
turns and interobject distances (see fig. M4-1). Each maze contained
five prominent objects--either letters (A--E) or numbers (1--5). The
mazes were computer modeled using WorldUp from SENSE8 on a Pentium Pro
200. Desktop participants viewed the VE on a 35 cm x 26.5cm monitor and
navigated with a four degree-of-freedom joystick (lacking pitch and roll).
Participants in the immersive condition interacted with the VE by moving
in a 6 foot x 6 foot curtained enclosure in the real world. Head
movements were tracked with a six degree-of-freedom tracker (Polhemus
Fastrak). The sensitivity of the tracker was calibrated so that
collisions between the virtual viewpoint and the walls of the virtual maze
corresponded to collisions between the participants’ body and the curtains
in the real world. This set-up allowed immersed participants to navigate
with their body; however, the scale of their physical movements was nearly
tripled in the VE.

Figure M4-1
Whenever they wished, participants were allowed to view their
current position in the maze from a bird’s eye perspective. These views
were shown with the participant’s current direction of travel facing up,
and with their current position and heading represented by a large blue
triangle (see. Fig. M4-2).

Figure M4-2
Procedure: Participants were randomly assigned to receive
either
a map or a real-world exploration of one maze and either a desktop or
immersed VE maze. Participants were made aware of the tasks that they
would be doing, and were given as much time as they wanted to learn each
maze configuration. After learning each maze, participants were tested on
their knowledge in the real world maze. Participants were asked to point
and estimate distances to one of the maze’s objects from each of the other
four. These data were analyzed using the projective convergence
technique.
After pointing and estimating distances in the maze, participants
made interobject distance estimations on a computer. The response times
for these estimations were collected and analyzed as a function of the
number of turns in the maze between them.
Results
We found very few significant differences
between
survey
knowledge acquired from direct exploration, map study, and VE exposure.
Participants took significantly longer to acquire their spatial knowledge
from a VE (M = 572 seconds, SD = 260 s.) than from maps (M = 194 s; SD =
155 s. ) or a real-world walk through (M = 277 s; SD = 102 s.). Moreover,
people took significantly more time to learn from an immersive VE (M = 488
s., SD = 222 s.) than from a desktop VE (M = 378 s., SD = 154 s.).
Despite the different amounts of time spent learning, there were very few
differences in the accuracy of the configurational knowledge acquired. A
2 (real/map) x 2 (desk/immersed) MANOVA using locational error,
consistency, mean angle error, and mean distance error as dependent
variables did not reveal any significant effects of learning environment
type. The effect of the environment on locational error and consistency
is shown in figure M4-3.

Figure M4-3
In general, the reaction time measure was noisy and difficult to
interpret. Correlations between response times and the number of turns
between objects were calculated for each participant. These correlations
averaged -.05, -.01, -.06, and -.02 for real world, map, desktop, and
immersive exposure respectively. Although there was no significant
difference between these conditions, confidence intervals for their
differences were relative wide. We have since concluded that this measure
of survey knowledge is not particularly sensitive to the route/survey
distinction.