Department of Computer Science
Naval Postgaaduate School
Email:
darken@cs.nps.navy.mil
In the past three years, my work has taken on a new perspective that I originally did not anticipate. During the early stages of my dissertation research at The George Washington University, I discovered that, while virtual environments seem to be similar to real environments in many of the most obvious ways, I often became disoriented in VEs and had serious difficulties navigating in them. This caused me to question how the not-so-obvious differences between VEs and the real world were affecting my ability to navigate. This also made me wonder what other differences existed that would limit the effectiveness of VEs in general. The original focus was on determining principles that could be used to design navigable virtual spaces. This, of course, required some understanding of related fields outside of my own (Im a computer scientist), such as cognitive psychology and geography. While I did not know it at the time, I was becoming involved in a much deeper subject; spatial cognition. While my interest has been, and continues to be, an understanding of spatial cognition as it applies to simulated spaces, the use of VEs has, in itself, presented opportunities for study I had not anticipated. Our current research agenda includes the study of navigation within VEs (e.g., How to build better VE applications through improved navigation.) as well as the study of VEs used as training aids for real world navigation tasks (e.g., How to use a VE to learn how to navigate real environments in general, or, how to use a VE to familiarize oneself with a particular real environment.).
What I bring to this workshop is a perspective
from the computing sciences that is relatively uncommon. Most of my
colleagues in my field do not have an interest in the use of
technology in the study of humans and our relationship to our
environment. In fact, most do not study humans at all in spite of the
fact that all the technologies we deal with exist for the purpose of
serving man. I contend that bridging the gap between "technologists"
and "humanists" is essential to a better understanding of both people
and the machines we use. Within the context of this workshop, a
better understanding of how people navigate VEs may enable us to
better understand how people navigate real spaces. Understanding how
people navigate involves an understanding of cognitive models of
spatial information -- oftentimes dynamic spatial information. From a
practical standpoint, it is important to note that VEs allow
manipulation of variables that cannot be manipulated in the real
world. For example, we commonly vary environmental fidelity (e.g.,
remove objects or detail from objects) in ways that cannot be done to
real spaces.
I have spent considerable time recently on
trying to construct a better model of human navigation from what we
have seen thus far. In particular, models that assume the
navigability of the space (mostly man-made environments) seem to be
easier to handle than more amorphous, or loosely structured spaces
(such as natural environments). This is a topic I would like to
discuss in this forum if possible.
In order for us to better understand how to
solve navigation problems in virtual environments, it seems obvious
that we need to understand more about how people navigate in the real
world. In some ways, however, this point isnt so obvious
because VEs can be so vastly different from the real world that it
could be argued that were talking about a different animal
altogether. I disagree with this position in that people live in the
real world, our navigation skills and strategies are born in the real
world, and therefore if a VE does not leverage this real world
foundation, then the task becomes so abstract that it cannot be
performed at all. Consequently, we are keenly interested in how
people acquire spatial knowledge and how that knowledge might be
represented in memory for use in navigation tasks. While we often use
VEs as a tool for study, our interests are in the general issues
behind spatial cognition and how individual differences,
environmental differences, and interactions with an environment
affect navigation.
It seems apparent that individual differences
play a major role in spatial knowledge acquisition. In addition to
effects due to spatial abilities (e.g., mental rotation,
visualization) which I will not discuss here, we have seen a great
variety of strategies attempted in many of our environments. These
strategies are clearly affected by past experience as well as the
task and environment in question. There is a key element here that I
have yet to see appear in a clear form in the literature. This has to
do with confidence; confidence in ones own skills and
capabilities, and confidence in the available tools (e.g., maps, GPS,
etc.). Individual differences in this one area are so vastly
different that I can safely say that a large part of the variance we
see in performance in some of our navigation tasks can be traced
directly to confidence.
Case in point, a Marine with thirteen years
experience in the field attempts an intermediate level sport
orienteering course and succeeds in finding only one of nine targets
in a full hour. For comparative purposes, in the same experiment, we
had an Air Force Cadet with absolutely no field experience complete
the entire course with time to spare. How can this be? Certainly the
Marine is not such a poor navigator that he cannot do this task. In
post trial interviews, we discovered that entering the task, he had a
very high level of confidence in his ability to perform this task. He
had some ballpark estimate as to how long he expected it to take to
locate the first target, then the second, etc. As he made error upon
error, and the time in which he expected to find the first target
came and went, his confidence began to waver. Initially, he tried to
place blame on the map. He commented on how the map was "wrong" or
was missing important items. The monitor (an experienced Army
Infantry officer) replied that he was absolutely certain that the map
was correct. Later, the Marine began to question himself and his own
abilities. There is no doubt that one officer performing very poorly
on a task he should have been very good at in front of his peer also
had an adverse effect. But as a person begins to question what he
knows about a space and consequently, his ability to do the task at
all, performance plummets. Why this happens and how it can be
alleviated, possibly through familiarization in a VE, is a topic of
interest to our group.
The next topic I want to touch on in this
position statement has to do with the effect environment has on
strategy and consequently, spatial knowledge acquisition. Being adept
at navigation in one type of environment does not guarantee good
performance in all types of environments. Yet, we have seen
individuals who excel in navigation tasks no matter what the
environment might look like (at least as compared to a group of
individuals). This is true even when spatial abilities are not a
factor. It is clear that some people approach navigation problems in
a more coherent, environment independent fashion, while others learn
the keys to success in one type of environment only. These
differences can be highly sensitive. A number of world-class
orienteers from the San Francisco Bay area orienteering club came
down to Monterey to participate in our first study in natural
environments. Some of them noted how their performance was not what
it typically is in their own area because the vegetation in Ft. Ord
(our testing area) is different from what they are used to. This had
an adverse effect on their strategies. For example, some planned
routes through coastal Chaparral (not a good idea) or were surprised
at the height of some of the Oak forested areas. Others from that
group were not affected at all. Their strategies did not depend on
vegetation or richness of contour. Our interest here is in
determining what makes an approach environment independent and
determining if this is a trainable skill.
The last part of this has to do with the actual
interface to the virtual environment. It is well known that spatial
knowledge acquisition is affected by travel mode. That is, spatial
information I attend to when on foot is different from when Im
on my bicycle, which is in turn different from when I drive my car,
which is different from when Im a passenger in a car. This
topic becomes more confused when we introduce interfaces to VEs which
often break all the rules of physical reality. The effects of
interfaces on navigation are largely unknown. Unfortunately, most of
the attention has been given to maneuverability issues of VE
interfaces. While it is important to be able to get from here to
there, it is equally important (if not more so) to understand what
"here" and "there" are and what the spatial relationships are between
them. These, also, are affected by the interface, often
adversely.
Most of this position statement has dealt with
spatial knowledge acquisition. However, we are really interested in the
bigger picture which also includes spatial knowledge representations and
the use of this knowledge in navigation tasks. We have been trying
recently to construct a model of spatial knowledge representations based
on our work in natural environments. It is obvious to us that the flat
landmark, route, survey knowledge (LRS) construct is oversimplified. There
clearly is some hierarchical construct at work here. Attempts to use
cluster analysis on route planning and execution in natural environments
have thus far been unsuccessful. However, we believe that proximity
between targets causes clustering to occur in some fashion resulting in
high performance within a cluster and poorer performance between clusters.
Thus far, we have not been able to show this consistently, at least to our
satisfaction.
The last item I will discuss in this statement
has to do with disambiguation of junctions. In orienteering, we refer
to a "parallel error" as an error made when the orienteer thinks he
is in one place that is spatially similar to where he actually is.
This usually occurs at junctions. One of the reasons why problems
occur with maps is that there often is no information by which to
disambiguate two junctions. If we went there and looked at each
junction, differences would appear, but they are not readily apparent
from only the map. We think this is where a VE can be a useful tool.
This actually falls more into the category of landmark knowledge than
anything else. It has to do with recognizing a place correctly given
whatever information is available. However, even in participants who
perform our tasks very well, we do not yet understand how this
disambiguation takes place. What do they encode and recall that works
for them? Why do other fail where they succeed? Do both successful
and unsuccessful participants recognize the junction as a unique
place or does one or the other use it as part of a path between two
other points? We believe that an understanding of this issue will
lead us toward a more accurate model of human navigation and
consequently, a better understanding of navigation in VEs.
Naval Research Laboratory, Tactical
Electronic Warfare Division 1991 - 1996. Director of the TEWD
Virtual Environment Laboratory.
M.S. Computer Science May 1993. Advisor: John
Sibert. Topic: Navigation and Orientation in Virtual
Space.
2. Position Statement
3. Curriculum Vitae
Naval Postgraduate School
Department of Computer Science, Code CS/Dr
Monterey, California 93943-5118
(408) 656-4072, (408) 656-4083 (fax)
darken@cs.nps.navy.mil
http://interact.nps.navy.mil/darken
D.Sc. Computer Science October 1995. Advisor: John Sibert. Topic:
Wayfinding in Large-Scale Virtual Worlds. Minor areas of study in
Psychology (Human Factors) and Artificial Intelligence