![[Previous Chapter]](/icons/button-back.gif){kind=icon}
![[Table of Contents]](/icons/button-contents.gif){kind=icon}
by Mark Draper
6.1 General Discussion
The three studies documented in this thesis mark the beginning of an exploration of VB influences on spatial awareness in VEs. Although a specific VB effect was not found, many lessons were learned to aid future research in this area. Also, significant findings did result that may aid future efforts in the areas of virtual context influences and object height/virtual reach estimation. What follows is a general discussion of the findings as they relate to VBs and spatial awareness in VEs, along with a presentation of thoughts/lessons learned. Lastly, recommendations are given to guide future research.
The most significant finding (statistically speaking) of this thesis is the importance of context in VEs. Whether it be as furniture in a room or as a tiled floor, the availability of context seems to aid virtual performance of spatial tasks as it does spatial performance in the physical world. The textured, tiled floor presented a surprise in its strength as a major contextual aid for search-and-replacement tasks. All subjects immediately gravitated to it as the primary source of spatial information and it was no doubt the major reason behind the ceiling effect that was observed in Studies 1 and 2.
Why did the VB fail to show an effect in any of the studies except for the ambiguous height-VB interaction in Study 3? This question is especially puzzling given the several positive comments by subjects regarding its potential use as a source of spatial information. Many subjects in each study claimed to use the FVB to aid their performance in the spatial task at hand (in ways that support the experimental hypotheses) and the majority preferred the FVB over the other VB configurations. Why then did no VB effect emerge statistically?
There are several plausible explanations. The first is that the hypothesized VB effect may be smaller then expected. Given the observed ceiling effects in the first two studies, the coarseness of the movement actuator in the third study, and the relatively small sample sizes in all the studies, it is quite possible that the experimental designs were not sensitive enough to reveal the subtle effects of a FVB. A replication of these efforts (after modifications to improve on the listed deficiencies) with an increased number of subjects would increase sensitivity towards observing small VB effects.
As previously mentioned, the FOV of the helmet is another major influencing factor on the failure of the VB condition to show an effect. Subjects must be able to see their VB sufficiently in any condition for its potential to be adequately tested. Subjects in these experiments only occasionally viewed their legs and torso in the course of the experiments. Their `awareness' of their VB configuration was often limited to seeing the virtual arm. Sadly, current HMDs cannot provide near the FOV comparable to the human visual system. The trick would be to find a way to sufficiently present a FVB image while living within the constraints of today's technology.
A related issue to the above was a general tendency by subjects not to want to look down at large pitch angles while in the virtual worlds. This could be due to several reasons including uneasiness with the technology, the potential for disorientation, the weight of the HMD, and concern that the HMD could slip off their heads (there was a tendency for the HMD mount to slip slightly during use). The end result of this tendency was to exacerbate the FOV limitations of the HMD and minimize the impact of the visual stimulus on trials involving the FVB.
A review of post-test comments indicated that some subjects may have resorted to switching strategies between trials in an attempt to optimize task performance in all three studies. Different strategies in the search-and-replace studies would result in different subject viewpoints, which in turn could result in different solutions to the spatial task as demonstrated by Toye (1986) in his work on spatial layout. Therefore, the potential existed for strategy to be a confounding variable in these experiments.
Body fidelity is another possible explanation. The initial premise taken during experimental design was that FVB movement fidelity is of less importance then the providing of an accurate static image of the body for use as a reference point. Although attempts were made at generating a FVB with increased movement, the resulting side-effects (i.e., lower update rate, increased time lag, etc.) were deemed far worse then the resulting higher FVB fidelity. Also, since this effort was clearly not focused on the concept of presence, VB fidelity was seen as secondary to an accurate static image. However, it is now believed that the lack of VB fidelity may have actually helped prevent its use. In general, subjects did not feel an association with the VB except for the arm and the detached shoulder occasionally distracted subjects while they performed the tasks.
Lastly, it could be quite possible that there is no effect of VB image on spatial awareness. Perhaps a reason so many subjects felt the FVB aided performance was because of how the question was worded (the question was positively instead of neutrally worded). However until all the above explanations have been adequately tested and eliminated as limiting factors, this assertion should not be accepted as truth.
One aside in the area of lessons learned. The FVB used in this study was developed and calibrated using males as subjects. In pilot-tests it became apparent that this FVB and the FVB tracker-sensor would need major tweaking to provide a consistent satisfactory image for female subjects. Female hips are generally higher and more curved then those of males, which caused the sensor belt to tilt and slip and cause the FVB to be placed at a high, tilted location. Eventually, the solution was found in packing material available in the lab. This material was used to align and hold steady the FVB sensor at a lower location on the hips, minimizing the problem. As a result of these issues, though, females were not actively recruited to serve as subjects (although all who signed-up did participate). The lesson here is that both a male and a female FVB may be required in future studies to avoid the one-size-fits-all strategy.
6.2 Recommendations for Future Research
Any future studies that attempt to test the influence of a VB on spatial awareness will need to make every effort to design experiments that have high power/sensitivity. The more powerful the experiment, the more likely that a VB effect will be identified if one exists. This increase in power could be obtained through a sound choice of task and by utilizing a much larger sample size than was used in these studies.
If search-and-replace tasks are used, efforts need to be made to reduce the ceiling effect found in the first two studies, while efforts should be made to hold strategy constant within subjects. The tiled floor should be replaced with one that has a less grid-like texture (i.e., woodgrain) to increase task difficulty by taking away a powerful contextual cue for spatialization. Time limitations might be increased to study the effect of stress/workload on VB influences. It is not recommended that the number of targets be increased very much however, as pilot studies for this thesis indicated that five targets would cause frequent severe errors to occur.
An interesting thought generated during this research was that the NVB is really a misnomer. The NVB used in this work is in essence a VB with an arrow as a virtual hand. Perhaps a `real' NVB configuration should be included in future search-and-replace studies where subjects learn cube locations with no arrow but they are provided the arrow for the replacement phase (to facilitate picking cubes). This approach would be a more literally correct NVB configuration.
Future studies should increase the fidelity of their FVBs over the present FVB configuration in terms of fit (maybe separate VBs for males and females) and movement. Using a more complete FVB like Jack (Badler, et al., 1993) would allow for a more veridical test of a FVB influence by increasing subject association with it and quite possibly utilization of it. Increased fidelity would also minimize distractions often caused when the FVB did not behave in an expected manner. A fully articulating FVB that actually `walked' using accurate virtual paces would also have the benefit of reacting correctly to any subject movements and the potential to greatly increase spatial knowledge acquisition by providing more ecological cues to the subject (i.e., Loomis, et al., 1992). This would also eliminate restrictions on subject movement implemented in Study 2 and bring future studies closer to the ultimate goal of VR of allowing intuitive interaction without limits.
While working the VB fidelity issue, why not put `clothes' on it? The VB image used in these studies was one color. This made it difficult for those subjects using the FVB to determine where the cube and body aligned on search-and-replace tasks. A simply-textured pair of shorts, a nice Polo shirt, and a dark belt would increase the number of reference points available for use on a VB.
An obvious recommendation when pursuing future VB issues would be to employ a HMD with the largest FOV possible, especially vertical FOV. However, even vertical FOVs of 90 degrees may not be enough to provide continuous access to the FVB image presentation. An interesting work-around strategy would be to redesign the FVB so that it almost always is presented within the subject's limited FOV. Perhaps the subject could be in a virtual vehicle, which acts as his/her virtual body, or maybe the subject could wear a device similar to those worn by stadium concessionaires (i.e., those that sell popcorn, beer, etc.) and have the VB image also contain this protruding image into the VE. The goal here would be to solve the VB awareness problem while living within today's technology constraints.
Future virtual reach estimation studies should attempt to replicate the height-VB interaction discovered in Study 3. If a similar interaction is found, attempts can begin to be made to dissect the spatial modifications caused by different VBs at defined target heights. If it is not, it can be better laid to rest as a type 1 error in the statistical analysis.
Future research may also want to examine the adaptation effects surrounding the existence and use of a VB in a VE. Previous research in prism adaptation may offer clues about how humans may adapt to a VB or even the entire VE, resulting in performance that is different than that obtained when experimental conditions were first introduced.
If context is the primary factor under consideration, perhaps a condition should be added that allows context in the memorization phase but not in the replacement phase of search-and-replace tasks. This would test the positive (or negative) carry-over effects of context on spatial awareness and could have direct relevance to several virtual training issues. Also, investigation of different floor patterns could provide important information for designers of virtual worlds for spatial tasks.
Some general guidelines for research will now be briefly mentioned, as they were learned the hard way by this author and may be of some benefit to future students. First, "suck and see" is the most effective way to carry out basic, exploratory research. This means trying out ideas and evolving research empirically versus spending a large amount of time planning one study with little time to make revisions. However, "suck and see" is not a substitute for sound logic, detailed design, and a thorough review of the literature that are also primary elements of competent research. Rather, it should be viewed as a tool for refinement and enlightenment that complements the other areas. Second, plan a schedule and then allow two to three extra months for hardware/software contingencies, on-the-job training, software development time, and "suck and see" time. This is especially true if the focus area has been relatively unexplored. Third, subjects will provide surprises and will always find a way around the `intent of the rules'. For example, one pilot subject attempted to separate the free floating right arm from the body for use as a detached virtual ruler! Lastly, it is important in most cases to maximize design efficiency and minimize actual study time when testing in VEs. Today's VR technology is far from optimized and poor visual images, a heavy helmet, and/or slow tracker performance will rapidly drain the subject. This could result in a requirement for frequent rest breaks, the potential for negative practice effects due to fatigue, or even the loss of subjects through motion sickness or discomfort.
6.3 Conclusion
The literature review and three studies documented in this thesis represent the first known attempt at exploring the relationship between VBs and spatial awareness in VEs. The task in these studies varied from a conventional search-and-replace effort to an ecologically-based estimation of maximum virtual reach. Factors manipulated included the VB configuration, target size, target height, and level of surrounding context in the environment. Although a specific VB effect was not found, it is believed that the general research question has become more focused as a result of this effort. Ideas were generated, deficiencies were documented, questions were raised, and lessons were learned hopefully to aid, inspire, and guide future research efforts in this area. It is the strong belief of this author that this area offers the potential for many significant findings that could shape the design and development of future VR worlds in many areas including simulation, training, and architecture. The exploration has just begun.