Exploring the Influence of a Virtual Body on Spatial Awareness

by Mark Draper

CHAPTER 5 - STUDY 3: VIRTUAL REACH ESTIMATION STUDY

5.1 Overview

This task involved perceived reachability to an object as the manifestation of spatial awareness (Figure 24). Reachability has been used for many ecologically-based distance estimation studies (Carello, et al., 1989) but this effort considered virtual reach to a virtual object as a function of VB configuration and object height. Given the obvious differences between virtual and real environments (body image, image quality, FOV, time delays, etc.), no assumption was made as to the overall accuracy of perceived virtual reach as compared to perceived real-world reach (or of actual virtual reach to actual real-world reach for that matter). However, it would be interesting to see if the existence of a VB would affect the relative performance of perceived virtual reach to actual virtual reach. Perhaps then some hypotheses could be made as to the importance of one's real body image on perceived reachability in the real world.

Figure 24: Study 3 - Virtual Reach Estimation Study

This task involved first learning the extent of one's virtual reach and then making reachability estimations to a floating target. The same VB configurations that were used in the previous study were used here (FVB, NVB), along with an additional `hand only VB (HVB)' configuration. The HVB was similar to the NVB except that the white arrow was replaced by a correctly scaled (approximately) right hand. Three different target heights were also included (low, medium, high). The rationale for the three target heights was as follows. At the low level there was a strong potential for a large section of a FVB to be within the same visual image as the target, at the medium level there is less potential for the FVB to be included in the same image as the target (unless the subject actively attempted to include both images), and at the high level there would be no chance for a FVB to exist in the same visual image as the target. Therefore, by varying target height, what was also varied is the potential influence of a VB on estimated reachability decisions.

It was hypothesized that the FVB configuration would result in more accurate reachability assessments for the low and possibly the medium target heights then the HVB or NVB. There would be no effect of VB configuration for the high target height. There would be no difference between the HVB and NVB conditions. A particular overall effect for target height was not hypothesized.

5.2 Methodology

5.2.1 Subjects

Nine subjects ( 8 males, 1 female) from the University of Washington volunteered to participate, ranging in age between 21 and 37 (average age = 28). A height restriction was imposed (due to limitations in varying the VB size) so that all participants had to be between 5'6" and 6'1" tall. Actual subject heights ranged from 5'7" to 6'0". All subjects were right-handed. All had normal or corrected-to-normal visual acuity with contacts. Seven of the nine subjects had experienced VR in the past (two had experienced it 2 to 5 times, five had experienced it 11 or more). Four subjects had just completed an earlier VR study and received a 15-minute break between studies.

5.2.2 Experimental Design

A 3 x 3 within-subjects, factorial design was used. A within-subjects design was necessary due to the large individual differences found in previous research on spatial estimation tasks and the relative stability/consistency of these estimations per individual over time. In addition, the effects of a VB on spatial awareness may not be large requiring the added sensitivity and power of a within-subjects design.

The two independent variables were VB configuration (FVB, HVB, NVB) and object height (low, medium, high). The dependent variable was the averaged absolute offset error (in inches) between perceived maximum virtual reachability and actual maximum virtual reach. A short post-test questionnaire provided a few subjective ratings for review (Appendix A).

Each VB/target-height condition was counterbalanced (cyclic square) across trial order to minimize potential general practice effects. Each subject repeated his/her particular trial order twice. The averaged result for each trial was used for the analysis. Appendix B shows the experimental ordering for this study.

5.2.3 Apparatus and Stimuli

The apparatus, experimenter's control station, and virtual room were the same as that used in the first study, with the exception that the room always remained empty except for the target.

The virtual target used as the stimulus for this study was a virtual tennis ball (vTB). It was 2.5 inches in diameter and colored canary yellow to be equivalent to a real tennis ball. The vTB would appear randomly anywhere within an area defined by 1) the pre-defined height level for that particular trial and 2) an imaginary line segment stretching diagonally across the room from one room corner to the opposite corner, truncated within 5 feet of each corner. The exact height for each vTB level would vary among subjects, as it was affected by different body heights and different arm lengths. To standardize the process of determining exact tennis ball height, the angle between a subject's extended right arm and his/her torso mid-line was set to three different values (30 degrees, 40 degrees, 115 degrees). The angles were set through prior empirical tests so that at the low level there was a strong potential for a large section of a FVB to be within the same visual image as the target (Figure 25), at the medium level there was less potential for the FVB to be included in the same image as the target, and at the high level there was no chance for a FVB to exist in the same image as the target. Once the subject's extended arm was set to each angle (without bending, stretching, or twisting), the height from that hand's fingertips to the floor was used to set the vTB height for that particular level. A custom-modified T-bevel was used to capture/record the three arm angles.

Figure 25: Perspective View (FVB/low target height condition)

The FVB and NVB configurations were as described earlier. The HVB configuration consisted of a standard gray right `hand' provided by dVISE and scaled to approximate the size of a human hand (Figure 26). This virtual hand moved where the real right hand moved but did not grasp or have moveable fingers. The HVB is typical of many standard VR self-representation images. The ability to virtually touch was limited to a one inch section at the far end of each VB configuration and these `touch zones' were equated/standardized across VBs to guarantee that each VB configuration had the same virtual reach extent. Figure 27 displays a perspective view of reach involving all three VB configurations.

Figure 26: HVB Configuration

Figure 27: VB Reach Images (from virtual eyepoint)

5.2.4 Task and Procedure

Each subject first received written and oral instructions on the task to be accomplished (Appendix C). After reading the instructions and asking any questions about the task, the subject was tested for normal visual acuity using the Snellen near-point acuity test on a Keystone Ophthalmic Telebinocular. The subject was then measured for body-height so that the correct VB size could be matched to the subject. To standardize the process of determining exact vTB height, the angle between a subject's extended right arm and his/her torso mid-line was set to three different values (30 degrees, 40 degrees, 115 degrees) and the height between the right hand's middle fingertip and the ground was measured at each angle. These data were then entered into the computer to set the three vTB heights for that subject. Once these measurements were taken, the subject was fitted with the HMD, handed the 3D joystick, and given instructions on how to move in VR. The subject was then given repeated instructions not to bend, stretch and twist during any practice or experimental trial. The experimenter continually monitored and corrected the subject if any of these unwanted movements were observed.

After offering time for the subject to get acquainted with his/her virtual surroundings, the subject was given a series of 9 practice trials to become familiar with his/her actual virtual reach, with three practice trials for each VB configuration (one occurrence of each target height in each VB configuration) occurring in random order. It was emphasized that this was the designated time for the subject to get educated on the extent of his/her virtual reach without bending, stretching, or twisting in any way. The subject would start in the corner of the room facing out towards the opposite corner. A target vTB was presented directly in front of him/her at a pre-defined height along an imaginary diagonal line connecting the two corners. The subject would move directly toward the target (making no off-line movements) and practice actually touching the vTB as many times as desired in each trial (after the minimum of three touches), again in the hopes of learning his/her maximum virtual reach. The vTB would flash red when touched, providing feedback to the subject. Each touch was also recorded automatically and the maximum reach for that trial was averaged with the maximum reach for each of the other two trials at that same vTB height to determine the subject's maximum virtual reach for that height. When the subject wanted to advance to the next practice trial, he/she would verbally report to the experimenter. The subject would then be instantly transported back to the corner of the virtual room and a new target configuration would be provided. After these 9 practice trials, the experimental trials would begin.

A subject began each trial in the same corner of the room as with the practice trials, with a particular VB configuration and looking outwards into the room towards the opposite corner. The vTB would then appear floating (as in the practice trials) at a set height. The subject then proceeded to move directly toward the vTB without moving off-line, and positioning himself/herself so as to be able to just touch the vTB without bending, stretching, or twisting. The subject would NOT actually attempt to touch the target during the experimental trials, only to position himself/herself at a maximum reach distance. When the subject stated that he/she was comfortable with the chosen position (there was no time limit), the perceived virtual reach was recorded. No feedback was provided to the subject during the experimental trials. This procedure was repeated for all 9 treatment conditions and then the entire order was repeated. At the end of the second trial, an average perceived virtual reach for each of the nine conditions was computed for each subject from which an offset error (the dependent variable) was generated.

5.3 Results

Abs. estimated maximum reach offset error (averaged) was the dependent variable analyzed for this study. The two separate estimates of maximum virtual reach for each condition were first averaged and the result was then used to calculate the abs. offset error from the actual maximum virtual reach for that target height. After detailing the results using this variable, a section is included to summarize the results of the post-test questionnaire.

5.3.1 Abs. Estimated Maximum Reach Offset Data

The data satisfied the normality assumption as is with no transformations or need to remove outliers. The mean abs. estimated maximum reach offset error for each independent variable and condition is shown in Table 10. A two-way repeated measures factorial anova revealed a significant main effect for height (F(2,16) = 9.29,

p < 0.002). However, the Mauchly Test for Sphericity indicated a violation of the assumption of homogeneity (W = 0.535, p < 0.12) for this test. Therefore, the Huynh-Feldt epsilon correction was used and the results were still significant (F(1, 12) = 9.29, p < 0.025). The VB-height interaction was also significant (F(4, 32) = 5.04, p < 0.003). However, the Mauchly Test for Sphericity indicated a violation of the assumption of homogeneity (W = 0.135, p < 0.17) for this test also. The Huynh-Feldt epsilon correction was used and the interaction remained significant (F(3, 27) = 5.04, p < 0.01).

Table 10: Mean Offset Errors - Abs. Estimated Max. Reach (inches) (Study 3)

   Height/VB      NVB    HVB      FVB    
Low Target        1.1     1.9     1.1    
Medium Target     2.3     3.3     2.0    
High Target       3.3     3.1     4.2    

This VB-height interaction (Figure 28) reveals a general trend towards larger errors as target height increased, however the interaction is disordinal. The HVB condition resulted in larger mean errors during the lower two target heights but the FVB appeared to have caused the largest errors in the high target condition. The NVB, acting as a sort of baseline, appeared to rise in a generally linear fashion. A post hoc contrast comparison using the Dunnett correction test was performed on the VB levels at each height. The first contrast, comparing the HVB to the other two VB conditions in the low target condition, was not significant. There was no significant difference between the three VB configurations at this level (given that the means of the FVB and NVB are exactly equal). The second contrast, testing the same comparison at the medium target level, was significant (F(1,8) = 6.18, p < 0.05). This indicates that the HVB condition resulted in significantly larger errors then the other two VB conditions. At the high target level, the contrast tested compared the FVB to the other two VB conditions. This contrast was also significant (F(1,8) = 7.11, p < 0.05), indicating that performance using the FVB was significantly worse then with the other two VB configurations at this height.

Figure 28: VB-Target Height Interaction (Study 3)

5.3.2 Post-Test Questionnaire

A short post-test questionnaire was administered after this study to gain input on subject thoughts, strategies, and preferences. Subjects felt fairly confident of their overall estimation accuracy (mean rating = 4.9/7.0). They felt less aware of their VB configuration per trial (mean rating = 4.1/7.0). Six of the nine subjects felt that the existence of the FVB aided their judgments of reach. They claimed it helped with depth perception, aided slightly in low-target height trials as a reference point, and allowed for a better feeling of relative size. The three remaining subjects either could not bring themselves to depend on the FVB or stated that it aided during practice trials only. Five of the nine subjects preferred the VB over the other two configurations, the remaining three had no preference. The subjects rated their overall satisfaction with the FVB as just slightly better then average (4.3/7.0).

Eight of the nine subjects felt that target height affected their performance. Some had trouble differentiating between the low and medium heights. Three subjects felt that the lowest height was most difficult to estimate, one felt that the high height was hardest, and one subject felt that the medium height was most difficult. Strategies varied from a feel or intuition, to using apparent size, to using the target location on screen, even to counting movement increments from the target. Two subjects stated that their strategy on the low heights was to use the FVB as a reference point for elevation and distance estimation (i.e., "used the tips of my virtual toes").

5.4 Discussion

This study marks the first known attempt at assessing virtual reach estimations with different VB configurations in a VE. It provided many insights into the nature of virtual reach and the impact of a FVB on these judgments. First the significant findings will be discussed, followed by a detailing of other issues, ideas, and questions generated as a result of this study.

A significant main effect for the height factor was found. However, since the height-VB interaction was also significant (and disordinal), exploring this relationship in more detail will provide a more accurate and complete picture of the findings.

The height-VB interaction (Figure 28) does indeed show a general trend of increasing offset error with target height. This most likely was due to the existence of a patterned background in the low condition. The proximity of the floor `texture' to the target provided an additional depth cue that was less available during the middle height trials and not available at all in the high trials. However, three subjects felt that the high targets were easiest to estimate, due most likely in the difficulty in judging between the two lower heights.

The actual interaction between height and VB is much more difficult to decipher, however. While the HVB resulted in poorest performance at the medium level, the FVB created the largest errors at the high target level (there was no significant difference between the VBs at the low level). An informal review of the directional offsets for these two conditions revealed no trends to aid in explanation. So then, why did this finding occur? Although no definitive answer could be obtained, a few potential explanations were developed. One explanation is that the mere existence of a specific VB image in a virtual world can slightly modify/distort one's mental model of spatial relationships within this environment (i.e., perhaps a prism adaptation effect). The exact nature of this distortion cannot be determined in this initial study, but it is likely subtle and dependent on many variables including VB configuration, the fidelity of the virtual world, the type of VB incorporated, the task involved, and the strategies used to complete the task. A second explanation is that a type 1 error occurred in testing this interaction, resulting in the inadvertent significant finding of VB configuration as it relates to the more plausible height main-effect. The reasoning for this is that many subjects did not look at their VB configuration at all on the high-height conditions, yet the FVB showed a significant effect. The strategies that were used (in most cases) at this height did not even involve the subject casually viewing his/her VB. Also, when these contrasts were tested using the more stringent Scheffe correction, all significant findings disappeared. Clearly, more empirical research is needed to more fully develop the relationship between VB and spatial estimations at different heights.

Some general notes about this experimental design are in order. First, the presence of a medium height level may have caused more problems then it was worth. Many subjects confused it with the low level (given the relatively low amount of depth cue information in this virtual world) and no subject actively attempted to combine the VB image with the target at this height. In fact, these confusions may have contributed to the VB-height interaction discussed above, providing more strength for the second explanation. Perhaps the eliminating of this level would result in more accurate estimations at the low level and an increased focus on the potential of the VB to act as a reference point at the low height.

A second design issue concerned the granularity of movement (forward and backward) through button presses on the 3D joystick. Many subjects complained about the coarseness of movement, with one press of the button resulting in a virtual movement of up to several inches. It was explained to the subjects that finer movement steps could be obtained by looking up or down when depressing the button. However this took away from the essence of the task and as a result subjects just moved back and forth several times until they were satisfied with their position. A follow-on study may consider having the subject remain stationary in a fixed position and present the object at different distances (i.e., Carello, et al., 1989) with the dependent variable being the distance at which a certain judgment threshold is crossed. The reason that this method was not used in the present study was the desire to include movement in the task, since one of the primary benefits of immersive VR is the ability to move in virtual worlds.

A third issue revolves around the unattached right arm that existed in all FVB conditions. This arm was fine for lower level heights but often provided a distraction at higher heights. A more refined FVB with an attached right arm and a moveable elbow (such as Badler's Jack software (Badler, et al., 1993)) would go a long way towards more accurately testing virtual reach and the influence of a FVB.

A fourth issue revolved around the setting of target heights for each subject. Although the chosen method of using a modified t-bevel worked satisfactorily, a more accurate tool for setting arm angle would definitely help standardize the visual conditions more completely between subjects.

A fifth issue concerns the lack of depth information in this study. Subjects received less depth information in this study than the previous studies for two reasons. The first is because the subjects were not allowed to view the target from several angles/perspectives. The second reason is that the stereo depth cue was diminished due to the requirement for subjects to estimate maximum reach from a distance of approximately 20 - 35 inches from the target (stereo information is most powerful when close to the target). This stereo cue potentially was further compromised if a subject's inter-ocular distance was much different than the default value used in this study (2.55"). This lack of depth information most probably contributed to the observed variance in the data.

Finally, it need be mentioned again that FOV limitations of the HMD may have decreased any potential FVB effects. Subjects' overall awareness of their particular VB configuration on every trial was fairly low (average rating = 4.1/7.0) and is due in large part to FOV limitations. Any increases in FOV would benefit follow-on reachability estimation studies.