Inside-Out Display Pilot Studies

Title: AIIP2: Inside-Out Displays and Optical Flow. Background: A sensitive measure for control-reversals would be useful as a means to investigate the possible relationship between presence and selected rest frames, and to measure the effect of particular manipulations on presence. AIIP2 examined the sensitivity of a particular control-reversal measure by testing its ability to detect the influence of a manipulation known to affect control reversals (the use or absence of moving textures on the ground representation of an inside-out display). Hypothesis: Either accuracy or reaction-time (RT) or both should be improved on an inside-out display roll-correction task when a moving texture is present on the ground representation. Methods: Twenty-two adults volunteered, none familiar with the hypothesis. The task consisted of correcting an inside-out depiction of an aircraft roll on a computer monitor by pressing either the left or right arrow key. Accuracy and RT were measured. The independent variables were presence/absence of a moving texture of the ground portion of the display; roll angle; and pitch angle. The roll/pitch combinations used were 60/-20, 100/-20, 130/-10 and 160/-10. There were two replications of each roll/pitch combination for each condition, for a total of 2x2x8 = 32trials per participant. There was a block on moving vs. static ground representations. (The block was to allow for a ``build-up of effect'' in the moving condition.) Order of angle combinations were randomized within each block. RT-accuracy studies have a data analysis problem if the two dependent measures vary in opposite ways across conditions. In AIIP2, accuracy was controlled for by only analyzing RT's for participants who picked the correct roll at least 14/16 times in both conditions. This was consistent with preliminary data suggesting that this accuracy level would generally be met, and with the expectation that any significance would be in the RT data, rather than in accuracy. To maintain counter-balancing across condition orders at the required accuracy threshold, successive participants were added as needed to either order. Results: In the condition order with the moving texture ground representation first, 8/8 participants met the accuracy threshold. In the condition order with the static representation first, only 7/14 participants did so. This difference is significant on Fisher's exact test (p < .02). For those above the accuracy threshold, there was a strong order effect for accuracy (p <.003) and a trend towards an order effect for reaction-times (p < .07) with the second condition performing better in each case. There was no difference on a paired t-test of median RT's between conditions for those above the accuracy threshold. On questionnaire data, 11 participants reported believing they performed better with the moving texture; 5 indicated no preference; and 6 stated that the moving texture was a counter-productive distractor. Conclusions: The difference in accuracy-threshold between the two condition orders, together with the strong order effect, imply that the static condition was measurably more difficult that the moving texture condition. This suggested that a control-reversal measure might be viable, with refinements to reduce the strong order effect, which may obscure differences between conditions.

Title: AIIP3: Inside-Out Displays and Foreground Occlusions I. Background: AIIP2 implied that the control-reversal measure could distinguish between moving-texture and static ground conditions. This study applied the measure to another pair of conditions: foreground occlusion/no foreground occlusion. Modifications were made to attempt to reduce the order effect found in AIIP1. Hypothesis: The use of a foreground occlusion, by making the inside-out display the visual background, should increase the likelihood that the ground representation of the inside-out display will define the selected rest frame. This should improve performance on a roll-correction task. Methods: Similar to AIIP2. Three adult volunteers, all male. One was familiar with (but did not believe) the hypothesis. The images of a plane and of a ground representation with a moving texture were displayed on a computer monitor. The images were in a circle, surrounded by a black annulus which covered the rest of the monitor. In front of the monitor was mounted an adjustable camera iris at eye level, surrounded by black tarp to remove peripheral cues. The diameter of the iris was adjustable. In the foreground occlusion condition, the inner edge of the black annulus was obscured by the foreground occlusion. In the non-foreground occlusion condition, the FOV was expanded so that one could see part of the black annulus. The monitor was viewed from a distance of about 22 cm. The FOV of the scene was about 30. In the foreground occlusion condition, the FOV was set at about 27. In the non-foreground occlusion condition, the FOV was about 45, with the outside 15 showing the black annulus. Several modifications were made to make the task more difficult than in AIIP2, to reduce the order effect. The first modification was to allow both the ground and the plane to roll, with the task being to find the shortest direction (left or right) to align the plane with the ground. The second modification was to increase the number of possible angles, selecting randomly with replacement from the set of plane rotation angles (55, 65, 75, 85, 95, 105, 115, 125, 135, 145, 155); from the set of ground rotation angles (60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160); and from the set of pitch angles (0, 10, 20). The third modification was that a longitudinal design was used with 8 blocks of 16 trials in each of the foreground occlusion and non-foreground occlusion conditions. Each participant had 2 sessions of 4 blocks each, conducted on separate days. The condition blocks followed an ABBA pattern, counter-balanced across participants. Results: Examining adjacent pairs of trial blocks with a paired two-tail t-test on error rate and RT, the error rate was significantly lower in the foreground occlusion condition. (Foreground occlusion: median = 1.2/16, STD = 1.5/16; non-foreground occlusion: median = 1.8/16, STD = 1.5; p < .04). There was no difference in RT's. Conclusions: The finding that there was a significant difference in errors, but not reaction-times, was consistent with AIIP2. A possible interpretation is that participants thought of the two conditions as being identical, and therefore responded at the same rate. The difference between the two conditions therefore showed up in the error rate. This interpretation is consistent with the fact that none of the participants reported feeling a difference in their performance between conditions.

Title: AIIE3: Inside-Out Displays and Foreground Occlusions II. Background: AIIP3 found an effect of foreground occlusion to reduce the error rate for a roll-correction task. However, this was based on a longitudinal study of only 3 participants. AIIE3 sought to replicate this result with a larger sample size. Hypothesis: The use of a foreground occlusion, by making the inside-out display the visual background, should increase the likelihood that the ground representation of the inside-out display will define the selected rest frame. This should improve performance on a roll-correction task. Methods: Nine adults volunteered (6 male, 3 female). None were familiar with the hypothesis. The procedure was identical to AIIP3, except that a study of the error distribution from AIIP3 was used to bias the trials towards the most difficult cases. (This was done with the intent to increase the low error rate from AIIP3, and thus to allow a difference between the conditions to show more clearly.) In AIIP3, The trials in which plane and ground had opposite rolls produced 80% of the errors. There were twice as many errors in pitch up as pitch down conditions (see Figure C.1. To try to increase the error rate, therefore, the random selection of angles was biased 80% towards opposing rolls, and 80% towards pitch 0. As in AIIP3, participants were run through 16 blocks of 16 trials. As there were 9 participants in AIIE3, the below results summarize 9x16x16=2304 trials. Results: Examining adjacent pairs of trial blocks with a paired two-tail t-test on error rate and RT, no significant difference was found between the foreground occlusion and non-foreground occlusion conditions on either error rate (Foreground occlusion: median = 1.6, STD = 2.0; non-foreground occlusion: mean = 1.7, STD = 2.4) or on median RT in seconds (Foreground occlusion: median = .29, STD = .18; non-foreground occlusion: median = .28, STD = .18). The possibility was considered that an effect existed in the early blocks which was obscured by a performance asymptote in later blocks. However, no difference between conditions was found in the first 8 blocks, nor in the first 4 blocks. Conclusions: It is noteworthy that the angle selection procedure described above did not succeed in increasing the error rate from AIIP3 to AIIE3. This suggests that the task is overly simple, and that a performance asymptote is quickly reached. If so, a better approach than used here might be to alternate trials with and without the foreground occlusion, rather than taking blocks of 16 trials in each condition. This was not done in AIIE3 for two reasons: the difficulty of quickly re-adjusting the apparatus, and to allow for a ``build-up of effect'' within each condition block. A further try (using the inside-out display metaphor) would be to use participants with no training on inside-out displays and record their intuitive response for correcting a roll as a function of the presence or absence of a foreground occlusion. One would hypothesize (if foreground occlusions do affect the selected rest frame) that there would be a stronger tendency to mis-interpret inside-out displays in the absence of a foreground occlusion.

  

Figure C.1: Inside-Out Display Pitch Representations