Metrics and Models for the Perceptual Design of Virtual Transparency - Progress (06/99~07/99)

---

PROGRESS AND PUBLICATIONS:

The following abstract was submitted to HFES/IEA 2000 to be held in San Diego, CA. The filter parameters used for this experiment will become standard for all applications in the laboratory.

Predictive compensation of head tracking can be valuable for reducing the consequences of the sensor, computation, and rendering pipeline delays inherent to virtual environment (VE) systems. A number of researchers (e.g., Azuma & Bishop, 1994; Friedmann, Starner, & Pentland, 1992; Liang, Shaw, & Green, 1991) have already demonstrated the feasibility of various implementa-tions of predictive software. However, despite considerable technical achievements, simple engineering measures like RMS predictor error are insufficient indicators of predictive compen-sation's impact on the user's perceptual experience. Thus, there is a need for systematic assess-ment of predictor effects on user performance. This work represents the first study of the impact of predictively compensated head tracking on the user's perceptual experience in a virtual envi-ronment.

Our approach is based on assessment of how predictive compensation affects user perception of a spatially stabilized virtual object that is observed in an immersing helmet mounted display (HMD). Our very fast VE system (Jacoby, Adelstein, & Ellis, 1996) makes possible the con-trolled addition of differing time delays which we then attempt to cancel by predicting ahead by the same delay increment. In our view, an ideal predictive compensator should make the dis-tinctive effects of the additional delay imperceptible to the user. Thus, our approach is to meas-ure subjects' ability to discriminate between a reference condition in which system latency is set to its baseline minimum (27±5 msec from position tracker to display) and a condition in which a fixed latency is deliberately added but fully compensated by prediction. Our null hypothesis is that the two conditions would be indistinguishable from each other.

Ten subjects participated in a two-alternative, forced-choice experiment using a discrimination technique previously developed to study operator sensitivity to changes of latency (Ellis, Young, Adelstein, & Ehrlich, 1999). The VE consisted solely of a virtual sphere fixed in an empty, dark space. Subjects were asked to yaw their heads back and forth through ~40º while paced by a 0.6 Hz metronome, causing the HMD image of the ball to move side to side in a manner corre-sponding to a static object. Six blocks of trials were conducted per subject at different incre-ments of artificially added latency (16.7 to 100 msec in 16.7 msec steps). For each latency con-dition, a Kalman filter predictor based on Azuma and Bishop's (1994) parameterization was set to compensate exactly for the added latency. Blocks were composed of five randomly ordered repetitions for each of the four possible pairings of minimum and compensated latency condi-tions. Subjects judged whether or not the stimuli in each pairing differed. A separate control study with the same latencies was conducted on eight different subjects to determine dis-criminability in the absence of predictive compensation.

We postulate that subjects rely on different cues to discriminate the presence of additional latency depending on whether or not the VE is predictively compensated. For the control experiment, the difference between the virtual sphere's rendered displacement and the subjects' expectation for its minimal spatial instability are simply the result of time lag and the consequent motion offset. In the main experiment, this difference arises not from lag, but from overshoot and noise artifacts induced by the computational limitations of predictive compensation. Ideally, if the overshoot and noise are imperceptible to the subject, the virtual sphere will appear to have perfect spatial stability. At minimum, these artifacts should prove to be less detectable than the lag of the uncompensated latency if prediction is to be a useful technique.

The results, plotted in Figure 1, show that discriminability grows monotonically from the level expected for random guessing as the amount of predictively compensated delay increases. A two-way ANOVA was performed to test the effect of the number of added latency increments and the presence of predictive compensation on an arcsine transformation of the proportion of correct responses. The arcsines produced the normal distribution needed for the analysis. The main effect of added latency on the proportion of correct responses was significant (F = 24.902; df = 5,80; p < .001), while the presence of predictive compensation alone was not (F = 2.330; df = 1,16; p < 0.146). Interaction between the two factors was significant (F =3.586; df = 5,80; p < .006). This latter result in conjunction with Figure 1 implies that artifacts introduced by predictive compensation are less discernable than the disruptions attributable to uncompensated time delays for the shorter but not the longer added latencies.

Figure 1: Percent correct discrimination (mean ± std error) as a function of latency added to the 27 msec system minimum.

REFERENCES

Azuma, A., and Bishop, G. (1994) Improving static and dynamic registration in an optical see-through display. Proceedings, Symposium on Interactive 3D Graphics (SIGGRAPH), Orlando, FL, pp. 197-204.

Ellis, S., Young, M., Adelstein, B., and Ehrlich, S. (1999) Discrimination of changes of latency during voluntary hand movement of virtual objects. To appear in Proceedings, 43rd Annual Meeting of the Human Factors and Ergonomics Society.

Friedmann, M., Starner, S., and Pentland, A. (1992) Synchronization in virtual realities. Presence, 1, 139-144.

Jacoby, R., Adelstein, B., and Ellis, S. (1996) Improved temporal response in virtual environ-ment hardware and software. Proceedings, SPIE Conference on Stereoscopic Displays and Applications VII, Vol. 2653, Bellingham WA, pp. 271-284.

Liang, J., Shaw, C., and Green, M. (1991) On temporal-spatial realism in the virtual reality envi-ronment. Proceedings, ACM Symposium on User Interface Software and Technology (UIST) 1991, Hilton Head, SC, pp. 19-25.

SIGNIFICANT EXTERNAL LIAISONS:

Exploratory development with World Tool Kit is being conducted through Raytheon and William Chapin to test the performance of NT systems on Pentium-based machines to determine if our rendering experiments may be ported to these systems in the future, potentially offering great savings in software and hardware costs.

ISSUES:

The joint research agreement with the local R&D corporation has not yet been signed by the corporation complicating preliminary research and development.