Speed perception as an objective measure of presence in virtual environments

Transcription

1 Speed perception as an objective measure of presence in virtual environments Guy Wallis 1 ; Jennifer Tichon 1, Tony Mildred 2 1. Peception and Motor Control Lab, School of Human Movement Studies,University of Queensland 2. RailCorp Training Facility, Petersham, Sydney gwallis@hms.uq.edu.au, jtichon@uq.edu.au, Tony.Mildred@railcorp.nsw.gov.au Abstract. Many industries are becoming interested in the use of simulated environments in training. Although some industries, such as the aeronautical industry, have well established parameters and protocols for conducting training, other industries are looking to establish programs tailored to their needs. One issue facing such industries is the validity and efficacy of their training. Training validity and efficacy are affected by a number of factors including the quality of the simulator, in terms of how immersed or present a user feels; the degree to which the simulator reproduces the particular real-world sensory experience; and the accuracy of the simulated interface (steering-wheel, treadmill, joystick, motion platform, etc.). This paper considers objective measures for the assessment of simulator quality and provides preliminary data from a study of three rail simulators. Work based on a speed perception task provides evidence of the measure s ability to discern differences between simulators revealing its sensitivity and reliability across users. The paper explains how such differences can help to motivate design aspects of new simulators, can prompt changes to existing systems, and constrain training scenarios to maximize their efficacy. 1. INTRODUCTION Traditionally, there have been a relatively small number of industries making use of simulator technology. Recent improvements in fidelity, affordability and industry awareness have seen that number increase dramatically, but often without a good understanding of how best to make use of it. In choosing a suitable platform for conducting simulator-based training, a company must consider simulator design including the range of sensory feedback, levels of interaction supported, and mobility. Equally, once the simulator is established careful thought must be given to the design of training scenarios which make best use of the simulator s strengths and avoid its weaknesses. Ultimately what one would like to know is how best to optimize learning outcomes. One company making extensive use of VR-based training is RailCorp. RailCorp provides passenger rail transport throughout NSW via its CityRail and CountryLink services and is responsible for the safe operation, crewing and maintenance of passenger trains and stations. It also owns and maintains the metropolitan rail network and provides access to freight operators in the metropolitan area. To provide safety training across its 15,000 widely distributed personnel the company utilizes a high-fidelity, large-screen simulator and two full in-cab simulators. All of these simulators are centrally located at the Australian Rail Training College in Petersham, Sydney. None of these simulators is in any way portable. This restricts the use of such machines to those staff who can be brought in from around the network for training. This can be both logistically and financially restrictive, and so the company also make use of a portable, video-based system which can be taken on the road to staff based across Sydney. Although the use of video-based training is convenient it restricts training to passive observation rather than interactive simulation. RailCorp have made extensive and effective use of their simulators but are open to suggestions as to how to further optimize their learning outcomes. Previous research clearly indicates that one crucial aspect of maximizing learning outcomes is the extent to which trainees feel immersed or present in the simulated environment [1]. They also confirm that interaction enhances retention of training experiences and plays important part of the perceived level of presence. Despite its undoubted importance in training, measurement of simulator quality is primarily based on subjective self-reports, longitudinal studies of safety outcomes, or physiological measures hypothesized to correlate at least with the level of presence. This paper describes some preliminary work undertaken to obtain a direct, objective measure of presence and simulator fidelity. Simulator Quality and Training Validity As described above, when designing a simulator and later designing training scenarios, it is crucial to know how successful experience in that simulator will be in

2 evoking the desired outcomes and responses in the realworld. Knowing whether to include simulator elements such as a motion platform or sound cues directly impacts on construction costs but also has implications for user presence and simulator fidelity. Understanding the true cost of excluding certain cues or limits to graphics quality can help maximise the quality of the virtual experience. This in turn will help maximise training outcomes. This knowledge can not only help produce the best simulator for a given budget, but can also help fashion an appropriate training regime by highlighting the circumstances under which the simulator works best. For example it might help set limits on simulator speed (both high and low), or set a limit to rates of heading change which are linked to the refresh rate of the system. Unfortunately there are no truly satisfactory measures of either presence or fidelity. Some attempts have been made, however, and they are reviewed in this section. Presence As Witmer & Singer [1] describe, there appears to be a link between an increased sense of presence and the quality of training outcomes. Specifically, the authors suggest that increased presence increases the similarity of the behaviour elicited in the virtual environment and that produced in the real environment. However, despite its importance, presence remains an imprecisely defined concept. One might think of it as the ability of a user to suspend disbelief in the simulated environment, or as the ability of a subject to commit attention to, or lose themselves in, the environment [2,3]. Such descriptions are a starting point, but leave presence extremely hard to quantify. That is not to say that attempts have not been made. The most common of these is to utilize subjective evaluation scales. Based on the theoretical work of Sheridan [4] and Held & Durlach [5], Witmer and colleagues [1,6] have developed and tested a seven point Presence Questionnaire (PQ). The questionnaire has gained a significant level of acceptance. Both it and other similar scales have been tested across a number of studies [7,8] including in the specific case of the RailCorp simulators being investigated here [9,10]. Ultimately, however, questionnaires are only as reliable as the subjective reports upon which they are based. They also say nothing about cues which supplement learning but which are beyond superficial, personal reflection. Alternatives to subjective measures have also been discussed. Physiological measures such as cardiac frequency, skin conductance (GSR), reflex motor behaviour and event evoked cortical responses have been suggested and in some case pilot work has been conducted. Authors generally propose that a sign of high presence would be that physiological reactions to the simulated environment are similar to those observed in a real environment. In a pilot study by Dillon, Keogh, Freeman and Davidoff [12], it was observed that cardiac frequency was higher during the presentation of rally driving sequences than calm boat sequences. Whilst such measures have the potential to circumvent the problems of self-assessment and subjective report, their use remains patchy, and detailed research of their suitability remains scarce [11]. More work is required in this area. Ultimately, however, it seems likely that such measures will, at best, only serve to validate large-scale emotional responses. Such responses may well be crucial in desensitisation work or stress inoculation, but may prove too coarse to measure all aspects of simulator quality. Fidelity A clearly related and yet distinct concept in assessing the quality of a simulator is what might loosely be referred to as fidelity. This is the degree to which the simulator behaves like its real counterpart. The term is most often used to refer to image quality which is affected by a number of factors such as refresh rate of the simulation, resolution (pixel count), render quality (illumination model, texture resolution), and field of view etc. These factors are certainly likely to impact on feelings of presence, but they may also serve to enhance training in a silent manner by introducing sources of information which are covertly integrated into a trainee s representation of the environment. A driver may well receive non-visual cues as to the state of the vehicle - cues such as the speed related rumble of a vehicle or the related sounds of the vehicle traversing the driving surface etc. The absence of such cues may be most noticeable at high simulated velocities and therefore affect presence at these speeds, but they may also play an important role in establishing a suitably information rich environment across the entire range of velocities. Speed perception Beyond the realm of the subjective questionnaires and simple physiological response measures, this study seeks an objective measure of simulation quality and validity which combines a sense of presence and fidelity. A review of the human perception literature suggests that one such measure might be velocity perception. Humans are quite adept as estimating their rate of forward motion, even at unecological speeds i.e. well beyond those for which evolution has equipped us. This is partly through the training gained by observation of a speedometer in fast moving, landbased vehicles such as cars or trains.

3 In order to make speed estimates the brain relies on integrating a range of sensory cues including proprioception, optic-flow, distance perception, auditory perception, indeed any cues which correlate reliably with speed. In the visual domain optic flow represents a significant source of information, but only in the presence of high contrast, high frequency images [13], [14]. Because of the polysensory nature of speed perception it may well offer a convenient metric for the simultaneous assessment of simulator fidelity and presence. There is every reason to think that an accurate impression of forward speed is a precursor to immersing subjects in the environment. Importantly, the measure will also help ensure that the simulator is designed to include a range of cues which affect behaviour in real-world scenarios. As mentioned above, such a measure should also help guide training programmes by identifying those conditions under which the simulator works most effectively. The study described here takes three types of simulator and tests whether speed perception is a sufficiently sensitive measure to determine differences between them. 2. METHODS Simulators There are three, fixed-base simulators located at the Petersham site. The largest and most sophisticated offers a 180 degree field-of-view from a control desk configured to reproduce the controls found in one of a series of engines run on the RailCorp network. The desk has force-feedback controls but no enclosure, see figure 1. The large field of view provides a strong drive to motion detection systems in the human eye which are highly sensitive to peripheral stimulation [15] such motion cues have been widely implicated in heading estimation [16]. The other two simulators are smaller, full cab devices with a view restricted to a flat, frontal portion of the simulated environment offering approximately a 40 degree of horizontal field-of-view, see figure 2. Although smaller and offering a relatively narrow field of view, the cabs natural screen layout provide a naturalistic reference-frame, something lacking from the large screen system. Figure 1 Wide-screen simulator As mentioned earlier, RailCorp also make use of video presentations. These consist of a wall-projected recording of the simulated environments utilized in the interactive simulators. It uses a standard XGA video projector to project an image with visual dimensions similar to the cab-based simulator. Participants Nine expert train drivers took part in the experiments. They had an average of 22.4 years experience as drivers. Design The drivers were asked to sit in one of the three simulators and passively view a section of track being negotiated at a fixed speed. Their task was to estimate their current speed in the absence of any instrument readouts or sound cues. Four simulated speeds (20, 40, 60 and 80 km/h) were used and they were simulated in pseudo-random order 12 times for each speed making a total of 48 trials. The same experiment was conducted once in each of the three simulators. The order of simulator runs was made at random to counteract any effects of learning. In each trial the drivers were permitted to view the environment for a five second period, during which they were required to write down their estimated speed in a response table. The delay between trials was variable and random but was at least 5 secs. This helped to reduce biases which might have arisen if a correlation existed between a specific intertrial delay and the difference in speed between specific trials.

4 corresponding to a roughly constant percentage error. A similar slope is observed for the video presentation but in this case the slope starts as a sizable overestimate at lower speeds. These errors are particularly large when considered as percentages of actual speed. The average error at Judgment Error (%) WS Cab Video Simulator Type Figure 3 Speed estimate error expressed as a percentage of true simulator speed. The three bars represent data for the three simulator types (WS: Wide-screen, Cab: Enclosed cab, and Video). Error bars represent standard error of the mean. Figure 2 Cab-based simulator 3. RESULTS The first analysis aimed to discover whether drivers were misperceiving their speed in a systematic manner in any of the three simulators. In order to make comparisons across speeds the responses were first normalized by expressing the error (in km/h) as a percentage of the true simulated speed. The averages for each simulator type are shown in figure 3. Drivers using the wide-screen simulator tended to underestimate their true speed. In contrast, drivers watching the video presentation tended to overestimate their speed. The difference between the two was statistically significant F(1,8)=14.7, MSe=863.5, ES=0.585, p= Estimates made in the cab simulators were identically variable to the other two simulators (s.e.m. 7 km/h) but were much closer to the true speed and did not differ significantly from zero error t 9 = , n.s. A follow-up analysis looked at the actual speed estimate errors as a function of speed. Results appear in figure 4. In this case, for both of the interactive simulators there appears to be a negatively sloped increase in the magnitude of error as speed increases 20km/h is as high as 35%. The data reveal that performance during video presentation is particularly inaccurate. They also suggest that performance in the cab simulator is better across all speeds than in the wide-screen simulator. 4. CONCLUSIONS Overall, the results indicate that the speed perception measure is sensitive to differences in a user s simulated experience. The measure was consistently able to discriminate performance in the three simulators and revealed an unexpected effect in the large-screen system, namely that observers tend to underestimate their speed, despite remarkably good performance in the same task when conducted in the cab simulators. One plausible explanation for this discrepancy is the lack of reference-frame in the large screen simulator. The presence of two fixed reference edges near to the centre of vision may well offer a cue for experienced drivers which is lacking in the large simulator. This is, of course an empirical question and further studies will reveal if a relatively inexpensive modification to the simulator can improve performance, particularly at higher velocities.

5 Performance in the video presentation was particularly poor. The measure may help motivate changes to the set-up used such as altering the viewing distance to suit Judgment Error (km/h) WS Cab Video Simulator Type Figure 4 Speed estimate error in km/h for each of the four test speeds (20, 40, 60, and 80 km/h). Error bars represent standard error of the mean. the recorded viewing perspective; as well as alterations to video image quality. These suggestions are simply speculative at this early stage, but overall one can conclude that the speed perception test can provide a sensitive, objective guide to a simulator s limitations, and help motivate future modifications. 5. FUTURE DIRECTIONS The work described here is an encouraging start but more work is required to validate its use in constraining simulator design and modifications. In the future it would be important to carry out controlled studies of subtle alterations to the simulator. The ultimate aim of this work is to develop a suite of tools based on subjective and objective measures of presence and simulator fidelity to constrain both simulator design and training protocols. REFERENCES 1. Witmer, B.G. & Singer, M.J. (1998) Measuring Presence in Virtual Environments: A Presence Questionnaire. Presence, 7, Steuer, J.S. (1992) Defining virtiual reality: dimensions determining telepresence. Journal of Communication 4, Coelho, C., Tichon, J., Hine, T.J., Wallis, G., & Riva, G (2006), The sense of presence in virtual reality technologies. In From Communication to Presence G. Riva & F. Davide (Eds), Amsterdam: IOS Press. 4. Sheridan, T.B. (1992) Musings on telepresence and virtual presence. Presence: Teleoperators & Virtual Environments, 1, Held, R.M. & Durlach, N.I. (1992) Telepresence. Presence: Teleoperators & Virtual Environments, 1, Witmer, B.G., Jerome, C.J. & Singer, M.J. (2005) The Factor Structure of the Presence Questionnaire, Presence: Teleoperators & Virtual Environments, 14, Schubert, T., Friedmann, F., & Regenbrecht, H. (2001) The Experience of Presence: Factor Analytic Insights, Presence: Teleoperators & Virtual Environments, 10, Lessiter J., Freeman J., Keogh E et al (2001) A Cross-Media Presence Questionnaire: The ITC Sense of Presence Inventory, Presence: Teleoperators and Virtual Environments, 10, Tichon, J. and Wallis, G.M. (2007), Improving simulations for train driver training in degraded rail conditions. Cyberpsychology and Behaviour, in press. 10. Tichon, J., Wallis, G., & Mildred, T. (2006) Virtual training environments to improve train driver s crisis decision making. SimTect Ijsselsteijn, H., Riddler, J., Freeman, J., & Avons, S.E. (2000) Presence: concept, determinants and measurement. Proceedings of SPIE, Human Vision and Electronic Imaging. San Jose, CA. 12. Dillon, C., Keogh, E., Freeman, J., & Davidoff, J.B. (2001) Presence is your heart in it? Proceedings of the 4th Annual International Workshop on Presence, p 21-23, Temple University, Philadelphia. 13. Snowden, R.J., Stimpson, N., & Ruddle, R.A. (1998). Speed perception fogs up as visibility drops. Nature, 392, Distler, H., & Bülthoff, H.H., (1996), Velocity perception in 3-D environments. Perception, 25 ECVP Abstract Supplement. 15. Warren, W.H. & Kurtz, K.J. (1992) The role of central and peripheral vision in perceiving the direction of self-motion Perception & Psychophysics 51, Lappe, M., Bremmer, F., & van den Berg, A.V. (1999) Trends in Cognitive Sciences 3,