Detecting and Understanding Breaks in Presence from Physiological Data: Work in Progress

Detecting and Understanding Breaks in Presence from Physiological Data: Work in Progress Andrea Brogni, Vinoba Vinayagamoorthy, Anthony Steed, Mel Slater Department of Computer Science University College London Gower Street, London WC1E 6BT, UK ABSTRACT In this paper we give an overview of our recent work on extending and applying our breaks in presence model. We have started using physiological sensors, initially to monitor user response and we have shown appropriate responses to stressful situations. We described how we have applied this to our study of agoraphobia. We then describe our explorations of the possibility of predicting breaks in presence form physiological monitoring alone. This has the potential for giving a non-intrusive measure for presence. It is also a technique that is easily transferable to other mixed-reality situations. Keywords: virtual environments, presence, physiological measures, breaks in presence. 1. INTRODUCTION Virtual environments are a type of simulation in which computer graphics is used to create a believable world that is interactive and responds to user input [Burdea and Coiffet 1994]. Our research explores the different interactions between users and virtual environments. For us, the main research question is to identify those features that create a more believable virtual environment. In situations where the user believes that the virtual world is real and behaves appropriately, we say that they are experiencing a sense of presence. The experience of a sense of presence has both subjective and objective aspects [Freeman et al. 2001]. In previous work we introduced the breaks in presence model [Slater and Steed 2000]. The breaks in presence model notes that during the virtual environment experience the user will occasionally react, not as if the virtual environment was believable, but as if they were in the real world. We would say they had experienced a break in presence (BIP). BIPs might occur because of external anomalies of the display (e.g. loss of tracking, screen blanking), internal anomalies in the virtual environment (e.g. lack of collision response, un-synchronised audio and video) or spontaneously (e.g. realisation that an experiment is taking place). By observing when BIPs occur we hope to build an estimate of presence by arguing that after BIPs, the user cannot be present in the virtual environment for a short period of time. In this paper we give an overview of our recent work on extending and applying our breaks in presence model. We have extended the breaks in presence model through the use of physiological monitoring. We have shown that users selfreported BIPs produce a characteristic physiological response. In recent work we have shown that from physiological monitoring it may be possible to predict when BIPs occurred even though the user does not indicate the BIP. We have applied the breaks in presence model to our studies of agoraphobia. In this paper we will also present early ideas about applying breaks in presence to mixedreality systems.

2. PRESENCE MEASUREMENTS The definition of presence is still a crucial point in the virtual environments field. The difficulty to define an experience in a real environment entails a difficulty to define an experience in a virtual environment. A common definition is the sense of being there in the virtual place [Slater and Wilbur 1997], [Witmer and Singer 1998], [Slater 1999]. Another definition is the perceptual illusion of non-mediation [Lombard and Ditton 1997], that involves continuous human perceptions and where it becomes dominant the knowledge of the existence of a medium in the user s communication environment. The sense of presence can be evaluated by taking various measures, including pre and post questionnaires, physiological measures, and reported breaks in presence (BIPs) during the experience [Slater and Steed 2000]. Measures of presence have been dominated by questionnaire measures. When used alone these have been argued to be unsatisfactory, on several grounds. Freeman et al. [Freeman et al. 1999] have shown the inherent instability of subjective presence questionnaires and have proposed behavioural measures. Slater [Slater 2003] has argued that presence questionnaires alone are unsafe on methodological grounds since it is difficult to rule out the possibility that the phenomenon to be measured is brought into being merely by asking questions about it. More generally presence questionnaires are administered after the event, and so cannot reflect the changing state of the participant during the ongoing experience. Physiological measures have been used in an attempt to overcome these problems with the traditional questionnaire approach. The method consists in recording the physiological response during the virtual experience. There are numerous physiological measurements that could be recorded during a study, using non-invasive devices. Usually the most common are the heart rate, the skin conductance and the skin temperature, but there are a lot of devices that allow the researchers to measure more complex signals, like Electro EncephaloGram (EEG), Electro CardioGram (ECG), Electro MyoGraphy (EMG), Blood Volume Pulse (BVP), Galvanic Skin Response (GSR) (also named Skin Conductance Response SCR), Heart Rate (HR) and respiration. Sheridan [Sheridan 1992] describes presence more like a mental manifestation, if this is true perhaps a measure showing the brain activity such as from an EEG might be more appropriate to capture all the possible processing states of the brain of the person experiencing VR technology. In 2002, Wiederhold at al. [Wiederhold et al. 2002] have used physiological responses to analyze the behaviour of phobic and nonphobic people in a virtual environment, particularly related with the fear of flying. Meehan, et al. [Meehan et al. 2002] used physiological responses to measure how the believability the experience of being at the edge of a pit. The hypothesis was that if the people felt present in the room, the virtual pit would have been able to evoke physiological responses similar to those of a corresponding real environment. They note that heart rate and skin conductance measure the arousal of the individual, therefore might only be used when such arousal is intrinsic into the task, i.e. fear. Other experiments have shown that GSR/SCR and HR could be used as objective measures in monitoring reaction in VEs [Jang et al. 2002]. A new approach to presence measures is based on the idea that a person experiencing virtual environments technology, at any one moment, interprets the stimuli coming from the environment as belonging either to the virtual or to the real world [Slater and Steed 2000]. A conceptual advantage of the BIP model is that it relies on data gathered during the experiment instead of only after the experience, as with the questionnaires. 3. BREAKS IN PRESENCE AND QUESTIONNAIRES The concept of BIPs was introduced in Slater and Steed. The idea was that rather than only

use a post-experience questionnaire to assess presence; presence might be assessed during the experience itself. The hypothesis was that during immersion in a VE a participant simultaneously perceives two streams of sensory data from the real world, and also from the virtual world. At any one moment these sense impressions would be the basis of a perceptual Gestalt, corresponding more or less to the world portrayed by the VE or to the real world in which the whole event was taking place. Sensory data corresponding to the non-favoured interpretation may be ignored, or incorporated into the prevailing Gestalt. For example, a wall in a CAVE-like system may not be noticed, a loud sound that does not belong to the virtual world may be incorporated into the flow of events within the virtual world interpretation, and so on. This is much like what happens in dreams, where sensory data from the physical world (a sound, a flashing light, something touching the sleeper s face) are incorporated into the ongoing dream in an attempt to keep the dream going for as long as possible. However, occasionally, as in dreams, the participant experiences a Gestalt switch from the virtual to the real interpretation. A break in presence is any event whereby, for the participant, at that moment, the real world becomes apparent, and for the duration of that event, the participant acts and responds more to the real world setting than to the virtual world. This model is an abstraction and like any abstraction it does not deal with the full complexity of reality. For example, it does not account for mixed perceptions where the participant simultaneously holds and even partially responds to both (real and virtual) interpretations, as noted by Spagnolli & Gamberini [Spagnolli et al. 2002]. In the first step of the research, we concentrated on the relationship between BIPs and reported questionnaire based presence. Before going through the actual experiment, all the volunteers have been trained on how to signal the BIPs. They were shown a series of Gestalt Images that could be interpreted in one of two ways, and their spontaneous switch in interpretation was used to explain the idea of change in place within a VE. They were instructed to press a button on the wand device whenever they experienced a change in their sense of place, from being in the VE to being in the real world of the ReaCTor (a CAVE TM -like system [Cruz-Neira et al. 1993]) and laboratory environment. During the virtual experience, any time the button on the wand device was pressed the relative elapsed time since the beginning of the experiment was recorded into a file. We simply counted the number of BIPs reported by subjects. Our hypothesis is that the greater the number of BIPs the lower the reported overall presence should be, on the average. The reported presence variable is negatively correlated with the number of BIPs. There is, therefore, a highly significant relationship between the number of BIPs collected during the experience and the presence questions administered after the experience, two variables which in terms of the way they are constructed, should, in principle from a purely procedural point of view, be independent [Brogni et al. 2003]. The results we have obtained are encouraging for further analysis, because the reported presence seems to be negatively correlated with the number of BIPs. The latter are reported during the experience, and not based on an after-the-event questionnaire, so, are more likely to be an on-line reading of the level of presence, which can give us information not only about the complete experience, but also about the single events during the experience. 4. BREAKS IN PRESENCE AND PHYSIOLOGICAL RESPONSES As discussed in the previous section, a BIP occurs when the participant stops responding to the virtual stream and instead responds to the real sensory stream. By respond we mean the total response of the participant including involuntary changes in physiological response, changes in eye movement patterns, changes in volitional

behaviour, and finally subjective verbal responses including questionnaire and interview responses. We hypothesise that BIPs should be detectable in physiological time series. If a person, however, truly became present in a VE then a BIP (a sudden switch in presence to physical reality) should be experienced as a profoundly shocking event. Of course, given today s VE systems, we do not expect such a profound sense of presence and therefore a very strong shock in response to a change in presence is unlikely. Nevertheless, it is useful to examine whether there is any support for the notion that a BIP registers as an event at the physiological level. We report on the results of an experiment where subjects reported BIPs and where SCR and HR data were recorded. We find that the evidence to date does not contradict the hypothesis that BIPs are observable as events in the SCR and heart rate time series. The approach that we have followed during the first analysis was not to attempt to characterise presence itself by physiological responses, but to examine whether there was a physiological signature corresponding to BIPs. A ProComp+ device was used to record SCR/GSR and BVP from which HR is derived. Treating BIPs as stimuli and the SCR and HR waveforms as response, we wished to examine whether there was any correlation between the stimuli and the response. The heart rate was normalised for each person by applying a de-trend routine (MATLAB) that takes out trend and absolute value. Figure 1 shows the mean waveform within ±4 seconds of a BIP. This (the solid black curve) is found by averaging the waveform in this range over all 60 participants over all BIPs. As we can see the mean heart rate appears to increase and reach a peak approximately 1 second before a BIP is signalled. The dashed (relatively flat) curve labelled random BIPs shows the mean waveform averaged over randomly placed stimuli points (500 iterations were used for the mean). I.e., pseudo-random BIP times were generated for each subject, and these were used in place of the true BIP times. The difference between this curve and the true BIP one demonstrates that the true BIP curve is not simply picking up an inherent property of the underlying time series itself, and that the behaviour of the time series in the neighbourhood of a BIP probably shows a genuine spike. Figure 1: Mean Wave Form for Heart Rate in the Neighbourhood of a BIP. The heart rate response may be caused by the action or intention of pressing the button to signal a BIP, rather than by the BIP itself. A further 10 subjects were recruited who were placed in the ReaCTor wearing all the same equipment as the original subjects, and who were given visual signals to carry out a number of different activities raising an arm, looking to the left or right, and so on. One of the activities signalled was to press the wand button. These wand button press times were recorded, and used as pseudo BIPs in order to carry out the same analysis for this control group as the original group. The resulting curve is shown as the dotted line labelled button press control. This shows an oscillation that is again quite different from that of the true BIP curve. Finally another 10 subjects were recruited. These also stood in the ReaCTor and were shown a red ball projected on the ReaCTor wall. They were asked to press the button only when the ball turned red. This instruction was communicated forcefully, in order to

produce some anxiety about the importance of pressing the button at the right time. This control group was used in order to generate some anxiety in the subjects. We were interested in the extent to which the heart rate response would be similar to a BIP in this case. Once again the time of the button presses were used as pseudo-bips in the analysis. The curve is labelled anxiety control in Figure 1. The result is a curve that has a similar shape to the heart rate BIP curve, but seems to be shifted about 4 seconds to the right. Clearly there is a heart rate peak but the slope increases at the onset of the button press rather than before the button press. case we see a similar pattern to the BIP waveform, with a rise in SCR reaching a peak after about 2.25 seconds. There was no control group for SCR where just the button pressing activity was measured. The evidence is not inconsistent with the hypothesis that BIPs are associated with observable physiological responses. These responses are probably not associated with the simple act of pressing the button to signal a BIP as evidenced by the heart rate data. The SCR response is especially interesting because this indicates a similar type of arousal as caused by an induced anxiety in the control group [Slater et al. 2003]. This is not inconsistent with our hypothesis that the experience of a BIP is likely to be a stressinducing event (since loss of presence in everyday life is something rarely experienced). Figure 2: Mean Wave Form for SCR in the Neighbourhood of a BIP. A similar analysis was carried out for SCR, which was measured only for a subsample of 20 subjects during the original experiment. Again the SCR results were de-trended and these normalised results were averaged. The result is shown in Figure 2. Here we see that mean SCR reaches a peak approximately 1.8s after the BIP (the solid line labelled GSR (SCR) ). If we compare with the random BIPs curve (the dotted line), we see that the pattern around the true BIPs is quite different from what is obtained from randomly placing stimuli through the time series (again 500 iterations were used). The 10 people recruited for the anxiety inducing control also had their SCR measured. The result is shown with the dash-dot line labelled anxiety control. In this 5. DETECTING ANOMALIES (BIPs?) An anomaly is a perceived fault in the simulation that signals the physical reality in which the experience is taking place. Examples include interference from cables, changes in frame rate or latency, incorrect rendering, glitches in texturing or z-buffering, and so on. These anomalies may or may not correspond to BIPs depending on how they are perceived and experienced by the participant (they themselves may not cause a BIP, and BIPs can occur spontaneously). In the previous works, we focused on the possible changes in the level of presence and the relationships with the physiological happenings. We have studied the signals BIP by the users of the VE systems. The concept of BIP is related to the VR and it needs to open to other possibility, that s why we start considering anomalies in general. We want to investigate whether such anomalies are associated with a characteristic physiological signature. If they are, then they become objective experiential or perceptual events. Usually, the designer of a VE system would want to avoid such anomalies, irrespective of whether the concept of presence was of interest in it.

Finally, we are interested in the inverse problem: if anomalies have a characteristic physiological signature, can the moments when anomalies occur be predicted by analysis of the corresponding physiological time series? If this were the case then we would have an automatic method of detecting when such anomalies occur, which would provide an overall measure of one aspect of the effectiveness of the VE. Figure 3: A volunteer in the Bar. background one after the other, and in addition there was background chatter as might be heard in a real bar. The entire experience lasted approximately 5 minutes. See Figure 4 for a typical scene. At four times the 4 projection walls became white, so that the bar and virtual people completely vanished visually for 2 seconds each. These were the induced anomalies. Physiological measures were recorded for each participant throughout. We consider whether the anomalies are signalled within the GSR data. What happened, if anything, in the neighbourhood of the anomalies in terms of the GSR data? The GSR waveform is extracted for ± 10s around each anomaly point, and averaged over all anomalies over all participants. Each extracted GSR wave has its origin set at zero at the start of the sequence, so that only changes are averaged rather than absolute values. The result is shown in Figure 5 as the black curve. On the average the GSR curve rises to a peak about 3s after the onset of the anomaly. Figure 4: The Bar scene. An experiment was conducted in a fourwalled Cave-like system, where 20 people experienced a lively scenario in a bar (Figure 3). There were 5 virtual characters in the bar, a barman and two couples, one pair standing near the bar, and the others sitting across the room. These virtual characters would be aware of the location of the participant and would often address remarks towards him/her. During the experience two songs played in the Figure 5:Average GSR waveforms. The black curve is the true signal and the coloured ones are the randomized anomalies It is possible to test the significance of this result by simulation, following the same procedures as above, except that instead of using the true anomaly times, we use 4 randomly generated times for each

participant, within the valid period of the bar experience. The coloured curves in Figure 5 show 100 such simulated curves. It is clear that none of them approaches the shape of the true generated curve. The next step is to consider if we can predict where the anomalies occurred, using an inverse relationship. We work with the hypothesis that the response to an anomaly might be embodied in the frequency domain therefore not transparent in the original time series. In order to investigate this we use a continuous wavelet transform [Mallat, 2001] of the GSR signal. The original signal is transformed by the continuous wavelet transformation into the wavelet coefficient C, using the Haar wavelet throughout. We need a more systematic approach to exploring the relationship between the anomalies and the GSR time series. Instead of using the GSR directly, we use its wavelet transformation coefficients, and we calculate a standard logistic regression [Mccullagh and Nelder, 1983], which restricts the fitted response variable to the correct range. form of wavelet coefficients at different scales and the moments in a time sequence where anomalies occurred (Figure 6). Considering the previous result, can we construct an equation from one data set and apply it to another data set, such that the equation well predicts the anomalies in the second data set? To do this we pool the data from the first 10 participants. Then the same regression analysis can be performed as in the previous section. Having fitted the regression equation we can now use that equation to estimate the anomalies for participants 11 to 20. The result is that in 8 out of the 10 cases the correlation between the predicted anomaly sequence and the true sequence was high. The two poor cases had correlations of -0.03 and 0.03. Of the remaining 8 cases, the lowest correlation was 0.19 and the highest 0.57. All of these have very high statistical significance (Figure 7). Figure 7: predicted anomalies for three participants in the set 11-12, based on an equation from participants 1-10. Figure 6: The horizontal axis shows time during the experiment and the vertical axis is the anomalies scale. The green curve represents the true anomalies and the blue one the estimated from the regression analysis. The analysis has shown that it is possible to find an equation that models the relationship between the GSR sequence expressed in the 6. THE AGORAPHOBIA EXPERIMENT At the moment, we are running the third study in a series of studies designed to uncover cues that play an importance in enhancing the realism and believability of simulations of urban environments. Previous studies highlighted two main factors in maintaining the believability of: believability of the

environment and the characters populating them. The purpose of these studies were three folds: Firstly to explore important cues in VEs that aids users to maintain a sense of presence. Secondly, to recreate a believable VE that stimulates physical responses as a real one. Finally, a subsidiary goal of this study is to drive platform work. This meant integrating, the PIAVCA 1 written in collaboration with BT Exact, DIVE 2 [DIVE], eye tracking and physiological monitoring into a stable system. One insight our work has provided is that pictorial realism is neither sufficient nor necessary to create a sense of presence. This is partially supported by work done by Longhurst et al [Longhurst et al. 2003], where studies showed artistically enhancing an image through the addition of dust, dirt and scratches increases the perceived realism of an image. What seems to be important is that the world and populating characters are consistent [Garau et al. 2003] and believable. The current study focuses on the perception of realism in urban virtual environments. It was designed to explore three aspects important to create a believable virtual environment: sense of space, realism of textures in VE and believability of characters in the VE. We are looking at interaction effects between different levels of visual realism and behavioural realism of the virtual environment and the inhabiting characters. The idea is to recreate a believable artificial environment that stimulates physical responses as a real one, but that can be individually controlled by the sufferer to experience only the features that can be handled by the patient. As the techniques to cope with the anxiety are learned trough a number of virtual sessions, the richness of the environment is increased to transform the virtual into a real world bringing the patient to be able to cope with the anxieties in the everyday experience. The previous studies in the series had confirmed the importance of the realism of the environment associated with the realism of the avatars. The world with these features has been rated with a high level of presence [Brogni et al. 2004]. The focus in the recent round of experiments has been in establishing whether physiological measures such as ECG, GSR and respiration, can be used to confirm that a person is suffering from stress in the virtual environment. By using stress as a surrogate for presence, we can investigate appropriate responses. We are currently attempting to build models that integrate this with our previous BIPSs studies [Brogni et al. 2003, Slater et al. 2003]. The ongoing study has been designed with three factors. These are the degree of visual realism with respect to the number of textures used in the virtual environment, the visual realism of the characters populating the environment and the behaviour realism Figure 8: Volunteer walking on the street. 1 Platform Independent Architecture for Virtual Character Animation (PIAVCA) was the system used to drive the animations of characters in our VEs. 2 Distributed Interactive Virtual Environment Due to time restriction and number of participants needed (80) for the complete study, it was decided that the study would be split into two phases. The first half of the

study concentrated on two factors, visual realism of the VE and the visual realism of the characters without accompanied behaviour realism that will be added in a second phase. A gender balance was maintained across the four conditions studied in the first phase. The reason for this is that there is evidence [Argyle and Cook 1976] that males and females can respond differently to nonverbal behaviours. The scenario is a common street with shops and people walking on the pavements (Figure 8 and Figure 9). In this study, the concept of presence and BIPS are being used as a tool to enable us pinpoint the aspects of VE that enhance realism and believability. A multitude of data was collected and is currently being analysed. These range from the physiologically monitored data and BIPS to subjective data collected using presence questionnaires (SUS) [Slater 1999] and in-depth post-experimental interviews. Figure 9: Characters on the street. The analysis of the physiological signals is addressed to verify the previous works about the anomalies and to define a tool for the identification of the BIPs during the VR experiences. GSR is analysed with the wavelet techniques and a high frequency (256 Hz) ECG is used to make Hear rate variability (HRV) analysis all over the BIPs occurrences. 7. CONCLUSIONS AND FUTURE WORK In this paper we introduce the concepts of presence and breaks in presence, and we have shown the results of our recent studies aimed to define a relationship between BIPs and physiological measures. In different rounds of studies, we have found some relationships between the BIPs and the physiological signals, especially HR and GSR. The arise of the HR just before the BIP happening and the following peak of the GSR after are clear indicators of a stress situation and the combined event can be associated to a break, or a decreasing, in the level of presence. The aim of the ongoing agoraphobia experiment is verify this possibility, but also evaluate the level of presence in the different condition of realism. The relationship between the realism of the avatar and the realism of the environment is the main point. If we extend the concept of BIP to the idea of anomaly, we can study the possible detection of these anomalies using only the physiological signals. An early study has shown that a detection of an induced anomaly is possible. In future work we intend to apply physiological methods to mixed-reality systems. The original concepts of breaks in presence was difficult to extend, because it was premised on the Gestalt switch between two environments, one virtual, one real, that both completely immersed the user. With mixed-reality system, immersion does not have the same implications, the illusion is that the real world is altered in some away. However the definition of presence as appropriate response is easily applicable in this situation. A mixed-reality display of a pit should still generate stress because it presents constrained but realistic cues. We can therefore use the physiological methods we have described here. We can also look for

mixed-reality anomalies. These may not be as shocking as the anomalies we have generated in the ReaCTor system because the perceptual cues involved will not be as immersive, but never the less, we might expect to be able to detect some effects from the user s noticing anomalies in the presentation of an object that is the focus of attention. If these methods prove successful, we will be able to start to demonstrate potential new uses of mixed-reality systems. We may also be able to start to investigate what disrupts users when using mixed-reality systems and thus develop better guidelines for presentation of such systems. AKNOWLEDGEMENTS This research was funded by UK EPSRC Equator project. The experiment described in Section 5 was undertaken in collaboration with the PRESENCIA project. (IST-2001-37927). We thank Angus Antley, Ashwin Beeharee, Doron Friedman, Maia Garau, Marco Gilles, Joel Jordan and David Swapp for their help. Finally, we would like to thank all the participants in the experiments. REFERENCES [Argyle and Cook 1976] Argyle M. and Cook M., Gaze and Mutual Gaze, Cambridge University Press, (1976) [Brogni et al. 2003] Brogni, A., Slater, M., Steed, A., More Breaks Less Presence, 6th Annual International Workshop on Presence ( PRESENCE 2003). [Brogni et al. 2004] Brogni A., Romano D. M., Steed A., and Slater M., "Presence through Virtual Worlds", Journal of Presence, To be Submitted, (2004) [Burdea and Coiffet 1994] Burdea, G., and Coiffet, P. 1994. Virtual Reality Technology. Wiley. [Cruz-Neira et al. 1993] Cruz-Neira, C., Sandin, D., and DeFanti, T. 1993. Surround-screen projectionbased virtual reality: The design and implementation of the CAVE. In Proceedings of SIGGRAPH 93, ACM - SIGGRAPH, 135 142. [DIVE] Swedish Institute of Computer Science, "The DIVE home page", Last accessed: 3 Nov. 2003, http://www.sics.se/dive [Freeman et al. 1999] Freeman, J., Avons, S.E., Pearson, D.E. and IJsselstijn, W.A. (1999) Effects of Sensory Information and Prior Experience on Direct Subjective Ratings of Presence, Presence: Teleoperators and Virtual Environments, 8(1), 1-13. [Freeman et al. 2001] Freeman, J., Lessiter, J., and IJsselsteijn, W. A. 2001. An introduction to presence: A sense of being there in a mediated environment. The Psychologist, British Psychological Society (April). [Garau et al. 2003] Garau M., Slater M., Vinayagamoorthy V., Brogni A., Steed A., and Sasse M. A., "The Impact of Avatar Realism and Eye Gaze Control on the Perceived Quality of Communication in a Shared Immersive Virtual Environment", SIGCHI,(April 2003) [Jang et al. 2002] Jang, D. P., Kim, I. Y., Nam, S. W., Wiederhold, B. K., Wiederhold, M. D., and Kim, S. I. 2002. Analysis of physiological rensonse to two virtual environments: Driving and flying simulation. CyberPsychology & Behavior 5, 1, 11 18. [Lombart et Ditton, 1997] Lombard, M, Ditton, T. (1997). At the Heart of it all: The Concept of Telepresence. Journal of Computer Mediated Communication, 3, 2. [Longhurst et al. 2003] Longhurst P., Ledda P., and Chalmers A., "Computer graphics, virtual reality, visualisation and interaction in Africa", Proceedings of the 2nd International Conference on Computer Graphics, Virtual Reality, Visualisation and Interaction in Africa, pp 123-132, (2003) [Mallat 2001] Mallat, S. (2001) A Wavelet Tour of Signal Processing, 2nd Edition, Academic Press. [McCullagh and Nelder 1983] McCullagh, P. and Nelder, J.A. (1983) Generalised Linear Models, Chapman and Hall. [Meehan et al. 2002] Meehan, M., Insko, B., Whitton, M., and Frederick P. Brooks, J. 2002. Physiological measures of presence in stressful virtual environments. In Proceedings of ACM - Transactions on Graphics, vol. 21, ACM - SIGGRAPH, 645 653. [Scholg et al. 2002] Scholg, A., Slater, M., and Pfurtscheller, G. 2002. Presence research and eeg. In Fifth Annual International Workshop PRESENCE 2002. [Sheridan 1992] Sheridan, T. 1992. Telerobotics, Automation and human supervisory control. The MIT Press, Cambridge, MA. [Slater and Wilbur 1997] Slater, M., and Wilbur, S. 1997. A framework for immersive virtual environments (five): Speculations on the role of presence in virtual environments. Presence: Teleoperators and Virtual Environments 6, 6, 603 616. [Slater 1999] Slater, M. 1999. Measuring presence: A response to the Witmer and Singer questionnaire. Presence: Teleoperators and Virtual Environments 8, 5, 560 566. [Slater and Steed, 2000] M. Slater and A. Steed A Virtual Presence Counter, Presence: Teleoperators and Virtual Environments, 9(5), October 2000, MIT Press, ISSN 1054-7460. [Slater et al. 2003] Slater, M., Brogni, A., Steed, A., Physiological Responses to Breaks in Presence: A

Pilot Study, 6th Annual International Workshop on Presence ( PRESENCE 2003). [Slater 2003] Slater, M. (2003) How Colourful was Your Day? Why Questionnaires Cannot Assess Presence in Virtual Environments, PRESENCE: Teleoperators and Virtual Environments, in press. [Spagnolli and Gamberini. 2002] Spagnolli, A., & Gamberini, L. (2002) Immersion/Emersion: Presence in hybrid environments. Paper presented at the Presence 2002: Fifth Annual International Workshop, Porto, Portugal, 9-11 October 2002. [Wiederhold et al. 2002] Wiederhold, B. K., Jang, D. P., Kim, S. I., and Wiederhold, M. D. 2002. Physiological monitoring as an objective tool in virtual reality therapy. CyberPsychology & Behavior 5, 1, 77 82. [Witmer and Singer 1998] Witmer, B. G., and Singer, M. J. 1998. Measuring presence in virtual environments: A presence questionnaire. Presence: Teleoperators and Virtual Environments 7, 3, 225 240.