The Relationship between Immersion and Psychophysiological Indicators

In a previous experiment, we introduced the concept of functional threshold, an objective indicator of the realism of a virtual experience (Hébert-Lavoie, Doyon-Poulin, & Ozell, 2022). A functional threshold is a sensory quality value beyond which perception is considered functionally degraded. It is a notion that complements the perceptual threshold, that is, a limiting sensory quality value for what humans can perceive.

Functional thresholds and perceptual thresholds can be used to manipulate immersion in an objective and meaningful way. Immersion is defined as the extensiveness of the simulation of sensory modalities by a system (Hein, Mai, & Hußmann, 2018). It is a factor of the concept of presence (Slater, 2009), which refers to the perceived realness of a mediated or virtual experience (Skarbez, Brooks, & Whitton, 2017). Using functional thresholds and perceptual thresholds, it is possible to constrain immersion between a minimal and a maximal bound. These constraints define the perceptual values between which presence can be evoked. The functional thresholds represent the minimal bounds, and the perceptual thresholds represent the maximal bounds. These thresholds bring objectivity and meaning to the manipulation of immersion. Rather than setting their experiment at arbitrarily chosen values of low or high immersion, researchers can use them as reference points.

This paper employed functional thresholds to study the relationship between immersion and psychophysiological indicators. Psychophysiological indicators are of interest to researchers in the field of presence, as they are considered an objective assessment of the concept (Schirm, 2021). However, while their sensitivity has been demonstrated, their relation to presence must be clarified, as they have been associated with both increases and decreases in presence (Grassini & Laumann, 2020). We suspect they might have been mistakenly considered a holistic measure of the concept. Therefore, we want to analyze how a manipulation of immersion—a factor of presence—affects psychophysiological indicators.

We have previously measured the functional thresholds for contrast sensitivity, size of the field of view (FOV), and visual acuity (Hébert-Lavoie et al., 2022). In the present study, we manipulated those visual qualities while participants executed a driving task to investigate the effect of immersion on psychophysiological indicators. This method of indirectly manipulating presence will allow us to analyze immersion's impact on psychophysiological indicators. The goal is to understand how they can be used as an objective way to quantify presence in a virtual environment. Hence, this paper aims to examine the effect of an objective degradation of immersion on psychophysiological indicators.

To contextualize the topic of this study, we will first discuss matters related to the concept of presence in a virtual environment, detail the role of the sensory modalities targeted by the study toward presence, and present the psychophysiological indicators and their relation to presence. Afterward, we will describe the methodology applied to reach our objectives and present its results. Finally, we will discuss the implication of psychophysiological indicators on immersion and conclude on their impact on presence.

Presence qualifies and quantifies a user's impression of a virtual environment (Souza, Maciel, Nedel, & Kopper, 2021). Higher presence contributes to better performance, engagement, and skill learning (Caldas, Sanchez, Mauledoux, Avilés, & Rodriguez-Guerrero, 2022; Riley, Kaber, & Draper, 2004).

Immersion represents the set of valid actions supported by a virtual environment system (Slater, 2009). It is an objective characteristic of a virtual environment designating the extent and the precision of the simulated sensory modalities (Hein et al., 2018). For example, to evaluate immersion provided by the visual modality, the following qualities must be assessed: visual accommodation, eye movements, stereopsis, perception of movement, size of the field of view, visual acuity, light and contrast sensitivity, and color perception (Hébert-Lavoie et al., 2022). Immersion provides the boundaries within which the place illusion can occur. Thus, the place illusion depends on the system used to generate a virtual environment (Slater, 2009).

Coherence refers to the aspects of a virtual environment that contributes to the plausibility illusion. It is defined as the ability of the environment to act as the user expects it to act (Skarbez et al., 2017). Coherence represents the set of reasonable circumstances that a scenario can evoke without introducing unreasonable circumstances. A reasonable circumstance is a state of a virtual scenario that is evident from prior knowledge of the individual experiencing it. Reasonable circumstances are different for every individual (Skarbez, Neyret, Brooks, Slater, & Whitton, 2017).

Contrary to coherence, immersion is an objective concept. The objective nature of immersion facilitates its measure and manipulation. Immersion can be quantified by measuring the qualities composing each sensory modality simulated by a system (Cummings & Bailenson, 2016). A virtual environment simulating more relevant sensory modalities has been shown to increase presence (Marucci et al., 2021). Still, this kind of evaluation is not meaningful in itself, as there is no clear information on the outcome of a given level of immersion.

To better indicate the expected effect of the value of sensory qualities, we have introduced the notion of “functional threshold” (Hébert-Lavoie et al., 2022). This notion is used as a perceptual boundary. It is defined as a sensory quality value below which perception should be considered functionally degraded. The other sensory boundary is the perceptual threshold, defined as the limiting sensory values resulting in perception. Hence, we expect that a virtual environment has a greater probability of evoking the place illusion if it involves sensory quality values between the functional and perceptual threshold (Hébert-Lavoie et al., 2022).

Thus, ideally, technologies that simulate the perceptual threshold of every sensory quality would be used to maximize presence. However, simulating the perceptual threshold of every sensory quality has yet to be achievable. In fact, it is believed that functional fidelity would be more attainable than perfect physical fidelity (Cummings & Bailenson, 2016). It is difficult to identify exact thresholds of immersion since their assessment is relative rather than absolute (Cummings & Bailenson, 2016). Still, functional thresholds give an estimation of the sensory quality values that are necessary to evoke presence.

Measuring immersion through the assessment of sensory quality values has several advantages that resemble a valid ideal measure of presence defined by Souza et al. (2021). It is relevant, with a direct link to presence; it is reliable and can be applied to any virtual environment; it allows the objective quantification of variation; it is non-intrusive and inexpensive, with the possibility to be applied asynchronously without the need for participants. Moreover, it is an objective characterization that considers the system used and that can be done prospectively without any user intervention. Its most prominent flaws are that it measures immersion rather than the whole concept of presence and that data related to functional and perceptual thresholds can be difficult to gather for the time being.

The following subsections present the main features of the three visual qualities used for the immersion's manipulation. For each, a functional threshold has been previously measured (Hébert-Lavoie et al., 2022).

The concept of contrast sensitivity is a means of taking into account light sensitivity without being affected by the time-varying nature of this visual quality (Wolfe et al., 2006). It enables the use of a context-independent measure of the phenomenon. Contrast sensitivity is equivalent to the slightest difference in luminance required to identify a target (Bennett, Bex, Bauer, & Merabet, 2019). When dealing with natural contrast stimuli, the recommended measurement is the root mean square (RMS) contrast expressed as the ratio of the standard deviation and the mean luminance of the image (Pelli & Bex, 2013). Contrast sensitivity is a visual quality. Manipulating it implies an impact on immersion and should theoretically affect presence. However, we could not find any paper addressing the relationship between contrast sensitivity and presence.

One's contrast sensitivity is dependent on visual conditions. When optimal, the perceptual threshold, that is, the minimum contrast that can be perceived in the fovea, is 0.2% (Thibos, Still, & Bradley, 1996). Furthermore, the functional threshold corresponds to 25.6% RMS contrast with a 95% confidence interval (CI) [24.4, 26.6] (Hébert-Lavoie et al., 2022).

The FOV corresponds to the portion of space that the eyes can capture when motionless (Wolfe et al., 2006). Generally, higher presence is evoked by values closer to the total FOV size (Duh, Lin, Kenyon, Parker, & Furness, 2002; Lin, Duh, Parker, Abi-Rached, & Furness, 2002; Slater, Spanlang, & Corominas, 2010). Larger FOV size also increases workload (Grinyer & Teather, 2022) and improves task performance (Ragan et al., 2015). However, in virtual environments with vection, causing a visual-vestibular cue conflict (Melo, Gonçalves, Narciso, & Bessa, 2021; Weech, Kenny, & Barnett-Cowan, 2019), a large FOV is a significant factor in cybersickness (Rebenitsch & Owen, 2016; Teixeira & Palmisano, 2021), a phenomenon generally negatively correlated with presence (Shi, Liang, Wu, Yu, & Xu, 2021; Weech et al., 2019).

When considering the amplitude in which the eyes can move while the head is fixed, to encompass the whole range of possible values that equipment such as a head-mounted display (HMD) should have, the perceptual threshold of the FOV is 290 degrees horizontally and 190 degrees vertically (Riecke, Nusseck, & Schulte-Pelkum, 2006). In comparison, the functional threshold of FOV is 96.6 degrees both horizontally and vertically (Hébert-Lavoie et al., 2022).

Visual acuity refers to the finest detail that can be detected or identified (Bennett et al., 2019). Generally, the notion of resolution acuity is used to measure visual acuity in a virtual environment. Resolution acuity corresponds to the smallest angular separation between two neighboring objects that can be detected (Wolfe et al., 2006).

Under optimal conditions (Iskander, Hossny, & Nahavandi, 2018), the human eye has a resolution acuity of 1 arcmin on average (Elbamby, Perfecto, Bennis, & Doppler, 2018; Kemeny, 1999; Kim et al., 2019; Livingston, 2006; Perroud, Régnier, Kemeny, & Mérienne, 2019). The identified functional threshold for visual acuity when the head is stable is 12.2 arcmin (Hébert-Lavoie et al., 2022).

Since the angular resolution of most current HMDs is 4 to 6 arcmin, the pixel density is too low to reproduce maximal visual acuity (Maxwell et al., 2018). Thus, objects are difficult to discern at distances where they would be clearly seen in real life (Blissing, 2020; Kemeny, 1999), such as writing on road signs. The lack of realism concerning visual acuity is reflected only to some extent in presence measurement (Felton, 2021). An increased screen resolution significantly increases the reported presence when there is a great difference between the tested resolution conditions (Duh et al., 2002; Hvass et al., 2017). However, smaller differences between conditions do not yield significant results (Dinh, Walker, Hodges, Song, & Kobayashi, 1999).

In summary, changing the values of visual qualities allows for manipulating immersion objectively and consistently. The contrast values that a virtual environment should offer to evoke the place illusion should be between 0.2% and 25.6%, the FOV between 96.6 degrees both horizontally and vertically, and 290 degrees horizontally and 190 vertically, and the resolution values should be between 1 arcmin and 12.2 arcmin. However, the presented functional thresholds were obtained for a task where the head was stable, and the participants only had to focus on one element of the virtual environment (Hébert-Lavoie et al., 2022). We do not know if these values would still be valid in a different context.

In recent years, there has been a growing interest in utilizing psychophysiological indicators to measure presence (Schirm, 2021). Certain psychophysiological indicators, such as basic heart rate, have been found to be associated with the coherence factor of presence (Skarbez, 2016). Psychophysiological indicators encompass a wide range of metrics, including cardiovascular response metrics and eye response metrics, which are among the most widely employed measures across various research domains (Tao et al., 2019).

Psychophysiological indicators are believed to be matched between a virtual environment and a real environment (Slater et al., 2010). They offer the benefits of being continuous, relatively non-invasive, real-time, and, to some extent, an objective assessment of presence (Grassini & Laumann, 2020; Schirm, 2021). Thus, they are helpful in investigating phenomena that are difficult to capture using self-report methods, such as questionnaire, and providing a more accurate analysis of the usefulness of a virtual environment. Using psychophysiological indicators to assess the concepts related to presence should provide a way to improve the validity, sensitivity, and diagnosticity of its measurement.

Generally, psychophysiological indicators are used to objectively evaluate cognitive activity, emotional responses, and cognitive-emotional states (Fairclough, 2009). Hence, their preferred use in evaluating presence is to assess related factors (Grassini & Laumann, 2020). Cardiovascular and eye responses have been used to evaluate phenomena such as frustration, anxiety (Fairclough, 2009), cybersickness (Chang, Kim, & Yoo, 2021; Kim, Kim, Kim, Ko, & Kim, 2005), stress, and mental workload (Lohani, Payne, & Strayer, 2019), which have been shown to affect presence (Felton & Jackson, 2022; Lackey, Salcedo, Szalma, & Hancock, 2016; Ling, Nefs, Morina, Heynderickx, & Brinkman, 2014; Ma & Kaber, 2006; Nalivaiko, Davis, Blackmore, Vakulin, & Nesbitt, 2015; Riley et al., 2004; Riva et al., 2007; Souza et al., 2021).

The assessment of these phenomena during a presence evaluation provide indications of the perceived realism of a virtual experience. However, phenomena related to presence are not a direct assessment of the concept itself. For instance, previous studies examining the relationship between presence and mental workload (Lackey et al., 2016; Ma & Kaber, 2006; Riley et al., 2004) have reported inconsistent results. In experiments involving a landmine neutralizing task (Riley et al., 2004) and an army room-clearing task (Lackey et al., 2016), a significant negative correlation between mental workload and presence was observed. However, in a free throw basketball task, mental workload was significantly positively correlated with presence (Ma & Kaber, 2006). Mental workload reflects the cost of mental resources necessary to accomplish a task (Galy, Paxion, & Berthelon, 2018; Hart & Staveland, 1988; Tao et al., 2019). Thus, in evaluating the perceived realism of a virtual scenario, mental workload can be used to assess whether two scenarios engage cognitive functions similarly with the same level of intensity (Milleville-Pennel & Charron, 2015).

Psychophysiological indicators represent the dynamic trend of mental workload and thus make it possible to estimate the instantaneous load (Paas, Tuovinen, Tabbers, & Van Gerven, 2003), that is, the number of mental resources available at a specific point in time (Xie & Salvendy, 2000). While most psychophysiological indicators were found to discriminate changes in mental workload, they were not universally valid in all task scenarios. Psychophysiological responses caused by mental workload are highly scenario-dependent, are affected by several task characteristics and individual differences (Tao et al., 2019), and, as previously pointed out, can also be attributed to several other phenomena. Hence, the use of multiple types of mental workload measures, such as self-report measures and psychophysiological indicators, is recommended to capture the concept. Various mental workload measures allow researchers to hypothesize about the factors contributing to the total load (Korbach, Brünken, & Park, 2018) and to obtain a more accurate assessment of mental workload (Miller, 2001).

There is a relation between cardiovascular responses and mental workload (Stuiver, Brookhuis, de Waard, & Mulder, 2014). Cardiovascular responses—the reactions of the heart and its related systems to psychophysiological events—are regulated by the sympathetic and parasympathetic nervous system (Berntson, Quigley, Norman, & Lozano, 2017).

Cardiovascular measures are thought to be sensitive to mental workload since sympathetic nerves take control of cardiac activity when individuals are under a high mental workload, causing distinguishable variation in the cardiovascular response (Tao et al., 2019).

Cardiovascular response measures have been validated in discriminating mental workload across varied research domains and thus are recommended for such use (Tao et al., 2019). Means of quantifying cardiovascular responses include basic heart rate measurements and heart rate variability (HRV) measures.

Basic heart rate refers to the relation between time and heartbeats. An increase in heart rate is associated with a consistent increase in mental workload during real-world and simulator driving tasks (Lohani et al., 2019; Mehler, Reimer, & Wang, 2011; Milleville-Pennel & Charron, 2015).

HRV is described as the change in the time intervals between adjacent heartbeats (McCraty & Shaffer, 2015). Since many factors influence HRV measures, an awareness of the recording context and subjects is essential. HRV is influenced by posture, age, sex, heart rate, type of activity (Shaffer & Ginsberg, 2017), health status (Berntson et al., 2017), and information regarding respiration (Miller, 2001; Shaffer & Ginsberg, 2017).

Generally, HRV metrics, such as RMSSD, SDNN, pNN50, LF power, and HF power increase with drowsiness, fatigue, and disengagement and decrease with workload, stress, and vigilance (Lohani et al., 2019; Shaffer & Ginsberg, 2017). LF/HF does not have the same dynamic toward these phenomena. A low LF/HF reflects parasympathetic dominance, which occurs when individuals engage in tend-and-befriend behaviors. A high LF/HF indicates sympathetic dominance, which is linked to fight-or-flight behaviors (Shaffer & Ginsberg, 2017).

HRV metrics were used to evaluate mental workload in car driving tasks. According to Mehler et al. (2011), RMSSD and LF power were reported to provide statistically significant differentiation, while SDNN, pNN50, HF power, and the LF/HF metrics all failed to differentiate conditions. Stuiver et al. (2014) used LF and HF power to differentiate between conditions. Both were lower with a higher mental workload but did not differ significantly between conditions. Shakouri, Ikuma, Aghazadeh, and Nahmens (2018) reported significant differentiation of mental workload conditions as measured by NASA-TLX, while RMSSD, LF, HF, and LF/HF did not yield significant results. Finally, in an aviation context, Alaimo, Esposito, Orlando, and Simoncini (2020) indicated that HRV metrics were consistent with the results obtained with NASA-TLX. Their results were not significant, but they reported that their conditions with a higher LF/HF, lower RMSSD, and SDNN corresponded to a higher NASA-TLX score.

Past studies have obtained mixed results on the effectiveness of HRV metrics as a measure of mental workload in a driving or driving-like context. The insensitivity of the HRV measures is supposedly due to an insufficient mental or emotional demand (Shakouri et al., 2018) or to a confound resulting from a continuing rise in heart rate caused by the initial reaction of the individual to the studied condition and a decrease in heart rate due to regulation effect (Stuiver et al., 2014).

Eye response metrics can capture mental workload changes induced by visually demanding tasks (Tao et al., 2019). Among the many types of eye response metrics (Mahanama et al., 2022), this paper will focus on a pupil size–related metric, that is, the low-frequency/high-frequency index of pupillary activity (LHIPA).

Studies have revealed that pupillary activity is correlated with mental workload (Mahanama et al., 2022) and drowsiness (Wilhelm, Widmann, Durst, Heine, & Otto, 2009). An increase in pupil diameter is reliably associated with an increase in mental workload, while a decrease is associated with drowsiness (Lohani et al., 2019). Pupil diameter measures are widely used to assess users' mental workload when interacting with human–computer interfaces (Mahanama et al., 2022).

The LHIPA is a wavelet-based algorithm inspired by the Index of Cognitive Activity (ICA) and Index of Pupillary Activity (IPA). It is sensitive to variations of mental workload manipulated by task difficulty. Like in cardiovascular response analysis, the low-frequency/high-frequency (LF/HF) ratio reveals changes in components of the autonomic nervous system. Decreases in pupil diameter indicate parasympathetic activation or sympathetic inhibition. Increases in pupil diameter are related to sympathetic excitation or parasympathetic inhibition (Duchowski, Krejtz, Gehrer, Bafna, & Bækgaard, 2020).

The LHIPA is more effective at detecting load than the similar IPA (Mahanama et al., 2022). It also presents the advantage of yielding significant results in working memory capacity and is negligibly affected by small angles off-axis distortion of the pupil diameter. However, it appears that it is not sufficiently sensitive to distinguish between some levels of task difficulty (Duchowski et al., 2020).

In a previous experiment, we measured the functional thresholds for contrast sensitivity, size of the field of view, and visual acuity (Hébert-Lavoie et al., 2022). Functional thresholds provide a way to meaningfully manipulate immersion, indicating the visual quality value below which a virtual environment should be considered degraded. It provides a way to analyze the effect of an objective degradation of immersion with more diagnosticity. Thus, this paper aims to study the effect of a manipulation of immersion on two psychophysiological indicators: cardiovascular response and eye response.

Basic heart rate has been shown to be related to the coherence factor of presence (Skarbez, 2016). However, there is no evidence establishing if it also pertains to immersion. Moreover, we have found no studies about the relationship between other psychophysiological indicators like HRV metrics and LHIPA and factors of presence. Because of the added perceptual load generated by degrading visual qualities, we expect the psychophysiological indicators to be sensitive to immersion manipulation below the functional threshold.

To deepen our analysis, we also assessed presence and mental workload through questionnaires. Since immersion is related to presence, we expect to observe a decrease in presence by varying the values of visual qualities below their respective functional thresholds. Also, while a link between mental workload and presence has been shown, to our knowledge, the relationship between mental workload and immersion has yet to be explored. It is unclear if an immersion degradation affects the mental workload of a virtual environment user.

To this end, this paper reports the results of an experiment where we measured the cardiovascular response, eye response, presence, and mental workload of participants driving a car in a virtual environment under various visual quality levels set above, at, or below their respective functional thresholds.

This research project was reviewed by Polytechnique Montréal's research ethics board and met the current research ethics standards on ethical conduct for research involving humans (CER-2122-35-D). Participants recruited were required to possess a valid driver's license (learner, probationary, or full license). They were all volunteers over 18 years of age, had no uncorrected vision problems, and did not have a known prevalence of being subject to cybersickness. Also, to ensure the validity of the cardiac measurements, participants had not consumed alcohol or caffeinated beverages and had not engaged in strenuous exercise for at least 12 hours before the experiment.

We recruited 67 participants (25 women, 42 men) for the experiment. Of these, 7 could not complete the experience due to cybersickness (6 women, 1 man). The following analyses exclude their data. Participants estimated their prior experience with virtual reality (VR) HMDs with a median experience of 2 hours (min = 0 hours, M = 32.5 hours, max = 1,000 hours).

The experiment's computer used an NVIDIA GeForce RTX 2080 SUPER GPU and an Intel Core i9-9900K CPU with 64GB RAM.

The participants used the HTC Vive Pro Eye HMD to complete the driving task. This HMD has a dual OLED display with a diagonal of 3.5 inches allowing a 110 degrees FOV, a refresh rate of 90 Hz, and a resolution of 1440 × 1600 pixels per eye for maximum resolution acuity of 4.6 arcmin. It generates a luminance between 0.04 and 130 cd/$m2$ and occupies the AdobeRGB colorspace (Clausen, Fischer, Fuhrmann, & Marroquim, 2019). We removed the HMD headphones for the experiment.

To drive the car during the experiment, participants used the Flashfire Suzuka Wheel 900A steering wheel and pedal assembly. The kit has a steering wheel that allows 900 degrees of adjustable rotation. It has twenty action buttons, including two handles behind the steering wheel. The gas and a brake pedal were positioned on the ground in front of our participants.

In addition, the participants used the HTC Vive Pro Eye controllers. They were strapped to their forearms to allow their hands to be tracked during the experiment while being able to touch the steering wheel without hindrance.

For the cardiovascular data gathering, we used the Polar H10.¹ It is a heart rate monitor with an adjustable strap that can be attached to a participant's chest. The device allows a heart rate measurement with a sampling frequency of 1,000 Hz. It has been reported to have a high correlation and agreement level with a gold-standard reference device (Umair, Chalabianloo, Sas, & Ersoy, 2021). Its Bluetooth connection enabled the transfer of heart rate data during the experiment. These data were used to evaluate the mental workload of our participants through the analysis of their cardiovascular response. We transferred the ECG raw data in a .csv file using the Elite HRV smartphone application, giving us access to the time elapsed between two successive R-waves, called the RR intervals.

We used the HTC Vive Pro Eye embedded eye-tracker for the eye response analysis. It has a 120 Hz gaze data output frequency and an accuracy of 0.5–1.1 degrees within 20 degrees of the center of the FOV.

During the experiment, participants could encounter a total of 27 cars on the road. A simple algorithm controlled these cars, allowing them to follow a predetermined route and avoid obstacles that may have been encountered. They were moving along the route at speeds varying between 10 and 35 km/h. They were not necessarily respecting the speed limit. Moreover, 33 pedestrians were walking or running along the road, sometimes crossing it, at speeds varying between 3 and 10 km/h. Other animated human characters enriched the environment. They were not considered pedestrians as they were motionless and positioned farther from the road. They were mainly standing up or seated and performed what appeared to be discussion animations.

To evaluate the presence and mental workload of our participants during the experiment, we used the presence questionnaire (PQ) and NASA-TLX, respectively. Both are empirically validated self-report measurement methods, providing data to support our findings. By assessing our participants' self-reported presence and mental workload, we seek to gain a more comprehensive understanding of the relationship between immersion and psychophysiological indicators.

We measured presence with a paper version of the UQO Cyberpsychology Lab revised PQ (UQO Cyberpsychology Lab, 2013), corresponding to the Witmer & Singer PQ version 3. It is a 24-item questionnaire based on the Witmer, Jerome, and Singer (2005) PQ using 19 items of the PQ version 2 (Witmer et al., 2005) and 5 items related to audio and haptics. The PQ (Witmer et al., 2005) is one of the most widely used (Hein et al., 2018) and shortest questionnaires (Souza et al., 2021) to measure presence. It presents the advantages of not being correlated with any specific use (Hein et al., 2018) and being sensitive to increases in immersion and coherence (Skarbez, Brooks, & Whitton, 2020; Skarbez, Brooks, & Whitton, 2018). Place illusion and plausibility illusion are approximately of equal importance in regard to scores on the PQ (Skarbez et al., 2018). Thus, its widespread use, reliability, and face validity ensure a certain level of comparability between studies.

We used NASA-TLX to measure our participants' mental workload subjectively. NASA-TLX is a subjective method of measuring mental workload using a multidimensional rating scale requiring operator judgment of several psychological and task-related variables (Hart & Staveland, 1988). It comprises a total of six factors. Three are task-related, namely physical demand, mental demand, and temporal demand, and three are operator-related, namely the individual's judgment of their performance, the effort required to perform the task, and the individual's level of frustration (Hart & Staveland, 1988). Physical, mental, and temporal demands are determined by the situation's complexity and seem to be indicators of both intrinsic and extraneous workload factors (Galy, Cariou, & Mélan, 2012). Effort is believed to be an indicator of germane load (Galy et al., 2018).

A review of 36 published NASA-TLX scores (Grier, 2015) allows for comparing mental workload. For the task of driving a car, the statistical distribution range found was 15.00 (min), 28.05 (25th percentile), 41.52 (50th percentile), 51.73 (75th percentile), and 68.50 (max). When considering any task, the following values were obtained: 6.21 (min.), 36.77 (25th percentile), 49.93 (50th percentile), 60.00 (75th percentile), and 88.50 (max). Thus, driving tasks impose a relatively lower mental workload than general tasks.

The questionnaire was implemented in our virtual environment. Each question appeared in a separate pane. Participants had to read the question and validate that they understood the scale, for example, from very low to very high or from perfect to failure. Then, they would rate each factor of their mental workload using their HTC Vive controller to move the cursor on the scale to a value between 0 and 20. For our analysis, we use the weighted NASA-TLX score.

During the experiment, we asked our participants to perform a driving task in the previously described virtual environment. It was required that they respect the Highway Code at all times. Each participant had to complete three scenarios during which they were exposed to one of seven visual conditions. Three scenarios were used to prevent the participants from being able to guess the events that would occur during their driving task. The different scenarios corresponded to three different starting positions on the road. Each participant did all three different scenarios in random order.

We measured the participants' situational awareness using freeze probes (see Appendix A) during the driving task. However, the present paper will not discuss the situational awareness--related results. Each scenario consisted of three trials ending with a freeze. The experimenter triggered the freezes at pre-selected times, occurring randomly every 3 to 6 minutes and 30 seconds while the driving task was executed, and the cardiovascular response and eye response data were recorded. During a freeze, the simulation was stopped, and the participants had to answer questions about their situational awareness. After the first and second freeze, participants resumed driving under the visual conditions displayed before the freeze. It took between 15 and 35 minutes to complete a scenario.

In summary, each participant performed three driving scenarios for which data was analyzed. All participants did the baseline scenario with no degradation and two degraded scenarios.

At the end of the 5 minutes necessary to record their resting heart rate baseline, we asked the participants to put on the HMD and follow the HTC Vive Pro Eye eye-tracking camera calibration procedure. The participants then performed a practice drive, allowing them to get accustomed to the car and virtual environment parameters and evaluate if they were prone to suffer from cybersickness. At this point, all visual qualities were set at their maximum value. At the end of the practice, we did a freeze probe, specifying that the questions seen in practice would not necessarily be the ones being asked during the experiment. After answering all probes, they were prompted to complete the NASA-TLX questionnaire in the virtual environment. Afterward, they removed the HMD, weighed each NASA-TLX factor for the driving task, and filled out the Presence Questionnaire (PQ) (Witmer et al., 2005).

The three driving scenarios were then executed. At the end of each scenario, the participants were questioned to make sure that they did not feel cybersickness symptoms. They were then asked to complete the NASA-TLX questionnaire, take off the HMD, and fill out the PQ.

To analyze the cardiovascular response, we processed the collected RR intervals using the Kubios HRV Standard 3.5.0 analysis package. The default settings were used to compute the basic heart rate values and the HRV statistics. Artifacts were corrected by interpolation with the Kubios HRV Standard beat correction threshold algorithm. We analyzed only the periods with less than 1% of their beats corrected. For each trial, we analyzed a fixed period of 2 minutes and 30 seconds, starting 30 seconds after the beginning of a trial. We investigated the effect of the degradation of contrast sensitivity, size of the field of view, and visual acuity on mean heart rate, HRV time-domain, and HRV frequency-domain metrics. The time-domain metrics were SDNN (ms), RMSSD (ms), and pNN50 (%); and the frequency-domain metrics were LF (⁠$ms2$⁠), HF (⁠$ms2$⁠), and LF/HF. The frequency-domain metrics were obtained based on a fast Fourier transform spectrum. Our cardiovascular metrics were log-transformed when necessary to obtain normally distributed data and residuals.

To analyze the eye response, the LHIPA was computed with an implementation of the Duchowski et al. (2020) algorithm. Eye-related data was obtained using the VIVE SRanipal SDK, which provided the position and pupil diameter of each eye during the experiment. To compute the LHIPA, the acquired data was preprocessed by removing data collected 200 ms prior to the start of a blink and 200 ms following the end of a blink.

Each participant experienced two of the six possible degradation condition scenarios and a scenario without degradation. For the cardiovascular and eye response analysis, since each scenario comprised three trials, we gathered 60 measurements (⁠$n=60$⁠) for each degradation condition and 180 measurements (⁠$n=180$⁠) for the condition without degradation.

The PQ and NASA-TLX questionnaires were completed at the end of each scenario. Thus, the questionnaires were completed twenty times (⁠$n=20$⁠) for each degradation condition and sixty times (⁠$n=60$⁠) for the condition without degradation.

Since our data were unbalanced and our participants had been subjected to repeated measurements, we used linear mixed models for our analysis. Our linear mixed models were fitted with the lme4 R package. We used the restricted maximum likelihood (REML) as an estimation method since it can give an unbiased estimate of the variance parameters given the number of participants (⁠$n=60$⁠) that took part in the experiment. We assumed an unstructured covariance matrix of random factors. Satterthwaite's method was used to compute the degrees of freedom for the t-tests and $p$-values. Every linear mixed model analysis significance was evaluated with an $α=0.05$⁠. Each model Intra-Class Correlation (⁠$ICC$⁠) was evaluated with the R sjstats package, using the adjusted value. Marginal (⁠$R2m$⁠) and conditional (⁠$R2c$⁠) R squared values were evaluated with the R MuMIn package.

We conducted Tukey-adjusted post hoc contrast tests using the emmeans R package, when necessary. These tests were conducted when we observed significant results related to the manipulation of a visual quality. The purpose of these tests was to determine whether the differences between the functional threshold and minimum value levels of the visual quality were statistically significant.

Each linear mixed model was built with the visual quality manipulation level as a fixed factor and a random intercept relating to each participant. As it is essential to consider subject-related factors (Shaffer & Ginsberg, 2017) in cardiovascular response analysis, we introduced age and gender as fixed effects in every cardiovascular response model and HR in every HRV-related model.

As shown in Table 1, variables that significantly impacted the mean heart rate were the size of the field of view at the functional threshold and the age of our participants. For the condition with the size of the FOV at the functional threshold, we recorded a significantly lower mean HR (⁠$M=71.94$⁠, $SD=12.41$⁠) compared to the condition without degradation (⁠$M=77.96$⁠, $SD=11.61$⁠), $t(476.25)=-2.92$⁠, $p=0.004$⁠. The estimate for this variable had a value of −0.0198 with a 95% CI [−0.0330, −0.0066]. Tukey-adjusted post hoc comparisons indicated that the mean heart rate was significantly lower when the FOV size was set at the functional threshold as compared with the minimum value (⁠$p<0.05$⁠, CI of the difference = −0.0382 to −0.0042). We also found that the mean HR was significantly lower for older participants.

Table 2 shows the results of the linear mixed models for RMSSD, SDNN, and pNN50. For these three models, we can see a significant negative impact of age and mean HR on the three dependent variables.

For the RMSSD model, apart from age and mean HR, the variables that significantly impacted RMSSD are the size of the FOV at the functional threshold and the minimum value. For the condition with the size of the FOV at the functional threshold (⁠$M=38.00$⁠, $SD=26.89$⁠), we recorded a significantly higher RMSSD compared to the condition without degradation (⁠$M=30.90$⁠, $SD=17.27$⁠), $t(480.27)=2.42$⁠, $p=0.02$⁠. At the minimum value (⁠$M=36.58$⁠, $SD=22.77$⁠), we also recorded a significantly higher RMSSD than the condition without degradation, $t(478.46)=2.26$⁠, $p=0.02$⁠. In addition, the estimate for the FOV at the functional threshold had a value of 0.0616 with a 95% CI [0.0119, 0.1112], and the estimate for the FOV at the minimum value had an estimate of 0.0557 with a 95% CI [0.0077, 0.1041]. Tukey-adjusted post hoc comparisons indicated that the RMSSD between the FOV size at the functional threshold and at the minimum value did not differ significantly (⁠$p=0.8555$⁠).

For the SDNN model, other than age and mean HR, the variables that significantly impacted SDNN were the contrast sensitivity and the FOV at the minimum value. For the condition with contrast sensitivity at the minimum value (⁠$M=42.83$⁠, $SD=21.95$⁠), we recorded an SDNN significantly different from the condition without degradation (⁠$M=41.94$⁠, $SD=18.03$⁠), $t(486.07)=-2.27$⁠, $p=0.02$⁠. At the FOV at the minimum value (⁠$M=45.33$⁠, $SD=18.40$⁠), the SDNN was also significantly different, $t(483.66)=2.12$⁠, $p=0.03$⁠. The estimate for the contrast sensitivity at the minimum value is −0.0688 with a 95% CI [−0.1277, −0.0094], and the estimate for the FOV at the functional threshold is 0.0640 with a 95% CI [0.0053, 0.1233]. Tukey-adjusted post hoc comparisons indicated that the SDNN between the contrast at the functional threshold and at the minimum value (⁠$p=0.0719$⁠), and between the FOV at the functional threshold and at the minimum value (⁠$p=0.5013$⁠) did not differ significantly.

Furthermore, the variables that significantly impacted pNN50, other than age and mean HR, were the contrast sensitivity and the FOV at the minimum value. For the condition with contrast sensitivity at the minimum value (⁠$M=12.91$⁠, $SD=17.21$⁠), we recorded a pNN50 significantly different from the condition without degradation (⁠$M=11.37$⁠, $SD=13.41$⁠), $t(435.57)=-2.10$⁠, $p=0.04$⁠. For the FOV at the minimum value (⁠$M=15.52$⁠, $SD=18.31$⁠), we recorded a pNN50 also significantly different from the condition without degradation, $t(433.73)=2.08$⁠, $p=0.04$⁠. The estimate for the contrast sensitivity at the minimum value is −0.1913 with a 95% CI [−0.3684, −0.0132], and the estimate for the FOV at the functional threshold is 0.1874 with a 95% CI [0.0128, 0.3646]. Tukey-adjusted post hoc comparisons indicated that the pNN50 between the contrast at the functional threshold and at the minimum value (⁠$p=0.6421$⁠), and between the FOV at the functional threshold and at the minimum value (⁠$p=0.7697$⁠) did not differ significantly.

Table 3 shows the results of the linear mixed models for LF, HF, and LF/HF. For these three models, we can see a significant negative impact of age and mean HR on LF and HF, and a significant positive impact on LF/HF.

For the LF model, contrast sensitivity at the minimum value was the only model variable with a significant impact on LF apart from age and mean HR. For contrast sensitivity at the minimum value (⁠$M=1430.86$⁠, $SD=1531.68$⁠), we recorded a significantly lower LF compared to the condition without degradation (⁠$M=1541.29$⁠, $SD=1530.10$⁠), $t(494.63)=-2.28$⁠, $p=0.02$⁠. The estimate for contrast sensitivity at the minimum value had an estimate of −0.2121 with a 95% CI [−0.3922, −0.0297]. Tukey-adjusted post hoc comparisons indicated that the LF between the contrast at the functional threshold and at the minimum value did not differ significantly (⁠$p=0.2573$⁠).

For the HF model, other than age and mean HR, the FOV at the functional threshold was the only model variable with a significant impact on HF. For FOV at the functional threshold (⁠$M=927.03$⁠, $SD=1852.05$⁠), we recorded an HF significantly higher from the condition without degradation (⁠$M=454.00$⁠, $SD=651.89$⁠), $t(484.80)=2.74$⁠, $p=0.006$⁠. The estimate for the FOV at the functional threshold is 0.2312 with a 95% CI [0.0663, 0.3957]. Tukey-adjusted post hoc comparisons indicated that the HF score between the FOV size at the functional threshold and at the minimum value did not differ significantly (⁠$p=0.1076$⁠).

For the LF/HF model, other than mean HR, the FOV at the functional threshold was the only variable with a significant impact on LF/HF. For FOV at the functional threshold (⁠$M=3.71$⁠, $SD=2.72$⁠), we recorded an LF/HF significantly lower from the condition without degradation (⁠$M=6.27$⁠, $SD=5.83$⁠), $t(489.69)=-2.75$⁠, $p=0.006$⁠. The estimate for the FOV at the functional threshold is −0.2395 with a 95% CI [−0.4092, −0.0695]. Tukey-adjusted post hoc comparisons indicated that the LF/HF was significantly lower when the FOV size was set at the functional threshold as compared with the minimum value (⁠$p<0.05$⁠, CI of the difference = −0.477 to −0.0419).

Table 4 shows the linear mixed model results for the LHIPA. Visual acuity at the minimum value was the only model variable with a significant impact on this model. For visual acuity at the minimum value (⁠$M=1.76$⁠, $SD=0.32$⁠), we recorded a significantly higher LHIPA compared to the condition without degradation (⁠$M=1.63$⁠, $SD=0.31$⁠), $t(472.90)=2.46$⁠, $p=0.01$⁠. In addition, the estimate for the visual acuity at the minimum value had an estimate of 0.1008 with a 95% CI [0.0209, 0.1810]. Tukey-adjusted post hoc comparisons indicated that the LHIPA between the visual acuity at the functional threshold and at the minimum value did not differ significantly (⁠$p=0.2891$⁠).

Table 5 shows the linear mixed model results for the PQ score. Other than the participants' prior experience in VR, the model was significantly impacted by the contrast sensitivity at the minimum value, the size of the FOV at the minimum value, and the visual acuity at the functional threshold and at the minimum value.

For contrast sensitivity at the minimum value (⁠$M=96.45$⁠, $SD=17.72$⁠), we recorded a significantly lower PQ score compared to the condition without degradation (⁠$M=105.20$⁠, $SD=16.46$⁠), $t(120.08)=-3.59$⁠, $p=0.0005$⁠. In addition, the estimate for the contrast sensitivity at the minimum value had an estimate of −8.12 with a 95% CI [−12.49, −3.77]. Tukey-adjusted post hoc comparisons indicated that the PQ score between the contrast at the functional threshold and at the minimum value did not differ significantly (⁠$p=0.3170$⁠).

For size of the FOV at the minimum value (⁠$M=100.15$⁠, $SD=14.25$⁠), we recorded a significantly lower PQ score compared to the condition without degradation, $t(120.11)=-2.85$⁠, $p=0.005$⁠. In addition, the estimate for the size of the FOV at the minimum value had an estimate of −8.44 with a 95% CI [−10.80, −2.08]. Tukey-adjusted post hoc comparisons indicated that the PQ score did not differ significantly (⁠$p=0.0578$⁠) between the FOV size at the functional threshold and at the minimum value.

For the condition with the visual acuity at the functional threshold (⁠$M=92.40$⁠, $SD=16.00$⁠), we recorded a significantly lower PQ score compared to the condition without degradation, $t(120.02)=-5.64$⁠, $p<0.0001$⁠. At the minimum value (⁠$M=82.15$⁠, $SD=14.74$⁠), we also recorded a significantly lower PQ score than the condition without degradation, $t(119.97)=-9.66$⁠, $p<0.0001$⁠. In addition, the estimate for visual acuity at the functional threshold had a value of −12.15 with a 95% CI [−16.51, −7.79], and the estimate for visual acuity at the minimum value had an estimate of −21.86 with a 95% CI [−26.23, −17.50]. Tukey-adjusted post hoc comparisons indicated that the PQ score was significantly higher when the visual acuity was set at the functional threshold as compared with the minimum value (⁠$p<0.05$⁠, CI of the difference = 2.667 to 16.8).

Table 6 shows the linear mixed model results for the NASA-TLX score. The model was significantly impacted by the visual acuity at the functional threshold and at the minimum value.

For the condition with the visual acuity at the functional threshold (⁠$M=45.38$⁠, $SD=15.29$⁠), we recorded a significantly higher NASA-TLX score compared to the condition without degradation (⁠$M=36.84$⁠, $SD=17.25$⁠), $t(120.35)=3.90$⁠, $p=0.0002$⁠. At the minimum value (⁠$M=54.74$⁠, $SD=20.26$⁠), we also recorded a significantly higher NASA-TLX score than the condition without degradation, $t(120.35)=7.56$⁠, $p<0.0001$⁠. In addition, the estimate for visual acuity at the functional threshold had a value of 9.49 with a 95% CI [4.80, 14.18], and the estimate for visual acuity at the minimum value had an estimate of 18.39 with a 95% CI [13.70, 23.08]. Tukey-adjusted post hoc comparisons indicated that the NASA-TLX score was significantly lower when the visual acuity was set at the functional threshold as compared with the minimum value (⁠$p<0.05$⁠, CI of the difference = −16.0 to −1.82086).

The objective of this paper was to study the effect of a manipulation of immersion on two psychophysiological indicators: cardiovascular response and eye response. To deepen our analysis, we also assessed presence and mental workload through questionnaires. For this purpose, we degraded three visual qualities while our participants executed a driving task in a virtual environment. During the driving task, we recorded their cardiac and eye responses and asked them to fill out the PQ and NASA-TLX questionnaires.

There were no significant differences between the contrast sensitivity degradation at the functional threshold and the condition without degradation for cardiac and eye response metrics. Thus, our experiment does not allow us to suspect an effect of degradation of the contrast sensitivity at this value on the studied psychophysiological indicators. However, as shown in Tables 2 and 3, there was a significant but relatively low effect on the degradation of contrast sensitivity at the minimum value on SDNN, pNN50, and LF. The three HRV metrics indicate a lower score in that degradation condition. In this condition, we could assume a higher mental workload, stress, or vigilance due to the low contrast rendering many elements hard to distinguish during the driving task. Participants had to be more focused to be able to reach their goals in the virtual environment. The comparisons between the degradation conditions at the functional threshold and at the minimum value did not yield any significant differences for the three examined psychophysiological indicators. This lack of significant differences suggests that psychophysiological indicators may not be a sensitive measure for assessing contrast variation in a virtual environment.

The linear mixed models of mean HR, RMSSD, HF, and LF/HF revealed a significant difference between the measures of FOV at the functional threshold and the condition without degradation. Mean HR and LF/HF were lower in the functional threshold FOV condition than in the condition without degradation. Additionally, RMSSD and HF were higher in the functional threshold FOV condition than in the condition without degradation. The same effect was observed for the FOV at the minimum value condition. RMSSD, SDNN, and pNN50 were significantly different in the condition with the FOV minimum value than in the condition without degradation. In all cases, the FOV at the minimum value had a relatively low effect on the significant HRV metrics. These results were unexpected as we anticipated that a smaller FOV, implying a lower immersion, and thus a lower presence, would affect the cardiovascular response indicating a higher mental workload, stress, or vigilance. We assumed that a reduced FOV would force our participants to make more head movements, spending more mental resources gathering the information needed to accomplish their tasks.

A first explanation would link these effects on psychophysiological indicators to an unconscious decrease in cybersickness. It is known that a virtual environment like ours, with camera displacement while the body stays static, can cause this kind of discomfort (Melo et al., 2021) due to visual-vestibular cue conflict (Weech et al., 2019). This fact is evidenced by the number of participants that had to withdraw from our experience because of nausea or some informal comments at the end of the experiment regarding slight dizziness. Hence, we suppose that our participants were unconsciously experiencing cybersickness, even with the measure put in place to mitigate this phenomenon. The psychophysiological indicators were significantly affected in the conditions where the FOV was degraded because a smaller FOV imposed a lower cybersickness. Alternatively, we could explain these results because there are less data to acquire and process with a smaller FOV, causing a decrease in mental workload. We suppose so because, when compared to the Grier (2015) NASA-TLX analysis, our driving task imposed a low workload. The condition without degradation has a median score of 35.32, which is lower than the 50th percentile of a driving task. Thus, the information required to carry out the task might have been readily available even with a reduced FOV.

The comparisons between the FOV degradation conditions at the functional threshold and the minimum value did not yield any significant differences for RMSSD, SDNN, pNN50, and HF. These findings suggest that these specific psychophysiological indicators may not possess sufficient sensitivity to effectively assess FOV variation in a virtual environment. However, a significant difference was observed for mean HR and LF/HF between the degradation conditions at the functional threshold and the minimum value. Unlike the aforementioned indicators, mean HR and LF/HF are capable of distinguishing between FOV at the functional threshold and the minimum value. This distinction may be attributed to two factors. Firstly, basic heart rate measures like mean HR have an advantage over HRV metrics in terms of sensitivity for tasks with a relatively low workload (Mehler et al., 2011). Secondly, LF/HF exhibits a heightened sensitivity to parasympathetic dominance compared to other individual indicators (Shaffer & Ginsberg, 2017). Nonetheless, it is important to note that our experiment does not allow us to draw definitive conclusions regarding the impact of FOV variation on cardiovascular response due to seemingly confounded results.

Visual acuity at the functional threshold does not significantly affect any psychophysiological indicators in our experiment. However, we acknowledge that the functional threshold for visual acuity in the present study differs from the experiment in which it was measured. In that previous experiment (Hébert-Lavoie et al., 2022), the functional threshold was determined in a task involving large objects, which did not require specific visual acuity. Thus, it was postulated that functional thresholds are context-dependent and that the task performed in a virtual environment should modulate the functional threshold. In our present experiment, since the task involved reading characters at a distance, the functional threshold should have been set accordingly, allowing its accomplishment without difficulty. However, the technology at our disposal did not provide a sufficient resolution for that purpose. Furthermore, the PQ and NASA-TLX scores suggest a potential effect of visual acuity manipulation on presence-related variables such as psychophysiological indicators. However, the scope of our conclusions related to the PQ and NASA-TLX score is limited due to a small number of observations, $n=20$ per degradation condition. Both measures exhibited significant differences between the condition without degradation and the visual acuity degradation conditions, indicating a lower presence and a higher mental workload. Also, the comparison between the visual acuity degradation conditions at the functional threshold and at the minimum value for the PQ and NASA-TLX scores was significant, implying sensitivity between different visual acuity levels. Thus, we expected the LHIPA to yield a significant difference between the condition with the visual acuity at the functional threshold and the condition without degradation.

The condition with the visual acuity at the minimum value was the LHIPA linear model's only significant variable, showing a relatively low effect of visual acuity at the minimum value on the measure. A low visual acuity makes many elements, such as speed limits, indistinguishable. As for the condition of contrast sensitivity at the minimum value, participants had to be more focused and vigilant in this condition to reach their goals in the virtual environment. Hence, this effect could imply a higher mental workload at the minimum value for this condition, as supported by the NASA-TLX score. No significant differences were found when comparing the visual acuity degradation conditions at the functional threshold and at the minimum value. Therefore, it can be concluded that LHIPA does not exhibit a high sensitivity to visual acuity manipulation.

To summarize, our findings suggest that not all psychophysiological indicators demonstrate equal sensitivity to the tested visual quality manipulations. Specifically, our data imply that SDNN, pNN50, and LF are sensitive to contrast sensitivity under the functional threshold. Our results also indicate that LHIPA is sensitive to a virtual environment with a visual acuity under the functional threshold. We cannot confirm if the degradation of FOV affects any psychophysiological measure, since the effect found in mean HR, RMSSD, SDNN, pNN50, HF, and LF/HF seems to be confounded. Nevertheless, these indicators may still offer insights into the effects of FOV reduction. Overall, it appears that psychophysiological indicators lack the necessary sensitivity to accurately measure an immersion manipulation. This is evident given the relatively low effect of immersion manipulation on psychophysiological indicators, and because most of our results fail to distinguish between the functional threshold and minimum value conditions.

For basic heart rate, the residual variability was explained by the participants' random intercept with an ICC of 0.92, indicating excellent reliability between our raters. The ICC of HRV metrics, ranging from 0.53 to 0.82, indicated moderate to good reliability between our participants. Hence, for these models, the variation was mainly due to differences between participants. On the other hand, the LHIPA model had an ICC of 0.39, indicating poor reliability. For this model, the variation between participants was less important than the variation in each participant measurement. Thus, a greater number of measurements for each participant would have helped reduce the uncertainty of the model.

Otherwise, it appears that the inclusion of random effects of participants was relevant. This is shown by the differences between the $R2m$⁠, which estimates the fraction of the variance explained by the fixed effects, and $R2c$⁠, which estimates the fraction explained by the fixed and random effects. Since $R2c$ is constantly greater than $R2m$⁠, including random effects improved our models' accuracy.

In summary, the linear mixed models used in this paper enabled us to deepen our comprehension of the influence of visual quality degradation on immersion. They helped us understand the context in which the psychophysiological measures can be applied.

Our results regarding the degradation of contrast sensitivity and visual acuity revealed that these immersion conditions only significantly impacted our measured psychophysiological indicators when they were at a value below the functional threshold. These visual conditions appeared to decrease presence, although this conclusion is limited due to a lack of statistical power.

Still, we do not think this presence decrease is only due to the immersion factor's degradation. Instead, we argue that these visual conditions affected the coherence of our environment, the other factor of presence. Coherence is defined as the ability of a virtual environment to act as a user expects it to act (Skarbez et al., 2017). We introduced unreasonable circumstances while manipulating immersion by degrading contrast sensitivity and visual acuity under the functional threshold. These unreasonable circumstances changed how our participants would typically respond and thus their expectations toward the virtual environment. For instance, when the visual acuity in the virtual environment was at the minimum value, the resolution was so low that our participants had to change their strategy to acquire relevant information. For example, they had to delay the moment they perceived the information or change their positioning to perceive it appropriately. Thus, we suspect that the unreasonable circumstances introduced by the degradation of visual qualities under the functional threshold were responsible for the psychophysiological indicators' significant results. The introduction of unreasonable circumstances implies that the resulting coherence degradation was responsible for these significant results. This assertion is in agreement with Skarbez's (2016) results. Skarbez indicated that inconsistent behavior of a virtual environment affected participants' heart rates. While the basic heart rate was not significantly different for our contrast sensitivity and visual acuity conditions, we observed a significant effect on the other studied psychophysiological indicators for which variations are associated with similar phenomena, such as HRV metrics and LHIPA. Thus, we consider that an immersion's degradation does not affect psychophysiological indicators unless they introduce unreasonable circumstances. While more work is needed to confirm such affirmation, we suggest that psychophysiological indicators could constitute an objective measure of coherence. Although we suspect a similar effect of the size of the FOV on our psychophysiological metrics, we could not show it with our experiment, as other factors affected our measures.

Consequently, since it measures a factor responsible for variations in psychophysiological indicators, we suggest that mental workload assessed by the NASA-TLX is also related to the coherence of a virtual environment. In our experiment, the visual acuity's degradation conditions seemed to impose a higher mental workload as subjectively evaluated with the NASA-TLX. However, we would need more statistical power to have more confidence in the significance of the difference between the NASA-TLX score of the condition without degradation and its score for the visual acuity's degradation conditions. Still, our visual acuity's degradation results allow us to argue that subjectively assessed mental workload is also affected by unreasonable circumstances. Moreover, we suspect we did not observe a similar effect in the contrast sensitivity condition under the functional threshold because participants felt they could still perform adequately. Some participants in the contrast sensitivity at the minimum value condition mentioned that they felt they were driving in fog. Their ability to create reasonable circumstances explaining the degradation helped them adapt their behavior without subjectively impacting their mental workload. Thus, we suggest that subjectively assessed mental workload also evaluates the coherence of a virtual environment.

The primary outcome of this paper is the suggestion that psychophysiological indicators, such as cardiovascular and eye response, are only sensitive to considerable degradation of immersion. Since they have been proven to affect presence, we have proposed that it was a measure of coherence rather than immersion. However, this affirmation must be taken carefully since we have yet to test all combinations. Other psychophysiological indicators might have unsuspected effects on our tested sensory quality. Moreover, the tested cardiovascular and eye metrics might significantly affect other sensory qualities.

As shown by the related work, cardiovascular response and eye response are related to variations in mental workload. However, this link is not direct and other phenomena such as cybersickness and anxiety can also cause these indicators to vary. The complexity of psychophysiological inferences (Fairclough, 2009) represent considerable limitations in the usage of psychophysiological indicators as a measure of presence.

A limitation of this study is the potential impact of luminance on the computation of LHIPA, as it was not controlled for during the manipulation of visual quality. This suggests that the results could have been affected by changes in pupil size due to variation in luminance. However, it should be noted that LHIPA has been shown to be more resilient to variations in luminance compared to other similar pupillary activity measures (Duchowski et al., 2020). Future studies should take into consideration the potential impact of luminance on LHIPA and aim to control for this variable.

Otherwise, cybersickness could have confounded our results. As previously discussed, a virtual environment with vection like the one we used can cause cybersickness (Teixeira & Palmisano, 2021). We believe that cybersickness might have had a minor effect on participants that completed the experience, but it was not important enough to be consciously acknowledged during its execution. Moreover, we do not believe that this phenomenon influenced conditions other than those where we manipulated the size of the FOV. Our participants received a clear directive to stop the experience if they felt cybersickness symptoms. Thus, other than for the degradation of the FOV, we do not suspect that this phenomenon affected our results in a way that would radically change our conclusions.

Also, since the chosen metrics were selected for a driving task, they might not be appropriate for every virtual environment activity. For example, to study mental workload, HRV metrics are only suited for tasks where participants are seated or mostly immobile (Shaffer & Ginsberg, 2017). Thus, researchers must carefully choose the psychophysiological indicators they use and be aware of the context in which they will be applied while studying presence in a virtual environment.

Moreover, we do not imply that psychophysiological indicator can completely capture the concept of coherence. Instead, we believe that cardiovascular and eye response indicators are sensitive to some variation in coherence and can be a helpful diagnostic tool to assess presence in a virtual environment.

In this experiment, we objectively manipulated immersion by setting a virtual environment contrast sensitivity, size of the FOV, or visual acuity values above, at, or below the functional threshold. The functional thresholds are sensory quality values below which perception is considered degraded. They represent a meaningful value to set visual qualities in a virtual environment since they allow using a reference value for immersion study. By using meaningful visual quality values rather than arbitrarily chosen ones, it is possible to better grasp the effect of immersion degradation in a virtual environment.

Our findings indicate that reducing contrast sensitivity and visual acuity below the functional threshold had a significant but relatively low impact on cardiovascular and eye response metrics. We noticed that when sensory qualities are set below the functional threshold, they introduce unreasonable circumstances that affect the users' behavior in the virtual environment. Based on these observations, we propose that psychophysiological indicators are not a direct measure of immersion but should rather be used to assess the coherence factor of presence objectively. Furthermore, we contend that immersion directly impacts coherence when sensory quality values are set below the functional threshold.

This work was supported by Mitacs, the Consortium for Aerospace Research and Innovation in Canada (CARIC), the Consortium de recherche et d'innovation en aérospatiale au Québec (CRIAQ), and by its partners under GRANT CARIC-CRIAQ MDO-1649-TRL4+.