Breathing training guided by an embodied agent in immersive virtual reality: a study of psychological and physiological effects 

By Christian Corrò & Luca Chittaro, University of Udine, Italy, christian.corro@uniud.it 

Abstract 

Slow deep breathing (SDB) techniques offer physical and mental health benefits that help mitigate stress, anxiety, and pain. However, economic and logistical barriers limit access to traditional breathing training programs. While self-help resources like videos and apps are available, they lack the personalized guidance and feedback provided by human trainers and tend to have high dropout rates. The increasing availability of consumer-grade virtual reality (VR) headsets presents new opportunities for immersive breathing training experiences guided by embodied agents, potentially enhancing engagement and adherence. While immersive VR (iVR) has shown benefits for relaxation, much of the existing research has not focused on using an embodied agent as a coach for SDB training, leaving unclear whether this approach can benefit from iVR. This paper addresses this question by comparing agent-guided SDB training in iVR (headset-based) vs. non-immersive VR (based on a computer monitor). We introduce and evaluate an embodied agent that provides users with real-time, personalized feedback on their respiratory frequency and depth. We evaluate both self-reported measures (including engagement, attention, presence, and perception of the embodied agent) and objective physiological signals, comprising respiration, electrodermal activity, cardiac activity, and brain prefrontal cortex activity (acquired through functional near-infrared spectroscopy, fNIRS). Results highlight several benefits of iVR over non-immersive VR. They showed that iVR increased users’ attention, engagement, sense of presence, social presence, and perceived anthropomorphism of the embodied agent compared to non-immersive VR. Physiologically, iVR improved depth of breathing at the target SDB frequency, enhanced cardiorespiratory coupling, and decreased sympathetic arousal. The study included an evaluation of a secondary, smaller personalization technique (name personalization: the embodied agent could or could not address participants by name), which did not yield any significant result. The paper describes and discusses these findings, highlighting the important role that iVR could play in enhancing the effectiveness of breathing training. 

1 Introduction 

Slow deep breathing (SDB) techniques are commonly included in complementary non-pharmacological treatments to mitigate stress, anxiety, and pain (Gholamrezaei et al., 2022; Jafari et al., 2017; Joseph et al., 2022; Nahin et al., 2015; Pancini et al., 2025; Perciavalle et al., 2017; Shao et al., 2024). SDB involves intentionally reducing the respiratory frequency to around 4–6 breaths per minute, promoting a state of relaxation and well-being (Lehrer et al., 2000). Its benefits are attributed to the modulation of the autonomic nervous system, particularly the enhancement of parasympathetic activity, which leads to improved autonomic balance and a reduction in stress responses (Zaccaro et al., 2018). SDB requires consistent training and practice to develop proficiency, particularly for beginners. For instance, learning how to properly use the diaphragm for breathing can be challenging for novices and typically requires personalized guidance to master. Unfortunately, logistical and economic barriers often prevent people from accessing SDB training programs, reducing their opportunities for regular training, learning, and practice. While self-help resources, such as books, videos, and apps (Chittaro & Sioni, 2014), have been created to assist with breathing training, these alternatives lack the personalized feedback and guidance of a human instructor (Zhu et al., 2017). As a result, they suffer high dropout rates, making it difficult for users to maintain consistent breathing exercises as part of their routine (Shamekhi & Bickmore, 2018). 

Virtual agents have been proposed as a solution to address these limitations as they can deliver personalized, real-time feedback through conversational interfaces and rich nonverbal cues (Shamekhi & Bickmore, 2018). These interfaces help simplify instructions, reducing the cognitive load associated with following guidance, which is essential in ensuring effective learning and consistent practice of techniques like SDB. When they are embodied, virtual agents offer a unique advantage in replicating both verbal and nonverbal communication typical of human counseling, while also providing key benefits such as enhanced privacy, reduced costs, and overcoming geographical barriers (Provoost et al., 2017). In the context of breathing training, an embodied agent can monitor the user’s respiratory rate and depth, providing immediate feedback and guidance to help the user maintain the correct breathing pattern and to guide them in correctly using the diaphragm. 

In recent years, immersive virtual reality (iVR)-based relaxation systems have gained attention due to their ability to immerse users in highly engaging, controlled environments that could facilitate deeper states of relaxation (Barton et al., 2024; Blum et al., 2019; de Zambotti et al., 2022; Pancini et al., 2025; Riches et al., 2021; Rockstroh et al., 2019; Weibel et al., 2023). In these virtual environments (VEs), users can practice SDB techniques by following visual and/or audio instructions. The sense of presence provided by iVR can help in reducing external distractions, facilitating a more focused practice (Kuvar et al., 2024; Pancini et al., 2025). The availability of consumer-grade virtual reality (VR) headsets presents new opportunities for immersive breathing training experiences guided by an embodied agent, potentially enhancing engagement and adherence. However, while VR-based relaxation systems have shown significant potential in enhancing user experience, much of the existing research has not focused on using an embodied agent as a coach for SDB training. Consequently, it remains unclear whether agent-guided SDB training can benefit from iVR. 

This paper addresses this question by comparing SDB training guided by an embodied agent in iVR (headset) vs. non-immersive VR (computer monitor). The embodied agent we propose and study is capable of providing personalized instruction with real-time feedback on users’ breathing behavior, including respiratory frequency and depth. 

Furthermore, this study explores a secondary smaller personalization technique: name personalization, in which the embodied agent addresses participants by name. This was included to explore whether such a simple personalization technique produces any measurable effect on engagement and attention, both of which are crucial for adherence and effective training outcomes (Ait Baha et al., 2023; Provoost et al., 2020). Unlike previous studies, we do not focus solely on users’ self- reported measures like engagement, presence, and perception of the embodied agent. Our study also provides an in-depth analysis of users’ physiological signals including respiration, electrodermal activity, cardiac activity, and brain prefrontal cortex activity (acquired through functional near-infrared spectroscopy, fNIRS). These measures allowed for a deeper understanding of the users’ psychophysiological responses and provided objective confirmation of the self-reported measures. 

2 RelatedWorkandMotivation 

Despite the widespread use of SDB as a complementary treatment strategy to manage stress, anxiety, and pain, the underlying mechanisms of SDB are not yet fully understood (Jafari et al., 2020). They involve a combination of cognitive (e.g., distraction from stressors), emotional (e.g., reduced arousal and improved emotional regulation), and autonomic modulations (Gholamrezaei et al., 2022). Below, we provide a brief overview of the physiological mechanisms of SDB necessary to understand the subsequent sections of this paper. The effects of SDB are closely linked to the autonomic nervous system (ANS), which governs involuntary physiological processes (Gholamrezaei et al., 2022; Russo et al., 2017). The ANS consists of two primary branches: the sympathetic nervous system (SNS) and the parasympathetic nervous system (PNS). The SNS, often referred to as the “fight-or-flight” system, prepares the body to respond to stress by increasing heart and breathing rate, inhibiting digestion, and redistributing blood flow to essential muscles, among other effects. While necessary for immediate threats, prolonged activation of the SNS is associated with chronic stress, anxiety, and various health complications (Sevoz-Couche & Laborde, 2022). In contrast, the PNS, known as the “rest-and-digest” system, slows heart rate and breathing rate, promotes digestion, facilitates relaxation, and supports recovery, among other effects (Sezer & Sacchet, 2025). SDB promotes balance within the ANS by stimulating the PNS and reducing SNS overactivity. This effect is mediated through the activation of the vagus nerve, a key component of the PNS that plays a crucial role in regulating the cardiovascular, respiratory, and gastrointestinal systems (Zaccaro et al., 2018). During SDB, the rhythmic expansion and contraction of the lungs stimulate vagal afferent pathways, which can enhance parasympathetic tone and suppress sympathetic arousal. As a result, SDB can induce a state of relaxation, marked by a reduced heart rate, lower blood pressure, and increased heart rate variability (HRV), a key indicator of autonomic balance. In addition to its physiological effects, SDB also impacts emotional and cognitive states by reducing the arousal levels associated with sympathetic overactivity. By inducing relaxation and emotional regulation, SDB can alleviate stress and anxiety (Fincham et al., 2023). 

SDB involves reducing the breathing frequency to around 0.1 Hz (i.e., 6 breaths per minute), but it is not as straightforward as it may sound. It is necessary to maintain sustained attention on one’s breath for an extended period to achieve the benefits described above. SDB requires consistent training and practice to develop proficiency. Guidance is essential to correct errors and maintain focus, particularly for beginners. Embodied agents, designed to mimic interactions with human trainers, can offer a viable solution for SDB training. They have been used in various types of health interventions (Provoost et al., 2017; Ter Stal et al., 2020), effectively promoting adherence to long-term treatments (T. Bickmore & Picard, 2005; T. W. Bickmore et al., 2010; Winkler et al., 2025). Embodied agents can also foster social 

presence, defined as the user’s psychological experience of “being with another” in a mediated environment (Biocca et al., 2003). When users experience social presence, they are more likely to develop trust and rapport, which are essential psychological constructs for effective intervention (Oh et al., 2018). Social presence can lead to increased users’ engagement, motivation, adherence, and overall satisfaction with the intervention (Provoost et al., 2017). In the context of SDB training, social presence could encourage users to engage more deeply with the exercises, practice consistently, and achieve better training outcomes. 

In summary, the use of an embodied agent as a coach for SDB training can be framed within a conceptual pathway that may link agent embodiment to users’ psychophysiological outcomes. By providing verbal and nonverbal cues that replicate human interaction, embodied agents may foster social presence (Biocca et al., 2003). Social presence, in turn, may promote trust and rapport, which are known to enhance engagement and adherence (Provoost et al., 2017). Enhanced engagement may reduce mind-wandering and increase attention to the training task (Kuvar et al., 2024), which could improve compliance with the breathing pattern and ultimately support the psychophysiological modulation associated with SDB. Pacing alternatives such as visual cues can provide temporal information but may lack relational and communicative properties that could sustain engagement over time. Audio guidance may go further, delivering verbal instructions, corrective feedback, and encouragement, and potentially fostering a degree of relational engagement. However, it does not convey nonverbal cues such as visible breathing demonstrations, facial expressions, and body language, which are considered central to the sense of social presence and co-regulation that characterize face-to- face coaching (T. Bickmore & Picard, 2005). Moreover, according to Social Learning Theory (Bandura, 1977), people can learn new behaviors through observational learning, that is, by observing and replicating the behavior of a model. The embodied agent may serve as such a model by enabling a visible demonstration that provides a motor reference for the user to follow. This may be especially relevant for beginners, who may have not yet internalized the breathing pattern and may therefore rely more on external modeling to guide their breathing behavior and maintain the correct breathing pace over time (Shamekhi & Bickmore, 2018). 

To the best of our knowledge, only one user study has specifically investigated SDB training facilitated by an embodied agent. In (Shamekhi & Bickmore, 2018), an embodied agent acted as a coach to guide users through SDB sessions. The virtual coach provided adaptive guidance based on real-time respiration data, offering feedback and adjusting instructions accordingly. The study demonstrated the effectiveness of this approach in promoting relaxation. However, it relied solely on self-reported measures, such as user satisfaction and self-reported anxiety, without an analysis of physiological responses. Additionally, the experience was based on non-immersive VR, using a computer monitor. Consumer-grade VR headsets open new possibilities for immersive breathing training. A growing body of research (Barton et al., 2024; Blum et al., 2019; Chittaro et al., 2024; de Zambotti et al., 2022; Hu et al., 2021; Pancini et al., 2025; Rockstroh et al., 2019; Savaş et al., 2024; Weerdmeester et al., 2022; Weibel et al., 2023) has demonstrated that iVR offers benefits in terms of enhanced engagement, increased motivation, relaxation, improved mood regulation and emotional regulation, and reduced anxiety by exploiting VEs that reproduce nature and biofeedback mechanisms, i.e., making users aware of their physiological signals, for example by visualizing them (Lüddecke & Felnhofer, 2022). 

Biofeedback in iVR has received growing attention, with wearable physiological sensors increasingly incorporated into iVR applications to enable real-time physiological monitoring and adaptive feedback delivery (Guillen-Sanz et al., 2024). Respiration is a relevant physiological signal in this context, monitored in iVR applications related to mental well-being and relaxation (Guillen-Sanz 

et al., 2024). In particular, game-based iVR systems have shown that coupling diaphragmatic breathing with biofeedback in an immersive environment can effectively reduce anxiety and physiological arousal, e.g., (Weerdmeester et al., 2022). In such systems, users can observe their own physiological state through real-time visualizations, which can foster self-efficacy, i.e., the belief in one’s ability to manage a given situation (Bandura, 1997), and internal locus of control, i.e., the belief that outcomes in their life are determined by one’s own actions (Rotter, 1966). These psychological mechanisms have been identified as key factors underlying the effectiveness of biofeedback-based experiences (Weerdmeester et al., 2022). Despite its potential, biofeedback remains underutilized in iVR, partly due to the technical complexity of extracting and processing physiological signals in real time from wearable devices and integrating them into common game engines (Guillen-Sanz et al., 2024). Moreover, while the real-time visualizations used in game-based iVR biofeedback systems may be engaging, they may not be easily interpretable and may not provide the explicit corrective guidance that beginners need (Gaume et al., 2016). The system proposed in this study differs from game-based iVR biofeedback systems in an important way. It integrates an embodied agent directly into the biofeedback loop, enabling the agent to deliver personalized, real-time corrective feedback through both verbal instructions and nonverbal cues, aiming to replicate the individualized guidance provided by a human trainer. The use of iVR for SDB training guided by an embodied agent remains largely unexplored in the literature, a gap that this study aims to help address. 

3 The Proposed System 

The proposed system was developed in Unity 2022.3.12f. It comprises three main components: the VE, the embodied agent and the breathing training mechanism. 

3.1 Virtual Environment 

The VE represents a natural landscape featuring a dirt path bordered by a lush meadow, with a setting sun visible in the background. In this VE, there are a variety of trees, including large oaks and birches, with logs placed for seating. This design choice was made because natural scenery has been shown to promote mood enhancements in VEs (Anderson et al., 2017; Gao et al., 2019; Valtchanov et al., 2010). The user sits on a wooden log bench, surrounded by the meadow’s trees and grass, in front of the embodied agent (Figure 1). 

3.2 Embodied Agent 

The 3D model of the embodied agent was created using the character creation tool Ready Player Me. Its appearance is designed to be a female in her mid-thirties. The embodied agent is capable of both verbal and nonverbal communication. 

3.2.1 Nonverbal communication 

The embodied agent exhibits a range of nonverbal behaviors, including iconic, emblematic, and deictic hand gestures, posture shifts, eyebrow raises for emphasis, and head nods. Additionally, it provides guidance for breathing exercises through its gestures, such as progressively raising its hands to signal inhalation for 4 seconds (Figure 1b) and progressively lowering them for 6 seconds during exhalation 

Nonverbal behaviors were recorded using an Xsens Link motion capture suit to provide natural movements. These behaviors are partly pre-determined and partly responsive to the participant’s breathing patterns. For example, the embodied agent displays positive facial expressions, such as subtle smiles and nods, when the user reaches the correct breathing frequency. It can also inflate and deflate its abdomen to visually cue proper diaphragmatic breathing technique. 

3.2.2 Verbal communication 

Since the quality of the instructor’s voice is crucial in relaxation practices (Shamekhi & Bickmore, 2018), the embodied agent uses a natural, pre-recorded voice to sound natural to the user. The agent can address participants by name. At the beginning of the session, the virtual agent can ask for the participant’s name. Once the system captured the participant’s name using Azure speech-to-text, synthetic voice clips with predefined feedback containing the name were generated. The synthetic voice was cloned from the above-mentioned natural voice using the Eleven Labs voice cloning service. The predefined feedback consisted of short utterances for each specific issue. All possible combinations of deviations (e.g., fast inhalation combined with shallow depth, or simultaneous deviations in both frequency and depth) were addressed. 

3.3 Breathing Training 

The SDB training involved diaphragmatic breathing with an inhale duration of 4 seconds and an exhale duration of 6 seconds, maintaining a frequency of 0.1 Hz, i.e., 6 breaths per minute. This specific breathing frequency has been shown to maximize HRV (Lehrer et al., 2000; Shaffer & Ginsberg, 2017). At the beginning of a session, the virtual agent welcomed the participant (Figure 1a) and asked for their name. It then provided a brief overview of the exercise with verbal instructions, such as “when you inhale, inflate your abdomen and when you exhale, deflate it.” Following this, the virtual agent instructed the participant to sit comfortably and breathe normally for 2 minutes to record the baseline physiological signals. After the baseline was recorded, the participant was asked to take some deep breaths for 30 seconds. This was done to calibrate the maximum depth of inhalation, which would be used as a reference value for feedback on breathing depth during the training. Then the virtual agent began breathing training. Breathing training was implemented as a structured sequence that guided the participant through the breathing pattern using both verbal and nonverbal cues. The training consisted of alternating phases of verbal instruction, i.e., counting the seconds during inhalation and exhalation, and nonverbal guidance, interleaved with verbal and nonverbal feedback on the participants’ breathing behavior. During the nonverbal guidance phase, the virtual agent displayed nonverbal cues, such as smiling or raising and lowering hands, to indicate the timing of inhalation and exhalation. This pattern of verbal and nonverbal phases was repeated throughout the session for a total training time of 6 minutes. The training concluded with the virtual agent thanking and greeting the participant. 

3.3.1 Biofeedback mechanism 

The biofeedback system acquires the respiratory signal in real time via a Thought Technology elastic girth sensor to measure diaphragmatic respiratory activity. The sensor is connected to a Thought Technology ProComp Infiniti encoder, which digitizes the signal at a sampling rate of 10 Hz. The digitized signal is transmitted to the application through a WebSocket connection with approximately 1 ms of latency, enabling continuous monitoring of the participant’s breathing behavior throughout the session. At the beginning of each session, a calibration phase is performed. Participants are first instructed to breathe normally for 2 minutes to establish a baseline respiratory signal. They are then asked to perform deep breaths for 30 seconds. The reference value for maximum inhalation depth is computed as the median peak amplitude across these breathing cycles, in order to reduce the influence of potential artifacts or anomalous breathing episodes on subsequent depth calculations. Two parameters are extracted and monitored in real time: respiratory frequency and depth. Respiratory frequency is estimated by detecting successive peaks and troughs in the incoming respiratory waveform to compute the instantaneous breathing rate on a breath-by-breath basis. The respiratory frequency used for feedback is calculated as the mean of the instantaneous frequencies derived from the last three breathing cycles. Respiratory depth is computed as the percentage of the current breathing amplitude relative to the individually calibrated maximum inhalation depth obtained during the calibration phase. During the training, the system continuously compares these parameters against the target values. Corrective feedback is triggered when the respiratory frequency deviates by more than 20% from the target frequency of 0.1 Hz, or when the breathing depth falls below 20% of the individually calibrated maximum. Feedback is delivered at a maximum rate of one instance per minute. These thresholds and rate were determined through pilot testing conducted prior to the study, which indicated that they represented a suitable balance between sensitivity to meaningful deviations from the target breathing pattern and avoiding overwhelming the participant with excessive corrections. When both respiratory 

frequency and depth threshold conditions are satisfied, positive reinforcement is provided. Feedback is delivered through the embodied agent via both verbal and nonverbal channels. This closed-loop system, in which the agent continuously monitors the user’s physiological state and adapts its guidance accordingly, constitutes the core biofeedback mechanism of the proposed system. 

4 UserStudy 

We conducted a 2×2 mixed-design study to assess the effects of VR type and of name personalization on engagement, attention, presence, perception of the virtual agent, SDB performance and physiological responses, including respiration, cardiac activity, EDA, and brain activity. Name personalization was the between-subjects variable, while VR type was the within-subject variable. The two levels of name personalization were: (i) Name: the virtual agent addressed participants by their name during the training, and (ii) noName: the virtual agent did not use the participant’s name. The two levels of VR type were: (i) immersive VR (iVR), where participants used a VR headset, and (ii) non-immersive VR (non-iVR), where participants used a computer monitor. 

4.1 Participants 

The study involved 36 participants (22 male, 14 female) with an age range of 22 to 56 years (M = 27.00, SD = 9.87). A priori power analysis was conducted using G*Power version 3.1.9.7 (Faul et al., 2007) to determine the minimum sample size required to test the study hypotheses. Results indicated the required sample size to achieve 80% power for detecting a medium effect, at a significance criterion of α = .05, was 34 for a 2×2 mixed-design ANOVA. The recruited sample size of 36 is thus adequate to test the study hypotheses. The participants were recruited by reaching out directly to graduate and undergraduate students at our university and did not receive any compensation for their participation. We asked them how many hours they had used a VR headset and to rate their familiarity with breathing exercises both in the past and within the last month on a 7-point scale (1 = not at all, 7 = very). Participants were divided into the two groups (Name and noName) in such a way that: (i) each group had 18 participants (11 male, 7 female); (ii) the two groups were similar in terms of age (Name: M = 26.10, SD = 9.97; noName: M = 28.10, SD = 10.49), the amount of hours of VR headset use (Name: M = 12.40, SD = 20.70; noName: M = 16.80, SD = 38.20), familiarity with breathing exercises in the past (Name: M = 2.08, SD = 1.44; noName: M = 2.70, SD = 2.26) and familiarity with breathing exercises in the last month, which was very low (Name: M = 1.42, SD = 1.65; noName: M = 1.20, SD = 0.63). Each of the four demographic variables was analyzed using a one-way ANOVA, which confirmed no statistically significant differences between the two groups. Gender distribution across the two groups did not significantly differ, as indicated by a chi-square test. 

4.2 Materials 

The system was run on a PC equipped with a 2.70 GHz Intel i7-1085H CPU, 32 GB RAM, and an NVidia Quadro RTX 3000 GPU. In the iVR condition, participants experienced the training session through a Meta Quest 3 VR headset connected to the PC via Quest Link. In the non-iVR condition, the training session was displayed in full-screen mode on an HP 24″ monitor with 1920×1080 resolution and 60 Hz refresh rate. The distance between the screen and the participant was 1.30 m. To minimize external influences on the autonomic nervous system, tests were conducted in a quiet room at a controlled temperature of 23°C. Participants used over-ear headphones with microphone (Sony WH- 1000XM4) in both conditions to minimize auditory distractions. In the non-iVR condition, potential visual distractions were reduced by positioning portable room dividers to the sides of the participant. The lights in the room were dimmed to match the lighting of the virtual environment. To record participants’ physiological data, we employed five sensors, following the placement suggestions described in (Andreassi, 2006): (i) an elastic girth sensor was positioned 1 cm above the navel to measure diaphragmatic respiratory activity; participants were asked to exhale fully, and the sensor was adjusted to be securely fitted but not uncomfortable; (ii) the two electrodes of the electrodermal activity (EDA) sensor were placed on the intermediate phalanges of the middle and ring fingers of the left hand; (iii) a blood volume pulse (BVP) finger clip sensor was placed on the distal phalanx of the index finger of the left hand, and (iv) two functional near-infrared spectroscopy (fNIRS) sensors were placed symmetrically on the participant’s forehead, in the electrode positions Fp1 (left) and Fp2 (right) of the international 10-20 system, to measure brain activity bilaterally in the prefrontal cortex (PFC). The data from the first two sensors (respiratory activity and EDA) were recorded by a Thought Technology ProComp Infiniti encoder with a sampling rate of 10 Hz. The data from the other sensors (BVP and fNIRS) were recorded using a BioSignalsPlux encoder (8-channel, 16-bit resolution version) and OpenSignals software (version 2.2.5) with a sampling rate of 500 Hz, because a sampling rate of 500 Hz is recommended for HRV analysis (Electrophysiology, 1996; Pinna et al., 1994). 

4.3 Measures 

All the questionnaires listed in the following were administered to the participants in digital form through a computer monitor using the PsyToolkit platform (Stoet, 2010, 2017). 

4.3.1 Self-reported measures 

1) Engagement: To measure participants’ engagement, we used the short form of the User Engagement Scale (UES) questionnaire (O’Brien et al., 2018). It comprises 12 items and four subscales: focused attention (the level of immersion and concentration, e.g., losing track of time), perceived usability (how easy and intuitive the system is to use, along with any frustrations experienced), aesthetic appeal (how visually pleasing the system is), and reward (interest and gratification gained from the interaction). Participants rated each item on a 5-point scale (1 = strongly disagree, 5 = strongly agree). Scores for each subscale were calculated by averaging the responses of their respective items. 

2) Attention: To assess participants’ attention, we used the Mindful Attention Awareness Scale – State (MAAS-S), a 5-item questionnaire that measures its short-term expression. Participants rated their experiences for each item on a 7-point scale (0 = not at all, 6 = very much). Example items include: “I was finding it difficult to stay focused on what was happening” and “I was preoccupied with the future or the past”. Higher scores, calculated by reverse scoring all items and averaging the responses, reflect greater levels of state mindfulness. 

3) Sense of presence: To measure participants’ sense of presence, we used the Igroup Presence Questionnaire (IPQ) (Schubert et al., 2001), a widely used questionnaire for assessing participants’ sense of presence in virtual environments. The IPQ consists of 14 items across three subscales: spatial presence (the feeling of being physically present in the virtual space), involvement (the level of attention and immersion in the environment), and experienced realism (how realistic the environment feels to the user). Additionally, the IPQ includes an independent item, being there, which assesses the overall sense of being present in the virtual environment. Participants rate each item on a 7-point scale (0 = strongly disagree, 6 = strongly agree), and scores for each subscale are calculated by averaging the values of responses of their respective items. 

4) Social presence: To measure participants’ social presence, we used the Social Presence Scale (SPS) (Bailenson et al., 2001), which assesses with 5 items the extent to which participants perceive the presence and awareness of others in a virtual or mediated environment. In the context of this study, social presence specifically refers to the degree to which participants felt the virtual agent was aware of and engaged with their presence during the SDB training. Participants rate each item on a 7-point scale (1 = strongly disagree, 7 = strongly agree), with higher scores, calculated by averaging the values of responses of all items, indicating a stronger sense of social presence. 

5) Perception of the virtual agent: To assess participants’ perception of the virtual agent, we used the Godspeed Questionnaire Series (GQS) (Bartneck et al., 2009). GQS is commonly used to measure users’ perception of robots and virtual agents. It includes 24 items distributed across five subscales: anthropomorphism (how human-like the virtual agent appears), animacy (perceived lifelikeness), likeability (how much the user likes the virtual agent), perceived intelligence (how smart the virtual agent appears), and perceived safety (how safe users feel interacting with it). Each item is rated on a 7- point semantic differential scale, with scores calculated by averaging the responses for each subscale. 

6) Relaxation: To measure participants’ relaxation levels, we used the Relaxation State Questionnaire (RSQ) (Steghaus & Poth, 2022), a 10-item questionnaire designed to assess short-term changes in subjective relaxation. The RSQ measures four subscales: muscle relaxation (focusing on physical sensations like reduced muscle tension), cardiovascular relaxation (assessing perceived changes in breathing and heart activity), general relaxation (overall feelings of calmness and relaxation), and sleepiness. The fourth subscale, sleepiness, is useful to differentiate relaxation from drowsiness. Participants rated each item on a 5-point Likert scale (1 = do not agree at all, 5 = totally agree), and scores were calculated by averaging the responses for each subscale. 

7) Brief interview: As a last step, each participant was briefly interviewed. The interview had two main objectives: first, to gather insights from participants by allowing them to share their thoughts on the overall experience, and second, to collect feedback on how to improve the experience by identifying its strengths and weaknesses from the participants’ perspective. 

4.3.2 Physiological measures 

1) Power of the target respiration frequency band: We adopted the method described in (Chittaro & Sioni, 2014) to calculate the signal power in the 0.09–0.11 Hz band of the respiratory power spectrum, recorded during SDB training. This measure represents the intensity of the respiratory component near the breathing frequency encouraged by the virtual agent (0.1 Hz, i.e., 6 breaths per minute). Higher values for this measure indicate deeper breaths at the encouraged frequency, which is the goal of the SDB training. 

2) Heart Rate Variability (HRV) refers to the physiological phenomenon of variation in the time intervals between consecutive heartbeats (Pham et al., 2025). It is a non-invasive measure that reflects the regulation of the cardiovascular system operated by the autonomic nervous system (ANS), providing insights into the balance between sympathetic and parasympathetic activity. HRV is widely recognized as an indicator of stress and relaxation levels; higher HRV is associated with increased parasympathetic activity and a relaxed state, while lower HRV indicates sympathetic dominance and heightened stress. 

In our study, HRV was assessed using data collected from the blood volume pulse (BVP) sensor that detects changes (corresponding to the cardiac cycle) in blood volume within the peripheral vasculature. The resulting BVP signal exhibits a characteristic waveform with peaks representing individual heartbeats, from which inter-beat intervals (R-R intervals) and heart rate (HR) can be derived. Instantaneous R-R intervals were calculated from the BVP waveform using a peak detection algorithm from the NeuroKit2 Python library (Makowski et al., 2021) to identify successive peaks, generating a continuous R-R tachogram. All data were visually inspected to ensure that only normal-to-normal (NN) intervals were analyzed; artifacts and ectopic beats were removed to maintain data integrity. To quantify HRV, we used both time-domain and frequency-domain measures. For the time-domain measure, we opted for the standard deviation of NN intervals (SDNN). SDNN reflects the total variability in heart rate over the recording period and is considered a robust measure of overall autonomic activity. We selected SDNN over other widely used time-domain measures, such as the root mean square of successive differences (RMSSD), because RMSSD can be problematic in the context of SDB due to its sensitivity to respiratory patterns (Ali et al., 2023). SDNN has been shown to be more reliable under SDB conditions (Ali et al., 2023). 

For the frequency-domain measure, we focused on Respiratory Sinus Arrhythmia (RSA), which refers to the natural variation in heart rate that occurs during the breathing cycle, with heart rate increasing during inhalation and decreasing during exhalation. RSA is primarily governed by efferent vagal activity from the parasympathetic nervous system, specifically through the vagus nerve’s modulation of the heart (Porges, 2007). This makes RSA a valuable indicator of parasympathetic tone and autonomic nervous system function during breathing exercises. As described in (Ali et al., 2023), to quantify RSA we employed frequency-domain analysis of the NN intervals. A Fast Fourier Transform (FFT) was applied to the NN intervals to compute the power spectral density (PSD) of heart rate oscillations, allowing the distribution of power to be quantified across specific frequency bands associated with autonomic regulation (Electrophysiology, 1996). Two components are usually distinguished from short-term cardiac recordings: a low-frequency (LF) component (0.04–0.15 Hz) and a high-frequency (HF) component (0.15–0.4 Hz). The LF component reflects a combination of sympathetic and parasympathetic influences (Chalaye et al., 2009). The HF component is primarily mediated by parasympathetic activity and is closely linked to RSA, reflecting heart rate fluctuations synchronized with the respiratory cycle (Berntson et al., 1997). Under normal breathing conditions, the HF component reflects the respiratory-related HRV, while the LF component is generally independent of respiration. However, during SDB at approximately 0.1 Hz (6 breaths per minute), respiratory- induced oscillations in HR shift from the HF band to the LF band (Stark et al., 2000). This shift results in a significant increase in LF power due to the resonant properties of the cardiovascular system (Lehrer et al., 2000; Van Diest et al., 2014) as seen in Figure 2. This phenomenon reflects enhanced cardiorespiratory coupling, parasympathetic activity and improved baroreflex function (Lehrer et al., 2003). Since the standard RSA calculation from the HF band is not applicable during SDB at 0.1 Hz, we adopted the Deep Breathing RSA (DB-RSA) computation (Ali et al., 2023), which explicitly accounts for this frequency shift. DB-RSA is calculated using the following formula: DB-RSA = ln(LFtraining − LFbaseline + HFtraining). This calculation involves determining the LF power during the SDB training session (LFtraining) and subtracting the LF power measured during the baseline period (LFbaseline) to eliminate the contribution of baseline autonomic activity, isolating the effect of the SDB training on the LF band. We then added the HF power during the SDB training (HFtraining) to capture the parasympathetic activity present in the HF range. The natural logarithm of this value provides the DB- RSA measure, reflecting the strength of the heart rate oscillations induced by SDB. 

By analyzing SDNN as a time-domain measure for HRV and DB-RSA as a frequency-domain measure, we aimed to capture comprehensive information about the autonomic regulation of HR in response to the SDB training under the two different experimental conditions (iVR vs non-iVR). 

Baseline values were subtracted from the SDNN and DB-RSA recorded during the training to separate physiological responses to the training from intrinsic physiological differences among participants (Andreassi, 2006). This baseline subtraction permitted us to isolate and evaluate the physiological responses of the SDB training. Measuring HRV provided a reliable indicator of parasympathetic activation. The increase in HRV during SDB reflects the activation of the parasympathetic nervous system, which is associated with relaxation, stress reduction, and overall well-being. Therefore, in our study, we used HR and HRV, including both SDNN and DB-RSA, as objective physiological measures to assess the participants’ heart activity responses to SDB training. 

3) Electrodermal Activity (EDA) is a reliable and widely used indicator of sympathetic arousal, providing insights into emotional and cognitive states (Boucsein, 2012; Critchley, 2002). Since EDA is not influenced by parasympathetic activity, it serves as a specific marker of sympathetic responses (Boucsein, 2012). EDA consists of two main components: the tonic component, known as skin conductance level (SCL), and the phasic component, known as skin conductance responses (SCRs). Both components are driven by sympathetic nervous system activity. For the analysis, we used the NeuroKit2 Python library (Makowski et al., 2021) to preprocess the raw EDA data and decompose it into SCL and SCR components. To enhance signal quality and reduce high-frequency noise, we applied a fourth-order Butterworth low-pass filter with a cutoff frequency of 5 Hz, which is a standard practice in EDA signal processing (Boucsein, 2012). Following the recommendations in (Boucsein, 2012), we calculated the mean SCL and identified SCR spikes. To extract non-specific skin conductance responses (NS-SCRs), i.e., spontaneous SCRs that occur without identifiable external stimuli, two measures were calculated: (i) NS-SCR frequency, representing the number of NS-SCRs per minute (spikes/min), and (ii) NS-SCR amplitude, indicating the average amplitude of NS-SCRs. Baseline values were subtracted from the data recorded during the training. By analyzing SCL, NS-SCR frequency and NS-SCR amplitude, we aimed to capture a comprehensive picture of the participants’ sympathetic nervous system activity in response to SDB. 

4) Prefrontal Cortex Activity: Neuronal activation increases the metabolic demands of brain tissue, leading to an increase of regional cerebral blood flow. This hemodynamic response results in an increase in the concentration of oxygenated hemoglobin (HbO) and a decrease in deoxygenated hemoglobin (HbR) within the activated cortical areas (Tinga et al., 2021; Villringer & Chance, 1997). We focused on the prefrontal cortex (PFC) because changes in PFC activity have been associated with variations in cognitive load and stress regulation (Arnsten, 2009; Fishburn et al., 2014), making it an interesting region to study during SDB training. By measuring PFC activity, we aimed to evaluate neural correlates of cognitive load reduction and enhanced relaxation possibly induced by SDB training. We employed functional near-infrared spectroscopy (fNIRS) to measure changes in HbO and HbR concentrations bilaterally in the PFC. fNIRS is a non-invasive optical imaging technique that quantifies cerebral hemodynamics by detecting changes in near-infrared light absorption, which are directly related to HbO and HbR concentrations in cortical tissue. The near-infrared light emitted by the sensor penetrates the scalp and skull to reach the cerebral cortex. The light is then scattered and absorbed depending on the concentration of oxygenated and deoxygenated hemoglobin, allowing for the measurement of brain activity in the underlying cortical areas. Using analysis software provided by BioSignalsPlux, the raw optical density signals were converted into concentration changes using the modified Beer–Lambert law, accounting for the differential pathlength factor to correct for the optical pathlength in biological tissue. To enhance signal quality and minimize physiological noise and motion artifacts, we applied a preprocessing pipeline using the same software. The fNIRS signals were filtered using a 5th-order Butterworth band-pass filter to attenuate low-frequency drifts (e.g., respiration-related fluctuations) and high-frequency noise (e.g., cardiac pulsations). Baseline correction was performed to eliminate slow signal drifts, and a moving average filter was applied to smooth the signals and remove artifacts. To comprehensively evaluate participants’ neural activation during the SDB training, we analyzed both cerebral oxygenation (HbDiff), calculated as HbO – HbR and total hemoglobin concentration (HbT), calculated as HbO + HbR. Previous research suggests that increases in HbT are linked to enhanced cerebral blood flow, while rises in HbDiff reflect improved oxygen delivery (Gentili et al., 2013; Tachtsidis et al., 2009; Tinga et al., 2021). We analyzed these hemodynamic responses to assess changes in PFC activity associated with SDB training. 

4.4 Procedure 

Participants were informed that the study investigated breathing training under different conditions: using a VR headset and a computer monitor. However, they were not informed of the personalization technique employed by the virtual agent. Consent for participation was obtained, and participants were informed that they could withdraw from the study at any time without providing a reason. Then, the experimenter applied the physiological sensors to the participant. The demographic questionnaire and RSQ questionnaire were then administered. The order of the presentation of the within-subject variable was randomized across participants to control for order effects. For the iVR condition, after the pre-test measurements, the experimenter assisted participants in wearing the VR headset. Over-ear headphones were placed on the participant, and the experimenter initiated the physiological data recording and the system. Participants underwent a 2-minute baseline period during which the virtual agent instructed them to sit quietly and breathe normally while physiological data were recorded. This baseline measurement served as a reference for subsequent analyses. Following the baseline, participants engaged in a 6-minute personalized SDB training session led by the virtual agent. Upon completion of the experimental condition, the recording of physiological signals was stopped, and post-test measurements were administered immediately. These included the RSQ, UES, MAAS-S, SPS, IPQ and GQS questionnaires in that sequence. After a 5-minute break, the RSQ pre-test questionnaire was administered again, and participants tried the other experimental condition, following the same procedure as the first. Finally, the experimenter removed the sensors and asked participants if they were available for a semi-structured interview. After the interview, participants were thanked for their participation. 

4.5 Hypotheses 

Based on previous research, which has consistently shown that iVR enhances user engagement, sense of presence, and social presence, we hypothesized that (H1) participants in the iVR condition should exhibit higher levels of engagement, sense of presence, and social presence compared to those in the non-iVR condition (W. Huang et al., 2021; Nowak & Biocca, 2003; Pancini et al., 2025). IVR has also been extensively applied in mindfulness training programs. These programs have been shown to improve relaxation, mindfulness levels and meditation experience, reduce anxiety and depression, enhance sleep quality, regulate emotions, and elevate mood (Ma et al., 2023). Given these findings, we hypothesized that (H2) participants in the iVR condition should report higher levels of self-reported relaxation and attention compared to those in the non-iVR condition. Additionally, iVR is expected to promote relaxation by enhancing parasympathetic activity and reducing sympathetic activity. These effects are consistent with the idea that iVR environments can facilitate a deeper relaxation response, by immersing users in more engaging, calming, and controlled environments compared to non- immersive setups, which may not elicit the same type of physiological response. We hypothesize that (H3) physiological measures related to parasympathetic activity, including HRV indicators like SDNN and DB-RSA, should increase in the iVR condition. Conversely, (H4) measures associated with sympathetic arousal, including SCL and NS-SCR frequency and amplitude, are expected to decrease, indicating reduced sympathetic activity. In this study, the investigation of PFC activity using fNIRS is exploratory in nature. 

4.6 Statistical analyses 

All statistical analyses were conducted with Jamovi version 2.6.2. Table 1 and Table 2 report the means and standard deviations of the self-reported measures and the physiological measures, respectively, while Figure 3 and Figure 4 graphically illustrate the means. For measures assessed only in the post- SDB training questionnaire, we employed a 2 × 2 mixed-design ANOVA, with name personalization (Name vs. noName) as the between-subjects variable and VR type (iVR vs. non-iVR) as the within- subjects variable. For measures assessed at two time points (pre and post-SDB training), such as the self-reported relaxation state measured with the RSQ questionnaire, we conducted a 2 × 2 × 2 mixed- design ANOVA with name personalization as the between-subjects variable, VR type as the within- subjects variable, and time (pre vs. post) as the additional within-subjects variable. Prior to conducting the mixed-design ANOVA, we verified that all necessary assumptions were met, as described in (Cohen, 2013). Effect sizes are reported as partial eta squared (η2𝑝). The between-subjects variable, i.e., name personalization, did not yield any significant differences across any of the dependent variables considered, nor were there any significant interaction effects involving it. Therefore, the following sections concentrate on the other independent variables. 

To address the risk of Type I error inflation associated with the number of dependent variables, we organized dependent variables into a hierarchy of endpoints and applied a familywise error rate correction. The primary endpoints were the power of the target respiration frequency band and DB- RSA, as they directly measured the core physiological outcomes of SDB training: the former captured breathing accuracy at the target frequency, the latter captured the cardiorespiratory coupling and resonance that SDB could produce. The secondary endpoints comprised the self-reported measures (engagement, attention, presence, social presence, perception of the embodied agent, and relaxation), the cardiac activity measures (HR and SDNN), and the EDA measures. Prefrontal cortex activity (HbT and HbDiff) was designated as an exploratory endpoint. To control the familywise error rate, we applied Holm’s sequential Bonferroni procedure (Holm, 1979), a step-down method that controls the inflation of Type I error while providing greater statistical power than Bonferroni correction (Eichstaedt et al., 2013). Following (Bender & Lange, 2001), families were formed based on the hierarchy of endpoints and on the construct or domain each set of measures addressed. The two primary endpoints formed one family. Among the secondary endpoints, the following families were defined: engagement (the UES subscales), sense of presence (the IPQ subscales), perception of the embodied agent (the GQS subscales), relaxation (the RSQ subscales), cardiac activity (HR and SDNN), and EDA (SCL, NS-SCR frequency, and NS-SCR amplitude). Attention and social presence each comprised a single scale and therefore did not require familywise correction (Eichstaedt et al., 2013). The p-values for HbT and HbDiff are also uncorrected, due to the exploratory nature of these analyses. All other p-values reported in the following sections are corrected. 

5 Results
5.1 Self-reported measures 

The reliability of all questionnaires was assessed, with Cronbach’s alpha values always greater than 0.78. The means for each subscale are provided in Table 1. 

5.1.1 Engagement
Analysis of the UES revealed a main effect of VR type on one subscale: Focused attention, F(1, 34) = 

5.32, p = .029, η2𝑝= .14. Mean score was significantly higher in iVR compared to non-iVR. 5.1.2 Attention 

Analysis of the MAAS-S revealed a main effect of VR type, F(1, 34) = 19.22, p < .001, η2𝑝= .36. Mean score was significantly higher in iVR compared to non-iVR. 

5.1.3 Sense of presence 

Analysis of the IPQ revealed a main effect of VR type on all IPQ subscales. For the Being There subscale, F(1, 34) = 87.47, p < .001, η2𝑝= .72, for Spatial Presence, F(1, 34) = 112.98, p < .001, η2𝑝= .77, for Involvement, F(1, 34) = 66.78, p < .001, η2𝑝= .66 and for Realism, F(1, 34) = 7.94, p = .008, η2𝑝= .26. For all subscales, mean scores were significantly higher in iVR compared to non-iVR. 

5.1.4 Social presence
Analysis of the SPS revealed a main effect of VR type, F(1, 34) = 5.83, p = .021, η2𝑝= .15. Mean score was significantly higher in iVR condition compared to non-iVR. 

5.1.5 Perception of the virtual agent
Analysis of the GQS revealed a main effect of VR type only on the Anthropomorphism subscale, F(1, 

34) = 18.33, p < .001, η2𝑝= .35. Mean score was significantly higher in iVR compared to non-iVR. 5.1.6 Relaxation 

Analysis of the RSQ revealed a main effect of time on both general relaxation and cardiovascular relaxation subscales. For general relaxation, there was a main effect of time F(1, 34) = 7.68, p = .009, η2𝑝= .26, and a significant interaction between VR type and time, F(1, 34) = 6.48, p = .026, η2𝑝= .16. Pairwise comparisons with Bonferroni correction revealed that only the iVR condition resulted in a significant increase in general relaxation from pre- to post-SDB training (p < .05). Additionally, iVR significantly increased relaxation at post-SDB training compared to the non-iVR condition (p < .05). For cardiovascular relaxation, a significant main effect of time was found, F(1, 34) = 10.16, p = .004, η2𝑝= .23, as well as a significant interaction between VR type and time, F(1, 34) = 7.164, p = .016, η2𝑝= .17. Pairwise comparisons with Bonferroni correction revealed that only the iVR condition showed a significant increase in cardiovascular relaxation from pre- to post-SDB training (p = .001). Additionally, iVR significantly increased cardiovascular relaxation at post-SDB training compared to the non-iVR condition (p < .05).

5.2 Physiological measures 

HR and HRV data from one participant
signal, caused by frequent hand movements during the recording session. SCL, NS-SCR frequency, and NS-SCR amplitude data were removed from 2 participants due to the sensors detaching before the end of the recording session. HbT and HbDiff data were excluded from 8 participants due to sensor displacement, particularly when wearing the VR headset. A Shapiro–Wilk normality test was performed for each of the physiological measures considered, including the power of the target frequency band, HR, SDNN, DB-RSA, SCL, NS-SCR frequency, NS-SCR amplitude, HbT, and HbDiff. All followed a normal distribution, with p-values above 0.26. The means for each measure are provided in Table 2. 

5.2.1 Respiration 

A main effect of VR type on the power of the target respiration frequency band was observed, F(1, 34) = 5.81, p = .027, η2𝑝 = .15. Mean power of the target respiration frequency band was significantly higher in iVR compared to non-iVR condition. This result indicates that iVR significantly enhanced the depth of breaths at the target frequency of 0.1 Hz (6 breaths per minute), which was the primary objective of the SDB training. 

5.2.2 Cardiac Activity 

No statistically significant differences were found for HR or SDNN, while a main effect of VR type on DB-RSA was observed, F(1, 33) = 6.34, p = .021, η2𝑝 = .16. Mean DB-RSA was significantly higher in iVR compared to non-iVR. 

5.2.3 Electrodermal Activity 

No statistically significant differences were found for SCL, while a main effect of VR type was observed on NS-SCR frequency, F(1, 32) = 7.74, p = .009, η2𝑝 = .22, and on NS-SCR amplitude, F(1, 32) = 5.12, p = .039, η2𝑝 = .13. Both means were significantly lower in iVR. These results indicate that iVR decreased both frequency and amplitude of NS-SCRs compared to the non-iVR condition, suggesting a reduction in sympathetic arousal spikes during training. 

5.2.4 Prefrontal cortex activity 

A main effect of VR type was found for HbT, F(1, 26) = 4.87, p = .036, η2𝑝 = .15, and for HbDiff, F(1, 26) = 4.61, p = .041, η2𝑝 = .14. Both means were significantly lower in the iVR compared to the non- 

iVR. These results suggest that iVR led to a reduction in blood flow and oxygenation in the prefrontal cortex during the SDB training compared to non-iVR. However, because these analyses are exploratory and the p-values are uncorrected, the results should be regarded as preliminary. 

5.3 Participants’ feedback 

The feedback from participants offered valuable insights into their experiences, particularly regarding engagement, attention, relaxation, and perception of the virtual agent. Most participants (N = 32) reported feeling more engaged and focused when using iVR, emphasizing the enhanced attention provided by the immersive experience. For instance, P1 stated “With the headset, I felt completely absorbed in the task. I was concentrating and following the exercise much better compared to the screen version”. Many participants (N = 21) noted that when using the computer monitor, they tended to think about unrelated topics and paid less attention to the instructions, which is consistent with mind- wandering behavior, i.e., the shift of attention from the task at hand to self-generated thoughts and feelings that are unrelated to the external environment or the current activity (Vago & Zeidan, 2016). For example, P6 noted that “When using the computer monitor, my mind often drifted to unrelated thoughts, and I found it harder to stay focused on the breathing exercise”. Regarding relaxation, the majority (N = 18) found iVR to be more calming than the non-immersive condition. P23 remarked, “The environment felt much more soothing using the headset, allowing me to relax deeply”. Several participants (N = 16) reported that the gestures of the virtual agent were more noticeable and helpful in iVR. For example, P14 noted, “Using the headset, I could see her movements more clearly. They felt like real movements from a person in front of me, not like watching a video on a screen”. Additionally, 14 participants mentioned that they noticed the abdominal movements of the virtual agent only in iVR. A good number of participants (N = 12) said they were pleasantly surprised by the corrective feedback they received. They reported that it made them feel that the virtual agent was aware of their actual breathing behavior. For example, P7 noted, “When she told me to slow down my breathing, I was surprised that she could actually understand whether I was following the exercise correctly”. Finally, as it often happens with people who are unfamiliar with SDB exercises, 12 participants reported that following the breathing pattern was difficult, in particular exhaling for six seconds. P12 noted, “I found it challenging to follow the breathing pattern, especially when it came to exhaling. I felt like the rhythm wasn’t right for me”. 

6 Discussion 

The results of both self-reported and physiological measures, together with users’ feedback, provide a comprehensive understanding of the positive impact of iVR compared to non-iVR on personalized SDB training led by a virtual agent. 

6.1 Engagement, Attention and SDB performance 

The hypothesis that participants in the iVR condition would exhibit higher engagement and attention compared to non-iVR was confirmed. Participants reported significantly higher levels of engagement in iVR compared to non-iVR, specifically in focused attention. This finding aligns with previous studies demonstrating increased engagement in iVR, e.g., (W. Huang et al., 2021). Furthermore, participants reported significantly higher attention during the training in iVR compared to non-iVR. In non-iVR, participants tended to think more about topics unrelated to the training and paid less attention to the instructions of the virtual agent. They thus experienced more internal distractions, which are spontaneous, task-unrelated thoughts that divert attention away from the primary task and are indicative of mind wandering. On the contrary, the iVR condition reduced internal distractions, enabling participants to stay more focused on the training. Physiological measures further support this finding: iVR led to a significant reduction in the frequency of NS-SCRs, and an increase in the power of the target respiration frequency band. NS-SCRs indicate increased sympathetic activity linked to spontaneous emotional arousal that can occur with external and internal stimuli (Boucsein, 2012). While using a VR headset in the iVR condition inherently reduced external distractions, most participants pointed out that iVR also helped reduce internal distractions. This indicates that the reduction of NS- SCR frequency in iVR is likely due to a decrease in both external and internal distractions. The correlation analyses further support this interpretation, showing that participants who reported higher attentional focus also showed better SDB performance and lower sympathetic arousal. iVR has been applied to mindfulness training, showing similar effects in enhancing attention and reducing mind- wandering, as reviewed in (Ma et al., 2023). Our findings align with these studies, suggesting that iVR enhances users’ ability to remain present and attentive during training. Furthermore, previous studies showed that increased presence, promoted by iVR, is linked with fewer instances of task-unrelated thoughts compared to non-iVR, leading to increased attention to instructions (Kuvar et al., 2024). This may contribute to explain why iVR led to a significant increase in the power of the target respiration frequency band, indicating a better SDB performance (participants took deeper breaths at the encouraged frequency, which is the primary goal of SDB training). Our findings indicate that iVR enhanced attention to SDB training resulting in improved adherence. It should be noted that the improvements in attention, the reduction in mind-wandering, and the lower NS-SCR frequency observed in iVR may reflect not only the enhanced sense of presence provided by the immersive environment, but also the sensory isolation afforded by the VR headset itself. Through visual occlusion of the physical environment, the VR headset reduces the availability of competing visual stimuli, potentially constraining attentional capture independently of presence. Moreover, since most participants had limited prior experience with iVR, the higher engagement scores observed in the iVR condition may partly reflect the novelty effect, i.e., the increased engagement elicited by first-time exposure to unfamiliar technologies (M.-H. Huang, 2003; Miguel-Alonso et al., 2024). 

6.2 Sense of Presence and Social Presence 

The hypothesis that participants in the iVR condition would experience higher levels of sense of presence and social presence compared to the non-iVR condition was also confirmed. Participants reported significantly higher sense of presence in iVR across all subscales of IPQ. This is consistent with previous research showing that iVR enhances the user’s sense of presence (Abadia et al., 2018; Schubert et al., 2001). Participants in iVR felt more present and involved in the VE and perceived it as more realistic compared to non-iVR. The increased sense of presence might have contributed to a more effective focus on SDB training. Indeed, participants pointed out that the gestures of the virtual agent were more noticeable and helpful in iVR compared to non-iVR. Additionally, nonverbal cues, such as the abdominal dilation and deflation of the virtual agent to guide breathing behavior, were only noticed by participants in the iVR condition, further aiding their ability to perform SDB correctly. They also experienced a stronger social presence with the virtual agent in iVR compared to non-iVR, and felt the virtual agent as more aware and responsive to their presence. 

6.3 Perception of the Virtual Agent 

Participants reported an improved perception of the anthropomorphism of the virtual agent in iVR compared to non-iVR. The increased immersion provided by iVR likely made the human-like qualities of the virtual agent more salient, making it appear more lifelike and engaging, as supported by previous research, e.g., (Nowak & Biocca, 2003). Previous research also indicated that greater anthropomorphism in virtual agents can increase user engagement and lead to more effective interactions (Nowak & Biocca, 2003). This enhanced perception of anthropomorphism of the virtual agent in iVR may have strengthened the connection between participants and the virtual agent, contributing to a more effective SDB training experience. 

6.4 SDB Effects 

The hypothesis that participants in the iVR condition would exhibit superior SDB effects was supported by both subjective measures and physiological measures. Participants reported significantly higher general relaxation and, more specifically, cardiovascular relaxation in the iVR condition compared to non-iVR. Physiologically, participants in the iVR condition showed a significant increase in DB-RSA. The increase in DB-RSA is consistent with enhanced cardiorespiratory coupling, i.e., the interaction between breathing and heart rate, whereby respiration modulates heart frequency primarily through vagal activity, producing synchronized oscillations in HRV, and resonance, i.e., the amplification of HRV oscillations when the breathing frequency approaches the resonance frequency of the baroreflex- cardiovascular system (Lehrer et al., 2000). The higher DB-RSA values in iVR compared to non-iVR suggest that participants were more effective in achieving the desired relaxation state. The improvement in DB-RSA may also be attributed to the more accurate execution of SDB. Figure 5 provides an example in which the intensity of the respiratory component in the iVR condition is more concentrated around the target breathing frequency (0.1 Hz), while there is more dispersion in non-iVR. This means that in iVR the SDB exercise was performed more accurately. Considering the same example, Figure 6 illustrates the synchronization between the respiratory waveform and HR oscillations, representing RSA. In the iVR condition, HR oscillations indicate greater coherence and amplitude. The example analyzed in detail in the two figures is representative of the broader trend observed in the study, which led to a statistically significant difference between the two conditions in these measures. The increase in the accuracy of SDB performance likely contributed to the enhanced cardiorespiratory coupling and resonance captured by DB-RSA. The increase in DB-RSA plausibly involved vagal mechanisms but does not constitute definitive evidence of increased parasympathetic activation, because breathing at approximately 0.1 Hz can itself produce resonance, which can raise DB-RSA independently of changes in vagal tone. In addition to the stronger cardiorespiratory coupling, participants in iVR experienced a reduction in sympathetic arousal, as highlighted by the significant decrease in both the frequency and amplitude of NS-SCRs. The correlation analyses corroborate these findings, confirming that participants with higher DB-RSA also reported greater subjective relaxation and lower sympathetic arousal, consistent with improved autonomic balance during SDB training. As mentioned before, the reductions in NS-SCRs reflect a lower level of sympathetic arousal spikes, suggesting that participants in the iVR condition were less distracted and more relaxed during the SDB training. Finally, as an exploratory analysis, PFC activity was examined during SDB training in iVR compared to non-iVR. More specifically, a reduction in both HbT and HbDiff was observed, indicating decreased blood flow and oxygenation in PFC. These reductions are consistent with a lower cognitive load in participants engaged in SDB training in iVR, although this interpretation should be treated with caution for several reasons. First, the observed effect sizes were small. Second, HbT and HbDiff data were excluded for 8 participants due to sensor displacement, particularly when wearing the VR headset. Third, alternative explanations cannot be ruled out: paced breathing at 0.1 Hz may produce hemodynamic fluctuations in the prefrontal cortex unrelated to cognitive effort, and general task familiarization may have reduced neural engagement independently of VR type. Consistent with previous studies using fNIRS (Leff et al., 2011; Sagari et al., 2015), decreases in prefrontal hemodynamic engagement may reflect reduced cognitive effort as the task becomes more practiced. Decreased PFC activity might also be linked to relaxation. Given the single-session design, these results should be considered preliminary and exploratory. An additional factor that may have contributed to the reduced PFC activation in iVR is a possible reduction in environmental distractors. Unlike a computer monitor, which leaves the surrounding room fully visible, a VR headset can potentially reduce peripheral stimuli that could capture attention and engage prefrontal areas. This aligns with participant feedback, as many reported experiencing more distractions when using the computer monitor. Thus, the lower PFC activity observed in iVR may reflect not only a more automated execution of the breathing task or a deeper relaxation state, but also a reduced top-down effort to suppress competing environmental stimuli. Overall, iVR appears to reduce internal distractors and enhance relaxation. 

7 Limitations and Future work 

This study has several limitations that should be acknowledged.
First, the study employed a single-session design. While the findings show that iVR has a positive 

impact on breathing performance, psychophysiological modulation, and subjective engagement, they do not provide evidence of skill acquisition, retention, long-term adherence, or durable autonomic changes. Longitudinal studies with longer intervention periods are needed to evaluate the long-term effects of iVR on SDB training, particularly regarding adherence, skill retention, and sustained physiological benefits, as well as its impact on relaxation, stress reduction, anxiety management, and overall mental health. 

Second, the sample size and its demographic composition may limit the scope of the findings. While the a priori power analysis indicated that the sample was adequate for detecting medium effects, smaller effects may have gone undetected. Furthermore, participants were recruited from students at our university, yielding a demographically young sample, which may restrict the generalizability of the findings, as different age groups may respond differently to iVR. Future research should include larger and more diverse samples, including clinical populations, to assess effects across different age groups, cultural backgrounds, and clinical conditions such as anxiety disorders. 

Third, the within-subjects design may have introduced learning and carryover effects. Although a 5-minute rest period was administered between sessions, this washout interval may have been insufficient to fully dissipate the physiological effects induced by paced breathing at 0.1 Hz. While counterbalancing has likely helped mitigate order effects, it does not preclude the possibility of carryover and learning effects. Cardiorespiratory resonance and residual parasympathetic tone can persist beyond the active breathing period, and participants’ familiarity with the breathing pace acquired in the first condition may have affected performance in the second, independently of VR type. Consequently, carryover and learning effects may have influenced the observed condition differences. Future studies should employ longer washout periods or consider monitoring the return of autonomic measures to baseline levels before initiating the second condition, or alternatively adopt a between- subjects design. 

Fourth, the novelty effect may have influenced some of the observed differences. Since most participants had limited prior experience with iVR, part of the heightened engagement in the iVR condition may reflect unfamiliarity with the technology rather than the immersive properties of iVR itself (M.-H. Huang, 2003; Miguel-Alonso et al., 2024). Novelty-related influences tend to diminish over time as users become familiar with the technology, and thus the magnitude of observed effects may change with repeated exposure. Given the single-session design, it was not possible to disentangle novelty effects from more stable benefits. Future studies should address this possibility, for example, by including familiarization sessions. 

Fifth, the absence of additional experimental conditions, such as a disembodied agent condition or a comparison with a real human instructor, limits the ability to disentangle the contribution of the embodied agent from that of the delivery medium. Future studies should include such conditions to allow for a more nuanced interpretation of the results. 

Sixth, the null findings concerning name-based personalization should be interpreted with caution. The study did not include a manipulation check to assess whether participants noticed the use of their name by the virtual agent, making it difficult to determine whether the absence of effects reflects a genuine lack of impact or a failure of the manipulation to function as intended. Additionally, the brief exposure duration may have been insufficient to activate the relational mechanisms typically associated with personalization, such as the development of rapport and trust with the virtual agent. Future studies should include explicit manipulation checks for personalization and consider longer exposure periods. 

Seventh, the use of an embodied agent of a single gender may have limited the ability to account for cross-gender interaction effects. Future studies should consider testing virtual agents of different genders to examine how the gender of the agent might affect social presence, engagement, and SDB training outcomes. 

Eighth, the use of multiple physiological sensors may have affected participants’ comfort, potentially interfering with SDB training performance. Investigating the feasibility of iVR-based SDB training in real-world settings, beyond the laboratory, is also important for assessing its practicality and impact on everyday stress management and well-being. 

Beyond these limitations, several directions for future work emerge from this study. The personalization of the target breathing frequency deserves attention. The 6 breaths-per-minute pace (0.1 Hz) used in the present study corresponds to the resonance frequency most commonly reported at the population level (Lehrer et al., 2000). However, individual resonance frequencies may vary, and breathing at a non-optimal pace may attenuate parasympathetic benefits. An initial calibration phase could be added to identify users’ personal resonance frequency, for example using the structured resonance frequency assessment protocol proposed in (Shaffer & Meehan, 2020). The virtual embodied agent in our system could help automate this process, by guiding each participant through the assessment steps. 

Additionally, real-time visualization of how slow breathing affects cardiac activity could represent a valuable addition to the system. Helping users understand the direct impact of their breathing on heart rate could strengthen the biofeedback loop and enhance users’ awareness and self-efficacy during SDB training. This might be achieved through a psychoeducation module in which users are shown the relationship between their breathing behavior and cardiac activity as a preparatory step for SDB training. Such a module could help users develop a more concrete understanding of the physiological mechanisms underlying SDB, potentially improving training outcomes by increasing motivation and self-efficacy from the outset. 

8 Conclusions 

This study has provided different types of evidence that iVR improves some of the outcomes of a single- session SDB training guided by an embodied agent. The iVR condition resulted in higher engagement, attention, sense of presence, social presence, and a greater perception of the anthropomorphism of the virtual agent. These subjective improvements were corroborated by physiological measures that indicated deeper, more accurate breathing at the target frequency, enhanced cardiorespiratory coupling and resonance, and reduced sympathetic arousal, hallmarks of enhanced relaxation. These findings highlight the potential of iVR as an effective tool for enhancing interventions aimed at stress, anxiety, and pain management. By providing an immersive environment, along with an embodied agent offering personalized guidance and feedback, iVR can help to bridge the gap between self-help resources, such as videos and apps, and in-person SDB training. Future work should explore the long-term effects of iVR-based SDB training, including sustained physiological benefits, relationship with the virtual agent, skill retention, and adherence over time, and determine whether the psychophysiological effects observed in a single session translate into clinically meaningful outcomes across repeated sessions and in clinical populations. 

References 

Abadia, R., Calvert, J., & Tauseef, S. M. (2018). Salient features of an effective immersive non-collaborative virtual reality learning environment. Proceedings of the 10th International Conference on Education Technology and Computers, ICETC ‘18, 268–278. https://doi.org/10.1145/3290511.3290558&nbsp;

Ait Baha, T., El Hajji, M., Es-Saady, Y., & Fadili, H. (2023). The Power of Personalization: A Systematic Review of Personality-Adaptive Chatbots. SN Computer Science, 4(5), 661. https://doi.org/10.1007/s42979-023-02092-6&nbsp;

Ali, M. K., Liu, L., Hussain, A., Zheng, D., Alam, M., Chen, J.-H., & Huizinga, J. D. (2023). Root mean square of successive differences is not a valid measure of parasympathetic reactivity during slow deep breathing. American Journal of Physiology – Regulatory Integrative and Comparative Physiology, 324(4), R446– R456. Scopus. https://doi.org/10.1152/ajpregu.00272.2022&nbsp;

Anderson, A. P., Mayer, M. D., Fellows, A. M., Cowan, D. R., Hegel, M. T., & Buckey, J. C. (2017). Relaxation with Immersive Natural Scenes Presented Using Virtual Reality. Aerospace Medicine and Human Performance, 88(6), 520–526. https://doi.org/10.3357/AMHP.4747.2017&nbsp;

Andreassi, J. L. (2006). Psychophysiology: Human Behavior and Physiological Response (5th ed.). Psychology Press. https://doi.org/10.4324/9780203880340&nbsp;

Arnsten, A. F. T. (2009). Stress signalling pathways that impair prefrontal cortex structure and function. Nature Reviews. Neuroscience, 10(6), 410–422. https://doi.org/10.1038/nrn2648&nbsp;

Bailenson, J. N., Blascovich, J., Beall, A. C., & Loomis, J. M. (2001). Equilibrium Theory Revisited: Mutual Gaze and Personal Space in Virtual Environments. Presence, 10(6), 583–598. Presence. https://doi.org/10.1162/105474601753272844&nbsp;

Bandura, A. (1977). Social learning theory (Vol. 1). Englewood cliffs Prentice Hall. http://www.asecib.ase.ro/mps/Bandura_SocialLearningTheory.pdf&nbsp;

Bandura, A. (1997). Self-efficacy: The exercise of control. Macmillan. 

Bartneck, C., Kulić, D., Croft, E., & Zoghbi, S. (2009). Measurement Instruments for the Anthropomorphism, Animacy, Likeability, Perceived Intelligence, and Perceived Safety of Robots. International Journal of Social Robotics, 1(1), 71–81. https://doi.org/10.1007/s12369-008-0001-3&nbsp;

Barton, A. C., Do, M., Sheen, J., & Byrne, L. K. (2024). The restorative and state enhancing potential of abstract fractal-like imagery and interactive mindfulness interventions in virtual reality. Virtual Reality, 28(1), 53. https://doi.org/10.1007/s10055-023-00916-7&nbsp;

Bender, R., & Lange, S. (2001). Adjusting for multiple testing—When and how? Journal of Clinical Epidemiology, 54(4), 343–349. https://doi.org/10.1016/s0895-4356(00)00314-0&nbsp;

Berntson, G. G., Thomas Bigger Jr., J., Eckberg, D. L., Grossman, P., Kaufmann, P. G., Malik, M., Nagaraja, H. N., Porges, S. W., Saul, J. P., Stone, P. H., & Van Der Molen, M. W. (1997). Heart rate variability: 

Origins, methods, and interpretive caveats. Psychophysiology, 34(6), 623–648. https://doi.org/10.1111/j.1469-8986.1997.tb02140.x&nbsp;

Bickmore, T., & Picard, R. W. (2005). Establishing and maintaining long-term human-computer relationships. ACM Transactions on Computer-Human Interaction, 12(2), 293–327. https://doi.org/10.1145/1067860.1067867&nbsp;

Bickmore, T. W., Puskar, K., Schlenk, E. A., Pfeifer, L. M., & Sereika, S. M. (2010). Maintaining reality: Relational agents for antipsychotic medication adherence. Interacting with Computers, 22(4), 276–288. https://doi.org/10.1016/j.intcom.2010.02.001&nbsp;

Biocca, F., Harms, C., & Burgoon, J. K. (2003). Toward a More Robust Theory and Measure of Social Presence: Review and Suggested Criteria. Presence: Teleoperators and Virtual Environments, 12(5), 456–480. https://doi.org/10.1162/105474603322761270&nbsp;

Blum, J., Rockstroh, C., & Göritz, A. S. (2019). Heart Rate Variability Biofeedback Based on Slow-Paced Breathing With Immersive Virtual Reality Nature Scenery. Frontiers in Psychology, 10, 2172. https://doi.org/10.3389/fpsyg.2019.02172&nbsp;

Boucsein, W. (2012). Electrodermal Activity. Springer US. https://doi.org/10.1007/978-1-4614-1126-0&nbsp;

Chalaye, P., Goffaux, P., Lafrenaye, S., & Marchand, S. (2009). Respiratory Effects on Experimental Heat Pain and Cardiac Activity. Pain Medicine, 10(8), 1334–1340. https://doi.org/10.1111/j.1526- 4637.2009.00681.x 

Chittaro, L., Serafini, M., & Vulcano, Y. (2024). Virtual reality experiences for breathing and relaxation training: The effects of real vs. placebo biofeedback. International Journal of Human-Computer Studies, 188, 103275. https://doi.org/10.1016/j.ijhcs.2024.103275&nbsp;

Chittaro, L., & Sioni, R. (2014). Evaluating mobile apps for breathing training: The effectiveness of visualization. Computers in Human Behavior, 40, 56–63. Scopus. https://doi.org/10.1016/j.chb.2014.07.049&nbsp;

Cohen, B. H. (2013). Explaining psychological statistics (4th ed). Wiley.
Critchley, H. D. (2002). Electrodermal Responses: What Happens in the Brain. The Neuroscientist, 8(2), 132– 

142. https://doi.org/10.1177/107385840200800209&nbsp;

de Zambotti, M., Yuksel, D., Kiss, O., Barresi, G., Arra, N., Volpe, L., King, C., & Baker, F. C. (2022). A virtual reality-based mind–body approach to downregulate psychophysiological arousal in adolescent insomnia. DIGITAL HEALTH, 8, 20552076221107887. https://doi.org/10.1177/20552076221107887&nbsp;

Eichstaedt, K. E., Kovatch, K., & Maroof, D. A. (2013). A less conservative method to adjust for familywise error rate in neuropsychological research: The Holm’s sequential Bonferroni procedure. NeuroRehabilitation, 32(3), 693–696. https://doi.org/10.3233/NRE-130893&nbsp;

Electrophysiology, T. F. of the E. S. of C. the N. A. S. of P. (1996). Heart Rate Variability. Circulation, 93(5), 1043–1065. https://doi.org/10.1161/01.CIR.93.5.1043&nbsp;

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175– 191. https://doi.org/10.3758/BF03193146&nbsp;

Fincham, G. W., Strauss, C., Montero-Marin, J., & Cavanagh, K. (2023). Effect of breathwork on stress and mental health: A meta-analysis of randomised-controlled trials. Scientific Reports, 13(1), 432. https://doi.org/10.1038/s41598-022-27247-y&nbsp;

Fishburn, F. A., Norr, M. E., Medvedev, A. V., & Vaidya, C. J. (2014). Sensitivity of fNIRS to cognitive state and load. Frontiers in Human Neuroscience, 8(1 FEB). Scopus. https://doi.org/10.3389/fnhum.2014.00076&nbsp;

Gao, T., Zhang, T., Zhu, L., Gao, Y., & Qiu, L. (2019). Exploring Psychophysiological Restoration and Individual Preference in the Different Environments Based on Virtual Reality. International Journal of Environmental Research and Public Health, 16(17), Article 17. https://doi.org/10.3390/ijerph16173102&nbsp;

Gaume, A., Vialatte, A., Mora-Sánchez, A., Ramdani, C., & Vialatte, F. B. (2016). A psychoengineering paradigm for the neurocognitive mechanisms of biofeedback and neurofeedback. Neuroscience & Biobehavioral Reviews, 68, 891–910. https://doi.org/10.1016/j.neubiorev.2016.06.012&nbsp;

Gentili, R. J., Shewokis, P. A., Ayaz, H., & Contreras-Vidal, J. L. (2013). Functional near-infrared spectroscopy- based correlates of prefrontal cortical dynamics during a cognitive-motor executive adaptation task. Frontiers in Human Neuroscience, 7. https://doi.org/10.3389/fnhum.2013.00277&nbsp;

Gholamrezaei, A., Van Diest, I., Aziz, Q., Pauwels, A., Tack, J., Vlaeyen, J. W. S., & Van Oudenhove, L. (2022). Effect of slow, deep breathing on visceral pain perception and its underlying psychophysiological mechanisms. Neurogastroenterology and Motility, 34(4), e14242. https://doi.org/10.1111/nmo.14242&nbsp;

Guillen-Sanz, H., Checa, D., Miguel-Alonso, I., & Bustillo, A. (2024). A systematic review of wearable biosensor usage in immersive virtual reality experiences. Virtual Reality, 28(2), 74. https://doi.org/10.1007/s10055-024-00970-9&nbsp;

Holm, S. (1979). A Simple Sequentially Rejective Multiple Test Procedure. Scandinavian Journal of Statistics, 6(2), 65–70. 

Hu, X.-S., Beard, K., Sherbel, M. C., Nascimento, T. D., Petty, S., Pantzlaff, E., Schwitzer, D., Kaciroti, N., Maslowski, E., Ashman, L. M., Feinberg, S. E., & DaSilva, A. F. (2021). Brain Mechanisms of Virtual Reality Breathing Versus Traditional Mindful Breathing in Pain Modulation: Observational Functional Near-infrared Spectroscopy Study. Journal of Medical Internet Research, 23(10), e27298. https://doi.org/10.2196/27298&nbsp;

Huang, M.-H. (2003). Designing website attributes to induce experiential encounters. Computers in Human Behavior, 19(4), 425–442. https://doi.org/10.1016/S0747-5632(02)00080-8&nbsp;

Huang, W., Roscoe, R. D., Johnson-Glenberg, M. C., & Craig, S. D. (2021). Motivation, engagement, and performance across multiple virtual reality sessions and levels of immersion. Journal of Computer Assisted Learning, 37(3), 745–758. https://doi.org/10.1111/jcal.12520&nbsp;

Jafari, H., Courtois, I., Van den Bergh, O., Vlaeyen, J. W. S., & Van Diest, I. (2017). Pain and respiration: A systematic review. Pain, 158(6), 995–1006. https://doi.org/10.1097/j.pain.0000000000000865&nbsp;

Jafari, H., Gholamrezaei, A., Franssen, M., Van Oudenhove, L., Aziz, Q., Van den Bergh, O., Vlaeyen, J. W. S., & Van Diest, I. (2020). Can Slow Deep Breathing Reduce Pain? An Experimental Study Exploring Mechanisms. The Journal of Pain, 21(9–10), 1018–1030. https://doi.org/10.1016/j.jpain.2019.12.010&nbsp;

Joseph, A. E., Moman, R. N., Barman, R. A., Kleppel, D. J., Eberhart, N. D., Gerberi, D. J., Murad, M. H., & Hooten, W. M. (2022). Effects of Slow Deep Breathing on Acute Clinical Pain in Adults: A Systematic Review and Meta-Analysis of Randomized Controlled Trials. Journal of Evidence-Based Integrative Medicine, 27, 2515690X221078006. https://doi.org/10.1177/2515690X221078006&nbsp;

Kuvar, V. K., Bailenson, J. N., & Mills, C. (2024). A novel quantitative assessment of engagement in virtual reality: Task-unrelated thought is reduced compared to 2D videos. Computers & Education, 209, 104959. https://doi.org/10.1016/j.compedu.2023.104959&nbsp;

Leff, D. R., Orihuela-Espina, F., Elwell, C. E., Athanasiou, T., Delpy, D. T., Darzi, A. W., & Yang, G.-Z. (2011). Assessment of the cerebral cortex during motor task behaviours in adults: A systematic review of functional near infrared spectroscopy (fNIRS) studies. NeuroImage, 54(4), 2922–2936. https://doi.org/10.1016/j.neuroimage.2010.10.058&nbsp;

Lehrer, P. M., Vaschillo, E., & Vaschillo, B. (2000). Resonant Frequency Biofeedback Training to Increase Cardiac Variability: Rationale and Manual for Training. Applied Psychophysiology and Biofeedback, 25(3), 177–191. https://doi.org/10.1023/A:1009554825745&nbsp;

Lehrer, P. M., Vaschillo, E., Vaschillo, B., Lu, S.-E., Eckberg, D. L., Edelberg, R., Shih, W. J., Lin, Y., Kuusela, T. A., Tahvanainen, K. U. O., & Hamer, R. M. (2003). Heart Rate Variability Biofeedback Increases Baroreflex Gain and Peak Expiratory Flow. Psychosomatic Medicine, 65(5), 796. https://doi.org/10.1097/01.PSY.0000089200.81962.19&nbsp;

Lüddecke, R., & Felnhofer, A. (2022). Virtual Reality Biofeedback in Health: A Scoping Review. Applied Psychophysiology and Biofeedback, 47(1), 1–15. https://doi.org/10.1007/s10484-021-09529-9&nbsp;

Ma, J., Zhao, D., Xu, N., & Yang, J. (2023). The effectiveness of immersive virtual reality (VR) based mindfulness training on improvement mental-health in adults: A narrative systematic review. EXPLORE, 19(3), 310–318. https://doi.org/10.1016/j.explore.2022.08.001&nbsp;

Makowski, D., Pham, T., Lau, Z. J., Brammer, J. C., Lespinasse, F., Pham, H., Schölzel, C., & Chen, S. H. A. (2021). NeuroKit2: A Python toolbox for neurophysiological signal processing. Behavior Research Methods, 53(4), 1689–1696. https://doi.org/10.3758/s13428-020-01516-y&nbsp;

Miguel-Alonso, I., Checa, D., Guillen-Sanz, H., & Bustillo, A. (2024). Evaluation of the novelty effect in immersive Virtual Reality learning experiences. Virtual Reality, 28(1), 27. https://doi.org/10.1007/s10055-023-00926-5&nbsp;

Nahin, R. L., Stussman, B. J., & Herman, P. M. (2015). Out-Of-Pocket Expenditures on Complementary Health Approaches Associated With Painful Health Conditions in a Nationally Representative Adult Sample. The Journal of Pain, 16(11), 1147–1162. https://doi.org/10.1016/j.jpain.2015.07.013&nbsp;

Nowak, K. L., & Biocca, F. (2003). The Effect of the Agency and Anthropomorphism on Users’ Sense of Telepresence, Copresence, and Social Presence in Virtual Environments. Presence: Teleoperators and Virtual Environments, 12(5), 481–494. https://doi.org/10.1162/105474603322761289&nbsp;

O’Brien, H. L., Cairns, P., & Hall, M. (2018). A practical approach to measuring user engagement with the refined user engagement scale (UES) and new UES short form. International Journal of Human- Computer Studies, 112, 28–39. https://doi.org/10.1016/j.ijhcs.2018.01.004&nbsp;

Oh, C. S., Bailenson, J. N., & Welch, G. F. (2018). A Systematic Review of Social Presence: Definition, Antecedents, and Implications. Frontiers in Robotics and AI, 5, 114. https://doi.org/10.3389/frobt.2018.00114&nbsp;

Pancini, E., Di Natale, A. F., & Villani, D. (2025). Breathing in virtual reality for promoting mental health: A scoping review. Virtual Reality, 29(1), 29. https://doi.org/10.1007/s10055-024-01096-8&nbsp;

Perciavalle, V., Blandini, M., Fecarotta, P., Buscemi, A., Di Corrado, D., Bertolo, L., Fichera, F., & Coco, M. (2017). The role of deep breathing on stress. Neurological Sciences, 38(3), 451–458. Scopus. https://doi.org/10.1007/s10072-016-2790-8&nbsp;

Pham, T., Johnco, C. J., Lau, Z. J., Makowski, D., & Forbes, M. K. (2025). Which Heart Rate Variability ( HRV ) Indices Should I Use for Psychophysiological Research? A Data‐Driven Approach to Identifying Clusters of HRV Indices. Psychophysiology, 62(10), e70164. https://doi.org/10.1111/psyp.70164&nbsp;

Pinna, G. D., Maestri, R., Cesare, A. D., Colombo, R., & Minuco, G. (1994). The accuracy of power-spectrum analysis of heart-rate variability from annotated RR lists generated by Holter systems. Physiological Measurement, 15(2), 163. https://doi.org/10.1088/0967-3334/15/2/006&nbsp;

Porges, S. W. (2007). The Polyvagal Perspective. Biological Psychology, 74(2), 116–143. https://doi.org/10.1016/j.biopsycho.2006.06.009&nbsp;

Provoost, S., Kleiboer, A., Ornelas, J., Bosse, T., Ruwaard, J., Rocha, A., Cuijpers, P., & Riper, H. (2020). Improving adherence to an online intervention for low mood with a virtual coach: Study protocol of a pilot randomized controlled trial. Trials, 21(1), 860. https://doi.org/10.1186/s13063-020-04777-2&nbsp;

Provoost, S., Lau, H. M., Ruwaard, J., & Riper, H. (2017). Embodied Conversational Agents in Clinical Psychology: A Scoping Review. Journal of Medical Internet Research, 19(5), e6553. https://doi.org/10.2196/jmir.6553&nbsp;

Riches, S., Azevedo, L., Bird, L., Pisani, S., & Valmaggia, L. (2021). Virtual reality relaxation for the general population: A systematic review. Social Psychiatry and Psychiatric Epidemiology, 56(10), 1707–1727. https://doi.org/10.1007/s00127-021-02110-z&nbsp;

Rockstroh, C., Blum, J., & Göritz, A. S. (2019). Virtual reality in the application of heart rate variability biofeedback. International Journal of Human-Computer Studies, 130, 209–220. https://doi.org/10.1016/j.ijhcs.2019.06.011&nbsp;

Rotter, J. B. (1966). Generalized expectancies for internal versus external control of reinforcement. Psychological Monographs: General and Applied, 80(1), 1–28. https://doi.org/10.1037/h0092976&nbsp;

Russo, M. A., Santarelli, D. M., & O’Rourke, D. (2017). The physiological effects of slow breathing in the healthy human. Breathe, 13(4), 298–309. https://doi.org/10.1183/20734735.009817&nbsp;

Sagari, A., Iso, N., Moriuchi, T., Ogahara, K., Kitajima, E., Tanaka, K., Tabira, T., & Higashi, T. (2015). Changes in Cerebral Hemodynamics during Complex Motor Learning by Character Entry into Touch- Screen Terminals. PLOS ONE, 10(10), e0140552. https://doi.org/10.1371/journal.pone.0140552&nbsp;

Savaş, E. H., Semerci, R., & Bayram, C. (2024). The effect of a biofeedback-based virtual reality game on pain, fear and anxiety levels during port catheter needle insertion in pediatric oncology patients: A randomized controlled study. European Journal of Oncology Nursing, 70, 102621. https://doi.org/10.1016/j.ejon.2024.102621&nbsp;

Schubert, T., Friedmann, F., & Regenbrecht, H. (2001). The Experience of Presence: Factor Analytic Insights. Presence: Teleoperators and Virtual Environments, 10(3), 266–281. https://doi.org/10.1162/105474601300343603&nbsp;

Sevoz-Couche, C., & Laborde, S. (2022). Heart rate variability and slow-paced breathing:when coherence meets resonance. Neuroscience & Biobehavioral Reviews, 135, 104576. https://doi.org/10.1016/j.neubiorev.2022.104576&nbsp;

Sezer, I., & Sacchet, M. D. (2025). Advanced and long-term meditation and the autonomic nervous system: A review and synthesis. Neuroscience & Biobehavioral Reviews, 173, 106141. https://doi.org/10.1016/j.neubiorev.2025.106141&nbsp;

Shaffer, F., & Ginsberg, J. P. (2017). An Overview of Heart Rate Variability Metrics and Norms. Frontiers in Public Health, 5. https://doi.org/10.3389/fpubh.2017.00258&nbsp;

Shaffer, F., & Meehan, Z. M. (2020). A Practical Guide to Resonance Frequency Assessment for Heart Rate Variability Biofeedback. Frontiers in Neuroscience, 14. https://doi.org/10.3389/fnins.2020.570400&nbsp;

Shamekhi, A., & Bickmore, T. (2018). Breathe Deep: A Breath-Sensitive Interactive Meditation Coach. 

Proceedings of the 12th EAI International Conference on Pervasive Computing Technologies for Healthcare, 108–117. https://doi.org/10.1145/3240925.3240940&nbsp;

Shao, R., Man, I. S. C., & Lee, T. M. C. (2024). The Effect of Slow-Paced Breathing on Cardiovascular and Emotion Functions: A Meta-Analysis and Systematic Review. Mindfulness, 15(1), 1–18. https://doi.org/10.1007/s12671-023-02294-2&nbsp;

Stark, R., Schienle, A., Walter, B., & Vaitl, D. (2000). Effects of paced respiration on heart period and heart period variability. Psychophysiology, 37(3), 302–309. https://doi.org/10.1111/1469-8986.3730302&nbsp;

Steghaus, S., & Poth, C. H. (2022). Assessing momentary relaxation using the Relaxation State Questionnaire (RSQ). Scientific Reports, 12(1), Article 1. https://doi.org/10.1038/s41598-022-20524-w&nbsp;

Stoet, G. (2010). PsyToolkit: A software package for programming psychological experiments using Linux. Behavior Research Methods, 42(4), 1096–1104. https://doi.org/10.3758/BRM.42.4.1096&nbsp;

Stoet, G. (2017). PsyToolkit: A Novel Web-Based Method for Running Online Questionnaires and Reaction- Time Experiments. Teaching of Psychology, 44(1), 24–31. https://doi.org/10.1177/0098628316677643&nbsp;

Tachtsidis, I., Tisdall, M. M., Leung, T. S., Pritchard, C., Cooper, C. E., Smith, M., & Elwell, C. E. (2009). Relationship Between Brain Tissue Haemodynamics, Oxygenation And Metabolism In The Healthy Human Adult Brain During Hyperoxia And Hypercapnea. In P. Liss, P. Hansell, D. F. Bruley, & D. K. Harrison (Eds.), Oxygen Transport to Tissue XXX (pp. 315–320). Springer US. https://doi.org/10.1007/978-0-387-85998-9_47&nbsp;

Ter Stal, S., Kramer, L. L., Tabak, M., op den Akker, H., & Hermens, H. (2020). Design Features of Embodied Conversational Agents in eHealth: A Literature Review. International Journal of Human-Computer Studies, 138, 102409. https://doi.org/10.1016/j.ijhcs.2020.102409&nbsp;

Tinga, A. M., Clim, M.-A., de Back, T. T., & Louwerse, M. M. (2021). Measures of prefrontal functional near- infrared spectroscopy in visuomotor learning. Experimental Brain Research, 239(4), 1061–1072. https://doi.org/10.1007/s00221-021-06039-2&nbsp;

Vago, D. R., & Zeidan, F. (2016). The brain on silent: Mind wandering, mindful awareness, and states of mental tranquility. Annals of the New York Academy of Sciences, 1373(1), 96–113. https://doi.org/10.1111/nyas.13171&nbsp;

Valtchanov, D., Barton, K. R., & Ellard, C. (2010). Restorative Effects of Virtual Nature Settings. Cyberpsychology, Behavior, and Social Networking, 13(5), 503–512. https://doi.org/10.1089/cyber.2009.0308&nbsp;

Van Diest, I., Verstappen, K., Aubert, A. E., Widjaja, D., Vansteenwegen, D., & Vlemincx, E. (2014). Inhalation/Exhalation Ratio Modulates the Effect of Slow Breathing on Heart Rate Variability and Relaxation. Applied Psychophysiology and Biofeedback, 39(3), 171–180. https://doi.org/10.1007/s10484- 014-9253-x 

Villringer, A., & Chance, B. (1997). Non-invasive optical spectroscopy and imaging of human brain function. Trends in Neurosciences, 20(10), 435–442. https://doi.org/10.1016/s0166-2236(97)01132-6&nbsp;

Weerdmeester, J., van Rooij, M. M. J. W., & Granic, I. (2022). Visualization, Self-Efficacy, and Locus of Control in a Virtual Reality Biofeedback Video Game for Anxiety Regulation. Cyberpsychology, Behavior, and Social Networking, 25(6), 360–368. https://doi.org/10.1089/cyber.2022.0030&nbsp;

Weibel, R. P., Kerr, J. I., Naegelin, M., Ferrario, A., Schinazi, V. R., La Marca, R., Hoelscher, C., Nater, U. M., & von Wangenheim, F. (2023). Virtual reality-supported biofeedback for stress management: Beneficial effects on heart rate variability and user experience. Computers in Human Behavior, 141, 107607. https://doi.org/10.1016/j.chb.2022.107607&nbsp;

Winkler, A., Kutschar, P., Pitzer, S., van der Zee-Neuen, A., Kerner, S., Osterbrink, J., & Krutter, S. (2025). Avatar and virtual agent-assisted telecare for patients in their homes: A scoping review. Journal of Telemedicine and Telecare, 31(2), 207–221. https://doi.org/10.1177/1357633X231174484&nbsp;

Zaccaro, A., Piarulli, A., Laurino, M., Garbella, E., Menicucci, D., Neri, B., & Gemignani, A. (2018). How Breath-Control Can Change Your Life: A Systematic Review on Psycho-Physiological Correlates of Slow Breathing. Frontiers in Human Neuroscience, 12. https://doi.org/10.3389/fnhum.2018.00353&nbsp;

Zhu, B., Hedman, A., & Li, H. (2017). Designing Digital Mindfulness: Presence-In and Presence-With versus Presence-Through. Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems, CHI ‘17, 2685–2695. https://doi.org/10.1145/3025453.3025590&nbsp;

Keywords: Immersive virtual reality, breathing training,