Immersive Virtual Reality Avatars for Embodiment Illusions in People with Mild to Borderline Intellectual Disability: User-Centered Development and Feasibility Study

Background: Immersive Virtual Reality (IVR) has been investigated as tool for treating psychiatric conditions. Especially the practical nature of IVR, by offering a doing instead of talking approach, could support people who do not benefit from existing treatments. Hence, people with mild to borderline intellectual disability (MBID, IQ = 50-85) might profit particularly from IVR therapies, for instance, to circumvent issues in understanding relevant concepts and interrelations. In this context, immersing the user into a virtual body (i.e. avatar) appears promising for enhancing learning (e.g. by changing perspectives) and usability (e.g. natural interactions). However, design requirements, immersion procedures, and the proof of concept of such embodiment illusion (i.e. substituting the real body with a virtual one) have not been explored in this group. Objective: Our work aimed to establish design guidelines for IVR embodiment illusions in people with MBID. We explored three factors to induce the illusion, by testing the (1) avatar’s appearance, (2) locomotion using IVR controllers, and (3) virtual object manipulation. Further, we report on the feasibility to induce the embodiment illusion and provide procedural guidance. Methods: We conducted a user-centered design with 29 end-users in care facilities, to investigate the (1) avatar’s appearance, (2) controller-based locomotion (i.e. teleport, joystick, or hybrid), and (


Introduction Background
Immersive Virtual Reality (IVR) has been investigated as a treatment tool for a variety of psychiatric disorders, for instance in people with psychosis, addictive disorders, and eating disorders [1,2]. Yet, till now, the clinical effectiveness has only been proven in anxiety disorders, as (randomized) controlled trials in other mental illnesses are still required. However, the practical nature as doing instead of talking approach makes IVR therapy appealing for people that do not profit from existing treatments, such as people with mild to borderline intellectual disability (MBID, IQ = 50-85). People with MBID constitute a diverse group with lower intellectual and adaptive capabilities (e.g. problems in planning, problem-solving, abstract thinking, and judgment), which negatively impacts the development of essential skills for independent living. By using the term MBID, we combine the groups mild intellectual disability (IQ = 50-69) and borderline intellectual functioning (IQ = 70-85), as they often encounter similar challenges in life, for instance regarding mental health treatments [3][4][5]. Previous work suggests that IVR could help to reduce learning barriers by making, for example, abstract concepts and interrelations graspable [6][7][8]. However, the application of IVR and knowledge on requirements in MBID remain low [9], as few scholars explored IVR interactions in this group [10][11][12]. However, turning with the own body and interacting via hand-based manipulations seemed to benefit usability. Hence, natural interactions seem decisive for people with MBID, which might be amplified by an immersion into a virtual body. This embodiment illusion could facilitate lifelike interactions and therefore improve the access to IVR for our group.
Embodiment illusions in IVR allow us to substitute the real body with a virtual body or certain body parts, such as arms or hands [13][14][15]. The phenomenon is often assessed by the Sense of Embodiment (SoE) toward the virtual body (i.e. avatar) [15,16]. The 'SoE toward a body B is the sense that emerges when B's properties are processed as if they were the properties of one's own biological body.' [15, p. 375]. Hence, the embodiment illusion is induced through three main factors: a sense of self-location, agency, and ownership [15,17]. The sense of self-location refers to the feeling of being inside the body [15], the sense of agency comprises the 'global motor control, including the subjective experience of action, control, intention, motor selection and the conscious experience of will' [18, p. 7], and the sense of body ownership involves the self-attribution to the avatar [15]. However, till now, the significance of each factor for the illusion remains ambiguous [15]. Yet, illusions of virtual body ownership (IVBO) were investigated to influence attitudes and behavior [19], which makes them promising for enhancing therapy outcomes in groups that hardly benefit from cognitively demanding paradigms, such as people with MBID.
Prior findings in people devoid MBID showed that embodying a black avatar can reduce racial bias and that embodying a child can influence implicit attitudes and object size perception in IVR [20,21]. Both are referring to the proteus effect, derived from the Greek myth of a 'shape shifter', describing the phenomenon that we (humans) tend to change our beliefs and behavior based on our (digital) self-representations [22]. Besides such implicit approaches, explicit learning could be applied, for instance as psychomotor therapy with a focus on the bodily signals (e.g. cravings, tensions) [7]. Yet, despite various studies published that report on design requirements for such IVBO, no research focused on people with MBID. Hence, this work aims to design IVR avatars for embodiment illusions in people with MBID. Since the spatial immersion into the avatar and implementation of plausible actions (e.g. control) evokes realistic behaviors in IVR [22], we decided to look into three important components for embodiment illusions: (1) avatar appearance, (2) controller-based locomotion, and (3) object manipulation.

Related work
In the following sections, we look into the related work concerning (1) avatar immersion, (2) controller-based locomotion, and (3) object manipulation. Given the lack of research in our target group, we report on the existing evidence to identify crucial factors for our initial prototype and immersion design. We conclude the introduction with a summary of potential benefits and barriers of embodiment illusions for people with MBID and specify our research questions. Then, we describe our user-centered design method with three iterations and report the results per iteration. Subsequently, we discuss relevant factors in the context of prior work, limitations, and directions for future research. Finally, we conclude our work with a summary of our contribution to the field.

Immersion into IVR avatars for IVBO
Several factors contributing to IVR avatar immersion have been investigated to influence the SoE, such as the point of view (POV, i.e. the perspective), body appearance, control, and haptics (i.e. the experience of touch) [15]. For instance, an egocentric POV has been shown to reliably induce the sense of self-location [23], while a third-person perspective tends to lower it [24,25]. The sense of agency is induced through the experienced control toward the virtual body [15], influenced by the visuomotor congruence between real body and avatar [26][27][28], while incongruences tend to lower it [29,30]. Lastly, the sense of ownership is influenced by the body appearance and has been induced through avatar models with different degrees of anthropomorphism [31]. Still, despite possibilities to feel ownership towards avatars that differ from oneself in terms of gender and morphological characteristics [32,33], matching gender, skin-tone, and clothes can boost the IVBO [34,35]. However, the SoE factors cannot be regarded isolated from each other, as interrelations were identified [36], such as appearance influences on agency [37], and control/haptics on ownership [38][39][40]. Here, recent findings showed that primarily visuo-proprioceptive congruence contributes to the sense of agency and ownership, as well as a better task performance [28]. Moreover, Fribourg et al. (2020) explored the user preferences for three vital factors (i.e. POV, control, appearance), showing the need for an egocentric perspective and high motor control to outweigh the avatar's appearance [36]. Yet, these findings seemed task-dependent, with POV being relevant for locomotion, and avatar appearance when manipulating (virtual) objects using the upper body [36,37,41].

Controller-based locomotion for IVR avatars
Controller-based locomotion as essential component for the immersion into IVR avatars can be divided into physical and artificial approaches [42]. Physical techniques can be more intuitive (i.e. room-scale), though an intensive bodily involvement and unnaturalness (e.g. walking in place) might cause the opposite effect. Contrarily, artificial techniques (e.g. teleport, joystick) tend to increase the cognitive workload and are prone to cause cybersickness (i.e. motion sickness) [42,43]. As space for natural locomotion if often limited, adding artificial techniques of continuous (e.g. controller-based) or non-continuous (e.g. teleport-based) nature could form a viable solution [11]. Continuous approaches are preferred in open settings, while non-continuous approaches are widely used due to their user friendliness [42]. However, few studies explored virtual locomotion in combination with IVR avatars, showing influences on task-performance and obstacle avoidance [44][45][46]. The virtual body can improve the walking behavior in IVR, with less collision and more precise paths using a realistic avatar [47], as well as more natural behavior [48]. Here, using walking animations that mimicked natural behavior were preferred over the user's real motions (i.e. walking-in-place), tough this could lead to unintended steps [49]. Nevertheless, few studies examined the effects of virtual locomotion on the SoE factors. Dewez et al. (2020) compared natural walking, walking-in-place, and virtual steering and found a similar SoE, with equal performance with or devoid of an avatar [46]. In line with previous findings, movement incongruences between the user and avatar animation did not break the embodiment illusion [24].

Object interaction for IVR avatars
Likewise avatar immersion and locomotion design, interacting with objects and user interfaces (UIs) constitutes an essential component for immersive self-representations. Here, using avatars influences the interaction with objects and vice versa [41]. The alignment of this reciprocity to design for both, a high SoE and usability remains understudied in the current literature, especially when combined with artificial locomotion. Previous work that utilized IVR avatars during interaction tasks reported performance enhancements over controllers or virtual hands [50], independent of the model's human-likeness when comparing a realistic avatar with a generic or robot appearance [51,52]. Still, as spatial biases were found in IVR [53], the body might operate as reference frame [54], as the object size perception can be altered when using avatars [55]. Here, avatars can produce occlusions during interactions, which can affect the usability negatively, especially when using more anthropomorphic models [37,56]. Yet, using congruent body feedback could circumvent this issue, considering that haptics (e.g. self-touch) benefits the SoE and manipulation performance [28,57,58]. Lastly, an identified usability barrier are objects that are out-of-reach or placed low. A solution devoid altering the avatar or breaking the embodiment illusion could be artificial interactions (e.g. raycasting) implemented into the avatar's hands [41,59], allowing to interact with objects without substantial bodily movements.

Goal of this work
In sum, embodying an avatar in IVR might improve the usability [47], spatial awareness [6,41], and (self) presence [27,60], though adverse effects can occur due to an increase in complexity [56]. Yet, till now, design requirements for such IVR avatars and the feasibility to induce the IVBO have not been explored in people with MBID. So far, IVBO were mostly studied in controlled laboratory settings that use motion-capture systems or several body trackers to congruently map the user's bodily movements onto the virtual body. Though these systems seem to provide the highest control, they lack consumer friendliness to enter care institutions, given the high costs and difficulty in using the equipment. Here, solutions based on inverse kinematics (IK) that use the three-point tracking of the head-mounted display (HMD) and related controllers could provide an alternative, as most interactions focus on the upper body [36]. Hence, we aim to explore guidelines for such IVBO in people with MBID, by conducting a user-centered development based on three factors that contribute to functional and plausible actions [22]: How to design IVR avatars for illusions of virtual body ownership (IVBO) in people with MBID? a. How to design a virtual embodiment illusion for people with MBID based on inverse kinematics (IK)? b. How to design a controller-based locomotion technique for people with MBID? c. How to design a controller-based (object) manipulation for people with MBID? II.
To what extent do participants experience a SoE during the examined interaction task? III.
To what extent do participants experience a sense of presence (SoP) in the immersive virtual environment (IVE)?

Research design
We conducted a user-centered design approach to explore the three factors for IVR avatar immersion (i.e. avatar, controller-based locomotion, manipulation), initial feasibility of procedures, and proof of concept of an IVBO in people with MBID. For this, we developed an IVR avatar prototype to identify design recommendations for IVBO using three consecutive iterations with end-users in Dutch care facilities. Throughout these iterations, we refined the IVR avatar system and immersion procedure according to the subjects' needs. Hence, in this work, we establish design recommendations for the three components, explored the SoE levels, and SoP in the IVE to support others in creating accessible IVR avatars.

Participants
In total, 29 adults with MBID were recruited through convenience sampling by local therapists from an addiction clinic for individuals with MBID and Dutch care facility for people with MBID. Exclusion criteria included having a history of migraine, epilepsy, visual or motor impairment, severe mental disorder (e.g., schizophrenia, psychosis, active substance use disorder), susceptibility to COVID-19, proneness to motion sickness, or inability to wear the HMD.

Interaction system
We built the interaction system with consumer hardware and available software. The game engine Unity3D (v.2019.4 LTS) was used with the Mecanim IK and the XR interaction toolkit (preview, v.0.94) packages to develop an IVBO based on three-point tracking of the HMD and the related two controllers. We implemented the three identified components for IVR avatars: (1) Customizable avatar, (2) controller-based locomotion, and (3) (object) manipulation.
The (1) customizable IVR avatar component ( Figure 1) included an egocentric embodiment illusion. The participants were able to enter their height and arm dimensions (i.e. by going into the T-pose), customize the gender (f/m) and select a skin tone. The (2) controller-based locomotion component ( Figure 2) involved a visuomotor experience to move in the IVE, divided into physical and artificial approaches. The physical approach comprised a basic room-scale locomotion (2x2m) with walking animation when moving. Three artificial locomotion approaches were implemented, divided into joystick locomotion with 45° snap-turn and walking animation, teleportation locomotion using raycasting with a projectile curve, and a combination of both (hybrid). Haptics were used at beginning and execution of teleportation by using the controller's vibration motors. Furthermore, a 'teleport travel' technique was implemented that enables the transition to the different interaction contexts by using a screen-space UI [61], which followed the user's rotation on the Y-axis when holding the A-button. The (3) (object) manipulation component ( Figure 3) included a synchronous visuomotor experience to grab, pick-up from lower areas, place, and throw virtual objects. We used hand animations for grabbing (grip-button) and pinching (trigger-button), including haptics when grabbing and releasing the objects by using the controller's vibration motors. Raycasting on both hands was implemented to pick-up lower placed or out-of-reach objects and interact with UIs.

Hardware and Immersive Virtual Environment
We used an Oculus Quest HMD with 6 degrees of freedom, 1440 x 1600 pixels per eye, a 72 Hz refresh rate, and 90-degree field of view (FOV), touch controllers, as well as a compatible IVR-laptop (Intel Core i7 9750H -CPU, 16 GB RAM, NVIDIA GeForce RTX 2060) with Oculus Link (beta, USB 3.1 cable).
The IVE encompassed an open-world mechanic (200x200m) to evaluate the system's components. In the first room setting (15x15x2.5m) participants customized the IVR avatar. In the second room, artificial locomotion techniques were evaluated by completing a maze (50x50x2.3m) with four destinations and obstacles. Based on common game design, we utilized a vantage point to support spatial understanding and reduce unease. Further, destinations were cued using light beams of different color with matching leading lines on the walls [62]. In the third interaction context (15x15x2.5m), we evaluated the manipulation of three different objects (i.e. big cylinder, cube, and small cylinder). Participants used object-spaced UIs with low hierarchy to customize the avatar, select locomotion approaches, and control interaction tasks. A plain design was used to lower bias through environmental influences. We implemented landmarks (pink-colored) to aid user's orientation in the IVE.

Measures
A semi-structured interview (Appendix 1) was conducted after each of the five interaction tasks: IVR avatar customization, teleport, joystick, and hybrid locomotion, and object manipulation. For the IVR avatar, we aimed to explore the first impression, customization choices, usability issues, ownership perception, and points for improvement. For locomotion, we explored the first impression, usability issues, and impression of the body during locomotion. The questions for each locomotion technique were identical. Concerning the object manipulation, we asked for the first impression, usability issues, enjoyable aspects, and perception of the body during interaction. Lastly, we evaluated the impressions and usability issues concerning the IVE, UI interactions, and intentions for using the system.
Considering the needs of people with MBID, we adapted questionnaires in language and complexity (by using a plain Dutch language) with an expert from the field. This implies that questions asking for two different concepts were reduced to a single one, e.g. 'I felt like the form or appearance of my own body had changed' to 'I felt like the form of my own body had changed' [16]. Also, complex formulations were simplified, e.g. 'Somehow I felt that the virtual world surrounded me' to 'I felt that the virtual world surrounded me' [63].

Procedure
To comply with COVID-19 precautions, the researcher disinfected the materials and IVR apparatus prior to the evaluation. Before the evaluation, disinfection of hands and forearms was required, a FFP2 mask was used by the researcher, and a distance of 1.5m was kept whenever possible.
Participants were welcomed and informed about the research procedure before starting the experiment to comply with the ethical principles in accordance with the Declaration of Helsinki. The ethical approval was given by the University of Twente's ethics committee (ID: RP 2020-164) and the care institution's scientific board. The researcher explained the IVR technology, controls, and possible adverse effects. After informed consent was given, the participants were immersed into the IVE. Besides visual in-game cues, verbal instructions were used to guide the user through the procedure.
Before assessing the prototype version, participants found themselves in the customization room with controller models enabled but deactivated avatar. First, users were asked to set the interpupillary distance using the HMD slider for proper vision. Then, a short acclimatization period was conducted to enhance the spatial understanding, which includes the basic room-scale boundaries and locomotion. After remaining questions were answered, the assessment of the different components was started.
The first task encompassed the avatar customization ( Figure 1). Participants were instructed to go into T-pose to conduct the scaling procedure, followed by the selection of gender and skin tone on the UI ( Figure 1B), which enabled the avatar ( Figure 1A). The participants were given a maximum of five minutes to explore the avatar, hand, and walking animations with room-scale locomotion. After about three minutes, the participants were asked to touch the non-dominant hand by using the dominant one to explore the selflocation of hands through a tactile sensation ( Figure 1D). Then, users were asked to remove the HMD, so the extended VEQ questionnaire and dedicated semi-structured interview questions can be assessed by the researcher. After completion, participants were asked to put on the HMD again to proceed to the next component.
The second task evaluated the controller-based locomotion techniques to move in the IVE (Figure 2). The participants were asked to 'teleport travel' to the vantage point by using the related UI ( Figure 2H). Then, participants enabled the predefined locomotion approach, i.e. joystick ( Figure 2E), teleport ( Figure 2F), or hybrid ( Figure 2G). Following an introduction to the technique, participants were asked to complete the maze. In case of severe motion sickness, participants were allowed to stop earlier to finish the remaining procedure. Upon completion, subjects were asked to remove the HMD to assess the extended VEQ questionnaire and corresponding interview questions. Then, end-users were asked to wear the HMD again to evaluate the remaining locomotion techniques by following the same procedure in an overall counterbalanced manner.
The third task included different manipulations of three objects in a room setting ( Figure 3). First, participants were asked to 'teleport travel' to the locomotion UI to enable the preferred technique. Then, participants were instructed to 'teleport travel' to the manipulation tasks and move to the interactables and related UI. The participants were instructed to (1) grab and release each object ( Figure 3I), (2) pick-up the object from the ground using raycasting ( Figure 3J), (3) place objects at another location based on cues ( Figure 3K), and (4) throw objects into a box ( Figure 3L). After completion, the participants were again asked to remove the HMD and the extended VEQ questionnaire and related interview questions were assessed.
Following the evaluation of the three avatar components, subjects were asked the IPQ questionnaire, demographics, and remaining interview questions. Finally, users were debriefed and encouraged to express remaining questions or concerns that were subsequently thoroughly answered. The participants received a small non-monetary gift as a sign of gratitude (≈ 10€).

Data analysis
Qualitative data were analyzed based on Braun and Clarke's thematic analysis approach [64]. To account for the research design, we divided datasets from the iterations into three segments each: (1) avatar customization, (2) artificial locomotion, and (3) manipulation. In these segments, coding was applied to the transcribed verbatim to identify themes by conducting a recursive process, using Atlas.ti (v. 9.1.4). The coding process was continuously discussed among the researchers (i.e. SL, JN, and RK). Quantitative data regarding the extended VEQ and IPQ subscales were described for each iteration, as well as on an aggregated level. The descriptive analyses were conducted using RStudio (v. 1.3.1093). Table 1 describes the sociodemographic and technological experience characteristics of the sample. In total, five subjects (17.2%) terminated prematurely due to severe motion sickness (n = 3), anxiety (n = 1), or usage inability (n = 1), resulting in missing experimental and demographic data. The remaining subjects (n = 24) had a mean age of 34.2 (±9.8) years and most identified as male (n = 23). The sample was equally composed from the three institutions (n = 8) and included participants with borderline intellectual functioning (IQ = 70-85, n = 13) and mild intellectual disability (IQ = 50-69, n = 11). The technology experience toward computer and videogames was rated high compared to VR. The following sections describe the process that led to our final prototype, as well as procedural considerations identified during the design process.

Findings from iteration 1
Most participants reported a positive first impression regarding the IVR avatar system. The scaling procedure was feasible, though visual in-game instructions were lacking. Here, the avatar's congruence with the own self-concept was essential, since users selected their own skin-tone and gender, reporting on the desire to replicate the own body image ( 'I'm really a slim puppet now. In real life, I have a bit of a belly.' [P4]). Further, technical issues were described, such as unrealistic wrist movements and arm glitches, as well as the absence of haptics (self-touch) properties ('You cannot grasp it, so it is not yours.' [P1]).
Regarding the joystick locomotion, participants reported usability issues when using the 45° snap-turn, further contributing to the prevalent cybersickness ('When you hit the wall, it gets really bad.' [P3]). Contrarily, cybersickness was absent during teleport locomotion ('The teleporting went better. I did feel better.' [P3]), though usability issues due to the limited flexibility and own pace ('At one point it went too fast. Then it seems too easy, but then you have to take a step back.' [P5]) were described. In contrast to the joystick approach, during teleport use, participants missed the human-like walking illusion ('Seems like I just really walk, so to speak.' [P6]). The hybrid approach (i.e. joystick and teleport) showed no added value, since participants relied on their preferred technique. Still, all approaches showed a need for control habituation and attention shift from avatar to task.
Lastly, for the (object) manipulation, the usability was rated positively, though participants reported issues with raycasting, either because it was always enabled, or because it was difficult to hit the objects on the ground. Due to the lack of intuitiveness ( 'Normally you can bend down and grab.' [P2]) and haptics, a need for control habituation was essential ('Because it feels very different from when you're actually grabbing something.' [P3]). Two participants reported on the interaction realism ('But you know, all the movements, the behavior is indeed real.' [P5]), while another missed realistic gabbing animation into the hands instead of controller. The UI interaction showed a good usability, despite need for repositioning to use the object-spaced UIs.
Regarding the IVR avatar, the snap-turn was refined from 45° to 15° to remove usability issues and alleviate cybersickness. To improve manipulations, the grabbing attachment was changed to the avatar's hands instead of controller anchor, and Raycast activations were reduced to objects below 50cm. To enable further customization, altering the model's body dimension (arm-, belly-, leg-, and feet size) was implemented ( Figure 1C). We also improved implementation issues that lead to a smaller scaling of the model. Further, we refined the accuracy of the hand IK-targets for better proprioception. Lastly, as the VEQ items seemed complex for our target group, we added two questions to the semi-structured interview (Appendix 1), asking for the sense of ownership after each interaction tasks and the perception of change after the evaluation procedure.

Findings from iteration 2
Accordant with the first iteration, replicating the own body image was paramount, self-touch and physical collision were lacking, and some animation issues (i.e. arm glitches, unrealistic leg movements) were reported. Contrary, participants reported on the clothing, either due to the incongruence ('Because the suit I was wearing didn't match with what I was wearing' [P15]) or illusion of wearing the virtual clothes ('I was really convinced in my head, that I was wearing it today' [P10]). Technical issues included a restricted view toward the lower body when bending ('When I look down like that, all of a sudden, I got a belly.' [P9]), which also hindered customization due to occlusion. Moreover, hand size adjustments and customization aids (i.e. presets) were lacking. Noteworthy, the human-likeness was described mostly positive, though a single user missed haptics and described the uncanny valley ( 'I actually found that a bit creepy. Because your hands actually looked like real hands.' [P10]).
Regarding the artificial locomotion, cybersickness and snap-turn issues remained during joystick walking. Further usability remarks included inaccurate wall collisions with lacking haptics, inaccurate physics (i.e. weight), an unrealistic foot tilting ('So when I walked fast, my feet just shuffled.' [P17]), preferences for walking using room-scale, and advanced movements (i.e. running, jumping, climbing). Like in the previous iteration, participants missed the human-like walking illusion ('I did walk by myself but also didn't.' [P13]) during teleport ('It was a little inhuman.' [P19]), though some habituated ('At some point when you do figure it out, yes, then it will probably be a little easier' [P11]). Still, other usability problems, such as the limited range, restricted mobility, fast pace resulting in errors, and activation issues were described. The required bodily turning was perceived ambiguous, with one subject suggesting adding a snap-turn. Consistent with the prior iteration, users relied on their preferred technique for hybrid locomotion, and all techniques showed a need for control habituation and attention shift from body to task.
Lastly, the manipulation usability was rated positive, though the need for habituation periods for movements and controls, i.e. switching between locomotion and object manipulation, controller assignment ('What is where ? A and B, joystick.' [P8]), and limited room-scale area ('If you had more space, you could just walk there.' [P9]), were described. Like in the first iteration, hitting objects with raycasting was troublesome. Further, one participant reported on missing haptics ('On the one hand, it feels very familiar and on the other, it's unrealistic that I don't feel.' [P10]). Here, most described the interactions as realistic, though few mentioned a lacking grabbing realism and unrealistic physics/collision resulting in occlusion. Despite needing habituation, all subjects showed a good usability toward the UIs.

Changes for iteration 3
For the third iteration, we simulated bending using a backward placement of the virtual body, aiming to increase the visuoproprioceptive congruence for a more natural behavior. Further, we refined the accuracy of the IK-targets and improved the smoothness of the body rotation, by including influences of the hand locations. Moreover, we removed arm scaling due to the preponderant symmetric nature of human bodies and difficult scaling procedures. Notable, we deactivated the HMD's energy saving option, since we discovered floating floor levels after reactivation, which we aimed to account for during the iteration. To improve the object manipulation realism and remove usability issues, the 'idling' hand animation was refined, and objects were picked-up with sphere-instead of raycasts (i.e. 'magnetic' toward object). To allow further customization, we added an option for hand size adjustment to the related UI. Lastly, we implemented a dynamic FOV reduction (i.e. vignetting) to alleviate cybersickness during artificial locomotion.

Findings from iteration 3
Like in the previous iterations, replicating the own body image through customization was key. Participants described the avatar as human-like ('It looks real, and I also felt that I was touching my own hand' [P24]) with congruent haptics, and two justified customization choices (i.e. skin-tone) in contexts of social meaning ('And it's not because I'm racist.' [P24]). Likewise the previous iterations, minor technical issues like an unrealistic wrist movement and an imprecise bending of legs remained.
Regarding the joystick locomotion, cybersickness and snap-turn issues remained but were reported less severe. Participants rated the embodiment positive, reporting on the human-like walking illusion, though one described an unrealistic foot tilting. Furthermore, the preference for turning using the own body and an inaccurate wall collision, and an attention shift to task were described. Consistent with the previous iterations, cybersickness was absent during teleport locomotion, though usability issues due to the fast pace, activation issues, and turning using the own body remained ('I had to turn but I couldn't walk.' [P22]). Also, control issues that rotate the user after teleporting were reported. Still, the avatar was rated positive, though participants described an ambiguous humanlikeness when teleporting, as well as attention shift from avatar to task. Accordant with the other iterations, all approaches showed a need for control habituation and the hybrid locomotion remained mostly unused.
In line with the previous iterations, participants reported a positive manipulation usability despite need for initial (control) habituation. Just one subject mentioned selection issues when using the object-spaced UI, while another preferred the screen-spaced approach over the object-spaced. Lastly, the embodiment during manipulations was rated positive and human-like, with one subject describing the feeling of haptics through the controller's vibration motors ( Table 2 shows the extended VEQ scores throughout our iterative development. The contextual differences indicate that the sense of ownership tends to increase with growing interaction capabilities, while the perception of change (in the perceived body schema) decreases. Conversely, the sense of self-location and agency scores remain relatively stable across measurements, with positive agency trends during interactions, whereas self-location feelings decreased. Interestingly, ownership and agency scores concerning teleport locomotion were lower than in other active contexts, which matches the qualitative data (see below). The qualitative data indicates that the IVBO was dependent on habituation (

Locomotion preferences and Sense of Presence
Most participants described a preference for joystick locomotion (n = 15). Fewer selected the hybrid (n = 5) or teleport locomotion (n = 4). The latter was predominantly chosen by users that experienced (severe) cybersickness. However, the drop-outs ( n = 5, 17.2%) did not indicate their preference, which should be considered carefully.
The general presence (6.12 ±0.90) and spatial presence (6.03 ±0.81) were rated high compared to moderate involvement (4.27 ±1.88) and lower realism (3.59 ±1.38) scores. The IVR environment was often described as unrealistic and boring, hence participants suggested improving the graphics and realism, including some agents, objects (e.g. chair, cars, plants), or games to make the experience more appealing. However, according to [P11], this might cause overstimulation and distress.

Discussion
The present research reports on the feasibility and related design guidelines for IVBO in adults with MBID, by conducting a usercentered design approach with three iterations. In contrast to previous research on IVR embodiment illusions, our avatar was tailored to the needs of our vulnerable group, by gradually adding interaction and customization abilities. In particular, we investigated the IVR avatar with related IK, as well as controller-based locomotion and (object) manipulation (Appendix 2).
In the following sections, we discuss the findings related to our research questions. First, we discuss the feasibility to induce the illusion, influence of interactions on the SoE, and guidance to enhance the immersion. Then, we discuss the design insights from our three IVR avatar components, i.e. (1) avatar appearance, (2) controller-based locomotion, and (3) object manipulation. Finally, we report on limitations of our work, guidance for future research, and provide a succinct summary of our contribution.

Immersing people with MBID into IVR avatars
Our findings indicate that adults with MBID can embody anthropomorphistic IVR avatars from an egocentric perspective [23], even when avatar dimensions slightly differ from the self [20]. As expected, the highest ownership scores were achieved during object manipulation, requiring the amalgamation of interaction and navigation, though adding locomotion that mimicked human walking was sufficient to enhance the IVBO compared to baseline [46]. Despite participants' desire to replicate the own body image through customization, this did not lead to effectual IVBO using our IVR. Conversely, adding body control was found decisive, suggesting that the sense of agency is vital for inducing ownership illusions in people with MBID [65][66][67]. This finding is further supported by a decreasing perception of change in the body schema during more extensive interactions, despite the unaltered avatar dimensions. Still, the obtained sense of agency and self-location scores showed variance in active contexts, possibly due to visuomotor incongruences or missing human-likeness during locomotion [24,46], occlusions during interaction [41], and extended insights into the limitations of the IK. Hence, we suggest further multisensory integrations [68], i.e. advanced IK, animations, and physical interactions (e.g. collision) with haptics to amplify the illusion [40,69]. Besides, implementing more appealing IVEs could improve the user involvement and realism, potentially enhancing the IVBO [27,60]. However, using IVR avatars for people with MBID required extensive habituation periods when inducing and ending the IVBO, e.g. by gradually adding control and supporting postacclimatization, as some participants described prolonged body sensations ('But for my own body I have to get used to it very much. Also when I take off the glasses, all at once bam, oh I'm here huh.' [P9]). This process proved to be time-consuming and complex, potentially hindering uptake and usability in practice.

Designing IVR avatars for people with MBID
The initial IVR avatar was developed based on the relevant literature and comprised models with high anthropomorphism from an egocentric perspective [34,70], customizable gender (m/f) [71], skin-tone [35], and body dimensions (i.e. model, arms) [41,48]. In contrast to other studies, we omitted a mirror to inspect the virtual body, given that negative influences on ownership were suggested in prior work [72]. Our results indicate that an extended avatar customization could increase ownership feelings for people with MBID, given the desire to replicate the own body image, in particular the body dimensions. However, precise replication methods remain complex and subject to future research [35,73], limiting the applicability in consumer settings. Yet, the replication fidelity in our design questions the need for personalization to induce the IVBO in our target group. Instead, identification with the virtual body through a customization procedure seemed paramount, by replicating major body image characteristics with generic presets ('I'm really a slim puppet now. In real life I have a bit of a belly.' [P4]), as used in commercial social IVR applications (e.g. Meta Horizon). Here, modifiable features (e.g. clothing) appeared more trivial than body image features (e.g. gender, skin-tone, corpulence). Still, the body as reference frame could impact agency and interaction usability [41], as well as the world perception [20,74], possibly resulting in unintended effects [19]. Furthermore, the lacking mirror and plain IVE could have reduced the incongruence awareness in our work [46], as IVR environments can influence the perception [53], and facial properties might backfire when not personalized [35]. Nonetheless, our IK system proved sufficient to induce the IVBO, though avoiding an impaired control was crucial, as occlusions showed more negative remarks than visual mismatches regarding the leg movements [41,46]. Hence, functions for bending should be implemented to achieve sufficient proprioceptive congruence with the user's body.

Designing controller-based locomotion approaches for people with MBID
We investigated the design and user preferences for artificial IVR avatar locomotion approaches of continuous (i.e. joystick) and non-continuous nature (i.e. teleport). Our findings in people with MBID indicate a favor for the joystick in contrast to the teleport or hybrid approach. This difference was explained by the human-likeness and fidelity during joystick locomotion, which can be supported by higher ownership and agency values. In contrast to others, we considered user preferences during our design process, used approaches of different nature, and explored the effects on the SoE [46]. Accordant with previous work, a natural walking animation was vital [46,49], and visuomotor incongruences between the model and stagnant user did not break the illusion [24,46,75]. Instead, users described a sense of agency via controller operation, which can be supported by the obtained SoE scores. Similar findings were observed in the work of Dewez et al. (2020), suggesting that control over the IVR avatar is paramount to visual congruence. A potential explanation for this walking illusion might be the user's attention shift toward navigation, reducing the awareness of visuomotor incongruences while providing a realistic movement illusion. Contrarily to the similar SoE levels when comparing continuous techniques in non-impaired populations [46], we found lower SoE scores when using the non-continuous teleport. However, cybersickness prevalence and resulting drop-outs during joystick locomotion indicate severe usability drawbacks. Hence, designing for cybersickness alleviation seems essential to achieve both, a high SoE and usability, for instance through an adaptable locomotion. Our findings suggest tailoring the FOV (e.g. vignetting), turning (e.g. snap-turn, bodily turning), pace (e.g. speed, range), and experience (e.g. avoid collision, stairs), to account for the needs of our group. Also, enabling control habituation was crucial, given that artificial approaches tend to increase the cognitive workload [42]. Lastly, the hybrid locomotion was redundant since users relied on their preferred technique. Still, it remains interesting to explore this approach in more experienced users, as it allows fast moving without cybersickness while providing fidelity for object manipulation.

Designing controller-based (object) manipulations for people with MBID
We further explored the controller-based IVR avatar interaction by letting users engage in object manipulation tasks with the customized body and preferred locomotion approach. Here, designing for an intuitive interaction was decisive, with realistic animations for virtual hands, physical collision, and related haptics, further supporting the suggested multisensory integration to enhance IVBOs. Though not implemented in our prototype, tailored hand animations could be used to avoid visual interpenetration with virtual objects [41]. Previous work has shown that users favor defined hand poses [76], though constraints through limited animations could reduce the SoE and impact performance [41]. For interactions with lower placed objects, implementing a sphere-cast that is 'magnetic' outperformed raycasting and bending. Though bending was attempted intuitively by our participants, it resulted in severe balance errors that can potentially cause injury. Contrarily, raycasting showed severe usability drawbacks for smaller objects, presumably because the visual and haptic feedback was merely activated when hitting the object. Notable, using artificial interactions did not entail negative remarks, presumed that the control remained unimpaired. Nonetheless, object interaction using full IVBO is understudied, particularly when combined with artificial locomotion. During our design for people with MBID, we observed usability issues when both were combined. Though control habituation could reduce these, providing a generous room-scale area for object manipulation seemed more user friendly, though physical walking was limited to 2x2m. Still, as space is mostly restricted, we encourage others to further explore requirements for an unobtrusive amalgamation of interaction and locomotion. Lastly, the operation of object-spaced and screen-spaced UIs showed a good usability with no negative effects, indicating potential for autonomous usage of such IVR applications by our target users.

Limitations and Future work
There are some limitations in our research that should be considered. First, our convenience sampling in the design process included mainly male subjects with some technology literacy, which might reduce the generalizability of findings to the diverse group with MBID. Secondly, we failed to achieve an accurate scaling method in the first two iterations due to technical issues, considering that the state-of-the-art system was still in beta-stage and applied outside the controlled laboratory setting. Still, the findings show valuable insights for our design and hint towards applications of implicit learning (e.g. proteus effect) [19], as the SoE was observed despite inaccurate avatar dimensions. Thirdly, severe cybersickness issues resulted in drop-outs, which might bias the obtained data, such as locomotion preferences. Fourthly, questionnaires were assessed verbally, possibly leading to an increased social-desirability bias, while paper-based approaches can increase complexness. Previous research from our group suggests that a Visual Analogue Scale implemented in IVR might be more appropriate [7]. Lastly, we utilized a plain IVE that might reduce spatial awareness, such as height and object size perception [20]. Hence, we suggest using spatial cues in future research, which should be carefully selected to avoid distress (i.e. overstimulation) in people with MBID.
Future work should build upon our findings to further refine our guidelines for IVR avatars for people with MBID to design for natural IVR interactions and learning (e.g. psychotherapy, health education, life skills training). Here, influences on the SoE should be investigated to evaluate the interaction design and confirm the feasibility of IVBO in diverse samples (e.g. technology literacy, intellectual and adaptive functioning). From a technical standpoint, exploring multisensory integrations (e.g. advanced IK, interaction animations, haptics, and physics-based manipulations) appears paramount to enhance the feeling of agency, as natural and unimpaired interactions seem pivotal for IVBO. Still, investigating advanced body replication methods as opposed to more generic presets seems important to understand the self-attribution to IVR avatars in people with MBID. Our prototyping in the care setting revealed that customization and habituation procedures were complex and tedious, potentially hindering the applicability of IVBO in people with MBID. Hence, using a balanced design, by conducting habituation periods (i.e. adapting locomotion and interaction) before avatar customization seems promising to reduce the required time for inducing IVBO. This implies neglecting properties that are more trivial in the given use context, such as clothing or facial features, which might be more relevant in social-or collaborative IVR. For this, using game design and narratives seems promising, as common visual interaction cues (e.g. light beams, leading lines, placement cues) and aids (e.g. vantage point, landmarks) showed an adequate usability. Lastly, we combined promising design components, though a plethora of other options can be explored, such as hand tracking and redirected walking to further alleviate cybersickness for natural interactions.

Conclusion
Our findings suggest that adults with MBID can embody gender-matched IVR avatars with a high anthropomorphism. To induce the IVBO, having a high sense of agency over the virtual body appeared crucial, ideally with corresponding multisensory feedback, like physics-based collisions and haptics. This is consistent with previous research on place (PI) and plausibility (Psi) illusions [22], suggesting that plausible interactions are vital for IVBO in our group. However, implementing artificial aids into the virtual body (i.e., sphere-casting, raycasting) was not perceived as disruptive, presuming that the control was not impaired. Customizing the avatar according to the subject's body image appeared to boost the illusion, though it was complex and tedious, impacting the practicability of IVBO, as individuals with MBID showed an extensive need for (control) habituation. Therefore, balancing the IVBO immersion, by focusing on habituation and lowering customization effort seems crucial to achieve both, a high SoE and usability. Future research should further investigate guidelines for IVR avatars in people with MBID by designing for natural interactions, including multisensory integrations and other interaction approaches (e.g. hand tracking, redirected walking). In addition, procedures and usage for implicit and explicit learning should be explored, for instance as tool for playful health behavior change interventions.