The Influence of Embodiment and Personalization of Photorealistic Avatars on Body Perception in Virtual Reality

This project is already assigned.

Motivation

People who suffer from eating or body weight disorders often have a distorted mental body image (Parzer et al., 2020). But body weight misperception has also been found in healthy, normal-weight people (Longo, 2017). For exploring, detecting and visualizing body weight perception, approaches using virtual reality (VR) systems are gaining importance. Thereby, many potentially factors seem to influence per- ception of the body and body weight in VR. Such factors include the own body weight, the observation perspective on the virtual person (Neyret et al., 2020), its realism and degree of personalization (Piryankova et al., 2014; Thaler et al., 2018), and the illusion of being embodied in it (Wolf et al., 2020; Wolf et al., 2021). The effect of all these factors should be known and taken into account when designing a VR application that is intended to represent the user’s real self, in order to represent the user in the VR environment in such a way that they perceive themself in the same way as in reality.

Goal

This master thesis is conducted within the research project “ViTraS” (Virtual Reality Therapy by Stimulation of Modulated Body Perception) (Döllinger et al., 2019). In obesity therapy, the research project “ViTraS” investigates approaches to support the treatment of eating and body image disorders with the help of VR and augmented reality (AR). It aims to shed light on the factors that influence body perception and, in particular, body weight perception in VR. The master thesis will explore the influence of one’s own body weight on the body weight perception of a virtual person (avatar) in dependence of mainly two factors: the influence of the avatar’s personalization (personalized vs. non-personalized avatar) and the embodiment between participant and avatar (being self-embodied vs being non-self- embodied). In the “being self-embodied” condition, the participant embodies an avatar that they can see through an egocentric perspective and through an allo- centric perspective via a virtual mirror. In the condition “being non-self-embodied” the participant sees an avatar from the egocentric perspective, which they does not embody. Furthermore, in this condition, the participant does not have a body of its own, since this study aims to maximize discriminability between the two conditions and having a body of its own would limit it. Embodiment must be clearly distin- guished from the feeling of embodiment at this point; this feeling is collected during the experiment as a mediating variable and manipulation check, to check whether the variation in the embodiment condition has achieved the desired effect. It is measured via questionnaires and reflects the subjective feeling of the participant.

In summary, the goal of this master thesis is to investigate the influence of these factors, personalization and embodiment on body weight perception, operationalized by body weight estimation, in VR. By systematically examining the influence of both subfactors on the estimation of the avatar’s body weight, it can be analyzed whether and how both factors affect the estimation and interact with each other.

Thaler et al. (2018) investigated the influence of one’s own body weight and the in- fluence of a non-embodied avatar’s personalization on body weight estimation. Their results show that the accuracy of estimating one’s own body weight is predicted by the personal body mass index (BMI). Participants with a lower BMI underestimated their body size and participants with a higher BMI overestimated their body size. The results were only evident when participants judged an avatar that matched their own personalization and shape, not when they estimated the body weight of an avatar with different personalization, but same underlying body shape. Wolf et al. (2020) extended the results of Thaler et al. (2018). They explored how the type of medium (VR vs. AR) affects the body weight perception of a non-personalized, embodied virtual person. They showed that participants’ BMI, regardless of the medium used, strongly influenced body weight perception of the participants’ em- bodied avatar, even when the avatar was not personalized. Furthermore, they were also able to show that participants with a lower BMI mostly underestimated the avatar weight and participants with a higher BMI were more likely to correctly estimate or even overestimate the avatar weight.

By comparing body size estimation of self and another personalization, it was found in previous studies that a relevant factor for biased body size estimation seems to be embodiment. Wolf et al. (2021) compared body size estimation of a self- embodied virtual person against only observing the virtual person. They concluded that in the self-embodied condition, the weight of the non-personalized, photoreal- istic avatar was significantly underestimated in contrast to the non-self-embodied condition. Is this also true when the avatar has a personalized appearance? If so, a future (therapeutic) application would be conceivable, in which the user (or patient) in a non-self-embodied condition faces themself, looks at “themself” from different perspectives and subsequently manages to estimate their body weight in real terms. Another conceivable application would be a combination of self-embodied and non- self-embodied conditions, in which the patient learns to estimate their body weight in real terms. The findings of Neyret et al. (2020) match the idea. They observed that female participants rated their real body as more attractive when they saw it from a third person perspective, instead of a first person perspective. Based on their results, they believe that, by perceiving their body in third-person perspective, patients could obtain a new and unbiased perception of their own body (as if it were the body of someone else). This new perception could re-orientate their attention to the real features of their body shape in a more accurate and objective way.

However, the mentioned setting of personalized avatar and non-self-embodied condition or even the photorealistic avatar itself could be perceived as creepy. The circumstance that the participant sees but does not embody themself or the appearance of the avatar itself could evoke an uncanniness, which can influence the body perception and thereby perhaps the body weight perception, too. To account for this factor of “creepiness” within the setting or of the avatar, it will be measured as well. In addition, it remains open to what extent the participant identifies with the virtual person in the room and attributes personal characteristics, such as body weight, to this avatar. It would also be possible that the participant sees and evaluates the avatar as another person in the room, independent of themself. According to Petzold (2003), identity is “gained through other- and self-attributing, cognitively and emotionally evaluative interactions and volitive acts” (Petzold, 2003, p. 116). This means that identity is formed, among others, from self-attributions. For this reason, the self-attribution to the avatar and its effects on body weight estimation are additionally recorded within the scope of the study.

The results of prior related work suggest that embodiment and personalization are factors contributing to body weight perception of a virtual person. However, according to current research, there is no study that systematically compares both components. For this reason, this master thesis aims to create a holistic picture of the effect of being (not) embodied to and having a (non-)personalized avatar on body weight estimation in VR.

Concept

Within the course of the master thesis, a user study will be designed and conducted. The study will include the generation of a personalized virtual person following the methodology of Achenbach et al. (2017) and subsequent an exposure to a VR mirror application. For the VR user study a suitable virtual environment must be developed for each condition using the game engine Unity (Unity Technologies, 2020). It must be ensured that all conditions remain as similar as possible to each other and thus comparable by keeping the procedure, set-up of the environment and placement of avatar and participant in the room as identical as possible. In addition, confounding factors should be kept particularly in mind to make a study as free as possible from survey bias.

The independent variables are the personalization of the virtual person (personalized avatar vs. non-personalized avatar) and embodiment between user and the represented avatar (observing a self-embodied virtual person vs. only observing a non-self-embodied virtual person). This results in a mixed 2x2 study design, whereby the factor “personalization” will be collected “within”. It is important to mention that the personalized avatar used will match the scanned participant in gender, body shape and texture. The non-personalized avatar will have the same gender and body shape, but a standardized texture. Bodies naturally differ a lot in shape and since shape cues are highly relevant for estimating body size, the shape is kept constant over the conditions. In addition, by keeping the body shape, the participant’s BMI is also maintained, and through this reference the weight manipulation will be performed during the study. The goal is to investigate the above-mentioned independent variables for a possible main effect and an interaction effect regarding body weight estimation. Furthermore, possible mediating effects of all measured dependent variables on body weight estimation will be examined. The theoretical model to be tested is shown in Figure 1.

Figure 1. Theoretical model that summarizes the relationships to be collected between all constructs and variables involved.

A quantitative survey consisting of different questionnaires and a qualitative survey via a semi-structured interview will be conducted to evaluate the theoretical model. During the survey the following five dependent variables will be measured. Again, if new variables that seem relevant to body weight estimation emerge from the literature review, they will also be integrated into the user study and the design will be expanded.

The pESQ (Eubanks et al., 2021) will be used to measure the feeling of embodiment in virtuo. Additionally, a post exposure measurement of the feeling of embodiment using the VEQ (Roth & Latoschik, 2020) will be performed as recommended by Eubanks et al. (2021), because the short pESQ will not provide as much sensitivity as longer questionnaires.

Presence will be measured using the IPQ (Schubert et al., 2001). In addition, as with the measurement of the feeling of embodiment, an in virtuo measurement will be performed.

Participants will estimate the body weight of their avatar via (1) an active modification task in which the participant must adjust the avatar’s body shape to a given weight, to their own body shape and to a social ideal body shape using gesture control; (2) a passive estimation task in which the participant must estimate the modified body weight of the virtual person several times in virtuo. Additionally, for each participant, body weight and height will be recorded using calibrated medical equipment to subsequently calculate the body weight misestimation from the participant’s weight and estimated avatar weight.

The application of a self-attribution questionnaire with eight items, which has not yet been validated, is intended to show the extent to which the participant ascribes attributes from themself to the avatar and vice versa. It is applied to possibly make a statement about how much the participant identifies with the avatar in the different settings. Furthermore, this would provide information about how well the concept of “identity” can be operationalized through personification and embodiment.

In addition, it should be evaluated how creepy the presented avatar is perceived and how creepy the setting is perceived to be (e.g., how creepy it is perceived to be when your personal avatar faces you in the “only observing” condition). One way to survey the perception of the avatar would be via the revised eeriness and humanness index from Ho and MacDorman (2017). A future literature review should provide the most appropriate questionnaires possible to survey the creepiness of the setting and the uncanniness of the avatar.

A semi-structured interview with predefined questions will be conducted post exposure to obtain information about the perception of the virtual avatar (appearance, recognition of own shape, movements, creepiness), the predictors and and the VR experience itself.

Following the data collection, the gathered data should be analyzed and interpreted using the correct statistical procedure. Which procedure is used for the evaluation depends on the structure and the chosen hypotheses of the study. The phase and milestone plan for the master thesis can be found in Figure 2.

Figure 2. Work schedule of the master thesis.

References

Achenbach, J., Waltemate, T., Latoschik, M. E., & Botsch, M. (2017). Fast generation of realistic virtual humans. Proceedings of the 23rd ACM Symposium on Virtual Reality Software and Technology.

Döllinger, N., Wienrich, C., Wolf, E., Botsch, M., & Latoschik, M. E. (2019). Vitras - virtual reality therapy by stimulation of modulated body image - project outline. Mensch und Computer 2019 - Workshopband, 606–611.

Eubanks, J. C., Moore, A. G., Fishwick, P. A., & McMahan, R. P. (2021). A preliminary embodiment short questionnaire. Frontiers in Virtual Reality, 2, 24.

Ho, C.-C., & MacDorman, K. (2017). Measuring the uncanny valley effect: Refinements to indices for perceived humanness, attractiveness, and eeriness. International Journal of Social Robotics, 9, 129–139 (cit. on p. 8).

Longo, M. R. (2017). Distorted body representations in healthy cognition. Quarterly Journal of Experimental Psychology, 70, 378–388.

Neyret, S., Bellido Rivas, A. I., Navarro, X., & Slater, M. (2020). Which body would you like to have? the impact of embodied perspective on body perception and body evaluation in immersive virtual reality. Frontiers in Robotics and AI, 7, 31.

Parzer, V., Taube, M., Ludvik, B., Svensson, P. A., Brix, J. M., & Sjöholm, K. Difference in body image perception between weight gainers and weight maintainers in the conventionally treated control group from the swedish obese subjects (sos) intervention study. In: In European and inter- national congress on obesity 2020 (T. editor, Ed.). Ed. by editor, T. 2020, September.

Petzold, H. (2003). Integrative Therapie 3 Bände: Modelle, Theorien und Methoden für eine schulenübergreifende Psychotherapie. Junfermann Verlag.

Piryankova, I. V., Stefanucci, J. K., Romero, J., De La Rosa, S., Black, M. J., & Mohler, B. J. (2014). Can I Recognize My Body’s Weight? The Influence of Shape and Texture on the Perception of Self. ACM Transactions on Applied Perception, 11(3).

Roth, D., & Latoschik, M. E. (2020). Construction of the virtual embodiment questionnaire (veq). IEEE Transactions on Visualization and Computer Graphics, 26 (12), 3546–3556.

Schubert, T., Friedmann, F., & Regenbrecht, H. (2001). The experience of presence: Factor analytic insights. Presence: Teleoperators & Virtual Environments, 10 (3), 266–281.

Thaler, A., Geuss, M. N., Mölbert, S. C., Giel, K. E., Streuber, S., Romero, J., Black, M. J., & Mohler, B. J. (2018). Body size estimation of self and others in females varying in BMI. PLOS ONE, 13(2), e0192152.

Unity Technologies. (2020). Unity [https://unity3d.com].

Wolf, E., Döllinger, N., Mal, D., Wienrich, C., Botsch, M., & Latoschik, M. E. (2020). Body Weight Perception of Females using Photorealistic Avatars in Virtual and Augmented Reality. 2020 IEEE International Symposium on Mixed and Augmented Reality (ISMAR), 462–473.

Wolf, E., Merdan, N., Döllinger, N., Mal, D., Wienrich, C., Botsch, M., & Latoschik, M. E. (2021). The embodiment of photorealistic avatars influences female body weight perception in virtual reality. 2021 IEEE Virtual Reality and 3D User Interfaces (VR), 65–74.

Contact Persons at the University Würzburg