The sense of realism in avatar animation is a widely pursued goal in social VR applications. A common approach to enhancing realism is improving the match between avatar motion and real-world human movement. However, experience with existing VR platforms may reshape users' expectations, suggesting that matching reality is not the only path to enhancing the sense of realism. This study examines how different levels of experience with a social VR platform influence users' criteria for evaluating the realism of avatar animation. Participants were shown a set of animations varying in the degree they reflected real-world motion and motion seen on the social VR platform VRChat. Results showed that users with no VRChat experience found animations recorded on VRChat unnatural and unrealistic, but experienced users in fact rated these animations as more likely to come from a real person than the motion-capture animations. Additionally, highly experienced users recognized the intent to imitate VRChat's style and noted the differences from genuine in-platform animations. All these results suggest users' expectations of and criteria for realistic animation were shaped by their experience level. The findings support the idea that realism in avatar animation does not solely depend on mimicking real-world movement. Experience with VR platforms can shape how users expect, perceive, and evaluate animation realism. This insight can inform the design of more immersive VR environments and virtual humans in the future.

Virtual and augmented reality devices track the body motion of users because it is fundamental to rendering virtual content anchored to space. However, this same body motion data can be used as a biometric to identify users. Research on the effectiveness of these biometrics are often difficult to compare between because the common evaluation metric, rank-1 accuracy, is relative to the number of identities within the set. In this work, I motivate, select, and justify the use of a previously introduced classification model evaluation metric, multiclass AUC, that is invariant to the number of classes (i.e., individuals) being identified, producing more effective comparisons across disparate datasets, activities, and participant pool sizes. I also generalize this metric with regular rank-1 accuracy to produce N-class accuracy, allowing future work to compare to past work when multiclass AUC is not reported. The common use of this metric will allow a finer view of patterns in identifiability of this motion data, ultimately resulting in clearer research conclusions when comparing across works.

Social virtual reality is an emerging medium of communication. In this medium, a user’s avatar (virtual representation) is controlled by the tracked motion of the user’s headset and hand controllers. This tracked motion is a rich data stream that can leak characteristics of the user or can be effectively matched to previously-identified data to identify a user. To better understand the boundaries of motion data identifiability, we investigate how varying training data duration and train-test delay affects the accuracy at which a machine learning model can correctly classify user motion in a supervised learning task simulating re-identification. The dataset we use has a unique combination of a large number of participants, long duration per session, large number of sessions, and a long time span over which sessions were conducted. We find that training data duration and train-test delay affect identifiability; that minimal train-test delay leads to very high accuracy; and that train-test delay should be controlled in future experiments.

The use of virtual and augmented reality devices is increasing, but these sensor-rich devices pose risks to privacy. The ability to track a user’s motion and infer the identity or characteristics of the user poses a privacy risk that has received significant attention. Existing deep-network-based defenses against this risk, however, require significant amounts of training data and have not yet been shown to generalize beyond specific applications. In this work, we study the effect of signal degradation on identifiability, specifically through added noise, reduced framerate, reduced precision, and reduced dimensionality of the data. Our experiment shows that state-of-the-art identification attacks still achieve near-perfect accuracy for each of these degradations. This negative result demonstrates the difficulty of anonymizing this motion data and gives some justification to the existing data- and compute-intensive deep-network based methods.

Social Virtual Reality (VR) typically entails users interacting in real time. However, asynchronous Social VR presents the possibility of combining the convenience of asynchronous communication with the high presence of VR. Because the tools to easily record and replay VR social interactions are fairly new, scholars have not yet examined how users perceive asynchronous VR social interactions, and how nonverbal transformations of recorded interactions influence user behavior. In this work, we study nonverbal transformations of group interactions around proxemics and gaze and present results from an exploratory user study (N=128) investigating their effects. We found that the combination of spatial accommodation and added gaze increases social presence, perceived attention, and mutual gaze. Results also showed an inverse relationship between interpersonal distance and perceived levels of dominance and threat of the recorded group. Finally, we outline implications for educators and virtual meeting organizers to incorporate these transformations into real-world scenarios.

Scholars who study nonverbal behavior have focused an incredible amount of work on proxemics, how close people stand to one another, and mutual gaze, whether or not they are looking at one another. Moreover, many studies have demonstrated a correlation between gaze and distance, and so-called equilibrium theory posits that people modulate gaze and distance to maintain proper levels of nonverbal intimacy. Virtual reality scholars have also focused on these two constructs, both for theoretical reasons, as distance and gaze are often used as proxies for psychological constructs such as social presence, and for methodological reasons, as head orientation and body position are automatically produced by most VR tracking systems. However, to date, the studies of distance and gaze in VR have largely been conducted in laboratory settings, observing behavior of a small number of participants for short periods of time. In this experimental field study, we analyze the proxemics and gaze of 232 participants over two experimental studies who each contributed up to about 240 minutes of tracking data during eight weekly 30-minute social virtual reality sessions. Participants' non-verbal behaviors changed in conjunction with context manipulations and over time. Interpersonal distance increased with the size of the virtual room; and both mutual gaze and interpersonal distance increased over time. Overall, participants oriented their heads toward the center of walls rather than to corners of rectangularly-aligned environments. Finally, statistical models demonstrated that individual differences matter, with pairs and groups maintaining more consistent differences over time than would be predicted by chance. Implications for theory and practice are discussed.

As the public use of virtual reality (VR) scales, understanding how users engage across various sources of VR content is critical. 360-video is popular due to its ease of both creation and access. There are, however, few studies of 360-videos, and they suffer from three limitations. First, most studies rely on small and homogeneous samples of participants. Second, they tend to examine only a single 360-video, or a handful of them in a few exceptional cases. Third, very few studies trace participants’ VR use over multiple experiences. The current study examined a large sample of participants (511) and a large set of 360-videos (80). Each participant experienced 5 of the videos, and we tracked head movement in addition to self-report data on presence, arousal, simulator sickness, and future use intention for each video. This design allowed us to answer novel questions relating to individual differences of participants and changes in experience over time, and in general to present results of VR use at a scale not seen before in the literature. Moreover, the results suggest that looking at patterns across stimuli provide unique insights which are missed when looking only within a single piece of content.

Synchrony, the natural time-dependence of behavior in human interaction, is a pervasive feature of communication. However, most studies of synchrony have focused on dyadic interaction. In the current work, we explore synchrony in three-person teams using immersive virtual reality. Participants spent about two hours collaborating on four separate design tasks. The tracking data from the VR system allowed precise measurement of head and hand movements, facilitating calculation of synchrony. Results replicated previous work that found nonverbal synchrony in dyads in immersive VR. Moreover, we manipulated the context of the task environment, an informal garage or a traditional conference room. The environment for the task influenced synchrony, with higher levels occurring in the conference room than the garage. We also explored different methods of extending synchrony from dyads to triads, and explore the relationship of synchrony to turn taking and gaze. This paper provides theoretical insights about nonverbal synchrony and how design work functions in triads and provides suggestions for designers of VR to support good collaboration.

Ungrounded haptic devices for virtual reality (VR) applications lack the ability to convincingly render the sensations of a grasped virtual object's rigidity and weight. We present Grabity, a wearable haptic device designed to simulate kinesthetic pad opposition grip forces and weight for grasping virtual objects in VR. The device is mounted on the index finger and thumb and enables precision grasps with a wide range of motion. A unidirectional brake creates rigid grasping force feedback. Two voice coil actuators create virtual force tangential to each finger pad through asymmetric skin deformation. These forces can be perceived as gravitational and inertial forces of virtual objects. The rotational orientation of the voice coil actuators is passively aligned with the real direction of gravity through a revolute joint, causing the virtual forces to always point downward. This paper evaluates the performance of Grabity through two user studies, finding promising ability to simulate different levels of weight with convincing object rigidity. The first user study shows that Grabity can convey various magnitudes of weight and force sensations to users by manipulating the amplitude of the asymmetric vibration. The second user study shows that users can differentiate different weights in a virtual environment using Grabity.