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.

In social virtual reality (social VR), users are threatened by potential cybercrimes, such as identity theft, sensitive data breaches, and embodied harassment. These concerns are heightened by the increasing interest in the metaverse, the advancements in photorealistic 3D user reconstructions, and the rising incidents of online privacy violations. Designing secure social VR applications that protect users while enhancing their experience, acceptance and trust remains a challenge. This article investigates potential identity management solutions in social VR, and their impacts on usability and user acceptance. We developed a social VR prototype with novel and established countermeasures, including motion biometric verification, and conducted a study with 52 participants. Our findings reveal diverse preferences for identity management and underscore the importance of authenticity, autonomy, and reciprocity. Key findings include: passive verification is favored for pragmatic user experience, while active verification is preferred for its hedonic quality; continuous or periodic verification strengthens users’ confidence in their privacy; and while user awareness promotes authentic engagement, it may also diminish the willingness to disclose personal information. This research not only offers foundational insights into the evaluated scenarios and countermeasures, but also sheds light on the designs of more trustworthy and inclusive social VR applications.

This paper introduces “Motion Passwords”, a novel biometric authentication approach where virtual reality users verify their identity by physically writing a chosen word in the air with their hand controller. This method allows combining three layers of verification: knowledge-based password input, handwriting style analysis, and motion profile recognition. As a first step towards realizing this potential, we focus on verifying users based on their motion profiles. We conducted a data collection study with 48 participants, who performed over 3800 Motion Password signatures across two sessions. We assessed the effectiveness of feature-distance and similarity-learning methods for motion-based verification using the Motion Passwords as well as specific and uniform ball-throwing signatures used in previous works. In our results, the similarity-learning model was able to verify users with the same accuracy for both signature types. This demonstrates that Motion Passwords, even when applying only the motion-based verification layer, achieve reliability comparable to previous methods. This highlights the potential for Motion Passwords to become even more reliable with the addition of knowledge-based and handwriting style verification layers. Furthermore, we present a proof-of-concept Unity application demonstrating the registration and verification process with our pretrained similarity-learning model. We publish our code, the Motion Password dataset, the pretrained model, and our Unity prototype on https://github.com/cschell/MoPs