Hybrid options at conferences, which support in-person and remote attendance, have increasingly become the norm in order to broaden participation and promote sustainability. However, hybrid conferences are challenging, where in-person and remote attendees often have disjoint, parallel experiences with limited opportunity to interact with each other. To explore the potential for facilitating social interaction between in-person and remote conference attendees, we designed and built SocialMiXR, a research prototype that uses WebXR technologies to align the physical and virtual worlds into one hybrid space for socialization. We deployed SocialMiXR in a three-day field study with 14 in-person and remote attendees of an engineering conference. Our qualitative results demonstrate that participants felt they were together in the same conference experience and formed meaningful connections with each other. At the same time, they faced difficulties balancing different realities and capabilities given their separate contexts. We discuss implications for the design of hybrid social experiences at conferences.

Physical restrictions of the real spaces where users are situated present challenges for remote XR and spatial computing interactions using avatars. Users may not have sufficient space in their physical environment to duplicate the physical setup of their collaborators. However, simply relocating avatars can break the meaning of one-to-one motions and non-verbal cues. We propose MRTransformer, a technique that uses weighted interpolations to guarantee that collaborators look at and point to the same objects locally and remotely, while preserving the meaning of gestures and postures that are not object-directed (e.g., close to the body). We extend this approach to support locomotion and near-space interactions such as grabbing objects, exploring the limits of social and scene understanding while enabling more flexible uses of inverse kinematics (IK). We discuss limitations and applications of this approach and open-source AvatarPilot to support broader use in mixed reality collaboration systems.

Physical restrictions of the real spaces where users are situated present challenges to remote XR and spatial computing interactions using avatars. Users may not have sufficient space in their physical environment to duplicate the physical setup of their collaborators. However, when avatars are relocated to accommodate space constraints, one-to-one motions may no longer preserve their intended meaning. We propose AvatarPilot, a system that uses weighted interpolations to decouple low-level avatar motion from higher-level semantic intent. This approach ensures that collaborators look at and point to the same objects locally and remotely, while preserving non-object-directed gestures and postures near the body. We extend this technique to support locomotion and direct interactions in near space, such as grabbing objects, enabling more flexible use of inverse kinematics (IK). We discuss applications and limitations of this approach and release AvatarPilot as an open-source toolkit to support future mixed reality collaboration systems.

Recent advances in 3D reconstruction technologies allow people to capture and share experiences in three dimensions, yet little is known about how people perceive identity and privacy in such reconstructed experiences. We present ReliveReality, an experience-sharing method that uses deep learning-based computer vision techniques to reconstruct clothed humans, estimate 3D pose, and recover surrounding environments using only an RGB camera. ReliveReality can be embedded into social virtual environments, allowing others to socially relive experiences from multiple viewpoints on desktop or in VR. We conducted a 44-participant within-subject study comparing ReliveReality to traditional recorded video and to a version with avatar blurring obfuscation. Our findings reveal how people identify with reconstructed avatars, how obfuscation affects reliving experiences, and how privacy concerns influence sharing preferences. We discuss design implications for supporting identity, privacy, and control in reconstructed social experiences.