Avatar fidelity and Co-presence in a collaborative VR educational task: a secondary analysis of the HuTwin dataset

Abstract

Social presence plays an important role in collaborative virtual reality (VR), yet it remains unclear whether more human-like avatars improve only the felt quality of interaction or also support better task performance. This question becomes especially important in collaborative settings, where richer avatars may increase technical complexity while offering uncertain practical benefit. We address this issue through a secondary analysis of the HuTwin dataset, an open-access dataset of dyadic collaborative VR interaction. The analysis focuses on a shared cooking task involving 40 participants working in pairs under different avatar and network conditions. We compare a stylized Chef avatar with a more human-like Humanoid avatar, and examine their effects on participants' co-presence ratings and task completion time. We also test whether network delay shapes these effects. The results show a differential pattern across subjective and objective outcomes. The Humanoid avatar increased co-presence ratings, but it did not significantly improve task completion time. Delay reduced co-presence ratings, impaired task performance, and attenuated the co-presence-rating advantage of the Humanoid avatar. In addition, exploratory descriptive analyses suggested that co-presence ratings and task completion time did not co-vary detectably in this sample. The findings suggest that richer avatar embodiment primarily enhances the interpersonal experience of collaboration rather than its behavioral efficiency, and that this benefit is attenuated under degraded network conditions. More broadly, the study indicates that experiential and performance outcomes should be treated as distinct dimensions in collaborative VR research, rather than as interchangeable indicators of collaborative success.