IEEE TVCG / VR
Co-Embodiment
Moving One Body Together, as Two
In VR, virtual co-embodiment blends the movements of two people to operate a single avatar. Does sharing a body with a master help skills transfer? When something goes wrong, whose fault is it? A line of research that keeps raising curious questions.
In VR, you can try out ways of having a body that would be impossible in reality. One of them is "virtual co-embodiment" — a technique in which the movements of two users are blended together (for example, as a fifty-fifty weighted average) so that they control a single avatar as one.
Background
In sports and music alike, bodily skills are notoriously hard to put into words — the traditional advice is to "watch and steal." So what would happen if you could literally share a body with a master and practice while feeling their movements from the inside? And when you share a body, how does the feeling that "I am the one moving" (the sense of agency) change — and how do you feel about responsibility when things go wrong? Virtual co-embodiment opens up questions about both skill transfer and the sense of self.
Research Question
How does training with virtual co-embodiment change motor skill learning? And when people act while sharing a body with another, how do they experience the sense of agency and responsibility?
Methods
We built a system that maps the weighted average of two people's movements onto a single avatar, and ran comparative experiments using motor learning tasks. We compared practicing in a co-embodied body with practicing alone, measuring the retention of learned movements and the sense of agency. In follow-up studies, we examined how skills transfer when the co-embodied partner is an AI rather than a human teacher, and — in a study titled "It's My Fingers' Fault" — the sense of agency and attribution of responsibility when the shared body makes a mistake.
Findings
Training with virtual co-embodiment turned out to leave the learned movements in the body longer than practicing alone. What's more, learners retain a moderate sense of agency — the feeling that "I am the one moving" — while they learn. Skills can be transferred even when the partner is an AI teacher, though the way they transfer differs from the human case. The boundary between "my body" and "my action" is far softer than we tend to assume. Virtual co-embodiment uses that softness to explore new forms of skill transfer and collaboration — and at the same time, it is an attempt to approach the old question of "what is the self?" from the engineering side.