It has always bothered me when in VR, a virtual body can be seen, but the movement feels wrong and the body parts out of place. What about a virtual head, that is perceived only as a reflection, and is always correct thanks to Positional Tracking? Wouldn't it be much easier to accept and identify with?
My experiment involved reflective windows that I didn't mention in advance to the participating guinea pigs (friends and family), instead only talking about some binaural sound sources in the scene that I also wanted to test. In most cases, the reflection was discovered after a short time and the head accepted as a self-image. ("Oh, so this is me?", "I see myself in the mirror!") Immediately the participants explored the situation and tested different movements.
Even the participants who didn't notice anything at first, as soon as they had been given a hint, they discovered the avatar and identified with it. The more masculine character was also accepted as an object of identification by the female participants (and generally was received well enough to show up a second time later in this portfolio). Gender affiliations do not seem as important for rudimentary self-identification at least at this level of abstraction.
Of course these are only anecdotal findings, but the observations and self-observations have convinced me that it should be relatively simple to to create a pleasant minimal self-representation in the VR space given a robustly tracked head and hands (in the way Oculus' Toybox does it for example) . It doesn't seem to be necessary to use especially realistic models, or to match the identity of the people closely, if the position and movement are adequately represented and virtual and real body of the person are roughly equivalent in size. There seem to be body parts that are more important and some that are less important(e.g. head and hands vs feet)
Interestingly, Oculus have conducted similar experiments with reflections and mirrors and apparently came to similar conclusions: Link to the lecture