Synthetic vs. captured

Two extremes exist when constructing a virtual world as part of a VR experience. At one end, we may program a synthetic world, which is completely invented from geometric primitives and simulated physics. This is common in video games and such virtual environments were assumed to be the main way to experience VR in earlier decades. At the other end, the world may be captured using modern imaging techniques. For viewing on a screen, the video camera has served this purpose for over a century. Capturing panoramic images and videos and then seeing them from any viewpoint in a VR system is a natural extension. In many settings, however, too much information is lost when projecting the real world onto the camera sensor. What happens when the user changes her head position and viewpoint? More information should be captured in this case. Using depth sensors and SLAM (Simultaneous Localization And Mapping) techniques, a 3D representation of the surrounding world can be captured and maintained over time as it changes. It is extremely difficult, however, to construct an accurate and reliable representation, unless the environment is explicitly engineered for such capture (for example, a motion capture studio).

As humans interact, it becomes important to track their motions, which is an important form of capture. What are their facial expressions while wearing a VR headset? Do we need to know their hand gestures? What can we infer about their emotional state? Are their eyes focused on me? Synthetic representations of ourselves called avatars enable us to interact and provide a level of anonymity, if desired in some contexts. The attentiveness or emotional state can be generated synthetically. We can also enhance our avatars by tracking the motions and other attributes of our actual bodies. A well-known problem is the uncanny valley, in which a high degree of realism has been achieved in an avatar, but its appearance makes people feel uneasy. It seems almost right, but the small differences are disturbing. There is currently no easy way to make ourselves appear to others in a VR experience exactly as we do in the real world, and in most cases, we might not want to.

Steven M LaValle 2020-11-11