(Avatar Physics and Genetics)
1 Introduction
Our world is rich and varied, not only because of the visual detail, color, texture, and movement we perceive in it, but also because of our biological and emotional connection to living things, human and non-human. Our world is also rich because of the emergent properties of physical process. A belief expressed in this paper is that a rich virtual reality experience can be achieved through building a foundation of naturalistic laws, whereby more complex, realistic, and expressive levels of detail can emerge easily. With this underlying premise, a prototype of a virtual human has been developed, built entirely out of software, appropriate for a large-scale virtual world.
Many of the established techniques for designing virtual humans are descendants of the tools of computer-aided design and 3D modeling, which is the prevailing paradigm for building virtual world objects. Techniques and principles from tradi-tional art and design have become incorporated into such modeling systems to make the process of human modeling more intuitive [3].
For constructing virtual worlds, standards such as VRML have spawned many technical variants, as well as a culture of users. While including functionality for designing virtual worlds, these paradigms rely heavily on polygonal model represen-tations. Emphasis on polygonal models and rendering is referred to here as the "computer graphics paradigm". In this paradigm, the representation of a human is influenced by the specific nature of 3D modeling and animation software. This paradigm has been augmented by many other fields as in the extensive hu-man modeling research of Thalmann [5]. This work represents a thorough study in representing human anatomy, motion, and clothing, including facial modeling. Of note is the detailed modeling of hands, which, like faces, are important for expression.
The approach to virtual human design described in this paper places importance on physics and genetics, and places rendering towards the end of the pipeline. This approach is intended to scale up such that each avatar is unique yet related to the same humanoid gene pool, able to render itself in 3D, and equipped with all the physical laws necessary to land in the virtual world "running".
1.1 Avatars
When you chat in a 3D online world or play one of many 3D computer games, you are operating a synthetic character: an avatar. An avatar is defined here as a synthetic human represented graphically in a virtual world, which is controlled by a real human, and which represents that human’s identity. For purposes of this discussion, it is assumed that a virtual camera is positioned near the avatar, (usually behind, in third-person view), and that the camera can be manipulated for alternate views, including first-person view.
We are approaching a time when many of us will have our own avatar online, perhaps multiple avatars. Each will be distinct from every other, able to be customized, able to interact with other avatars and increasingly complex virtual worlds. It will even be possible for avatars to be bred and to reproduce their virtual genes online for future generations. Most of the basic research has been done to achieve this. All that is needed are the appropriate technologies and interfaces.
What makes for an interesting and effective avatar? It depends on its purpose, of course. In the case of a virtual world where communication is important, facial features and expressiveness must be well supported. In the case of action games, physics and interaction with the world must be well supported. In the case of hardcore violent action games, viscera and bodily fluids may be necessary - although that is not the focus here. In fact, this is a rather non-visceral approach to what it means to be or to interact with a human. Having said that, however, the experience of moving around on a terrain or floor, having a sense of balance, and other physical aspects of being a human in the world, are important in this representation. Thus, these avatars can be seen as supporting two major needs: social and physical. Where genetics comes into play is mostly in the service of social needs: the diversity and manipu-lable aspects of genetic customization are important components to having a unique avatar in a virtual world, and in being a part of a large, diverse community.
1.2 Parametric Motion
In setting out to design a virtual human, this method assumes the problem as a matter of simulation. Physically based modeling techniques have become common tools in computer game engineering, and are useful ingredients for building rich, immersive virtual worlds. Techniques inspired by the biological bases of motion control are proven useful as well [1]. In the work of Hodgins [2], there is detailed attention to human motion, in which biomechanics data are used to drive physically-based models.
Motion Capture systems are very useful for visualizing the motions of humans, and are used widely in computer games. But while motion capture data provide realistic motions, they are not adaptive, like the cerebellum, and cannot be scaled easily to handle any situation, such as complex terrains, collisions with varieties of objects, and variations in user control.
Techniques for "gluing together" separate motion scripts have become sophisticated, but the amount of glue and kinds of glue necessarily grows with each new inclusion of a motion script. In contrast, a parametric, reactive system for handling the motions of humans has potential. While a large amount of work must be done up front to get it right, the payoffs are big.
1.3 The Body: Three Aspects
There are three main aspects to the avatar’s physical body:
Bones
The body geometry is based on a hierarchical tree structure of bones (the skeletal system) comprised of rigid rods connected at joints. At the base of the hierarchical tree is the pelvis node. Each bone possesses a complete orientation matrix with 3D rotation values.
Motion Control
All animation procedures operate on the avatar’s skeleton, and consist of translating and rotating the pelvis (i.e., the avatar), and rotating individual bones. Motion is controlled by a combination of physics, procedural animation, and user input. Facial points constitute non-skeletal geometry, and are also dynamic.
Surface
An outer surface is rendered around the bones, to visualize skin and clothes. A com-bination of rendering techniques have been prototyped - a few are discussed below.
There are of course many more components than these. It is important to note that the order of these components is significant, and critical to the notion of designing a human "from the inside out". In many human modeling systems, the outer surface is the first thing that is modeled, using a 3D modeling application. This is useful for artistically specifying a desired shape, but causes difficulty when the animation sys-tem requires a skeleton. In a technique described by Thalmann [5], after designing the polygonal surface of a human model, the bones had to be "tediously" placed within the confines of the vertices.
The approach developed here makes a trade-off: by creating a completely code-based model, it is not possible to design the outer shape of the avatar with the ease that advanced modeling systems provide. However, it is possible to design in "parameter space", in such a way that the geometry and the motion are determined on a more fundamental level. By working hard to set up intuitive interfaces, and carefully-chosen parameters, what is provided is a rich palette for creative control.
2 Physics
