(Animated Artificial Life)
3.5 Rendering
The invention called image remains a standard vessel for perceivable stuff which is measured as being either "realistic," "not realistic," or something in-between. This standard for which the adjective realism can be applied, has, in my opinion, obscured developers of new computergraphic media (like artificial life and computer games), in the pursuit of virtual realities. A different, and very potent, modality of realism can be rendered.
Some visually-oriented readers may never reach this section because throughout the paper there is a lack of rich visual illustration: the creatures are almost all stick figures. That is because the domain I have been exploring doesn't require any more than skeletal expressions of articulated form and motion. If these illustrations could be set to motion, however, one would immediately see where the depth lies: not simply in the x, y, or z dimensions, but in the four dimensions of space and time. One cannot see physics if there is no movement. Also, a realistic, deep physics can make up for a lack of visual detail. When this living motion is eventually clothed, the dynamism can still remain within, as determined by the physical and biological laws employed.
Part of the methodology here is to take advantage of something that the eye-brain system is good at: detecting movement generated by a living thing. Since it requires high frame rates for viewers to resolve physically-based motion, sacrifices need to be made to achieve this, due to available computer speeds. I have chosen not to spend much computational energy on heavyweight texturemaps, lighting models, and surface cosmetics, when more computation can be spent on deep physics, for animation speed (to show off the effects of subtle movements that have evolved). No pasting 2D images of Disneyesque eyes and ears on the creatures. Instead, as the simulation itself deepens (for instance, if light sensors or vibration sensors are able to evolve on arbitrary parts of bodies, as determined by evolution) primitive graphical elements visualizing these phenotypic features will be rendered. They may be recognizably eye and ear-like, or they may not. The important thing is to visualize what is there, rather than what is not. By not being obscured by cosmetics, the essential dynamics shows through.
3.5.1. Techniques Used
While I have chosen to compromise surface rendering for the sake of more direct visual expressions and faster animation, there are a few techniques worth mentioning, used to visualize the creatures.
For all the articulated stick figures, black 3D lines are drawn to represent body parts. Line occlusions are not dealt with: since the lines are all the same color, this is not a consideration. The interesting (and important) 3D visualization comes in the drawing a shadow. This is the most salient graphical element which gives the viewer a sense of the 3D geometry. The shadow is simply a "flattened-out" copy of the collection of body lines (without the vertical component and translated to the ground plane), projecting the figure vertically onto the ground. The shadow is drawn with a thicker linewidth, and in a shade slightly darker than the ground color. The shadow is drawn first, followed by the body part lines (painter's algorithm). In some simulations, a series of shadow objects are drawn in increasingly darker colors and increasingly thinner widths, to create a composite shadow object which appears to have blurry boundaries.
The progression in designing increasingly complex bodies is accompanied by a similar progression in rendering. When line-segment-based body parts were variably fattened, the thicknesses of the lines were likewise increased, such that they appeared like rectangles of varying widths, oriented along the lengths of the body parts. The swimbots in Gene Pool are 2D examples of this technique.
Fig. 16. Representing body parts with spheres and cone sections.
In more advanced 3D bodies to be used in future simulations, each joint connecting body parts possesses a unique radius. The shape of a body part is determined by the radii of the connecting joints at either end of the part. Each body part is modeled as a portion of a cone connecting two spheres of different radii, as shown in figure 16. To render these parts, polygons oriented towards the viewpoint are drawn. Their shapes roughly correspond to the projections of the solid parts onto the viewplane. Since spheres and connecting cones are able to be expressed in parametric form, it is not difficult to generate 2D polygons representing their projections. In this illustration, spheres are shown as disks of a darker color for clarification.
With this technique, one polygon per body part is drawn as a silhouette, along with an associated joint-sphere silhouette, instead of many polygons showing a faceted surface. This enables faster animation speeds. Custom shading techniques are currently being worked out at this time.
next section:
Discussion
Discussion
