Behavioral Animation - Computer Animation: Algorithms and Techniques - page 351

The classic example of competing goals is that of a prey-predator model in which one set of char-
acters is trying to eat the other. Very simple prey-predator models can create interesting animation if
the rules of behavior are carefully chosen. Of course, if the two character groups have the same capa-
bilities, then the resulting interaction may result in some pretty boring interaction. In general, interest-
ing behavior might result when there are two or more classes of objects that have different qualities of
motion so that, depending on the geometric arrangement of the characters and the environment, a mem-
ber of one class or the other may win in a given situation. For example, one set of agents could be slower
but have better acceleration, a better turning radius, a better turning velocity, and better reasoning capa-
bilities. There is no absolute model to use when modeling character behavior because balancing the
models is dependent on the desired motion.
At a minimum, the reasoning component for prey-predator models could be based on simple attrac-
tion/repulsion. One character class is attracted to the other (e.g., trying to eat it) while the other class is
repulsed by the first (e.g., trying to avoid being eaten). While a simple force field model can produce
interesting motion, incorporating some predictive reasoning ability in one or both character classes can
make situational behavior more realistic. For example, the ability to compute the trajectory of a char-
acter from its current position and speed can produce more interesting and realistic behavior.
11.2 Knowledge of the environment
Behavioral animation is all about cognitive interaction with the environment combined with the lim-
itations imposed by simulation of physical constraints. One of the important issues when modeling
behavior is the information a character has access to about its environment.
At the simplest level, a character has direct access to the environment database. It has perfect and
complete knowledge about its own attributes as well as the attributes of all the other objects in the
environment. A character can “see” behind walls and know about changes to the environment well-
removed from its locale. While this might be a handy computational shortcut, such knowledge can
produce behaviors that are unrealistic. Characters reacting to events that are obviously hidden from
view can destroy believability in the character.
More realistic behavior can be modeled if the character is enabled with locally acquiring knowledge
about the environment. Such local acquisition is affected by simulating character-centric sensors, most
commonly vision. Modeling vision involves determining what can be viewed from the position and
orientation of the character. Modeling other senses might also be useful in some applications. Touch,
for example, can be modeled by detecting collisions and might be useful for navigating through dark
environments. Modeling sounds might be useful in a forest setting. However, since vision is by far the
most common sense modeled in computer animation, the following discussion will be restricted to
incorporating a character's sight into behavioral animation.
For a character that senses the environment, added realism can be created by providing the capa-
bility of remembering what has been sensed. Simulated memory is a way for senses to be recorded and
accumulated. Current positions, motions, and display attributes (e.g., color) of other objects in the envi-
ronment can be captured and used later when reasoning about the environment.
11.2.1 Vision
Vision systems limit knowledge about the environment by (1) modeling a limited fov and (2) comput-
ing visual occlusions. A simple form of vision only models the field of view and ignores occlusions.
Using the character's position and (head) orientation, an fov can be easily calculated. Any object within