Behavioral Animation - Computer Animation: Algorithms and Techniques - page 354

state. In addition, there is an internal state associated with the object made up of time-varying urges,
desires, and emotions, as well as (usually static) rules of behavior.
It should not be hard to appreciate that modeling behavior can become arbitrarily complex.
Autonomous behavior can be described as various levels of complexity, depending on how many and
what type of cognitive processes aremodeled. Is the objective simply to produce some interesting behav-
ior, or is it to simulate how the actual object would operate in a given environment? Howmuch and what
kinds of control are to be exercised by the user over the autonomous agent? Possible aspects to include in
the simulation are sensors, especially vision and touch; perception; memory; causal knowledge; com-
monsense reasoning; emotions; and predispositions. Many of these issues are more the domain of AI
than of computer graphics. However, besides its obvious role in rendering, computer graphics is also
a relevant domain for which to discuss this work because the objective of the cognitive simulation is
motion control. In addition, spatial reasoning is usually a major component of cognitive modeling
and therefore draws heavily on algorithms associatedwith computer graphics. Applications include mil-
itary simulations (e.g., [ 21 ] ), pedestrian traffic (e.g., [ 37 ] ), and, of course, computer games (e.g., [ 15 ]).
Autonomous behavior is usually associated with animal-like articulated objects. However, it can be
used with any type of object, especially if that object is typically controlled by a reasoning agent. Obvi-
ous examples are cars on a highway, planes in the air, or tanks on a battlefield. Autonomous behavior
can also be imparted to inanimate objects whose behavior can be humanized, such as kites, falling
leaves, or clouds. The current discussion focuses on the fundamentals of modeling human behavior.
Internal state
Internal state is modeled partially by intentions. Intentions take on varied importance depending on the
urges they are meant to satisfy. The instinct to survive is perhaps the strongest urge and, as such, takes
precedence over, say, the desire to scratch an itch. Internal state also includes such things as inhibitions,
identification of areas of interest, and emotional state. These are the internal state variables that are inputs
to the rest of the behavioral model. While the internal state variables may actually represent a continuum
of importance, Blumberg and Galyean [ 1 ] group them into three precedence classes: imperatives, things
that must be done; desires, things that should be done, if they can be accommodated by the reasoning
system; and suggestions, ways to do something should the reasoning systemdecide to do that something.
Levels of behavior
There are various levels at which an object's motion can be modeled. In addition to that discussed by
Blumberg and Galyean [ 1 ] , Zeltzer [ 44 ] and Korein and Badler [ 17 ] discuss the importance of decom-
posing high-level goals into object-specific manipulation of the available degrees of freedom (DOFs)
afforded by the geometry of the object. The levels of behavior are differentiated according to the level
of abstraction at which the motion is conceptualized. They provide a convenient hierarchy in which the
implementation of behaviors can be discussed. The number of levels used is somewhat arbitrary. Those
presented here are used to emphasize the motion control aspects of autonomous behavior as opposed to
the actual articulation of that motion (refer to Figure 11.12 ).
Internal state and knowledge of the external world are inputs to the reasoning unit , which produces
a strategy intended to satisfy some objective. A strategy is meant to define the what that needs to be
done. The planner turns the strategy into a sequence of actions (the how ), which is passed to the move-
ment coordinator . The movement coordinator selects the appropriate motor activities at the appropriate
time. The motor activities control specific DOFs of the object.