Behavioral Animation - Computer Animation: Algorithms and Techniques - page 339

At the highest levels of abstraction, the animator takes on the role of a director. At his command are
intelligent characters who know how to get the job done, they only need to be told what to do. It is the
character's job to find a way to do it. The animator ceases to animate and only directs the action in a
very general way. This chapter is about behavior and how it is modeled in computer animation appli-
cations. The implied objective is that the behavior be realistic, or at least believable, so that character
appears to be autonomous .
There are two general aspects of behavioral animation: cognitive modeling and aggregate behavior.
Cognitive modeling can range from a simple cause-and-effect-based system to an attempt to model
psychological changes in an agent's mental make-up under environmental influences. This aspect
of computer animation is in the realm of artificial intelligence (AI). Aggregate behavior refers to
how the individuals making up a group are modeled and how their behavior contributes to the quality
of the group's motion as a whole.

Cognitive modeling
In general, the task of modeling intelligence and cognition is the problem domain of AI. Aspects of
behavior modeling are addressed in computer animation because of the spatial aspects and the imagery
produced, and not by focusing on the accuracy of the mental processes being modeled. At all levels of
cognitive modeling, the character senses the environment, processes the sensory information possibly
with respect to his/her internal state and traits, and then responds with actions that interact with the
synthetic environment and possibly with changes to the internal state. At the basic level, a simple rea-
soning mechanism, such as use of a decision tree or case-based reasoning, suffices. More sophisticated
approaches might model personality and mood of the character. For example, a rule-based approach is
often used and can prove effective, although there are problems of scale. A common approach is to use
a database of simple actions (e.g., simple SIMS games) as a library of behaviors and have appropriate
behavior kick in when conditions are satisfied. Simple rules of the form (

) can
determine activity with the implication of executing the activity when the condition is met: if
(

cond , behavior

< > < >

cond_i is TRUE) then do activity_i. Of course, the condition can be arbitrarily complex as can

< >

the activity.
When more than one precondition is met, some arbitration strategy must be in place. A simple
method is to say that the first condition satisfied is activated, so order of rules sets the priority of rules.
For a simple example, consider the behaviors of walking, running, standing, and turning. One of the
major issues with this approach is transitioning from one activity to another. The simple solution is to
have a neutral behavior to start and end each behavior, for example, standing.

stand ! walk ! turn ! run ! stand

A more sophisticated approach is to blend from one activity to another with either precomputed
transitions or transitions computed on the fly. Adding more activities requires that more rules are added
and more transitions are needed. Meta-rules can be added to reason about the rules such as when to
apply, which are compatible, and so forth.
Higher levels of cognitive modeling often refer to themselves as modeling intelligence and
individuality. Instead of directly reacting to environmental conditions, a cognitive structure is created
that can more realistically sense the environment, alter the mental state, and reason about conditions
using internal representations while filtering everything through mental states.