At a higher level of abstraction, we are interested in animating the behaviors of animals with an intermediate level of behavioral complexity--somewhere in between the complexity of invertebrates and of primates such as humans. In this regard, our research is an instance of behavioral animation, where the motor actions of characters are controlled by algorithms based on computational models of behavior [Reynolds1982, Reynolds1987]. Consequently, an animator is able to specify motions at a higher level, i.e. the behavior level, as opposed to specifying motion at the locomotion level as is done in physics-based modeling. The animator is therefore concerned with the modeling of individual behaviors. Behavioral animation approaches have been proposed to cope with the complexity of animating anthropomorphic figures [Zeltzer1982], animating the synchronized motions of flocks, schools, or herds [Reynolds1987] and interactive animation control [Wilhelms1990].
The seminal work in behavioral animation is that of Reynolds Reynolds87. Creating vivid animations of flocks of birds or schools of fish using conventional keyframing would require a tremendous amount of effort from an animator. This is because, for example, while the overall motion of birds in a flock is highly coordinated, individual birds have distinct trajectories. In keyframing, the animator would have to script each bird's motion carefully in each keyframe. By contrast, Reynolds proposed a computational model of aggregate behavior. In his approach, each animated character, called a ``boid'', is an independent actor navigating its environment. Each boid has three simple behaviors: separation, alignment and cohesion. A boid decides which behavior to engage in at any given time based on its perception of the local environmental conditions, primarily the location of neighboring boids. The motions of the individual boids are not scripted; rather, the organized flock is an emergent property of the autonomous interactions between individual boid behaviors.
Although Reynolds' model successfully achieves behavioral realism, it pays little attention to locomotion realism. Because a kinematic model is used to control each boid's locomotion, the resulting motion of individual boids can be visually unrealistic and may not scale well to more elaborate motion. Additionally, the behavioral model is limited by its simplicity. In particular, since the goal is to animate flocking behavior, each boid is capable by design only of behaviors that are useful to flocking. Our artificial fishes are ``self-animating'' in the sense of Reynolds' work, but unlike his procedural boid actors, they are more elaborate physical models that also have much broader and more complex behavior repertoires.
|Xiaoyuan Tu||January 1996|