We have demonstrated the soundness of the design of the artificial fish's behavior system. Although arbitrarily more complex designs could be conceived and implemented with additional effort, our current design is sufficient for its purpose. That is to say, we have established a computational model of animal behavior, that can be effectively used for computer animation and, can serve as a touchstone for developing ALife theory.
It should be realized that, although we have validated the behavior
control scheme (see Fig. 
) via an artificial
fish agent, a scheme such as ours is by no means restricted to the
control of fish behavior. Rather, the aspects of behavior captured
in our model, i.e. perception, internal motivation, external stimuli
and how they interact to produce intentions and to select actions, are
generally applicable to all animals. This suggests many potential
extensions of our behavior control scheme to the modeling of other
creatures. This is best demonstrated by the work of Blumberg and
Galyean Blumberg95, where a behavior control paradigm
similar to ours, was used to direct the behavior of a kinematic
graphical dog creature.
On the one hand we realize that a crucial step in our behavior control
scheme is the construction of effective motor controllers. Realistic
visual effects depend on how well the motor controllers can control
the underlying model of the animal, to produce realistic locomotion.
On the other hand, abstracting motor controllers is a necessary step
towards the modeling of higher level behaviors. Currently, motor
control in complex creatures, such as articulated figures, remains an
unsolved problem. However when the underlying control problem for
such creatures is solved, we believe the design of our behavioral
control scheme can provide the starting point for the modeling of the
behaviors of the animal in question. Thus, as we have stated in
Chapter , the unavailability of motor controllers
for a particular animal, should not imply a lack of generality of the
control scheme we have proposed.
For animals that have complex cognitive abilities, additional control layers may be added on top of the current control scheme. The current control scheme can serve as a reactive behavioral control layer for the more basic behaviors of the animal, such as collision avoidance, foraging and mating. Moreover, many higher animals possess actuators (e.g. limbs) that are beyond the necessity of locomotion, for example, the arms and hands of primates and humans. These actuators may provide a large number of motor skills relevant to manipulation which are mutually exclusive to those used for locomotion. Consequently, the chances of being able to fulfill more than one intention at a time are greatly increased. This suggests that it may be advantageous to permit multiple intentions to be active simultaneously. For instance, a synthetic human actor should be able to fulfill both the intention of searching for someone in a crowded room and the intention of not spilling a cup of coffee.
Let us consider drawing a spectrum of various behavior control schemes, with the strictly winner-takes-all and the free-flow selection processes occupying the two extremes. It is reasonable to assume that a desirable control scheme would be a compromise between the extremes. In particular, such a scheme should allow multiple intentions to be active simultaneously while still maintaining reasonable simulation speed. This can be done by restricting the allowable number of active intentions. In addition, in order to generate behavior-oriented, harmonic movements, a more elaborate fusion mechanism at the motor level would be necessary to collect and combine the resulting motor commands.
| Xiaoyuan Tu | January 1996 |