We believe that the best way of achieving our goals in the long run is to pursue the challenging approach of constructing artificial animals. The properties and internal control mechanisms of artificial animals should be qualitatively similar to those of their natural counterparts.
There are several properties common to all animals. The most salient one is that all animals are autonomous: They have physical bodies that are actuated by muscles, enabling them to locomote; they have eyes and other sensors to actively perceive their environment; they have brains which interpret their perceptions and govern their actions. Indeed, autonomy is the consequence of possessing a brain capable of controlling perception and action in a physical body. The behavior of an animal is a consequence of its autonomous interaction with its environment to satisfy its survival needs. No external control is required for animals to cope with their dynamic habitats, yet their autonomy does not prevent higher animals from being influenced or directed (consider trained circus animals, or human actors).
Artificial animals should be self-animating actors that emulate the autonomy of real animals. In our artificial life approach to computer animation, we build animal-like autonomy into our graphics models; not only to minimize the amount of animator intervention while supporting modifiability and interactability, but also to obtain behavioral realism. Our main concern is how to model the locomotion, perception and behavior capabilities of animals and how to integrate these models effectively within a life-like artificial animal. Our research in this respect shares common goals with ALife research, where artificial animals are often referred to as ``animats''. Previous animats have been models of simple creatures and the behaviors simulated usually pertain to genetic reproduction and ``natural selection'' [Langton1987, Varela and Bourgine1991]. We attempt to develop artificial life patterned after animals that are more evolved and have a significantly broader range of behavior.
In the following sections, we will identify the essential properties and mechanisms that allow real animals to locomote effectively, to perceive, and hence to behave autonomously. From this we derive design methodologies for achieving realistic animated locomotion and behavior with minimal animator intervention.
|Xiaoyuan Tu||January 1996|