(Animated Artificial Life)
3.2 Background
The spirit of animated artificial life existed before computers, in the form of wind-up toys, mechanical automatons, zoetropes, and Rube-Goldbergian carnival attractions. These physical artifacts are, by necessity, designed in a mostly top-down manner. Robotics research has begun to incorporate learning and evolutionary computation in the control of physical machinery, so that behaviors can adapt to complex unpredictable environments(bottom up design. Architectures that exhibit emergent functionality [Maes, 1990], through intensive interaction with a dynamic environment, demonstrate this design approach. It can be argued that if any form of artificial life is to be considered "alive," its environment must be real, not virtual. It must be physical and truly situated in the world, no smoke and mirrors.
Though less able to pass the "is it alive" test, computer simulations occurring in virtual spaces of computer memory enable greater artistic flexibility, and supply a context for basic research in adaptive behavior. A number of physically-based locomotion systems have been designed for animation utilizing forward dynamics, rather than kinematic methods, for achieving realistic animal motion [McKenna, Zeltzer, 1990] [Raibert, 1991]. Early examples of using artificial life principles in computer animation include Boids [Reynolds, 1987], in which collective behaviors (flocking, herding, etc.) emerge from many interacting agents. [Badler, et. al] describes physically-based modeling techniques and virtual motor control systems inspired by real animals used to automate many of the subtle, hard-to-design nuances of animal motion. In task-level animation [Zeltzer, 1991] and the space-time constraints paradigm [Witkin and Kass, 1988] these techniques allow an animator to direct a character on a higher level.
3.2.1. Genetic Algorithms in Animation
Models of Creation are made implicit every time an artist or engineer designs a body of work and establishes a methodology for its creation. Darwinian models are different (though not absent of a creator). In Darwinian models, important innovation is relegated to the artifact. Invention happens after initial design. Surprises often result. Genetic algorithms [Holland, 1975][Goldberg, 1989] are designed to take advantage of lucky mutations. They are serendipity engines. Genetic algorithms have also been shown to enhance the non-linear design subprocesses of "explore, evaluate, and refine," as described by [O'Reilly and Ramachandran, 1998], for architectural design. More relevant is the notion that genetic algorithms may be good candidates as assistants for prototyping in the design of animated motion [Ventrella, 94a].
The genetic algorithm has been used for evolving goal-directed motion in physically-based animated figures, including a technique for evolving stimulus-response mechanisms for locomotion [Ngo and Marks, 1993]. A holistic model of fish locomotion with perception, learning, and group behaviors, which generates beautifully realistic animations, was developed by Terzopoulos, Tu, and Grzeszczuk [Terzopoulis, 1994]. The virtual creatures of Sims [1994 a,b] are the most celebrated examples from the growing zoology of artificial life entities entering the scene. Sims has developed a comprehensive model for defining a creature genetically, using the genetic programming paradigm [Koza, 1992]. Sims's virtual creatures incorporate deep physics and 3D shaded modeling, and are an excellent application for the Connection Machine, a massively parallel technology, for representing a massively parallel phenomenon(evolution).
next section:
3.3 A Collection of Artificial Life Experiments
3.3 A Collection of Artificial Life Experiments
