In this dissertation we presented the results of research spanning the fields of computer graphics and artificial life. With regard to computer graphics, we have proposed, implemented, and demonstrated an animation framework that enables the creation of realistic animations of certain natural ecosystems with minimal intervention from the animator. In our approach, the virtual creatures are self-animating, as are real animals and humans. Thus, the strength of our approach to animation lies in the fact that it turns the role of the animator from that of a graphical model puppeteer to that of an virtual nature cinematographer, a job not unlike that done by nature cinematographers of the National Geographic Society. Our artificial life approach has advanced the state-of-the-art of computer animation, as evidenced by the unprecedented complexity and realism of the behavioral animations that we have been able to achieve without keyframing. With regard to artificial life, we have successfully modeled complete animals of nontrivial complexity. The convincing simulation results validate our computational models, which capture the essential features of all biological animals--biomechanics, locomotion, perception, and behavior.
In particular, we have developed a physics-based, virtual marine world inhabited by life-like artificial life forms that emulate the appearance, motion, and behavior of fishes in their natural habitats. Each artificial fish is an autonomous agent with a deformable body actuated by internal muscles, with eyes, and with a brain that includes behavior, perception and motor centers. Through controlled muscle actions, artificial fishes are able to swim through simulated water in accordance with simplified hydrodynamics. Their functional fins enable them to locomote, maintain balance, and maneuver in the water. Though rudimentary compared to real animals, their brains are nonetheless able to capture many of the most important characteristics of animal behavior and carry out perceptually guided motor tasks. In accordance with their perception of the virtual world and their internal desires, their brains arbitrate a repertoire of behaviors and subsequently select appropriate actions. The piscine behaviors the fishes exhibit include collision avoidance, foraging, preying, fleeing, schooling, and mating. The easy extensibility of our approach to the modeling of additional behaviors is suggested most evidently by the complex patterns of mating behavior that we have been able to emulate in artificial fishes.
With regard to the implementation, we have pursued a bottom-up, compositional approach in which we started by modeling the basic physics of the animal and its environment. Upon the simulated physics substrate, we effectively modeled the animal's means of locomotion. This in turn positioned us to model the animal's perceptual awareness of its world, its motivation, and last but not least, its behavior. The compositional nature of our approach to synthesizing artificial fishes was proven crucial to achieving realism. Partial solutions that do not adequately model physics, locomotion, perception, motivation and behavior, and do not combine these models intimately within the agent will not produce convincing results.
In addition to realism, computational efficiency has been one of the most important design criteria of our implementation. The fidelity of our models was carefully chosen to achieve satisfactory computational efficiency. We have strived successfully to achieve visually convincing animations with low computational cost. Using a Silicon Graphics R4400 Indigo Extreme workstation, a simulation of ten artificial fishes, fifteen food particles and four static obstacles can run at about 4 frames/sec, including wireframe rendering time and user interface running time. With a Reality Engine graphics board on the Silicon Graphics ONYX workstation, the same simulation with hardware-supported GL, fully texture mapped surface rendering runs at about 3 frames/sec. Considering the complexity of the animations, the simulation speed we have been able to achieve is more than satisfactory. The main tradeoffs that we have made in order to gain high simulation speed is the simplification of the virtual environment and the algorithms used for simulating the fish's perceptual capability.
|Xiaoyuan Tu||January 1996|