Johnson, Michael Boyle. “WAVESworld, a Testbed for Constructing 3D Semi-Autonomous Animated Characters.” Thesis, Massachusetts Institute of Technology, 1995. https://dspace.mit.edu/handle/1721.1/29096.

Abstract

This dissertation describes a testbed for experimenting with the issues surrounding designing, developing, debugging, and delivering three-dimensional, semi-autonomous animated characters. This testbed sits atop an object-oriented framework for constructing and animating models. This framework facilitates iterative construction of the parts and props that comprise a character. It also provides facilities for writing and wiring together agents, which are processes that measure and manipulate models over time to produce a character's behavior. This framework encourages and facilitates encapsulation and reuse at many levels, which benefits collaborative character construction. This testbed can be used to compose three-dimensional, photorealistic animatable characters, where characters are composed of variously interconnected agents and a model, where a model is a set of objects encapsulating shape, shading and state information over time. It's possible to quickly build reusable, composable, blendable behaviors, where a behavior is the result of some set of processes operating on a model over time. One especially useful result of using this framework to develop media is its facility in acting as a very rich and compact storage medium for photorealistic scenes. This storage representation builds directly atop the RenderMan Interface, an industry standard for describing photorealistic scenes. In our object-oriented representation, though, since we maintain some level of continuity in our scenes over time, such scenes can have 3D models that change over time, where different parts of the models in the scene can be changing at different rates. Especially interesting is that these scenes need only a very modest run-time system for playback at arbitrary frame rates with respect to the scene time. Assuming the underlying components of the scene were sampled appropriately, the scene can be played back at arbitrary spatial and temporal frequency. In other words, the scene can be treated as continuous media. With appropriate sampling, the representation is not lossy. For a large class of scenes, this allows orders of magnitude of compression of the amount of data which need be stored or transmitted. This framework is extensible, so compound components of a scene can be encapsulated and efficiently stored and transmitted.