Abstract

This thesis seeks to address the problem of building things with behavior and character. By things we mean autonomous animated creatures or intelligent physical devices. By behavior we mean that they display the rich level of behavior found in animals. By character we mean that the viewer should “know” what they are “feeling” and what they are likely to do next. We identify five key problems associated with building these kinds of creatures: Relevance (i.e. “do the right things”), Persistence (i.e. “show the right amount of persistence), Adaptation (i.e. “learn new strategies to satisfy goals”), Intentionality and Motivational State (i.e. “convey intentionality and motivational state in ways we intuitively understand”), and External Control (i.e. “allow external entity to provide real time control at multiple levels of abstraction”). We argue that these problems can be addressed in a computational framework for autonomous animated creatures by combining key ideas from Ethology and Classical Animation. We have developed a toolkit based on these ideas and used it to build “Silas”, a virtual animated dog, as well as other creatures. We believe this work makes a significant contribution toward understanding how to build characters for interactive environments, be they virtual companions in immersive story-telling environments, or interesting and adaptive opponents in interactive games, or as the basis for “smart avatars” in web-based worlds. While our domain is animated autonomous agents, the lessons from this work are applicable to the more general problem of autonomous multi-goal, adaptive intelligent systems. Based on results generated from our implementation of Silas and other autonomous creatures we show how this architecture addresses our five key problems.