Creating a Cognitive Agent in a Virtual World: Planning, Navigation, and Natural Language Generation

Creating believable virtual humans for use in interactive video games and other computer graphics applications is a serious challenge. Much research has focused on how to create human models that exhibit realistic appearance and movement. This dissertation investigates how to create virtual humans that act like real people. In particular, we develop human agents that make plans, navigate through complex environments, and communicate with one another. Despite their autonomous behavior, our agents can be tightly controlled by content designers who wish to script their virtual world behavior. Many virtual environments, particularly those used in interactive games, have tight restrictions on memory and frame rate, and we show how judicious offline computation can yield significant runtime performance gains. We demonstrate a virtual world with agents that can plan, navigate, and communicate in English.