Goal-based personalities and social behaviors in believable agents

Agents are considered "believable" when viewed by an audience as endowed with behaviors, attitudes, and emotions typical of different personalities. Our work is aimed at realizing believable agents that perform helping behaviors influenced by their personalities, which we represent as different clusters of prioritized goals and preferences over plans for achieving goals. The article describes how such a model of personality is implemented in planning with the PRODIGY system and in execution with the RAP system. Both systems are integrated in a plan-based architecture where behaviors characteristic ofdifferent "helping personality types" are automatically designed and executed in a virtual world. The article also shows examples of the kinds of plan produced by PRODIGY for different personalities and contexts and how such plans are executed by RAP when a helping character interacts with a user in a virtual world.