Realizing believable agents: an integration of the ‘author‐based’ and the ‘model‐based’ approaches

Article type: Research Article

Authors: Rizzo, Paola

Affiliations: IP‐CNR – Institute of Psychology, National Research Council of Italy, Viale Marx 15, I‐00137 Rome, Italy E‐mail: paola@ip.rm.cnr.it

Note: [] The author is currently supported by a scholarship from the Committee 12 (Information Science and Technology) of the National Research Council of Italy. She is extremely grateful to Amedeo Cesta and Maria Miceli for their indispensable suggestions and comments, and to three anonymous reviewers for their very relevant and helpful remarks.

Abstract: A major research problem regarding believable agents is how to develop and execute their behavioral libraries. This work identifies two different approaches: the ‘author‐based’ one, depending on the designer's ability to hand‐code each agent's behavioral features, and the ‘model‐based’ one, grounded on a model which, starting from a set of primitives provided by the designer, automatically generates the agents' typical actions. This paper proposes to integrate the two methods by means of a two‐phase/two‐step strategy, that partially relieves the designer of the burden of hand‐coding all the behavioral libraries, while still allowing a good control over the characters' performance, and enabling the runtime creation and storage of new behaviors. A case study concretely illustrates how such strategy is implemented by means of a hybrid planning architecture, coupled with a goal‐based model of personality, in order to realize characters that interact with the user according to their personalities.

Journal: AI Communications, vol. 13, no. 3, pp. 145-168, 2000