Modeling Believable Virtual Characters with Evolutionary Fuzzy Cognitive Maps in Interactive Storytelling

To generate believable virtual characters in real-time is a key issue to improve users’ engaging experience in the interactive storytelling. In real life, the emotions and behaviors of characters evolve inductively according to the mutual causal relationships, with some stochastic variations. This is not addressed well in the virtual environment with conventional models, e.g. rule-based expert system, Fuzzy Cognitive Map and so on. In this paper, we use a computational model, namely Evolutionary Fuzzy Cognitive Map (E-FCM), to model the attributes of characters (such as emotions and behaviors) as concepts with the dynamic causal relationships among them. As an extension to FCM, E-FCM models not only the fuzzy causal relationships among the variables, but also the stochastic causal relationships, and asynchronous activity update of the concepts, so that the variables evolve in a dynamic manner with their respective evolving time schedules. As a result, the characters are presented with more realistic and dynamic emotions and behaviors, which enhances the user experience at last. Introduction Interactive storytelling in the virtual environment has gained a lot of interests from both industry and academy research as a new genre of interactive entertainment. In Aristotle’s Poetics (Aristotle 300BC), character is the most important factor other than plot in a story. People can remember a story easily though a vivid character, such as Hamlet in the drama “Hamlet”. Dynamic and realistic virtual characters are very important for users to gain an immersive or engaging experience. Two most important aspects of character modeling includes: emotion modeling and behavior modeling. Because the characters inhabit and present in the virtual environment, the behaviors and emotions of the virtual characters are evolving in real-time. Prior works (Mateas 1997; Mateas & Stern 2003; Cavazza, Charles, & Mead 2002; Magerko & Laird 2003), have studied believable characters with reasonable planning of their behaviors and the modeling of the behaviors or emotions. Compared to conventional storytelling, the character modeling in interactive virtual storytelling become complex, which have the following properties: Copyright c © 2009, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. 1. Complex Causal Relationships: The causal relationships are complex causal related, which include the mutual causal relationships between the characters and environment and the causal relationship between the emotions and behaviors, especially when the story scenario is complex and in large scale and a lot of context, emotion and behavior variables need to be modeled. 2. Dynamic: In order to bridge the experience in virtual world and real world, the emotions and behaviors of characters keep on changing as the story is going on. The characters need to respond rationally to the story changes, and show correctly from the behaviors and emotions. 3. Randomness: The virtual characters do not perform in a deterministic way always. There are some stochastic behaviors of the variables. Currently, there have been a lot of researches done on modeling the dynamic causal relationships among a set of variables, e.g. rule-based expert system, Fuzzy Cognitive Maps etc. Fuzzy Cognitive Map (FCM) (Kosko 1986) by Kosko is an efficient inference engine to model such complex causal relationships. Kosko and Dickerson (Dickerson & Kosko 1994) also make a simple study to model the characters (Fish, Shark etc) in a virtual world. However, as a generic model, FCM is not powerful or robust enough to model a dynamic and evolving virtual world. Based on FCM, numerous extensions are proposed to enhance its capabilites. Miao et. al. (Miao, Liu, & Miao 2001) proposed Dynamic Causal Network to model the concepts quantitatively with time variables. Moreover, the causality between two variables might be probabilistic rather than deterministic, beyond the fuzziness. Song et. al. (Song et al. 2006) proposed probabilistic events to model the uncertain concepts. In order to describe the general rulebased (AND/OR) inference, rule-based FCM is also proposed (Carvalho & Tom 2000). In addition, Evolutionary Multilayered Fuzzy Cognitive Maps (ECNFCM) (Mateou, Andreou, & Stylianou 2006) is also proposed as an inference tool for a real-time system with evolutionary strategy. However, the models above are mostly used as inference engines rather than real-time simulation modeling. Currently, there is not a solution that combines all the above features of the FCM extensions in order to model real-time character variables, e.g. behaviors and emotions.