People's day-to-day lives are infused with emotion. A wealth of empirical research has revealed a complex interplay between emotions, cognition and behavior. Emotional states can impact our beliefs, decision-making, action-selection, memory, attention, as well as physical attributes such as muscle tone, respiration, and heart rate. Face-to-face communication is likewise infused with emotion. People intentionally or unintentionally convey emotion through word choice, intonation, gesture, body language and facial expression, and observers readily detect these signals and use them to resolve ambiguities, infer intent, predict future behavior and assess the trustworthiness of the speaker and the information being conveyed. Developers of lifelike animated agents must contend with the fact that human observers will expect appropriate emotional responses from their lifelike characters. People will attribute emotion to these characters whether they are modeled or not, and will likely be disturbed by discrepancies from these expectations, potentially compromising the success of the intended application.
In this talk I will briefly review some of the key findings in the emotion psychology literature as they relate to lifelike agent design. I will then discuss my own work, in collaboration with Stacy Marsella, on developing computational models of emotions for lifelike animated agents. In particular, I will present a general computational framework of emotional appraisal and coping as a means of promoting behavioral consistency in human-like autonomous agents. Building on the psychological theories of Craig Smith and Richard Lazarus, we illustrate how their perspective on emotion facilitates the integration of such disparate reasoning modules as perception, planning, and dialogue management into a coherent appraisal of the agent's relationship to its physical and social environment, and illustrate how these appraisals can inform the generation of external behavior. These points are realized in an implemented system that has been applied to a significant real-world problem, the Mission Rehearsal Exercise virtual training environment.