Intelligent agents or non-player characters (NPCs) in human interaction scenarios are required to react in ways that are consistent to the training scenario. In many cases these NPCs are controlled by a simple state machine and scripts. The use of these techniques leads to responses that are of poor quality and machine driven. In our work we aim to design a model for incorporating emotional enhancements (concentrating on negative primary emotions such as anger and sadness and also looking at secondary emotions such as anxiety and fear) into intelligent agents. The model will allow NCPs in virtual training applications to simulate human behaviour in crisis management. In this paper, we review current research on models of emotions for intelligent agents in crisis simulations. We identify and discuss key challenges in the area, the psychological theories and contexts in which the models are used and promising research directions.
展开▼