Modelling stress levels in combat NPCs

Affective Artificial Intelligence (AI) has been explored to introduce emotions into non-player characters (NPCs) and enhance believability. However, there is a gap studying NPCs whose responses are affected by their stress level, especially in combat scenarios. Adding emotions such as anger, fear, and shock to danger response systems could make behaviour engaging and less predictable. This paper introduces an affect model based on human acute stress responses to calculate the NPC's stress level. The model is integrated into a designer framework to trigger various actions in combat, and allow customisable emotional responses, including movement and appearance. The framework was evaluated by players who blindly tested two versions of a bespoke game: control version with basic NPCs, and affective version with stress-equipped NPCs. It was also evaluated by designers who used the tool to create their own levels. Questionnaires’ responses demonstrate the efficacy of the stress model via players’ satisfaction scores around immersion and enjoyment where 80% of the participants reported playing more than expected, and 90% declared they would want to play more levels. Designers’ responses show a high system usability score of 76.5. These results also suggest the potential of emotion modelling as customisable game packages.