Improving Interaction With Non-Player Characters Through Physiological Data

This work-in-progress paper proposes using physiological data to estimate players' emotional states so that non-player characters (NPCs) in video games can respond to how players feel rather than relying solely on their prior choices. The authors argue that conventional NPCs lack access to the emotional cues humans use in conversation, such as speech patterns and gestures, and that physiological signals offer a channel humans do not naturally possess. The central thesis is that feeding such data into NPC dialogue can improve the quality of player-character interaction, with rapport serving as the evaluative measure.

The proposed system is built as a prototype in Unity, connected to an Arduino device using the eHealth toolkit. Galvanic skin response (GSR) is used as an estimate of arousal and electromyography (EMG) as an estimate of valence, with the resulting values mapped onto the two-dimensional Russell arousal-valence plane. Players take the role of a resident of a post-apocalyptic setting, navigating a short narrative experience with morally charged decisions. Baseline physiological measures are recorded during calm phases before key choices, and NPCs are then presented with four dialogue options and corresponding gestures based on the detected emotional position.

The evaluation design is a between-subjects experiment in which one group's NPCs respond to physiological data while a control group's dialogue depends only on prior choices, with rapport measured through a questionnaire adapted from earlier virtual-agent research. The paper reports no participant counts or results, as the study is presented as an intended demonstration of feasibility rather than a completed evaluation. The authors acknowledge that further work is needed to determine how access to physiological data could reshape NPC interaction design more broadly.