Physiological correlates of stress induced by virtual humans in a naturalistic virtual reality scenario

The ability to induce various levels of stress is central to understanding and modeling stress related responses. Virtual reality holds much promise for eliciting stress in a real-life yet controllable environment.; this requires compelling, virtual social environments with interactive virtual humans. In this study we use semi-automated virtual humans to simulate a realistic stressful event from everyday life - a job interview. Furthermore, using this environment we derive a dynamic physiological index of ongoing stress. Twenty-seven participants underwent an interactive VR interview while their electrocardiogram, skin conductance, respiration, and head motion were recorded. Our goal was to determine whether these signals were predictive of participants' momentary stress level, which were reported in a post experiment session. For that aim, we extracted 21 physiological features and tested their correlation with post-hoc reported stress per event. Our analysis indicated that the most informative feature was the magnitude of heart rate deceleration, which positively correlated with the reported stress level (r=0.61, p=0.008, averaged over all participants). We discuss the implications of our findings for applications based on the combination of interactive virtual humans in VR with real time physiology.