EmoLoop: An adaptive VR biofeedback system for stress regulation

Virtual reality (VR) and biofeedback offer new opportunities for regulating affective states. This study investigated an adaptive VR-based biofeedback system that modulated participants’ field of view (FOV) based on breathing rate, embedded within a restorative, nature-based VR environment. Participants completed baseline and stress-inducing Stroop tasks, followed by a restorative VR session. Physiological responses (heart rate, electrodermal activity, breathing rate) and subjective stress ratings were recorded. The Stroop reliably induced stress, while the restorative VR environment reduced physiological arousal. While between-group differences did not reach statistical significance, the adaptive biofeedback group exhibited greater physiological reductions on average during VR exposure, alongside lower usability and higher anxiety. These findings highlight a central HCI challenge: adaptive physiological feedback can support stress regulation but may undermine usability if mechanisms are overly subtle or opaque. A methodological contribution of this work is error-triggered retraining strategies for stress classification models to address temporal drift in physiological signals. Error-triggered retraining analysis using live VR data demonstrated effective adaptation to physiological drift, establishing the foundation for online implementation during VR biofeedback. Overall, this study underscores both the promise and the challenges of integrating adaptive biofeedback and machine learning into VR. Future VR interventions should balance immersion with transparency and leverage adaptive ML to ensure personalization, usability, and sustained effectiveness in stress regulation.