ScholarWorks@성균관대학교

초록

Increased information overload and cognitive fatigue have made it increasingly difficult for individuals to sustain attention and regulate emotions in everyday life. While mindfulness practice offers a proven approach to restoring attentional focus and emotional balance, its effectiveness can be undermined by frequent mind wandering. In this paper, we thus introduce a real-time robotic mindfulness intervention system that integrates a multi-biometric mind-wandering detector with the social robot Pepper. The system continuously monitors user state by analyzing biometric indicators and provides timely, context-aware verbal and non-verbal interventions generated via a GPT-based model. Our system estimates the mind-wandering (MW) level in real-time, and if the MW level exceeds a threshold, Pepper generates personalized feedback using off-the-shelf ChatGPT. In our experimental study with 45 participants, the robot-assisted group achieved higher Mindful Attention Awareness Scale (MAAS) scores and lower MW levels than both the no-intervention and audio-only groups. MAAS and MW showed a strong negative correlation (r = -0.78, p <.001). These results demonstrate that the robot's embodied presence and multimodal feedback effectively support attention recovery and enhance mindfulness engagement.

키워드