An LLM-Powered Robotic Coach to Support University Students' Mental Well-Being

Abstract

University students often experience high levels of stress and anxiety, which may adversely impact their well-being and academic performance. However, many do not access traditional mental health support due to barriers such as stigma and limited resources. Recent advances in social robotics offer new opportunities to expand access to mental health support. This study presents an autonomous robotic mental well-being coach powered by a Large Language Model (LLM) with Retrieval-Augmented Generation (RAG) to help university students manage stress and anxiety through interactive, evidence-based interventions. Results from a user study with 27 students across three weekly sessions showed a significant reduction in STAI-6 anxiety measures, along with improvements in self-reported moods in each session. Additionally, participants' readiness to change behaviour increased significantly after the study. User experience ratings were positive, and perceived competence, personalization, and naturalness of the interaction increased significantly across sessions. The findings suggest that the LLM-powered robotic coach has the potential to enable long-term, responsive interactions to support mental well-being. Overall, the results highlight the promise of socially interactive robots incorporating LLM-based dialogue and RAG to support university students' mental well-being and stress management.