Embodied Empathy: A Multimodal AR and LLM-Powered System for Self-Attachment Psychotherapy with Self-Initiated Humour

This study explores how to bridge the gap in mental health support by creating a more immersive, embodied digital therapy experience. The researchers developed a mobile application that operationalizes the Self-Attachment Technique (SAT) and the Self-Initiated Humour Protocol (SIHP). Unlike traditional text-only chatbots, this system uses a multimodal approach: it features customizable 3D childhood avatars, augmented reality (AR) to place these avatars in the user's physical space, and an LLM-driven virtual therapist that performs automated emotion mirroring. The researchers conducted an eight-day user study with 16 non-clinical participants to evaluate the system's feasibility, user engagement, and the impact of these specific modalities on the therapeutic experience.

The study reveals that personalization is central to therapeutic success. Users reported that creating a childhood avatar and viewing it in AR significantly strengthened their emotional bond with their "inner child." The integration of an LLM allowed for more flexible, natural dialogue compared to rigid, rule-based chatbots, leading participants to prefer a proactive "facilitator" persona over a reactive one. Furthermore, the researchers found that voice output (Text-to-Speech) was a powerful driver of perceived empathy, whereas voice input (Speech-to-Text) provided less utility. While emotion mirroring—where the avatar reflects the user's detected emotional state—was highly effective for engagement, the study highlights that technical precision is vital; misclassifications or overly intense animations can disrupt the therapeutic flow and break the user's sense of immersion.

As digital mental health tools move away from simple, scripted chatbots, this research provides a blueprint for "embodied empathy" in AI-driven therapy. By demonstrating that users seek proactive, emotionally resonant interactions, the study highlights the importance of combining generative AI with non-verbal cues like spatial presence and facial animation. These findings offer actionable design principles for developers looking to move beyond text-based interventions toward more holistic, supportive digital environments that can better sustain long-term therapeutic engagement.