Lab member Kayla Matheus defended her dissertation “Embodiment, Interactions, and Autonomy for Social Robots in Mental Health” today.
Abstract
Mental health challenges have increased substantially worldwide in recent years, affecting daily functioning, emotional regulation, and quality of life. Behavioral interventions (e.g., deep breathing, affirmations, exposures) are foundational to many evidencebased treatments for conditions such as anxiety and depression. While these behaviors are effective and simple in principle, practicing them outside of clinical settings remains a persistent challenge due to motivational barriers, cognitive load, and the demands of daily life. Accordingly, there is a growing need for technologies that can support the real-world practice and application of therapeutic interventions.
Social robots offer a promising interface modality for addressing this need. As physically embodied social agents, they can provide motivational cues and situated support in the environments where practicing a behavioral technique is most relevant. However, a number of open questions remain about how robot embodiment, interactions, and autonomy should be shaped for vulnerable populations in daily life. This dissertation accordingly asks: How can we design, develop, and validate social robots that effectively support mental health behaviors in real-world contexts?
To address this question, we target anxiety as a clinical domain and deep breathing as a foundational therapeutic behavior that supports both acute distress and long-term nervous system regulation. Our work centers on the human-centered design and system development of Ommie, a custom social robot engineered in collaboration with experts in psychology and neuroscience to support deep breathing through guided haptic interaction. A core behavior of the robot involves users placing their hands on the robot’s body to tactilely feel its expansion and contraction in the cadence of deep breathing. We observe that this interaction affords grounding and focusing effects distinct from that of verbal guidance.
In order to evaluate the efficacy and applicability of Ommie, we perform four studies across a range of populations and real-world contexts. Three single-session studies examine usability, acceptability, and state anxiety outcomes with anxious university students, pediatric patients exposed to procedural anxiety, and seniors with dementia and their care partners. A fourth longitudinal study deploys a fully autonomous, field-ready version of Ommie into the homes of adolescents with clinical anxiety. This study explores engagement, relationship formation, self-efficacy, and anxiety-related outcomes over multiple weeks. To support future adaptive interaction with long-term use, we additionally develop and evaluate a machine learning system for classifying deep breathing phases from multimodal robot data.
Together, our research contributes design insights, empirical and qualitative evidence, and technical systems that advance the application of human-robot interaction for mental health. Our work additionally represents the first longitudinal deployment of a custom-designed social robot for mental health support in the home. Overall, this dissertation demonstrates how embodied AI can augment existing psychological care and bridge the gap between clinical intervention and everyday life.
Advisor: Brian Scassellati
Other committee members:
Marynel Vázquez
Dylan Gee (Yale Dept. of Psychology)
Hae Won Park (MIT Media Lab)