Social Robotics and Large Language Models for Disability: A Scoping Review

Abstract This paper examines the intersection of social robotics and large language models (LLMs) in applications and interventions for disabled people. We adopt a hybrid approach drawing on methodologies from quantitative scoping and critical narrative reviews and synthesize research from 2014 to 2024, highlighting how these technologies are individually and jointly employed to address needs across neurological, physical, cognitive, and sensory disabilities. The aim of the review is to identify how human-robot interaction (HRI) and LLMs previously have been used for disabled people, and assess the implications of combining the two technologies for disability-related interventions. The review identifies existing studies, including their overarching goals, methodologies used, and setting, the role, form and capabilities of the agents and the targeted disability groups. Furthermore, it critically evaluates the social and ethical implications, including concerns about the framing of disability within technology design. By analysing the potential harms and benefits of integrating LLMs with physically embodied social robots for disabled people, this work discusses how such advancements might reinforce or challenge existing systemic inequities in disability-focused interventions in social robotics research. Finally, this paper offers a conceptual contribution to adaptation in social robots for disability, based on the findings in the review.