Integrating a Large Language Model Into a Socially Assistive Robot in a Hospital Geriatric Unit: Two-Wave Comparative Study on Performance, Engagement, and User Perceptions

Lauriane Blavette ^1,2,3, Sébastien Dacunha ^1,2,3, Xavier Alameda-Pineda ⁴

⁰Institut National de la Santé et de la Recherche Médicale, Optimisation Thérapeutique en Pharmacologie OTEN Unité Mixte de Recherche-Santé 1144, Université Paris Cité, Paris, France.

Jmir Human Factors +

|

December 3, 2025

View abstract on PubMed

Summary

Integrating large language models (LLMs) into socially assistive robots (SARs) significantly improved interaction success and user experience in a geriatric care setting. This advancement enhances the potential of SARs for older adults, addressing key challenges in healthcare.

Area Of Science

Geriatric Medicine
Human-Robot Interaction
Artificial Intelligence in Healthcare

Background

Older adults in resource-limited settings face complex medical and psychosocial needs.
Socially assistive robots (SARs) offer practical support, with large language models (LLMs) enhancing their dialogue capabilities.
Acceptability of SARs depends on minimizing errors and adapting to user characteristics.

Purpose Of The Study

To evaluate the impact of integrating an LLM into a SAR dialogue system in a hospital geriatric unit.
To compare system performance and interaction success across two experimental waves.
To explore user characteristics' influence on performance, acceptability, and usability.

Main Methods

28 older adults participated in a single-session evaluation of a SAR over 8 months.
Interactions were recorded across two waves: basic dialogue system and LLM-based system.
System performance, user engagement, acceptability, and usability were quantitatively and qualitatively analyzed.

Main Results

LLM integration significantly increased error-free interactions (27.8% to 70.2%) and interaction success (25% to 74.5%).
Acceptability and usability scores were significantly higher with the LLM-based system.
Emotional engagement correlated positively with interaction success; age negatively impacted physical engagement and acceptability.

Conclusions

Dialogue quality improvements from LLM integration enhance interaction success and user experience in geriatric care SARs.
Behavioral engagement is influenced by both system performance and individual user traits.
Multimodal behavioral analysis and self-reported measures are crucial for user-centered robot design in clinical settings.

Keywords:

behavioral engagement geriatric care human-robot interaction large language models multimodal analysis older adults socially assistive robot