Personalizing older adults’ cognitive training robot games with LLM support

As populations age, innovative solutions are needed to support cognitive health in older adults. This paper presents a platform that delivers personalized cognitive training games through a humanoid robot, leveraging a Large Language Model (LLM) to generate game content and users’ profiles from autobiographical memories. Four serious games were designed and developed as cognitive-stimulation activities, with content dynamically tailored to user profiles to provide personalised exercises. We conducted a four-week feasibility and user-experience evaluation study with 20 older adults (65+) and interviews with 10 caregivers to assess system’s usability, emotional impact, and perceived value. Results show that participants found the robot-delivered games engaging and emotionally meaningful, with significant improvement in perceived user control and high usability ratings. Caregivers especially valued the LLMs’ potential for scalable personalization. This work highlights the feasibility of integrating LLMs and social robots for cognitive activities in aging populations, providing indications on how to achieve this.