Foundation model robots show promise but struggle with reliability in elder care

A new perspective paper from researchers Zhiwen Qiu, Wei Liu, and Yuexing Hao reviews the state of foundation model-based robots in elderly and patient care. As global populations age, demand for automated care assistants is rising. These robots typically integrate large language models as conversational and reasoning layers within voice-centered, socially assistive embodiments. However, the study finds that multimodal grounding—the ability to connect language to physical actions and sensory inputs—remains rudimentary, and physical autonomy is severely limited. Most systems are essentially chatbots with a robot body, not true embodied agents.

Empirical evaluations cited in the paper report positive user engagement and usability benefits, but persistent reliability failures undermine trust. Hallucinations, conversational breakdowns, and lack of error recovery are common across the interaction pipeline. Evidence for actual care impact is concentrated on proximal outcomes like cognitive engagement and participation, with very limited validation of clinical or care-related changes (e.g., fall reduction, medication adherence). The authors argue future research must shift toward care-specific evaluation standards, accountable human oversight, and integration into existing care workflows to move from promising prototypes to responsible, deployable technologies.