Large language model-enabled multi-agent self-organised approach for personalised rehabilitation assistive devices design

With the increasing prevalence of age-related diseases among the elderly, escalating disability rates resulting from chronic illnesses, and heightened public health awareness, there is a growing demand for personalised design of rehabilitation assistive devices (RADs). Conventional product design methods have shown significant limitations in addressing the diversity and individuality of user needs. Therefore, leveraging large language models (LLM) to self-organise multi-agents, an innovative method for designing personalised RADs, called LLM-PDAgents, is proposed in this manuscript. Initially, a knowledge graph (KG) was constructed with the relationship among ‘phenomenon-symptom-function-behaviour-structure', establishing a chain-of-thought for product design knowledge within multi-agents. Subsequently, a thinking mechanism for task planning in the self-organised process of multi-agents was developed, which emulates the processes of planning, execution, and monitoring in product design. Ultimately, a multi-tiered progressive reflection strategy was devised to iteratively refine the product design, fostering a continuous optimization on the design solution until user satisfaction is achieved. Taking the real-world design of knee joint RADs as an example, LLM-PDAgents significantly improved the procedure and quality of the whole design process, and its outcomes showed merits on economic viability, manufacturing feasibility, user pertinence, and generated costs, indicating an efficient and reliable approach for the personalised design of RADs.