Presentation
Human Performance Modeling with Natural Language (HPM-NL) for Upper-limb Prostheses: Generative pre-trained transformer (GPT)-based rapid HPM under low hallucination
DescriptionAccurately predicting how people use prosthetic hands is key to improving their design and usability. Traditional human performance modeling (HPM) methods—like GOMS, ACT-R, or QN-MHP—are effective but often difficult to use, requiring expert knowledge and extensive time. We present HPM-NL, a novel, free tool that uses GPT-based natural language processing to quickly estimate task completion time and workload for upper-limb prosthesis tasks. Unlike conventional approaches, HPM-NL relies on empirical data from 30+ years of research, minimizes hallucination risk, and combines strengths from multiple HPM frameworks. It generates results in under a minute from simple text inputs, making it accessible to both clinicians and engineers. A validation study with graduate students compared predictions from HPM-NL to both Cogulator (a CPM-GOMS tool) and real human-subject data. Results showed that HPM-NL predictions closely matched actual data across tasks such as clothespin relocation and target acquisition, while Cogulator showed more discrepancies. HPM-NL also provides step-by-step breakdowns and visual feedback for better interpretability. This approach reduces modeling burden and opens the door for more rapid and scalable usability testing in prosthetic design.
Event Type
Case Study
Industry/Practitioner Content
Lecture
TimeThursday, October 16th9:50am - 10:10am CDT
LocationGrand Hall L
Human Performance Modeling