Automated lifting risk assessment with LLM-based agent and AI influence in physical ergonomics

Repetitive occupational lifting tasks across various industries often result in biomechan- ical overload, increasing the risk of work-related musculoskeletal disorders (WMSDs). Properly evaluating the risks associated with these tasks is essential for preventing WMSDs. Workers suffering from these disorders frequently experience reduced mobility, chronic pain, and require prolonged treatment, which can ultimately force them to leave their profession. Conventional lifting risk assessments rely on trained practitioners to perform on-site evalua- tions, assessing worker posture, task frequency, and the weight of objects involved. Even with the application of advanced machine learning models and the development of sensor-based methods for lifting risk assessment, ergonomists are required to interpret the data, conduct analyses and offer actionable recommendations to mitigate risk. In Study A, we proposed a novel self-explainable pipeline for predicting lifting risks with a conventional monocular camera and large language model (LLM)-based agent for perception and decision-making. In this approach, the LLM-based agent acts as an occupational health expert, delivering informed guidance on safe lifting practices. The proposed framework is a fully automated system with capability of calculating risk with the NIOSH lifting equation from video and offering clear, actionable advice supported by explanations. Automation in manual work perviously perform by physical ergonomists and AI-ergonomist collaboration remains un- derexplored. In Study B, we explored several key principles regarding the automation in ergonomics. We answered how will automation of physical ergonomists’ tasks reshape their roles and how to ensure that automated decision-making while not compromise safety.