Ergonomic Risk Assessment in Human-Robot Collaboration for Assembly Using Interpretable GAM and Wearable IMUs

During the process of manual assembly in human-robot collaboration, the repetitive physical load generated in the dynamic task area can lead to musculoskeletal disorders, thereby affecting the health of workers and production efficiency. Therefore, studying the dynamic ergonomic risk assessment is of great significance for improving the safety of the collaborative workspace. In this paper, a stratified interpretable modeling framework was established based on the kinematic data from wearable sensors for the first time, and the influence of individual motion strategies on ergonomic risk levels was elucidated. Combined with the characteristics of worker heterogeneity, a Cluster-specific Generalized Additive Model (CGAM) was obtained and verified. Based on the above research results, the continuous risk score of the specific “Trunk-Dominant” and “Arm-Dominant” patterns is computed by the stratified algorithm. Research has shown that under the set experimental tasks, the predictive accuracy of the model is high and the deviations are minimal. The coefficient of determination ( R 2 ) reached 0.79, and the Root Mean Square Error (RMSE) was 0.54. Comparative analysis of the adaptive robot intervention shows a significant reduction in the maximum risk score. This study proposes a new method for calculating ergonomic risk in the actual dynamic collaboration area considering individual movement variability, which has important theoretical significance for human-centric robot control strategies.