Estimating dynamic external hand forces during overhead work with and without an exoskeleton: Evaluating an approach using electromyography signals and random forest regression

Abstract

We developed a model to estimate hand contact forces during dynamic overhead tasks completed with and without passive arm-support exoskeletons (ASEs). One approach to assessing ASE effectiveness is evaluating shoulder joint forces through inverse dynamics, which requires data on both external kinetics and body kinematics. However, obtaining the former (e.g., hand contact forces) is challenging. To address this, our model estimates these forces using electromyographic (EMG) signals. For model development, we used data from a study in which participants completed dynamic overhead task simulations under various conditions, both with and without three ASEs. A random forest regression was used to map EMG signals to time series of hand contact force, considering task conditions and biological sex. Overall, the model produced reasonable force estimations, with errors generally consistent across conditions and regardless of ASE use. However, the model tended to underestimate peak forces, especially for upward vs. forward exertions and among males vs. females. Overall, the proposed model has the potential to support musculoskeletal modeling for assessing the effect of ASE use on workers. We provide several suggestions for improving future model performance.