Model-based vs. measured muscle activity in dynamic lifting tasks with and without back-support exoskeletons

Abstract

Back-support exoskeletons (BSEs) offer the potential to reduce the risk of work-related musculoskeletal disorders of the lower spine. Although surface electromyography (sEMG) is commonly used to evaluate BSE effects on muscle activity, there are practical limitations to monitoring multiple trunk muscles while wearing a BSE. Musculoskeletal models offer an alternative, providing estimates of muscle activity that may complement or substitute direct measurements. We evaluated estimates of muscle activity using the gait-full-body model from the AnyBody™ Modeling System (AMS) and the OpenSim full-body thoracolumbar model. These evaluations were done for two bilateral lumbar extensors (longissimus and iliocostalis), vs. normalized sEMG (nEMG) data obtained from 18 participants who completed symmetric and asymmetric lifting tasks with and without two different BSEs. Comparisons were done using maximum normalized cross-correlation (MNCC), root-mean-square error (RMSE), and peak activity (95th percentile). Both AMS and OpenSim yielded strong associations with measured nEMG (mean MNCC: 0.90–0.95), though moderate errors were found (mean RMSE: 0.09–0.15). Model estimates captured general reductions in peak muscle activity with BSE use, consistent with nEMG, but with varying magnitudes. Muscle activity estimates had smaller MNCC and larger RMSE in BSE conditions, suggesting limitations in current simplified human–BSE interaction models. While both modeling tools show promise for estimating trunk muscle activity during occupational tasks, further refinement is needed—particularly to improve accuracy during complex movements and BSE-assisted scenarios. These findings support the potential utility of musculoskeletal models in ergonomic assessment and exoskeleton evaluation but underscore the need for cautious interpretation.