An Improved Calibration Method of EMG-driven Musculoskeletal Model for Estimating Wrist Joint Angles

Lumped-parameter musculoskeletal model based on surface electromyography (EMG) promises to estimate multiple degrees-of-freedom (DoFs) wrist kinematics and might be potentially applied in the real-time control of powered upper limb prostheses. In this study, we proposed a new parameter calibration method based on the lumped-parameter musculoskeletal model. Compared with the existing calibration method in the lumped-parameter musculoskeletal model, this paradigm used an improved method of calculating estimated joint angles in optimization and a reduced training dataset (data from only single-DoF movements) to optimize model parameters. Surface EMG signals were then mapped into the kinematics of the wrist joint using the optimized musculoskeletal model. In the experiments, wrist joint angles and surface EMG signals were simultaneously acquired from able-bodied subjects while performing 3 movements, including flexion/extension (Flex/Ext) only, pronation/supination (Pro/Sup) only, and 2-DoF movements. The offline tracking performance of the proposed method was comparable to that of the existing calibration method with averaged r = 0.883 and NRMSE = 0.218. Moreover, the results demonstrated significant superiority of the proposed method over the existing method with less amount of data for parameter tuning, providing a promising direction for predicting multi-DoF limb motions with only single-DoF information.