An EMG-driven musculoskeletal model for estimation of wrist kinematics using mirrored bilateral movement

Myoelectric control methods aim at driving electric prostheses in restoring functionality in daily life for amputees. To achieve the simultaneous and proportional estimation of multiple kinematics of the wrist joint, model-free approaches are broadly developed but rely on the abundant training data and the numerical functions that omit neuro-mechanical transformation. The musculoskeletal model-based approach entails the underlying muscular transformation from neuro-commands to the limb motion. However, the model-based approach for simultaneous kinematics estimation of the distal joint, i.e., wrist joint, is often overlooked. This paper proposes an electromyography (EMG)-driven musculoskeletal model to estimate the wrist joint kinematics involving wrist flexion/extension and radial/ulnar deviation. The proposed approach computes the internal force/joint torque and integrates the wrist kinematics using the forward dynamics. The mirrored bilateral movements are utilised to optimise the internal physiological parameters, which improve the feasibil-ity for the practical application of myoelectric control for the unilateral transradial amputee. Experiments are conducted on six able-bodied subjects, involving a series of wrist movements in free space. Results demonstrate the proposed model-based approach provides high estimation accuracy in the contralateral case with mean coefficient of determination of 0.86 and 0.82 for wrist flexion/extension and radial/ulnar deviation, respectively. The proposed approach and setup give reliable and feasible meaning for the simultaneous and proportional control of multiple wrist kinematics.

Automated summary

This paper proposes an electromyography (EMG)-driven musculoskeletal model to estimate the wrist joint kinematics involving wrist flexion/extension and radial/ulnar deviation. The proposed approach computes the internal force/joint torque and integrates the wrist kinematics using the forward dynamics. Experiments demonstrate that the proposed model-based approach provides high estimation accuracy in contralateral case with mean coefficient of determination of 0.86 and 0.82 for wrist flexion/extension and radial/ulnar deviation, respectively.