Simultaneous and continuous estimation of upper limb kinematics of shoulder press movements: state-space EMG model

In this paper, a state space electromyography (EMG) model is proposed to predict multi-DOF upper limb joint angles continuously and simultaneously. The two main approaches to estimating multi-joint kinematics are numerical and model-based methods. The high computational cost of the numerical algorithms causes delays. On the other hand, the open-loop model-based systems would not be effective to estimate the joint angles due to the accumulated errors that are not compensated. To overcome this problem, a relation between EMG signal features and joint angles is trained and developed offline to form measurement equations. Then, the integration of the Hill-based musculoskeletal model and the trained measurement equations is used in a state-space model. By using Extended Kalman Filter (EKF), the multi-joint angles are estimated continuously and simultaneously. In this paper, the shoulder abduction–adduction and elbow flexion–extension angles of the human arm during dumbbell shoulder press are estimated using the EMG signals. The average root mean squared error (RMSE) of the estimation and real joint angles of the shoulder and elbow to be 0.26 and 0.27, respectively, indicates that the proposed model-based method can be used to estimate the joint angles continuously and simultaneously reducing the computational time of numerical estimations.