Estimation of Joint Stiffness via a Musculoskeletal Model Driven by Motor Neuron Twitch Properties

Joint stiffness estimation involves torque-based experiments together with system identification techniques. The musculoskeletal models recently used to estimate joint stiffness include bipolar electromyography data to drive the model. They do not provide information on individual alpha motor neuron properties which is essential to improve the personalization of the neuromusculoskeletal model. The use of bipolar electrodes limits the resolution to extract the moto-neuron properties. In this study, the NMS models were driven by the activation dynamics of EMG envelopes and the activation dynamics of motor units. The normalized EMG envelopes were regarded as the activation profiles of EMG Envelopes. The activation dynamics of motor units were decomposed from uni-polar HD-EMG data using a blind source separation technique, and the MU distributions were sampled to formulate the activation profiles. The results were validated at the torque and stiffness level by the experimental torque and stiffness estimation using the system identification technique. The torque estimations by both models were better than the stiffness estimations. At the torque level, the model driven by motor units produced improved results while it was vice-versa for stiffness. Overall, this method can be used to predict the torque, but enhancements should be made to increase the stiffness estimation.