Bidirectional joint torque prediction with EMG of multiple channels: Both agonist and antagonist muscles are necessary

In numerous rehabilitation applications, joint motion is frequently required to be estimated through human muscle activity for motor control paradigm to impaired patients. Predicting joint torque in different degrees of freedom involves measurement of voluntary electromyography (EMG) on multiple muscle groups. From viewpoint of kinesiological discipline, agonist and antagonist muscle groups are respectively in charge of joint movement in different flexion/rotation directions. As a result, modeling and estimation of bidirectional joint motion follows to measure EMG signals of both agonist and antagonist muscles. Due to high possibility of drop off of EMG electrodes during motion, acquisition of some EMG channels may be missed and predicting joint torque with rest EMG channels occurs. This paper discusses the bidirectional joint torque prediction issue with complete and incomplete channels of EMG signals. The prediction results indicate that, EMG of both agonist and antagonist muscles are necessary to fulfill a promising bidirectional torque prediction, otherwise the prediction would be much degraded owing to lack of EMG on either agonist or antagonist muscle group.