Inverse estimation of muscle activations from joint torque via local multiple regression

Abstract

The signal measured with an electromyogram (EMG) is the summation of all action potentials of motor units active at a certain time. According to previous literature, one can establish the relationship between torque and EMG/activations in a forward way, i.e., employing EMG of multiple channels to estimate the joint torque. Once the relationship is established, the torque can be predicted with EMG recordings. However, in some applications of neuroprosthetics where we need to make muscle control, it is required to inversely have an insight regarding the muscle activations under a specific motion scenario from the corresponding torque. Motivated by this point, this paper investigates inverse estimation of muscle activations in random contractions at the ankle joint. Local multiple regression is exploited for finding the relationship between muscle activations and torque. Such technique is able to rebuild the relationship between muscle activations and joint torque inversely based on experimental data obtained from five able-bodied subjects, and the resultant optimal weight matrix can indicate each muscle's contribution in the production of the torque. Further cross validation on prediction of muscle activations with joint torque with optimal weights shows that such approach may possess promising performance.