EMG – based Force Estimation for Dynamic Muscle Contractions in Physical Human-Robot Interaction

Electromyography is a method in the field of electrodiagnostic medicine for monitoring and assessing the electrical activity generated by skeletal muscles themselves. As reviewed, the study of human force estimation based on Electromyographical signals is such a crucial challenge, since it is further complicated by the dynamic nature of the human subject. Consequently, this paper aims to develop an effective algorithm to roughly approximate the human hand applied force during executing rectilinear-motion-machine interaction, in which a test influence variable, namely friction force against the object movement, was additionally exerted. This mathematical algorithm will be further implemented in a newly designed rehabilitation robot using EMG muscle force estimation instead of applying a costly multi-axis force/torque sensor. Artificial neural networks and Support vector machine approaches were successfully applied to distinctly classify the electromyography signals of human’s hand muscles detected by eight-channel EMG electrodes. After the set of tests was carried out, root mean square error was individually utilized to evaluate the quantitative performance of each technique. The experimental results illustrated that both approaches were considered acceptable for EMG–based force estimation for dynamic muscle contractions by indicating that the human applied force was validly estimated based on the EMG signals. Additionally, it can be implied that the more the resistant force applied against the object movement, the lower the force model estimated performance.