Human–machine interface of rehabilitation exoskeletons with redundant electromyographic channels

Abstract

A method for controlling an exoskeleton by means of instructions obtained by decoding an electromyography (EMG) signal is considered. The method allows to minimize errors when positioning the exoskeleton in the verticalization mode. The EMG signals are segmented into intersecting or non-intersecting windows and for each segment obtained in the previous step, they receive many signs of the EMG signal (vector of informative signs). The vector of informative features is fed to the neural network classifier, which controls the controller of the exoskeleton servomotors. The vector of informative features is obtained through a multilevel comparator, and the number of levels of which determines the dimension of the vector of informative features. The EMG classifier includes a comparator unit, a multiplexer, an informative feature calculation unit, a first neural network, a memory unit and a second neural network, the outputs of which control the servo motor controller. In order to adapt the exoskeleton control system to the patient, additional channels for classifying EMG signals are introduced into the human–machine interface. Each channel of the EMG signal is associated with a specific muscle or group of muscles that control the movement of the same limb joint. The servo motor controller uses a third neural network to aggregate these signals into a single control signal. The neural network control method with redundant EMG channels has been tested on the exoskeleton at the moment of controlling the verticalization of the patient.