Assistance of knee movements using an actuated orthosis through subject's intention based on MLPNN approximators

In this paper, we present an actuated orthosis aimed to assist the subject's knee joint movements through wearer's intention. The shank-foot orthosis system is considered as a black box and is controlled using a Multi-Layers Perceptron Neural Network (MLPNN) controller. This controller avoids the modeling of the “wearer lower limb-orthosis” system, and then avoids the long and complex procedure to identify the model parameters since MLPNN is able to represent model nonlinearities. In this study, the MLPNN is used to approximate the inertia, viscous and damping friction effects as well as the gravitational effect of the system. We propose to control the Shank-foot-orthosis system through the estimated human intention by measuring muscular activities of the quadriceps muscle. For that purpose, a second MLPNN (MLPNN estimator) is trained to give the desired subject's movement as functions of the ElectroMyoGram (EMG) signals measured at the quadriceps muscle. The proposed method avoids the use of classical approaches to model muscles, activation and contraction dynamics that are often nonlinear. EMG muscular activities represent the system input while knee joint angle represents the system output. The proposed approach is validated experimentally with a healthy subject. Promising results in term of desired trajectory tracking while ensuring system stability are obtained.