A robust adaptive neural control scheme to drive an actuated orthosis for assistance of knee movements

In this paper, we present an actuated orthosis intended to assist the subject׳s movements for flexion/extension of the knee. The equivalent system “subject׳s lower limb-actuated orthosis” is considered as black-box and is driven by a Multi-Layer Perceptron Neural Network (MLPNN) controller. This controller is adaptive, does not require the dynamic model of the system and is able to take into account all its uncertainties. The latter include the nonlinearities due to the subject׳s lower limb/actuated orthosis coupling, the modeling and identification errors as well as the parameter uncertainties resulting from the system׳s dynamics. Stability of the “subject׳s lower limb-actuated orthosis” system, using the proposed approach, is mathematically proved based on the Lyapunov theory. Performances of the proposed MLPNN controller are compared to those of the PID (Proportional Integrator Derivative) controller for the track of desired position and velocity trajectories. These comparisons include the trajectories errors, the capacity of each controller to assist the torque produced by the subject and the robustness of the system against external disturbances. To illustrate the efficiency of the proposed controller, real-time experiments were conducted on five voluntary subjects.