Model reference adaptive control using a neural compensator to drive an active knee joint orthosis

This paper presents an adaptive control approach of an actuated orthosis for the human knee joint rehabilitation. The objective of the proposed technique is to help patients to follow the guidelines of movement imposed by the therapists in terms of position and velocity. This is achieved by a system consisting of a mechanical orthosis actuated by an electrical driven motor. No needed prior knowledge concerning patients (height, weight, etc.). To prove the stability of the system, composed of the shank and the orthosis, in closed loop, we consider known its dynamic model structure. A Radial-Basis-Function Neural Network (RBFNN) is used to approximate online, a part of unknown dynamics and other unmodeled effects. In the goal to avoid abrupt transitions that can harm the wearer, we have used a reference model that can be constructed by an expert. The stability study conducted according to Lyapunov's approach guarantees that the proposed control remains stable even in the presence of bounded or assistive disturbances. The good performances of the proposed controller allow us to conclude with its effectiveness for trajectory tracking. In this work and for safety reasons, an adequate dummy has been used to perform real tests and detect any possible anomaly.