Direct Adaptive Fuzzy-Based Neural Network Controller for a Human-Driven Knee Joint Orthosis

This paper presents a control error based direct adaptive Neural Network (NN) controller applied to a lower limb knee joint orthosis during flexion/extension movements. The proposed approach requires neither pre-knowledge of the exact human-orthosis system nonlinearities nor it’s exact parameters. Unlike the available NN control approaches that rely on the tracking errors to derive the adaptive weights, our approach represent an alternative way on which we introduce the control error for online updating of the NN weights. A Fuzzy Inference System (FIS) is exploited to estimate the unknown control error. Then, the NN weights are tuned directly using back-propagation algorithm based on a quadratic criterion of the control error independently from the tracking error. In terms of stability, the tracking error has been proved to converge exponentially to an arbitrary small set despite the presence of external disturbances. Simulations are conducted to evaluate the effectiveness of the proposed control approach.