Adaptive backstepping control for a class of nonlinear systems via multilayered neural networks

The paper presents an adaptive backstepping approach for systems in strict-feedback form using multilayered, nonlinear-in-the-parameters neural networks. A benefit this approach is that the construction of the controller is greatly simplified by obviating the need to construct a regressor or basis functions for the neural network. In addition the network is adapted solely online, with no off-line training. The neural network architecture is very simple, and scales easily with the number of backward steps taken in the control design.