A neural network controller for feedback linearization

For a class of continuous-time nonlinear systems, a neural network-based controller which feedback linearizes the system is presented. For an unknown, state-feedback linearizable system, the controller achieves tracking performance and the semi-globally uniformly ultimately boundedness of the closed-loop signals is shown in the sense of Lyaponov. Modified Hebbian learning rules are used for online learning of ideal neural network weights. No off-line learning phase for NN is needed and initialization of the network weights is straightforward.