An adaptive neural network algorithm for parameter identification of linearizable nonlinear systems

Nonlinear system parameter identification is a challenging problem. In the area of robot control, for example, a good model of the nonlinearities improves performance for model-based controllers. An appealing approach to robot control is the method of input-output linearization whereby the nonlinear robotic system is made globally linear equivalent through an inverse dynamics compensator. The difficulty with this technique, however, is that an exact identification of the inverse dynamics parameters is required for global input-output linearization. In this paper, an adaptive neural network architecture is developed to model the inverse dynamics of a nonlinear system. Thus global linearization is possible and linear or nonlinear control may be designed, based upon the linear equivalent structure. An example using the model of the Mars Mission Research Center experimental rover is discussed and the neural network estimator with a nonlinear controller is presented to illustrate the feasibility of this approach.