Stable neural controller design for unknown nonlinear systems using backstepping

Despite the vast development of neural controllers in the literature, their stability properties are usually addressed inadequately. There is a lack of systematic approach in choosing neural network structure, initial weights, and training speed, due to the insufficient understanding of the controller behavior. Consequently, these choices can only be made via try and error in order to achieve stability. In this paper, we propose a stable neural controller design for a class of unknown, minimum phase, input-output feedback linearizable nonlinear system with known relative degree. The control scheme uses the backstepping design technique, and guarantees semi-global stability. Meanwhile, the controller preserves the nice performance properties of the standard backstepping controllers