Despite the vast development of neural controllers in the literature, their
stability properties are usually addressed inadequately. With most neural
control schemes, the choices of neural-network structure, initial weights,
and training speed are often nonsystematic, due to the lack of understandin
g of the stability behavior of the closed-loop system. In this paper, we pr
opose, from an adaptive control perspective, a neural controller for a clas
s of unknown, minimum phase, feedback linearizable nonlinear system with kn
own relative degree, The control scheme is based on the backstepping design
technique in conjunction with a linearly parameterized neural-network stru
cture. The resulting controller, however, moves the complex mechanics invol
ved in a typical backstepping design from offline to online. With appropria
te choice of the network size and neural basis functions, the same controll
er can be trained online to control different nonlinear plants with the sam
e relative degree, with semiglobal stability as shown by simple Lyapunov an
alysis, Meanwhile, the controller also preserves some of the performance pr
operties of the standard backstepping controllers. Simulation results are s
hown to demonstrate these properties and to compare the neural controller w
ith a standard backstepping controller.