A neural net controller for robots with Hebbian tuning and guaranteed tracking

A neural network controller structure is developed for general unknown serial-link robot manipulators. The control structure is based on a 3-layer neural network learning on-line using modified Hebbian rules. Under some mild assumptions, a Lyapunov proof guarantees that both tracking error and weight estimate errors are bounded and some specific bounds are given. Using Hebbian tuning rules in each layer of the neural network brings a relatively simple adaptation structure and offers computational advantages over gradient descent based algorithms. Without a preliminary off-line training phase, the network weights are easily initialized to enable on-line learning in real-time.