Design of neural network controller for robots using regressor dynamics

In this paper, a neural network structure for control of robot manipulators with unknown dynamics is proposed. The proposed structure takes advantage of the regressor dynamics of robot manipulators which linearizes the nonlinear robot dynamics in terms of its dynamic parameters. This linearized model enables the design of a neural network controller for robot manipulators based on the Adaline network structure with a modified Nguyen-Widrow off-line training algorithm to identify robot unknown dynamic parameters and a parameter adaptive control algorithm to perform on-line regulation. The resulting control scheme is computationally efficient and has very high potential in real-time applications. The proposed control scheme is illustrated through simulation and comparison studies