Stable Adaptive Control Of Robot Manipulators Using Neural Networks

Stable adaptive control for nonlinear systems using neural networks has been paid more and more attention in recent years. The typical approach is to combine direct and indirect neuro-adaptive control and variable structure methods to obtain improved system performance. However, most of the accomplishments in this area are for continuous time systems, and the variable structure control employed in their neural network-based adaptive control schemes is static. A new design idea is proposed in this paper to design a stable adaptive controller of sampled-data nonlinear systems using neural networks for integrating a neural network method with the adaptive implementation of the variable structure control with a sector. By applying the approach to the stable neuro-adaptive control of rigid and flexible link robotic manipulators, a set of stable sampled-data adaptive control approaches of robotic manipulators using neural networks are obtained. Under the assumption that the joint angle speed is not available, how to design a stable neuro-adaptive controller of robotic manipulators is a new problem for the neuro-adaptive control of robot with unknown nonlinear dynamics. This paper solves this problem by combining the robust control and the neural network method, and a neural network approach to controller- observer design of a robot is obtained. The main work in this dissertation is summarized as follows. (1) Stable adaptive control for sampled-data nonlinear systems using neural networks for integrating a neural network method with the adaptive implementation of the variable structure control with a sector is proposed, and the complete control structure and learning algorithm are given. On the other hand, for feedback linearization, how to use neural networks to approximate the general controller structure u(x)=D^-1(x)N(x) is also a problem by many researchers. To solve the problem, different techniques exist in the literature that assure local and global stability with an additional knowledge. The approach used here is to estimate D^-1(x) instead of D(x), which avoids many complications. This paper offers rigorous proofs of performance in terms of tracking error stability and bounded neural network weights, and good performance is illustrated in comparison with the neuro-adaptive control approach for continuous-time nonlinear systems based on the static variable structure control. (2) Stable indirect and direct adaptive control approaches of robotic manipulators using neural networks are developed by applying the stable adaptive control approach for sampled-data nonlinear systems using neural networks to the robot trajectory tracking control. The only difference is that the symmetry of the robot inertia matrix is here used to obtain exact decoupling instead of the approximation decoupling. (3) This paper also extends the neuro-adaptive control approach for continuous-time nonlinear systems based on static variable structure to the motion control of robotic manipulators, and a systematic performance comparison between stable neuro-adaptive control approaches of robotic manipulators based on the discrete time variable structure with a sector and ones based on the static variable structure are made. (4) Theoretical results on the indirect, direct neuro-adaptive control of robot manipu- lators based on the variable structure with a sector are given under the assumption that the bound on neural network ap...