Some Remarks on Characteristics of Direct Neuro-Controller with Regard to Adaptive Control

Abstract

Many neural network studies have been performed in order to apply both flexibility and learning ability of the neural networks to robot controllers. Although, many researchers have been performed, the stability of a direct neural controller has not been studied, and it advantage has not been clarified in comparison with another control theory such as the adaptive control and learning control etc. Therefore, this paper aims at clarifying stability condition of the direct neural controller in the condition of the plant time lag effect.As the first step of this research, we clarify the asymptotic local stability condition by taking quadratic parameter error as the Lyapunov function when the linear neural network is used as the controller. These theoretical results show that the adaptive control is tightly relate to the neural controller using two-layer perceptron type neural networks.Next, numerical simulation is executed for the direct neural controller applied to the time lag system using the proposed stable learning algorithm. The simulation results indicates that nonlinear function of neuron can compensate the effect of nonlinear and parasitic term in plant. Experiment results using one freedom force control system show that the proposed learning algorithm is effective to the actual plant.