A hybrid neuro-fuzzy system for robot control

This paper presents a new method to designing a neurofuzzy controller for a robot. Because control signals from a fuzzy logic controller are determined by response behaviors rather than its analytical models, open-loop responses of the robot are described by a set of two-order systems. Then, the parameters of the fuzzy controller, which are related to this system, are off-line optimized by the Nelder and Mead´s simplex algorithm. Next, a neural network is used to train the mapping relationship between the open-loop responses and the optimized parameters of their corresponding fuzzy controllers. In order to control a two-link manipulator with nonlinear dynamics, its open-loop responses are first tested, and its optimal fuzzy logic controller is then determined by perceiving such responses using a neural network based on trained patterns. The advantage of this method is that one does not need to care about the convergence problem during the adaptation process when it is used to design a neurofuzzy controller