Robot Manipulator Control Using Fuzzy Gaussian Potential Neural Networks
Mohammad Teshnehlab* and Keigo Watanabe**
*Graduate School of Science and Engineering, Saga University, 1 Honjo-machi, Saga, 840 Japan
**Department of Mechanical Engineering, Saga University
This paper describes the complete flexible design of a fuzzy gaussian potential neural network (FGPNN) having the ability to learn expert control rules of fuzzy controller. The proposed structure consists of gaussian potential function (GPF) which is utilized in the antecedent as the membership function, and the flexible bipolar sigmoid function (FBSF) is utilized in the conclusion part. The GPF enables a reduction in the number of labelings in the antecedent, and the FBSF leads to a reduction in the learning load in the conclusion and captures the linearity and/or nonlinearity of the system in the conclusion. The proposed construction reduces the complexity to a simple design in the antecedent, especially for large-scale inputs, thus shortening the time for learning with learning sigmoid function parameters (SFPs) in the conclusion part only. Finally, the simulations of two-link manipulator will be provided for both the conventional and proposed FGPNN controller in order to evaluate the newly designed controller.