Controlling multi-link manipulators by fuzzy selection of dynamic models

A method for the identification of complex nonlinear dynamics of a multi-link robot manipulator using Runge-Kutta-Gill neural networks (RKGNNs) in the absence of input torque information is proposed. The RKGNNs constructed using shape adaptive radial basis functions are trained by an evolutionary algorithm. Due to the fact that the main function network is divided into sub-networks to represent the dynamic properties of the manipulator, the neural networks have greater information, processing capacity and can be tested for properties such as positive definiteness of the inertia matrix. Dynamics of a three-link manipulator are identified using only their input-output position and velocity data, and promising control results are obtained to prove the effectiveness of the proposed method in capturing highly nonlinear dynamics of a multi-link manipulator.