Neural and neurofuzzy FELA adaptive robot control using feedforward and counterpropagation networks (Q1277265)

This paper presents an application of neural and fuzzy-neural controllers to adaptive control of rigid robotic manipulators. The main idea of the proposed approach lies in implementing the feedback error learning architecture of the controller. The controller designed in the paper includes the computed torque algorithm based on the nominal dynamics of the manipulator, and a neural network algorithm aimed at computing the real inverse dynamics of the manipulator despite modelling inaccuracies. The neural part of the controller has been implemented either as a feedforward network or a back propagation network equivalent to a fuzzy control algorithm. An attempt at proving local stability of the control algorithm is made. Simulation experiments performed on the model of the PUMA manipulator show that both neural controllers work satisfactorily and provide comparable quality of control.