Neuro-fuzzy based approach for hybrid force/position robot control

This paper presents a neuro-fuzzy control approach for MIMO systems. Motivated by the hybrid force/position control of robot manipulator problem, a systematic design procedure for fuzzy rules generation and optimization is proposed. The proposed neuro-fuzzy controller is constructed with respect to three phases. In the first one, which is called parameters learning phase, the neuro-fuzzy system is considered as a feedforward neural network and the backpropagation learning algorithm is then applied for parameters identification in order to map input/output data. In the second phase, a new clustering algorithm based on the inclusion concept is used for optimal clusters identification. Finally, the fuzzy rule base is generated and optimized. A 2-DOF planar manipulator force/position control simulation is presented and the results discussed.