Neuro-fuzzy adaptive torque control of a SCARA robot

Abstract

A robust neuro-fuzzy controller estimating the payload and torque required to follow a desired 4-3-4 trajectory is presented. The inverse dynamic model of the robot for no payload is used as the feedforward of the neurofuzzy controller. The payload is then estimated online using neuro-fuzzy adaptive torque control (NFATC) to compensate for the torque generated by the inverse model and to feed the required torque to drive the robot. The developed method is simulated when the measured velocity is contaminated with random noise. The results are compared with the performance of a model reference adaptive controller (MRAC). The neuro-fuzzy adaptive torque controller (NFATC) has performed at least an order of magnitude better than the MRAC controller. The results also indicate that the NFATC has a better robustness for increased payloads and measured noise than MARC.