Robust adaptive neural fuzzy controller with model uncertainty estimator for manipulators

In this paper, a new robust adaptive neural fuzzy controller (RANFC) for manipulators is proposed. The proposed controller uses the fuzzy logic inverse dynamic model output (approximated torques/forces) as a model-based control term; a decentralized PID controller as a feedback term to enhance closed-loop stability and improve transient performance; and multi-layer feedforward neural networks (NNs) as a slow learning tool to generate new rules or modify the membership function of existing rules to compensate for the unmodelled dynamics such as flexibility and friction. Moreover, in order to guarantee the robustness and stability of the controller, a new adaptation scheme is introduced to compensate for the measurement noise, external disturbances with random pattern, and effects of rapidly time-varying parameters. Another key feature of this scheme is that a priori knowledge of the bounds of uncertainties is not required. The global stability and robustness of the proposed controller are established using Lyapunov's approach and fundamentals of sliding mode theory. The simulation results illustrate the strength of the proposed controller in the presence of the model uncertainties and external disturbances.