High accuracy adaptive motion control for a robotic manipulator with model uncertainties based on multilayer neural network

Abstract

Abstract A robotic manipulator is widely used nowadays, and its high‐accuracy motion control has become a hot topic. However, since it is a very complicated control plant, there exist many structural and unstructured uncertainties in its mathematical model, which would degrade the manipulator control performance when the traditional model‐based control method such as feedback linearization control is adopted. In this paper, a practical method that combines feedback linearization control with multilayer neural network (MNN)‐based model uncertainties observer is proposed for high‐accuracy motion control of a robotic manipulator. The proposed controller not only accounts for the parametric uncertainties but also for the external disturbance. A multilayer neural network capable of online learning is designed to estimate all of the model uncertainties, which can be compensated through the feedforward cancellation technique. A weight adaption law for the online learning neural network is derived based on Lyapunov stability theory. Finally, a feedback linearization method with the MNN‐based observer is designed to stabilize the whole system. The stability proof shows that bounded stability of a manipulator system with the proposed controller is guaranteed. Extensive simulation results show that this control strategy achieves high‐accuracy tracking control in the presence of the manipulator parametric uncertainties and external disturbance.