A Novel Iterative Second-Order Neural-Network Learning Control Approach for Robotic Manipulators

Iterative Learning Control (ILC) is known as a high-accuracy control strategy for repetitive control missions of mechatronic systems. However, applying such learning controllers for robotic manipulators to result in excellent control performances is now a challenge due to unstable behaviors coming from nonlinearities, uncertainties and disturbances in the system dynamics. To tackle this challenge, in this paper, we present a novel proportional-derivative iterative second-order neural-network learning control (PDISN) method for motion-tracking control problems of robotic manipulators. The control framework is structured from time- and iterative-base control layers. First of all, the total systematic dynamics are concretely stabilized by a conventional Proportional-Derivative (PD) control signal in the time domain. The control objective is then accomplished by using an intelligent ILC decision generated in the second layer to compensate for other nonlinear uncertainties and external disturbances in the dynamics. The iterative signal is flexibly composed from various information on the iterative axis. On one hand, the previous iterative control signal is inherently reused in the current iteration but with an appropriate portion based on reliability of the current control performance. On the other hand, the iterative-based modeling deviation remaining is treated by a functional neural network that is specially activated by a second-order learning law and information synthesized from the current and previous iterations. Stabilities of the time-based nonlinear subsystem and overall system are rigorously analyzed using extended Lyapunov theories and high-order regression series criteria. Effectiveness of the proposed controller was intensively verified by the extensive comparative simulation results. Key advantages of the proposed control method are chattering-free, universal, adaptive, and robust.