A Self-learning Nonlinear Variable Gain Proportional Derivative (PD) Controller in Robot Manipulators

Abstract

This paper proposes a nonlinear variable gain Proportional-Derivative (PD) controller that exhibits self-constructing and self-learning capabilities. In this method, the conventional linear PD controller is augmented with a nonlinear variable PD gain control signal using a dynamic structural network. The dynamic structural network known as Growing Multi-Experts etwork grows in time by placing hidden nodes in regions of the state space visited by the system during operation. This results in a network that is "economic" in terms of network sileo The proposed approach enhances the adaptability of conventional PD controller while preserving its' linear structure. Based on the simulation study on variable load and friction compensation, the fast adaptation is shown to be able to compensate the non-linearity and the uncertainty in the robotic system.