PD Control of a Manipulator with Gravity and Inertia Compensation Using an RBF Neural Network

Abstract

Dynamic compensation can improve the accuracy of trajectory tracking for industrial manipulators. For irregularly shape or flexible manipulators, however, it is difficult to measure the position of the center of mass (COM) so that its dynamic model cannot be expressed explicitly. This paper proposes a proportional derivative (PD) controller with radial basis function neural network based gravity and inertia compensation (RBFNN-GIC). The RBFNN is utilized to estimate the gravity disturbance and to enable identification of COM to calculate the compensated inertia term. The proposed strategy based on the dynamic model can be used on any robot arm whose COM, gravity and inertia are difficult to obtain. To demonstrate the optimization and effectiveness of proposed PD controller, comparative experiments between the proposed control scheme and the traditional data-fitting method least mean square method (LMS) are conducted on a 3 degree of freedom (DOF) robotic manipulator.