Stochastic parameter identification and control for LQR feedback control in robot periodic motion

Abstract

For high-precision and high-speed control of robots, control systems are needed to be designed based on their dynamics models. The dynamics model requires identification of the dynamics parameters. However, due to un-modeled dynamics and noises, the conventional parameter identification methods such as LS (least square) method obtain only approximations whose optimality strongly depends on the control objectives. This paper proposes a method to identify values of the dynamics parameters suitable for use in control system design. In this method, the dynamics parameters are considered to be stochastic variables, and identified so that their variance is made small for large influence on control performance shaping its covariance to follow the desired one. By considering feedforward and feedback control system design with Linear quadratic regulator, the desired covariance matrix is introduced. Experiments using a planar 3-link manipulator show that the proposed method identifies the appropriate parameters, and the designed controller achieves highly accurate positioning of the end-effector.