Decoupling Observer for Contact Force Estimation of Robot Manipulators Based on Enhanced Gaussian Process Model

Abstract

This paper aims at addressing the challenge of contact force estimation of robot manipulators with incomplete model information. A novel decoupling observer is proposed based on an enhanced Gaussian process (EGP) model. Specifically, the dynamic model of the robot manipulator is decomposed into the nominal part established using Euler-Lagrange theory and a residual dynamics term with no priori information. To improve the model accuracy, the Gaussian process regression (GPR) technique is adopted to develop a data-driven compensation term for the residual dynamics. Compared to the purely data-driven models, the advantage of the proposed EGP model lies in its computational efficiency as the information contained in the nominal model has been exploited. Based on the statistical information of the residual dynamics learned via GPR, a novel decoupling observer is presented for real-time force estimation. Due to its capability of decoupling the contact force from residual dynamics and system noises, the proposed observer approach is no longer dependent on an accurate dynamic model of the robot manipulator. Simulation and experimental results demonstrate that the proposed scheme outperforms the state-of-art methods.