Learning impedance regulation skills for robot belt grinding from human demonstrations

Abstract

The purpose of this paper is to transfer the impedance regulation of manual belt grinding to robot belt grinding control.,This paper presents a novel methodology for transmitting human impedance regulation skills to robot control in robot belt grinding. First, according to the human grinding experimental data, the skilled worker’s arm impedance regulation is calculated. Next, the human skills are encapsulated as the statistical learning model where the kernel parameters are learned from the demonstration data by Gaussian process regression (GPR) algorithms. The desired profiles of robot are generated by the task planner based on the learned skill knowledge model. Lastly, the learned skill knowledge model is integrated with an adaptive hybrid position-force controller over the trajectory and force of end-effector in robot belt grinding task.,Manual grinding skills are represented and transferred to robot belt grinding for higher grinding quality of the workpiece.,The impedance of the manual grinding is estimated by k-means++ algorithm at different grinding phases. Manual grinding skills (e.g. trajectory, impedance regulation) are represented and modeled by GMM and GPR algorithms. The desired trajectory, force and impedance of robot are generated by the planner based on the learned skills knowledge model. An adaptive hybrid position-force controller is designed based on learned skill knowledge model. This paper proposes a torque-tracking controller to suppress the vibration in robot belt grinding process.