Admittance learning strategy using generalized simplex gradient methods for human–robot collaboration

Abstract

Human-Robot Collaboration (HRC) systems are becoming widespread in industrial applications owing to the advantages of enhancing work productivity and reducing production expenses. In this study, we focus on HRC systems that involve physical interactions between humans and robots. The relationship between force and position at the robot's end-effector is generally modulated using impedance or admittance control techniques to implement these systems. Furthermore, varying the target impedance of robots has been shown to enhance their performance in HRC tasks. This report presents a novel approach to admittance learning strategy aimed at minimizing human effort during physical human-robot collaboration tasks. A damping generation scheme based on Gaussian basis functions is introduced, enabling the generation of a diverse range of smooth damping profiles via the modulation of the weights of these functions. The Gaussian design is based on a frequency analysis of human movement, with weights adjusted via gradient descent to minimize the interaction force. A learning algorithm based on the generalized simplex gradient approximation technique is proposed to accommodate the noisy evaluation function, utilizing data from previous iterations to enhance estimation accuracy. The effectiveness of the proposed method is experimentally demonstrated through comparison to conventional methods, as well as trials involving a complex task.