Structure-Consistency Gaussian Process-Based Impedance Control for Manipulators With Unknown Dynamics

Impedance control is one of the fundamental control approaches for contact-rich robotic tasks. However, to apply the impedance control, the robot dynamics needs to be completely known, which could be a tricky problem for real applications. This article presents a structure-consistency Gaussian process impedance control approach to solve this problem, in which the Gaussian process is introduced to capture the unknown part of the robot dynamics. Different from existing approaches, the Gaussian process structure is specially designed to make it consistent with the robot dynamic model, which could improve the impedance control accuracy and stability. Moreover, we also provide an optimization-based approach to denoise the training set, especially for the robot acceleration signals, which could reduce the training complexity of the Gaussian process. Finally, comparative experiments are conducted to validate the proposed approach.