Gaussian Process-Enhanced Impedance Iterative Learning for Robot Interaction Control

Impedance control is an interaction control approach that allows robots to perform ideal characteristics similar to a mass-spring-damping model for safety and naturalness considerations. Iterative learning (IL) is effective in learning desired impedance parameters for robots under unstructured environments. In this paper, we propose an enhanced impedance IL method to speed up learning convergence and reduce the space complexity of data storage, where a sparse online Gaussian process is used to model a variable impedance strategy and is updated in the same iteration by removing similar data points from previous iterations while learning impedance parameters in multiple iterations. Numerical results based on a robot manipulator with 7 degrees of freedom have shown that the proposed method accelerates iterative convergence compared to the classical impedance IL method.