Variable Admittance Control using Preferential Bayesian Optimization in Human-Robot Collaboration

Abstract

In physical human-robot collaboration (PHRC) systems, impedance or admittance control is commonly used for safety and flexibility. Moreover, tuning and optimization of the impedance or admittance parameters can enhance the PHRC system’s performance. However, a challenge encountered in parameter optimization within PHRC systems is the selection of an appropriate evaluation function. This paper introduces Preferential Bayesian Optimization as a method for parameter optimization in PHRC settings. Unlike traditional approaches, this method does not require an explicit definition of the evaluation function and instead relies solely on human preference information during the optimization process. The effectiveness of this approach is verified through its application of tuning a variable admittance controller for a 7-degree-of-freedom robot arm.