文章基本信息

标题：Human-Robot Collaboration with Variable Autonomy via Gaussian Process
本地全文：下载
作者：Takeshi Hatanaka ; Kosei Noda ; Junya Yamauchi 等
期刊名称：IFAC PapersOnLine
印刷版ISSN：2405-8963
出版年度：2020
卷号：53
期号：5
页码：126-133
DOI：10.1016/j.ifacol.2021.04.091
语种：English
出版社：Elsevier
摘要：AbstractIn this paper, we investigate a scenario of human-robot collaboration mainly motivated by agricultural applications, and present a novel control architecture with variable autonomy based on Gaussian Process (GP) regression. The system consists of three components, namely a robot manipulator, system operator, and computer. The operator monitors an image, acquired by the robot, capturing possibly multiple candidates for a target to be reached by the end effector of the robot. He/she chooses a target to be reached among these candidates, and determines a velocity command for the end effector, while the computer also identifies a target through image processing. We first design an automatic controller and a mechanism to tune autonomy online. To this end, we build two GP models from the data on a training system without any uncertain factor that may disturb the operation: one is built from the data of an expert for the system operation and the other is from the operator’s own data. The former model is utilized as the automatic controller and the autonomy determination mechanism is designed based on the variance information provided by the GP models so that the controller supports the manual control only when the expert’s model is reliable, the computer correctly identifies the human target, and the real environment does not contain any uncertain factor that affects manual control. We finally demonstrate the mechanism through experiments from the viewpoints of the performance and human workload.
关键词：KeywordsHuman-robot collaborationsGaussian process regressionRobot controlTeleroboticsHuman-centered designLearning control