Learning Task-Parameterized Skills From Few Demonstrations

Jihong Zhu; Michael Gienger; Jens Kober

doi:10.1109/LRA.2022.3150013

Learning Task-Parameterized Skills From Few Demonstrations

Jihong Zhu^*, Michael Gienger, Jens Kober

^*Corresponding author for this work

Electronic Engineering

Research output: Contribution to journal › Article › peer-review

Abstract

Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task-parameterized learning improves the generalization of motion policies by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demonstrations in different situations. To comprehensively create different situations is non-trivial thus renders the method less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. This paper presents a novel concept to augment the original training dataset with synthetic data for policy improvements, thus allows learning task-parameterized skills with few demonstrations.

Original language	English
Pages (from-to)	4063-4070
Number of pages	8
Journal	IEEE Robotics and Automation Letters
Volume	7
Issue number	2
Early online date	11 Feb 2022
DOIs	https://doi.org/10.1109/LRA.2022.3150013
Publication status	Published - 1 Apr 2022

Bibliographical note

Publisher Copyright:
© 2016 IEEE.

Keywords

Imitation learning
learning from demonstration
physically assistive devices

Access to Document

10.1109/LRA.2022.3150013

Cite this

@article{ad589154a67f4a0b962a59135fc0519d,

title = "Learning Task-Parameterized Skills From Few Demonstrations",

abstract = "Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task-parameterized learning improves the generalization of motion policies by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demonstrations in different situations. To comprehensively create different situations is non-trivial thus renders the method less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. This paper presents a novel concept to augment the original training dataset with synthetic data for policy improvements, thus allows learning task-parameterized skills with few demonstrations.",

keywords = "Imitation learning, learning from demonstration, physically assistive devices",

author = "Jihong Zhu and Michael Gienger and Jens Kober",

note = "Publisher Copyright: {\textcopyright} 2016 IEEE.",

year = "2022",

month = apr,

day = "1",

doi = "10.1109/LRA.2022.3150013",

language = "English",

volume = "7",

pages = "4063--4070",

journal = "IEEE Robotics and Automation Letters",

issn = "2377-3766",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

number = "2",

}

TY - JOUR

T1 - Learning Task-Parameterized Skills From Few Demonstrations

AU - Zhu, Jihong

AU - Gienger, Michael

AU - Kober, Jens

PY - 2022/4/1

Y1 - 2022/4/1

N2 - Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task-parameterized learning improves the generalization of motion policies by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demonstrations in different situations. To comprehensively create different situations is non-trivial thus renders the method less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. This paper presents a novel concept to augment the original training dataset with synthetic data for policy improvements, thus allows learning task-parameterized skills with few demonstrations.

AB - Moving away from repetitive tasks, robots nowadays demand versatile skills that adapt to different situations. Task-parameterized learning improves the generalization of motion policies by encoding relevant contextual information in the task parameters, hence enabling flexible task executions. However, training such a policy often requires collecting multiple demonstrations in different situations. To comprehensively create different situations is non-trivial thus renders the method less applicable to real-world problems. Therefore, training with fewer demonstrations/situations is desirable. This paper presents a novel concept to augment the original training dataset with synthetic data for policy improvements, thus allows learning task-parameterized skills with few demonstrations.

KW - Imitation learning

KW - learning from demonstration

KW - physically assistive devices

UR - http://www.scopus.com/inward/record.url?scp=85124711729&partnerID=8YFLogxK

U2 - 10.1109/LRA.2022.3150013

DO - 10.1109/LRA.2022.3150013

M3 - Article

AN - SCOPUS:85124711729

SN - 2377-3766

VL - 7

SP - 4063

EP - 4070

JO - IEEE Robotics and Automation Letters

JF - IEEE Robotics and Automation Letters

IS - 2

ER -

Learning Task-Parameterized Skills From Few Demonstrations

Abstract

Bibliographical note

Keywords

Access to Document

Other files and links

Cite this