Learning and Correcting Robot Trajectory Keypoints from a Single Demonstration

Iñigo Iturrate*, Esben Hallundbæk Østergaard, Martin Rytter, Thiusius Rajeeth Savarimuthu

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

Kinesthetic teaching provides an accessible way for non-experts to quickly and easily program a robot system by demonstration. A crucial aspect of this technique is to obtain an accurate approximation of the robot’s intended trajectory for the task, while filtering out spurious aspects of the demonstration. While several methods to this end have been proposed, they either rely on several demonstrations or on the user explicitly indicating relevant trajectory waypoints. We propose a method, based on the Douglas-Peucker line simplification algorithm that is able to extract the notable
points of a trajectory from a single demonstration. Additionally, by utilizing velocity information in the task space, the method is able to achieve a level of precision that is sufficient for industrial assembly tasks. Along with this, we present a user study that shows that our method enables non-expert robot users to successfully program the robot for an assembly benchmark, and that they find this method intuitive.
Original languageEnglish
Title of host publicationProceedings of the 3rd International Conference on Control, Automation and Robotics
PublisherIEEE
Publication date2017
Pages52-59
ISBN (Print)978-1-5090-6089-4
ISBN (Electronic)978-1-5090-6088-7, 978-1-5090-6086-3
DOIs
Publication statusPublished - 2017
Event3rd International Conference on Control, Automation and Robotics - Nagoya, Japan
Duration: 24. Apr 201726. Apr 2017
Conference number: 3

Conference

Conference3rd International Conference on Control, Automation and Robotics
Number3
CountryJapan
CityNagoya
Period24/04/201726/04/2017

Fingerprint Dive into the research topics of 'Learning and Correcting Robot Trajectory Keypoints from a Single Demonstration'. Together they form a unique fingerprint.

Cite this