Towards learning from demonstration for industrial assembly

Hernandez Moreno, Victor

Towards learning from demonstration for industrial assembly

Hernandez Moreno, Victor

Permalink

Publication Type:: Thesis
Issue Date:: 2024

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (7.26 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Hernandez Moreno, Victor
dc.date.accessioned	2024-12-10T02:38:37Z
dc.date.available	2024-12-10T02:38:37Z
dc.date.issued	2024
dc.identifier.uri	http://hdl.handle.net/10453/182453
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	The industrial assembly sector is confronted with a confluence of challenges arising from demands for mass customisation, labour shortages, and competitive pressure. To address this unique combination, automation systems appear beneficial that can rapidly adapt to new tasks and be operated by non-technical personnel. The Learning from Demonstration (LfD) concept emerges as a promising solution, enabling non-expert operators to teach robots new tasks without the complexities of traditional programming. This concept also empowers robots to apply learnt tasks in changed situations, fostering flexibility for competitive solutions. Despite being a well-explored field in research, the effective deployment of LfD in industrial settings remains an unmet challenge. This thesis delves into the application of prominent LfD methods towards industrial assembly tasks and investigates how the concept can be leveraged to benefit this sector. Through a comprehensive analysis of LfD solutions in research and a comparison with industrial practices, key obstacles hindering the seamless integration of promoted solutions in an industrial environment are identified. These challenges include issues of practicability, task complexity and diversity, generalisability, performance evaluation, and integration concepts. With the goal of developing a framework that enhances applicability in the industrial assembly industry, this thesis promotes improvements in the three phases of characteristic LfD: human demonstration, robot learning, and robot reproduction. In contrast to the prevailing kinaesthetic teaching application, a guiding graphical interface is developed based on the Hausdorff approximation planner (HAP) framework, providing human operators with insights into the robot's kinematic constraints during demonstration. The robot learning phase is enhanced by combining the primarily employed trajectory-based Dynamic Movement Primitives (DMPs) method with the well-established Methods-Time Measurement (MTM-1) industrial taxonomy for extended generalisability across custom skills. Addressing challenges during reproduction in changing environments, a reactive control approach is presented that employs a novel multibody approximation scheme. This scheme informs potential fields generating wrenches of repulsion and attraction for robust reproduction given new environmental situations. A unified framework incorporating these novel methods is established and demonstrated through a physical demonstrator, allowing for real-world evaluation of the proposed methods. This thesis contributes a comprehensive framework to promote increased applicability of LfD for the industrial assembly sector, addressing key challenges in prevailing LfD research approaches, and providing a pathway towards effective deployment of robotic solutions in a competitive and evolving industrial landscape.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/182453/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2024 Victor Hernandez Moreno
dc.rights	au.edu.uts.lib/cph
dc.title	Towards learning from demonstration for industrial assembly	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

The industrial assembly sector is confronted with a confluence of challenges arising from demands for mass customisation, labour shortages, and competitive pressure. To address this unique combination, automation systems appear beneficial that can rapidly adapt to new tasks and be operated by non-technical personnel. The Learning from Demonstration (LfD) concept emerges as a promising solution, enabling non-expert operators to teach robots new tasks without the complexities of traditional programming. This concept also empowers robots to apply learnt tasks in changed situations, fostering flexibility for competitive solutions. Despite being a well-explored field in research, the effective deployment of LfD in industrial settings remains an unmet challenge. This thesis delves into the application of prominent LfD methods towards industrial assembly tasks and investigates how the concept can be leveraged to benefit this sector. Through a comprehensive analysis of LfD solutions in research and a comparison with industrial practices, key obstacles hindering the seamless integration of promoted solutions in an industrial environment are identified. These challenges include issues of practicability, task complexity and diversity, generalisability, performance evaluation, and integration concepts. With the goal of developing a framework that enhances applicability in the industrial assembly industry, this thesis promotes improvements in the three phases of characteristic LfD: human demonstration, robot learning, and robot reproduction. In contrast to the prevailing kinaesthetic teaching application, a guiding graphical interface is developed based on the Hausdorff approximation planner (HAP) framework, providing human operators with insights into the robot's kinematic constraints during demonstration. The robot learning phase is enhanced by combining the primarily employed trajectory-based Dynamic Movement Primitives (DMPs) method with the well-established Methods-Time Measurement (MTM-1) industrial taxonomy for extended generalisability across custom skills. Addressing challenges during reproduction in changing environments, a reactive control approach is presented that employs a novel multibody approximation scheme. This scheme informs potential fields generating wrenches of repulsion and attraction for robust reproduction given new environmental situations. A unified framework incorporating these novel methods is established and demonstrated through a physical demonstrator, allowing for real-world evaluation of the proposed methods. This thesis contributes a comprehensive framework to promote increased applicability of LfD for the industrial assembly sector, addressing key challenges in prevailing LfD research approaches, and providing a pathway towards effective deployment of robotic solutions in a competitive and evolving industrial landscape.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/182453