Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

An agent-based reinforcement learning approach to improve human-robot-interaction in manufacturing

Oliff, Harley ORCID: https://orcid.org/0000-0003-4154-8737 2020. An agent-based reinforcement learning approach to improve human-robot-interaction in manufacturing. PhD Thesis, Cardiff University.
Item availability restricted.

[thumbnail of PhD]
Preview
PDF (PhD) - Accepted Post-Print Version
Download (12MB) | Preview
[thumbnail of Cardiff University Electronic Publication Form] PDF (Cardiff University Electronic Publication Form) - Supplemental Material
Restricted to Repository staff only

Download (3MB)

Abstract

This work is aimed at the understanding and application of several emerging technologies as they relate to improving the interactions which occur between robotic operators and their human colleagues across a range of manufacturing processes. These interactions are problematic, as variation in performance of human beings remains one of the largest sources of disturbances within such systems, with potentially significant implications for productivity if it continues unmitigated. The problem remains for the most part unaddressed, despite these interactions becoming increasingly prevalent as the rate of adoption of automation technologies increases. By reconciling multiple areas encompassed by the wider domain of intelligent manufacturing, the presented work identifies a methodology and a set of software tools which leverage the strengths of neural-network-based reinforcement learning to develop intelligent software agents capable of adaptable behaviour in response to observed environmental changes. The methodology further focuses on developing representative simulation models for these interactions following a pattern of generalisation, to effectively represent both human and robotic elements, and facilitate implementation. By learning through their interaction with the simulated manufacturing environment, these agents can determine an appropriate policy, by which to autonomously adjust their operating parameters, as a response to changes in their human colleagues. This adaptability is demonstrated to enable the intelligent agents to determine an action policy which results in less observed idle time, along with improved leanness and overall productivity, over multiple scenarios. The findings of the work suggest that software agents that make use of a reinforcement based learning approach are well suited to the task of enabling robotic adaptability in such a way, and the developed methodology provides a platform for further development and exploration, along with numerous insights into the effective development of these agents.

Item Type: Thesis (PhD)
Date Type: Completion
Status: Unpublished
Schools: Engineering
Uncontrolled Keywords: Intelligent Manufacturing; Reinforcement Learning; Human-Robot-Interaction; Agent-Based; Human Factors; Adaptability.
Date of First Compliant Deposit: 25 November 2020
Last Modified: 05 Nov 2022 03:48
URI: https://orca.cardiff.ac.uk/id/eprint/136620

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics