A Collaborative Industrial Robotic System with Machine-Learned Trajectory Planning and Control

The manufacturing industry is undergoing a paradigm shift from mass production to mass personalisation, driven by increasingly diverse customer demands. This transition necessitates the development of smart manufacturing systems capable of adapting to customised products and varying production levels. To support such a transition, a hybrid manufacturing mode where human collaboratively works with industrial robots is proposed, which combines the high precision, repeatability and strength of robots with the flexibility and adaptability of human workers. This thesis presents a system implementation for collaborative industrial robots that addresses critical gaps in current industrial automation systems. By enhancing robots' ability to understand, predict, and adapt to human behaviours while maintaining high levels of precision and efficiency. This work paves the way for more flexible, safe, and productive human-robot collaboration in smart manufacturing settings. This developed collaborative industrial robots control system utilises advanced machine learning technologies, focusing on three key components: The cyber-physical workstation provides a real-time virtual representation status and planned actions of the collaborative robt and other components, demonstrated its effectiveness in visualising complex grasping and handover tasks, enhancing human operators' understanding of robot behaviour. The object 6D pose estimation pipeline employs advanced computer vision techniques to recognise and accurately determine the position and orientation of any objects based on their CAD models, outperformed state-of-the-art methods, reaching an average recall of 75.8 on a widely used public dataset. The proactive trajectory planning algorithm utilises reinforcement learning to predict human arm movements within the shared workspace, enabling the robot to dynamically adjust its trajectory and avoid collisions. It achieved a 96.5% success rate in collision avoidance across a diverse test set of 3,675 scenarios. To validate the practical applicability of the implemented system, comprehensive case studies were conducted which integrated all three components into a cohesive machine-learned control system. The findings and content of some part of thesis has been published or documented in international conferences and journals, underscoring the contribution of this work made to the society.

Name thesis:degree_name

PhD

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Mechanical and Mechatronics Engineering

Grantor dc:publisher

ResearchSpace@Auckland

Year dc:date.issued

2024

Author dc:creator

• Xia, Wanqing

Advisors dc:contributor.advisor

• Xu, Xun
• Xu, Weiliang

• Human-Robot Collaboration
• Smart Manufacturing
• Mass Personalisation
• Reinforcement Learning
• Computer Vision
• Object 6D Pose Estimation
• Proactive Trajectory Planning
• Collision Avoidance
• Machine Learning

Statement dc:rights

• Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Licence dc:rights.uri

• https://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm

Handle dc:identifier.uri

https://hdl.handle.net/2292/72760

OAI identifier oai:identifier

oai:researchspace.auckland.ac.nz:2292/72760