Reinforcement learning framework for window hardware installation

• Author / Creator

  Huang, Tzu-Hao

• The repetitiveness and precision of manufacturing tasks has increased the need for robots in the automation of the manufacturing industry; however, the complex and varied nature of manufacturing production lines poses challenges in terms of applying the rule-based automation approach. This has contributed to the trend of employing artificial intelligence-driven robotic systems. Previous applications relied on accomplishing a single automated task with its specifically designed AI model, and thus failed to provide a scalable AI solution that could accomplish a variety of tasks. In the present case study, window hardware installation was simulated with a reinforcement learning solution using the soft actor critic algorithm to improve model learning efficiency and scalability. Agent training techniques, such as rewarding shaping and curriculum training, were introduced into the model’s learning configuration. The proposed curriculum guided reinforcement learning structure provides the training agent a gradual and effective way to comprehend assigned tasks. The proposed approach further increases the potential for artificial intelligence-driven robotic hardware installation systems to be more sensitive and flexible to environment change and target hardware variations with the implementation of training guidelines. The model’s performance is demonstrated by refined self- driven motion planning and hardware sensitive decision-making. This application indicates that a robust and scalable artificial intelligence model can be realized by thoughtful agent incentives and learning pathways, and the case study demonstrates the framework required to facilitate such a possibility.

• Subjects / Keywords

  • Robotic
  • Window manufacturing
  • Reinforcement learning
  • Curricular learning

• Graduation date

  Spring 2022

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-gy5z-cg07

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.