Toward Enabling Robotic Visual Perception for Assembly Tasks

Licentiate thesis, 2024

Industry faces an urgent need for prospective solutions to scale up assembly automation, a challenge that requires immediate attention. In contemporary manufacturing, industrial robots need more intelligence to qualify for increasingly demanding flexible automation tasks. Research in artificial intelligence, computer vision, and robotics paints a promising picture of the future, where intelligent robots play a significant role in fostering sustainable and resilient manufacturing. However, academia and industry have yet to realize the potential of intelligent robots in production fully.

This thesis plays a pivotal role in advancing the development of intelligent robots for flexible automation tasks, a crucial area of research in automation and robotics. Toward this goal, this thesis investigates perception, a prerequisite of intelligence, and mainly focuses on visual perception, a critical contactless perception approach. A multi-method research approach, comprising a qualitative literature study and a quantitative experimental study, was adopted to explore the challenges and prospective technical solutions to enabling robotic visual perception for assembly tasks.

The research has identified four key challenges in enabling robotic visual perception for assembly tasks, particularly in developing and integrating vision systems in practical production. Additionally, the research has proposed six prospective directions for developing technical solutions, focusing on computer vision algorithms, dataset and benchmark, practical evaluation, human-robot collaboration, and product design.