A systematic literature review of computer vision applications in robotized wire harness assembly

Artikel i vetenskaplig tidskrift, 2024

In the current automotive industry, human operators perform wire harness assembly manually, which causes significant quality, productivity, safety, and ergonomic problems. Robotic assembly is a critical facilitator in addressing these problems. However, it remains challenging to implement for robotizing the assembly of wire harnesses. Wire harness assembly is a specific scenario of deformable linear object manipulation. Robotizing this assembly task demands robots to flexibly adapt their actions to the dynamically changing industrial environment based on robotic perception results. Existing research suggested the significance of robotic visual perception in the robotic assembly of wire harnesses. Implementing computer vision techniques is fundamental to enabling robots’ visual perception capabilities. Nonetheless, the industry has yet to introduce vision–based solutions to robotize wire harness assembly fully or partially. Through a systematic literature review, this article identifies fifteen scientific publications in vision–based robotized wire harness assembly. The results show various computer vision applications regarding wire harness components and assembly operations studied in previous research. Nevertheless, this article recognizes two significant challenges for computer vision applications in robotized wire harness assembly: (1) fulfilling production requirements on robustness and practicality and (2) exploiting the intrinsic physical features of wire harnesses for visual recognition. This article also advocated five prospective research directions toward more efficient and practical vision-based robotized wire harness assembly: (1) developing learning-based vision systems to exploit intrinsic features and multi–modality data of wire harnesses; (2) adapting vision systems proposed for robotizing assembly operations in manufacturing wire harnesses; (3) assessing the practicality, robustness, reliability, and sustainability of vision systems; (4) inquiring vision–based human–robot collaboration; and (5) exploring new product designs for facilitating visual recognition.