KI-basierte Detektion von Komponenten und Segmenten in Leitungssätzen

Abstract

Wiring harnesses are branched deformable linear objects with complex structures that pose challenges for robotic automation, particularly in determining suitable gripping points for path planning. This article investigates AI-based detection of wiring harness components and segments to address these challenges. A custom wiring harness, modeled after a door wiring harness, is designed to generate datasets for training neural networks. Object detection is used to identify branches and connector types, segment detection involves solving the sub-tasks of segment segmentation, classification into two classes, and topological assignment. Several established frameworks are used for these tasks. Optimal training parameters are determined through grid search.