Deep Learning-Based Trajectory Planning for Reflective Workpiece Alignment in Robotic Welding

Recent advancements in robotic welding and vision-aided automation have transformed industrial processes by providing high precision, rapid detection, and adaptability compared to traditional methods. This study introduces a trajectory generation approach based on Deep Iterative Matching, leveraging RGB-D data from industrial cameras (Zivid 2) to calculate welding paths, initial positions, and correct placement errors. A real dataset of aluminum workpieces was created to train deep neural networks (DNNs) to handle measurement corruption caused by reflections. By integrating intelligent solutions, this research aims to improve automation reliability, repeatability, and efficiency, advancing welding technologies in complex industrial environments.