Self-adaptive seam detection framework for unmanned structural steel welding robots in unstructured environments

Abstract

Existing vision-based seam detection frameworks for structural steel welding robots perform well in static, predefined workspaces but struggle in dynamic, unstructured real production environments. Therefore, a self-adaptive seam detection framework is proposed that enables welding robots to interpret the actual welding environment through an estimation of the target workpiece pose. An enhanced pose estimation algorithm is developed to generate a reliable workpiece pose estimation based on solely one RGB image. Based on this real-time workpiece pose, the system autonomously determined the necessary robot movements toward an optimal position and orientation for subsequent high-precision structured light sensor measurements. The seam extraction and welding trajectory planning were then completed through an automated process. Experimental results demonstrate that the proposed framework not only enables cost-effective, fully automated weld seam detection in dynamic unstructured environments, but also achieves 72 % higher efficiency than conventional methods while eliminating human intervention.