A 3D Reconstruction and Relocalization Method for Humanoid Welding Robots

Abstract

Welding robots represent pivotal equipment in intelligent welding for manufacturing and maintenance. Presently, most welding robots are stationary single-arm units, exhibiting limited flexibility and efficiency, thereby compromising welding quality and productivity. Consequently, there is an urgent need to develop a new generation of humanoid welding robots (HWR) endowed with autonomous mobility and dual-arm collaborative capabilities. Key to this advancement are pose estimation and three-dimensional (3D) reconstruction methods, which traditionally focus on mapping and navigating unfamiliar environments, often struggling to adapt to the routine welding and maintenance scenes of large-scale equipment. This paper introduces a novel approach to 3D reconstruction and relocalization tailored for HWR, facilitating rapid localization of welding areas and transmission of point cloud maps. Initially, a vision-based 3D reconstruction system is proposed, encompassing pose estimation, 3D reconstruction, and target detection, enabling self-localization and precise targeting for HWR. Subsequently, a novel method for 3D point cloud map segmentation based on 2D features and 3D point clouds matching is introduced to expedite the transmission of point cloud maps. Finally, a relocalization and point cloud map updating method grounded in prior knowledge is proposed, facilitating seamless welding operations by HWR in routine maintenance scenes. The effectiveness and superiority of the proposed methodology are validated through comparative tests with existing methods using actual HWR.