Vision-Guided Cobot-Assisted Adaptive Metal Sheet Bending

Abstract

Metal forming processes are significantly influenced by variations in sheet material properties and tool wear, which pose challenges to the repeatability of automated systems that rely on pre-set, rigid (i.e., non-adaptive) models. While skilled human operators can compensate for these variations by leveraging advanced visual and force feedback, automated solutions often lack real-time adaptability. Existing closed-loop control systems have demonstrated significant potential in addressing challenges such as variable springback. However, their high hardware costs have hindered widespread industrial adoption, particularly in adaptive bending applications. This study proposes a cost-effective hardware architecture that integrates a conventional predictive model with a 2D vision system and a collaborative robot (cobot) to overcome these barriers. By reducing the need for extensive material characterization, the system enables plug-and-produce functionality and ensures consistent output through real-time feedback and control, requiring minimal setup. Deployed in an industrial setting, the system evaluated multiple vision-based strategies, with ongoing research focusing on a marker-based, CAD-driven approach. Experimental results revealed an 80% reduction in processing time. Moreover, the proposed system achieved a final bend angle deviation of ±0.05°, significantly outperforming the 0.1°–0.5° error of human operators.