Abstract: To address the challenge of intelligent upgrading of robotic welding, this study innovatively proposes an autonomous optimization strategy for welding parameters based on visual perception of groove morphology, focusing on enhancing the dynamic adaptability of the welding system to large-scale grooves. By constructing a geometric constraint-driven swing model and integrating a laser vision sensor to achieve real-time capture of groove topography, the system can dynamically adjust the swing trajectory based on groove characteristics. The study used a dual verification method of comparative experiments and contour accuracy analysis, and the results showed that the swing model effectively reduced the heat density of the weld, while accurately controlling the weld width within the range of 11.0-12.1 mm, significantly improving the stability and forming quality of the welding process.