Abstract In order to improve the welding quality of robotic gas metal arc welding (GMAW), a welding system for seam tracking is developed based on a purpose-built vision sensor. By analyzing the features of robotic GMAW, a novel software program is developed for welding seam tracking. The program includes specific modules, such as the welding power control, the intelligent parameter setting, the image capturing and processing based on improved algorithms, the welding expert database, robot communication and path planning modules. To evaluate the feasibility of the developed seam tracking method, the accuracy and the real-time nature are verified using experiments on different types of typical weldments. The results prove that the proposed welding seam tracking method is able to achieve a good tracking accuracy for most welding applications.