Abstract: In practical production, robotic welding is mostly controlled by "teaching-reproduction" and "offline programming". Both of these methods lack the ability to recognize changes in actual working conditions, resulting in poor welding quality and reduced welding efficiency. The adaptive GMAW technology and corresponding equipment were developed based on the weld vision sensing system, and the general intelligent welding software was developed. The contour of the workpiece was extracted by a linear structured light sensor. The results show that the vision sensing system can conduct the global 3D scanning of the workpiece, obtain the 3D point clouds and intelligently identify the weld position. By combining the welding process parameters database, the workpiece identification and positioning, welding task decomposition and welding task execution based on three-dimensional stereovision are finally realized.