In response to the challenges about the low efficiency of welding in the aerospace field, particularly the complex intersection line welding, this paper proposes a teaching-free autonomous welding method for robots based on an optimized RANSAC and ICP point cloud algorithm for intersecting lines. First, based on the intersecting line theoretical model, the point cloud data of the collected workpiece is preprocessed. The optimized RANSAC algorithm is then used to select a set number of points as sample points to fit a surface, calculating the distance from all data points to the cylinder. Points that fall on the fitted surface are selected according to a preset threshold, and a new cylindrical surface equation is fitted using the selected points. Next, the ICP algorithm iteratively searches for the closest point pairs between two point clouds, calculating the optimal rigid transformation to align the source point cloud with the target point cloud in the target coordinate system. Finally, experimental results validate the proposed teaching-free autonomous welding method for complex spatial curves, using the optimized RANSAC and ICP point cloud optimization algorithms.