The aim of this thesis is to develop a robotic seam tracking system equipped with a visual range finder. The visual range finder, which consists of a CCD camera and a diode laser system with line generating optics, developed to recognize the types of weld joints and detect the 3-dimensional location of weld joints. In practical applications, however, images of the weld joints are heavily degraded due to spatters, arc glares, surface specularity, and welding smoke. To overcome the problem, this thesis proposes a syntactic approach which is a class of artificial intelligence techniques. In the approach, the types of weld joints are inferred based upon the production rules which are linguistic parsing rules consisting of a set of line and junction primitives of laser stripe image projected on weld joint. The production rules eliminate several noisy primitives to create new primitives through the merging process of primitives. After the recognition of weld joint, arc welding is started and the location of weld joints is repeatedly detected using a spring model-based template matching in which the template model is a by-product of the recognition process of weld joint. To show the effectiveness of the proposed approach, a series of experiments on joint type identification and robotic tracking were conducted for four different types of weld joints. The results show that the proposed method is very robust to visual noise and variations in weld joint conditions.