A spatial path generation algorithm based on visual sensory data and its application to the seam tracking robotic system which can operate even under poor sensing conditions are presented. At first, the concept of the robotic system with a wrist-mounted high performance vision sensor is presented, then an accurate spatial path generation algorithm using sensory data with a reliability coefficient is described. In the algorithm, piecewise polynomial functions are generated from sensory data that is acquired ahead of an end-effector, and the spatial path is recovered by connecting the piecewise polynomial functions successively under the minimum error condition as well as the boundary conditions. Computer simulations and experiments on a spatial path on a curved surface show that the proposed system operates effectively even under disturbances such as sensing noise, irregular tack-welding-beads and a burst of lack of sensory data, often appearing in practical applications.