As demand for production automation increases, there is a growing shortage of workers to teach robots how to move. Industrial robots are equipped with a look-ahead control system to avoid increasing motion time due to sequential stops at each taught pose, but it is difficult to predict the actual path of the robot, and skilled experience is required for use. Therefore, we propose a method for generating teaching poses that allows the robot to move under the look-ahead control of an industrial robot while matching the target path in a short time. We focused on the fact that the actual path under look-ahead control does not satisfy the taught transit poses, and that the actual path approaches a linear path when taught without transit poses. As a result of verification using an actual robot, it was shown that the actual path matched the target path better than with conventional technology, and the motion time was reduced by up to 92% from 11.3 [s] to 0.9 [s]. Furthermore, by reducing the number of teaching poses, it is also possible to reduce the motion time and the maintenance man-hours.