Arc sensor is indispensable to arc welding robot systems for compensating the joint misalignment such as mismatch of the workpiece, ill-conditioned positioner and thermal deformation during welding. Furthermore, the amount of these mismatches cannot be preivously expected, and changes from time to time. There are many kinds of seam trackers for correcting the welding path of the robot, where non-contact type sensors arc prevalently used in arc welding robot systems. In this study, an arc sensor was developed for GMA and FCA welding robot system. Since the arc sensor uses the arc characteristics during welding, the operating principle of the arc sensor must be adjusted according to the welding condition. Especially in GMA welding with the $CO_{2}$ shielding gas, the welding arc is not stable because of the short circuit and non-axial globular transfer mode of the molten droplet. In this study, the 2nd order least square curve fitting algorithm was adopted and the applicability of this algorithm was investigated for robot welding systems. For easy usage of the arc sensor, the operating parameters for arc sensor were limited to eight which can be easily determined by the operator.