In this paper, we demonstrate through user studies that mobile robot trajectories that imitate human-to-human approach trajectories are perceived more socially acceptable in the face-to-face interaction scenario than those imitating point-to-point trajectories. We generate robot trajectories to/from a human standing at an arbitrary location by applying inverse optimal control to a human-to-human trajectory dataset. The cost function used in a previous work for modeling human point-to-point trajectories does not represent human-to-human trajectories due to the circular paths often observed around the target human. We therefore propose a new cost function motivated by the social force model. The user study confirms that the resulting trajectories are more preferred with statistical significance than baseline.