Two-wheeled service robots capable of following a person are an active research topic. These robots will support and follow customers like tourists to many places. It is necessary that an appropriate distance be maintained between the human and the robot. In addition, robots need to not only approach but also turn toward the human to provide services when the human stops walking. Therefore, the control system should change its property depending on the situation. However, many of the previous researches report an algorithm having only one property. Thus, this research proposed a motion control system of a two-wheeled service robot that could turn around while simultaneously following a human. To achieve this, we use nonlinear model predictive control (NMPC) and evaluation function weights depending on the relative distance between robots and human.