<p>In a human-robot interaction, a robot has to move to a place where the robot can obtain precisely information of human such as positions, postures, and voice. It is because the human recognition accuracy depends on positional relation between human and the robot. In addition, the kind of necessary sensor information changes depending on the interaction task content. Therefore, we should change a path for improvement of the human recognition accuracy to for easiness to perform tasks. Accordingly, we need to design a path planning method considering some sensor characteristics, the human recognition accuracy and tasks simultaneously. Consequently, we define "Recognition Reliability" to evaluate the human recognition accuracy quantitatively. And this paper presents a path planning method using the potential field that integrates multimodal information such as the sensor characteristics, the recognition reliability and the task contents simultaneously.</p>