Recently, blueberries are becoming popular as functional food, and the cultivation area continues to increase in Japan. In America, a blueberry harvesting machine which shakes down blueberries is introduced. However, to harvest sweet high-value bluberries, it is necessary to pick by expert farmers' hands for estimating ripeness of each berry at the time of harvest. Therefore, we think that a robot which can pick ripe blueberries one by one like expert farmers do is necessary. In this paper, we propose a robot arm approach method of preventing the robot's end-effector from colliding with obstacles based on 3D vision and imgage processing, and we build a foundation of an automatic ripe blueberry harvesting system with a wheel moving arm robot.