【正文】 wide, and long. The 3D vision sensor is attached to the manipulator to scan from different viewpoints by the motion of the manipulator. A vacuum is used to suck the fruit into the sucking pipe of the end effector. Cherry trees cultivated by the method of single trunk training distribute their fruits around the main trunk. In order to harvest a fruit while avoiding obstacles, such as leaves and trunks, the end effector needs to approach the fruit from the outside of the trunk. For this reason, in this study, we manufactured an articulated manipulator with an axis of updown traverse and three axes of rightleft turning, so that the fruits could be harvested in any direction (Fig. 2). The updown traverse requires paratively large force caused by the gravity. Therefore, it is driven by an AC servomotor (Yaskawa Electric, SGMAH01BAA2C, rated power 100W, rated torque , rated speed 3000min?1) and a screw mechanism (lead 10mm). Three axes of the right–left turning do not require large torque. Axes of the ?rst and second right–left turning are driven by small AC servomotors (Yaskawa Electric, SGMAHA5BAA21, rated power 50W, rated torque m, rated speed 3000min?1) and harmonic reduction gears (reduction gear ratio 100). The remaining axis of right–left turning is driven by a small DC motor with reduction gears. The manipulator is designed to be able to move round the circumference of the tree trunk so that not 3 only fruits on the front side of the trunk but also the fruits on the other side of the trunk could be harvested. Since the fruits are located around the tree trunk, if the vision sensor scans from one viewpoint, fruits beyond the trunk are hidden. To scan from different viewpoints, the 3D vision sensor was attached to the second arm. The movement of the manipulator changed the location and direction of the 3D vision sensor so that the dead angle bees small. 4 The 3D vision sensor is equipped with a light projector, a photo detector, and a scanning device (Fig. 3). The light projector consists of an infrared laser module, a red laser module, cold mirrors, a half mirror, and two fullre?ecting mirrors. The photo detector consists of two PSDs, a lens, and a red optical ?lter that weakens the in?uence of sunlight. The scanning device consists of a galvanometer scanner and a stepping motor. The galvanometer scanner scans vertically and the stepping motor scans horizontally. Red and infrared laser beams are united in the same optical axis by a cold mirror that transmits infrared light and re?ects visible right. The beam is further split into two beams (each still including both wavelengths) by a half mirror. These two beams scan the object simultaneously. Light of the two beams re?ected from the object is focused onto two PSDs. The distance from the 3D vision sensor to the object is calculated by a triangulation method using the ratio of the currents of both electrodes of the PSDs. The laser beams emit blinking signals in order to eliminate the in?uence of sunlight. By this method, re?ected light is separated from the sunlight, thus resulting in continuous light. Infrared light with wavelengths about 700–1000 nm is re?ected well by all parts of the cherry tree. On the other hand, red light at about 690 nm is not re?ected well by unripe fruit, leaves, and stalks, but is re?ected well by red ripe fruit. In this study, an infrared light beam of 5 830 nm and a red light beam of 690 nm were used. The infrared laser beam (830 nm) measures the distance to each part of the cherry tree from the 3D vision sensor and the red laser beam(690 nm) detects the red fruit to be harvested. As mentioned above, the laser beam is split into two beams. The 3D vision sensor scans these two beams simultaneously, and two pixels were measur