Determining robot grasping-parameters by estimating "Picking Risk"

The momentum for developing home service robots against the background of a declining birthrate and aging society has been increasing in recent years. As there are various kinds of objects in homes, it is difficult for home service robot to pick target objects because they may be surrounded by other unrelated objects. This is because a robot must grasp target objects safely without hitting or breaking them. These objects are "obstacles" for the robot. This paper proposes a method of determining the optimal grasping parameters for robots in such complicated environments where objects are densely located. All objects in an environment in which a robot operates can generally be divided into three types: a target object, obstacles, and the robot hand itself. Therefore, the "risk" of colliding with an object, called the "picking risk" is computed by estimating the relative collision status of these objects. In addition, the optimal parameters can be determined by minimizing the picking risk. We tested and confirmed that the grasping success rate was 82% in experiments that used 47 depth images of daily objects such as cups or spoons.