Study on an algorithm for grasping an unknown object by robotic hand

点击翻译按钮获取中文摘要

When a robotic hand approaches an unknown object and grasps it stably, there are two primary considerations from the viewpoint of sensing. One is to detect contact location between robotic hand and unknown object and the other is to detect the surface geometry of object. The contact point provides a good starting point for sensing the surface geometry. The main results through this research grant are summarized below: 1. Contact point sensing With a proper combination of compliant joint and position-controlled joint, a link system can change its link posture while maintaining contact between the link and an object. It was shown that by selecting two different link posture during posture changing motion, we can compute an intersecting point which leads to an approximate contact point. When a link hits an object with relatively high speed, dynamic effects appears and as a result, the link system may bounce at the point of collision. It was also shown that even under such a case, we can obtain an approximate contact point if the link again contacts with the object with a different link posture. We also showed that an exact contact point is obtained if we can assume a finger-tip tactile sensor. 2. Object shape sensing A finger-tip tactile sensor was mounted at the tip of the newly developed three-jointed robot. The proposed sensing algorithm is composed of two parts, namely, one is to move the sensor along the tangential direction of object, and the other is to maintain the sensor attitude so that the most sensitive area of the sensor can always contact with the object to be detected.