Abstract
In an environment, where a human and a robot coexist, it is necessary to control the robot corresponding to the change in the object status by means of tactile sensation. On the other hand, Tacit Learning (TL) is formulated based on creature’s nerve systems to adapt itself to various environments. In this research, our objective is to develop a TL system capable of performing object grasping and tracing motion, which is applied to a 2-linkfingered hand equipped with tactile sensors. Since the output is limited to either positive or negative in the original TL, both positive and negative outputs are required to control the joint of the robot. We assumed two kinds of clusters (artificial neurons’ cluster) specializing in output of positive and negative values, and newly introduced an algorithm for determining which cluster is used according to the positive or negative value of the sensor. In a series of simulation experiments, we conducted two tasks: one of them was tracing a surface of a circular object with one finger; the other was picking up an object with two fingers. Simulated results show that our system can follow a suitable contour of the object and that its grasping force is adequately adapted according to shearing force direction and magnitude.
