人工触覚による対象物形状の特徴抽出

抄録

This paper deals with a simple tactile sensor model for extracting the feature of an object's surface and the experiment for classifying the shapes of grasped objects.
The proposed model of tactile sensor consists of two layers: the surface layer and the sensor layer. The surface layer is made of an elastic material of which the mechanical properties are determined by the object's surface. These properties of the surface layer are expressed theoretically as the blurring function. It is important that the operation of blurring function is considered as an analogical concept to the receptive field in the visual sense.
The sensor layer derives information from the multi-element sensors which are placed at equal intervals.
The experiment on object classification by one time grasping was performed by using an artificial hand with the tactile sensors. The artificial hand with the five fingers is structurally identical to the skeleton of the human hand. Moreover, the joints of finger are independently actuated by micro-pneumatic cylinders except the first joint of the index finger which is controlled by a positional servo mechanism.
The grasping process from the initial stage to the final stage of grasping the object is computer-controlled to follow the given sequence of a pattern of finger joints. Upon contact of tactile sensors with the object's surface when the object are tightly grasped by the fingers, the computer receives the data of sensor outputs and determines the types of surface at the contact point. Moreover, by the object's surface feature, classification is performed by a learning machine.