Recognition of 3D objects by a 3-fingered robot hand equipped with tactile and force sensors

This paper presents an algorithm that recognizes and localizes 3D objects using a 3-fingered robot hand, where an optical tactile sensor and a force sensor are mounted on each finger. Both sensors are capable of measuring the position and normal vector of the test object at the contact point. For efficient matching, the objects are represented by a distribution graph of surface description vectors and a hierarchical table. The measurements of a position and an orientation are described by a possibility sphere and a possibility cone, respectively, whose sizes represent the error characteristics of sensors. The matching object models are selected by fusing sensory data based on theses cones and spheres. When there exist multiple matching object models, the next sensing pose is selected in the multiple interpretation image so that the next sensing operation can discriminate as many remaining object models as possible. The use of hierarchical tables and possibility cones simplifies the matching and the determination of a next sensing pose