Recognizing Cylindrical Surface Using Impedance Perception

In a previous paper, we proposed the impedance perception technique, by which the stiffness matrix that constrains the motion-force relation of the robot's end-effector is estimated on-line, and the uncertainties of the estimates are evaluated. Based on this technique, this paper proposes a method of extracting information on local properties of a cylindrical curved surface, including normal direction, primary directions, curvature, and stiffness and friction coefficients, from the stiffness matrix obtained under the situation where the end-effector is slid on the surface. This technique can be implemented as an encapsulated perception function independent from control strategies, and thus it can be used for both autonomous and remote-controlled robots, and for direct monitoring of human manipulations. Results of preliminary experiments are presented.