Run-time Self-classification of External Force using Virtual Point Mass Approximation for Object Manipulation

Abstract

Many tasks assigned to a manipulator include interactions with its operating environment or manipulating objects, which are detected as forces and moments by the force sensor. It is, however, not easy to detect when a pure external wrench occurred in interactions or manipulations since signals measured by the force sensor consist of the inertial effect of the end-effector and manipulating objects as well as the effect of interactions. In order to separate these combined effects, a self-classification method for the 6-axis force sensor is proposed in this paper by relating the wrench and the virtual point mass. With the proposed method, wrenches due to the end-effector and objects can be classified in run-time without any prior information for them, and thus a pure external wrench can also be distinguished from them. The effectiveness of the proposed self-classification method is verified through experiments.