ニューラルネットによる非接触型インピーダンス制御のパラメータ学習

抄録

Although an impedance control is one of the most effective control method for a manipulator in cntact with its environment, the end-effector of the manipulator does not move until an external force is exerted. In order to control the manipulator motion before contact with the environment, a non contact impedance control has been proposed, which can regulate not only the end-point impedance but also virtual impedance between the manipulator and the environment using visual information. The characteristics of the non-contact impedance control, however, is determined by the virtual impedance which must be designed according to given tasks. The present paper proposes a method to regulate the virtual impedance parameters using neural networks by reducing an energy function during iterative execution of the given task. The proposed method is implemented using a direct-drive robot in a planar task space to show effectiveness of the method.