On-Line Learning of Non-Contact Impedance and its Appliation to Contact Tasks

Impedance control is one of the most effective control methods for interaction between a manipulator and task environments. The force resulted by the interaction, however, does not occur until the end-effector of the manipulator touches its environment. A non-contact impedance control method has been proposed in order to achieve the impedance control before the contact. The method can regulate not only the end-point impedance but also the virtual impedance such that the end-effector is surrounded by a virtual object which can touch the environment before a real contact. This paper proposes a learning method using neural networks to regulate the virtual impedance parameters for given tasks. The validity of the method is verified through computer simulations and experiments with a conventional robot manipulator.