Learning of robot tasks via impedance matching

摘要

The paper is aimed at presenting a physical interpretation of practice-based learning (so-called "iterative learning control") for robotic tasks from the viewpoint of "bettering impedance matching". At first, the concepts of impedance and impedance matching that are inherent to linear electric circuits are generalized for a class of nonlinear dynamics including robotic tasks by means of passivity. It is then shown in the simplest case when the tool endpoint is free to move that a simple iterative scheme of learning enables robots to make a progressive advance in a sense of zero-impedance matching at every trial of operation. In the case of impedance control when a soft and deformable finger-tip presses a rigid object or environment, it is shown that, for a given desired periodic force, physical interaction between the soft fingertip and the rigid object, the robot learns steadily the desired task by monotonously increasing the grade of impedance matching pertaining to dynamics of the robot task with controller dynamics.