Page 160 - Human Inspired Dexterity in Robotic Manipulation
P. 160

Dynamic Manipulation Based on Thumb Opposability  157


                 As shown in Fig. 8.6A, a conventional impedance controller adjusts the
              absolute impedance of the object. The object impedance can be replaced
              with the equivalent joint impedance to uniquely determine the equilibrium
                                            T
              point by using the transformation J K x J , where J x is the object Jacobian
                                            x    x
              and K x is the object stiffness. Meanwhile, as shown in Fig. 8.6B, the blind
              grasping is intuitively expressed as the spring connecting between the finger-
              tips to adjust for the relative force while it does not determine the absolute
              motion of the object. Thus, the proposed system is redundant and ill-posed.
              Therefore, the system has an infinite number of equilibrium points and the
              proposed controller might generate self-motion. The self-motion can be
              suppressed by the stability theory on manifolds [12]. Intuitively, this method
              uses an appropriate damping to maintain the state within a bounded region
              on the grasp manifold, as shown in Fig. 8.7, instead of an explicit null-space
              motion.
                 However, the real effect of the redundancy resolutions on grasp stability
              is uncertain. Therefore, we executed a comparison between the conven-
              tional impedance controller and the opposed controller with the KITECH
              robotic hand that has four fully actuated four-DOF fingers [5], as shown in
              Fig. 8.8. In both cases, the hand stably grasped an object at the equilibrium
              when no external force was applied. However, the hand showed different
              behaviors to external disturbances.
                 The joint impedance control preserves the position of both fingertips to
              grasp the object. However, when the left finger was pulled by the distur-
              bance, the distance between the fingers became larger than the object’s
              width and, as a result, the hand dropped the object.
                 Meanwhile, the blind grasping controller preserves the relative distance
              between fingers. When an external disturbance was added to one of the


                            Constraint manifold
                                                 Stable grasp
                                                 manifold

                                                  Lower priority task
                                                      manifold

                                            Neighborhood around
                                            nominal equilibrium

                              State space

              Fig. 8.7 Manifold.
   155   156   157   158   159   160   161   162   163   164   165