Page 165 - Human Inspired Dexterity in Robotic Manipulation

162 Human Inspired Dexterity in Robotic Manipulation

object is performed based on the approach vector obtained in the previous
step. At the next step of (3) grasping, the blind grasping introduced in
Section 8.3.3 was employed and executed by using of the grasp point and
type obtained from the grasping strategy engine. As shown in Fig. 8.9,it
is possible to apply the principle of superposition, which enables in-hand
manipulation such as translation and rotation as needed. After this step,
the autonomous robotic picking is completed by (4) lifting the grasped
1
object, and this experiment is presented with the video. We were able
to complete this experiment quickly and easily, with the advantage of blind
grasping, which can achieve stable-object grasping even with imperfect
object information including uncertainty.

8.5 DISCUSSION

This chapter introduced a passivity-based control method for grasping and
manipulation of a robotic hand inspired by thumb opposability. The
opposed motion can be generated by putting a virtual spring between the
fingertips. The physical observation of this opposed force revealed that
thumb opposability naturally generated a recovery motion disturbed from
the equilibrium point of stable grasp without the foreknowledge and sensing
of an object. Therefore, this controller is called blind grasping. The opposed
impedance of blind grasping played a vital role in stabilizing a grasped object.
The hand system has a redundant DOF because the controller has fewer
controlled DOFs than the control DOF. The underactuation usually has
an undesired property because to generate a self-motion. However, the
underactuation rather worked to increase the robustness of the system,
which was experimentally confirmed by the comparison between the con-
ventional joint impedance controller and blind grasping. Controllers for
positioning and orienting a grasped object can be easily superposed to the
blind-grasping controller to realize in-hand manipulation during the grasp.
The controller for in-hand manipulation was implemented with the super-
visory control and semiautonomous systems, to appropriately select a task for
manipulation. Experimental results showed that these high-level controllers
easily utilized the blind-grasping controller for realizing in-hand
manipulation.

1
See https://www.youtube.com/watch?v¼ZtP-I_Bpibs&t¼4s.