Page 155 - Human Inspired Dexterity in Robotic Manipulation

152 Human Inspired Dexterity in Robotic Manipulation

level also depends on the intervention of the operator. For example, in the
case of a supervisory control in [4], an operator selects a task to determine a
motion controller from a set of prepared controllers and directs the motion
of a grasped object during a precision grasp. In the case of the semiautono-
mous system in [5], the robotic system autonomously detects a target object
and grasp, and manipulates it. In the cases, the motion controllers (the block
surrounded by a thicker line in Fig. 8.1B) have the responsibility for main-
taining the stability of grasping and manipulation. We are interested in the
latter case. Therefore, in the following section, a grasp controller is designed
to stably realize grasp and manipulation.

8.3 MODELING AND DYNAMIC CONTROL OF PRECISION
GRASP

8.3.1 Contact Models
The contact condition between fingertips and an object is important for
describing grasping and manipulation and is classified by a friction model
and the shape of the fingertips. A friction-cone model determines the admis-
sible contact force to prevent the fingertips from slipping on the object sur-
face. Frictional-point contact and soft-finger models have been frequently
used for treating precision grasp. The frictional-point contact model restricts
the contact position but can freely change the orientation. The soft-finger
model restricts rotation around the surface normal in addition to the posi-
tion. The frictional point contact has been frequently used to describe cone-
shaped fingertips that do not change the contact position by the rolling
contact as long as the contact force stays within the friction-cone constraint,
as shown in Fig. 8.2A. Thus, the relative position among the fingers are
preserved, which simplifies the controller design of multifingered hands
because the controller only has to kinematically maintain the relative posi-
tion to manipulate the object. Thus, the frictional point contact model
has been frequently used in early periods.
However, real fingertips of conventional robotic hands have a finite cur-
vature and roll over surface of an object while changing the relative position
and the contact normal, as shown in Fig. 8.2B. Thus, kinematic analysis of
the contact becomes more complex. A typical solution to control grasping
and manipulation under rolling contact is to calculate a trajectory and the
internal force based on the geometry of the fingertips and object. However,
the control problem is challenging because it is difficult to obtain the geo-
metrical information of objects a priori. This paper introduces a passivity-
based control method of the precision grasp and manipulation with