Page 26 - Human Inspired Dexterity in Robotic Manipulation

22 Human Inspired Dexterity in Robotic Manipulation

captured while the subjects are moving their hand to grasp daily necessities,
such as a cup, dish, screwdriver, etc. As shown in Fig. 2.5, PC1, PC2, and
PC3 account for the flexion-extension of all finger joints, flexion-extension
of the MP joint, and adduction-abduction of the MP joint, respectively.
The inverse kinematics driven posture synthesis [18] is the most versatile
because MoCap data is not necessary for posture synthesis. In this method,
target positions of the thumb tip, fingertips, and when necessary, the palm
are assigned to the object. Inverse kinematics then tries to find the hand pos-
ture that minimizes the error between the target positions and hand land-
marks while avoiding the interference between hand and object surface.
So far, the target positions must be manually assigned, but it is possible to
generate them automatically by categorizing the pattern of target points
according to the attributes of the grasped objects. Also, deformation of
the hand surface shape due to the contact is not considered.
The Virtual Soldier Santos system, developed at the University of Iowa,
offers a more general solution for posture synthesis by introducing the attri-
butes of objects, such as surface shape, weight, and task. When the user
chooses an object and a task, the system selects the type of grasp (power
grasp, precision grasp, etc.) from several candidates. The number of fingers
and hands to use is calculated based on the object shape, hand size, and grasp
type. The power grasp is generated by inverse kinematics based on the
assumption that each link of the second to fifth fingers makes contact with
the object. The precision grasp is also generated by inverse kinematics based
on the assumption that the fingertips of all fingers contact with the object
[19]. This method can generate a different hand posture for the same object
by defining a different task.

2.4.2 Mechanical Analysis

Figs. 2.6 and 2.7showexamples of the inversekinematics-based graspposture
synthesis that was described in the previous section. In the case of the power
graspshowninFig.2.6,sixpairsoftargetpositionsandlandmarksareassigned;
thumb tip,fourfingertips,andpalm.The targetpositionfortheindexfinger is
especiallyputontheshutterbutton,becausetheuserofthecameraisexpected
to pushthis button byan index finger. In the case of theprecision graspshown
in Fig. 2.7, three pairs are assigned; thumb tip, index fingertip, and middle
fingertip. Once the grasping posture is synthesized, it is possible to compute
mechanicalinteractionbetweenhandandobjectbyintroducingthetheoryof
grasp and manipulation established in robotics.