Page 22 - Human Inspired Dexterity in Robotic Manipulation

18 Human Inspired Dexterity in Robotic Manipulation

following procedure is proposed [1]. In this method, the genetic model is
simplified (decimated) by reducing the number of meshes first. The deci-
mated generic model is then deformed so that the landmark positions and
hand dimensions of this model are identical to those of the target. Finally,
deformation of the decimated generic model is transferred to the original
generic model.
Joint centers are estimated by using MoCap data, or by using marker
positions on the skin surface measured under static conditions [14]. In the
latter method, a database of “correct” data are taken for a certain number
of people which is necessary to calculate a regression equation for estimating
the coordinates of joint centers from coordinates of marker positions
attached to the hand with the palm and fingers extended. Link lengths
are estimated by using hand dimensions taken under several positions when
the hand has different joint angles [15] or calculated as distances between two
adjacent joint centers, which are estimated by the previously mentioned
methods.
The accuracy of the generated individual hand model is evaluated by
computing errors in hand dimensions. When the correct dimensions are
used, the mean error is at most 0.5 mm. However, when the estimated
dimensions are used, the accuracy of the model depends heavily on the accu-
racy of the estimation [14]. Differences between the measured surface shape
of the target subject and that of the individual digital-hand model were esti-
mated to be 5.0 mm for the maximum under the cylinder grasping posture
[14], rather large compared to the error in the dimensions. Deformation of
the surface shape due to the change of hand postures needs to be simulated
by a physically more realistic method in the future.
Shape characteristics of the generic model are directly transferred to the
generated individual hand model of the target subject. Thus, the detailed
shape of the target subject such as the shapes of fingernails and medial
bending of the DIP joint will not be reproduced in the deformed generic
hand model.

2.3.2 Representative Digital Hand Models
The range of inter-individual variation in a body dimension is often
described using 5th and 95th (or 1st and 99th) percentile values. However,
the calculated range is misleading when more than one body dimension are
simultaneously considered. This is because it is not probable that the same
person has 95th percentile values of, for example, hand length, hand width,