Page 24 - Human Inspired Dexterity in Robotic Manipulation

20 Human Inspired Dexterity in Robotic Manipulation

have not been validated because the representative hands are virtual and no
one has the same hand dimensions as the representatives. They should be
investigated by comparing individual hand models obtained from measured
joint centers, surface shapes, and landmark positions with that created from
the PC scores and eigenvectors.
The representative hand models of virtual individuals at the center and
boundary of the distribution is used for screening design candidates of the
hand-held product. This allows us to eliminate prototype fabrications and
user experiments from the product design process, especially in its early stage.

2.4 ANALYSIS BY DIGITAL-HAND MODEL

2.4.1 Motion Analysis and Posture Synthesis
Hand postures and motion are synthesized from MoCap data, the posture
database, or inverse kinematics computation given contact constraints
between hand and object.
Among these three methods, the MoCap based one is the most intuitive.
In this method, the individual digital-hand model is fit in the MoCap data
by minimizing the error between the measured marker positions and corre-
sponding landmark positions assigned to the digital-hand model (Fig. 2.4).
However, this method is applicable only to the individual model, as no one
has the same hand dimensions as the representatives. The database and
inverse kinematics driven posture synthesis are thus important to synthesize
posture and motion for the representative hands.
The hand posture database is constructed from measured MoCap data. In
the simplest case, a hand posture is synthesized by computing the weighted
sum of the existing data captured in different postures. The motion profile of
the hand from initial to final posture is also synthesized by similar interpo-
lation [17]. This method ignores the coupling between DIP and PIP joint
movements, or that between the joint movements of the fourth and fifth
fingers. This will sometimes lead to the synthesis of realistic, but infeasible
posture and motion.
Similar to the concept of the boundary family, another method for con-
structing the posture database is to introduce a statistical approach [5].By
conducting a PCA for the measured MoCap data, the hand posture is repre-
sented by a smaller number of parameters. In this method, it is possible to
synthesize a realistic hand posture from the PC scores and the eigenvectors.
Also, it is possible to clarify the representative hand postures that appear in
our daily life (Fig. 2.5), if this method is applied to the hand motion data