Biomimetic Robotics - Engineered Biomimicry - page 85

or activity detection. Activity detection refers to
the process of segmenting an isolated word out
of the continuous data stream. This step is fol-
lowed by feature set extraction and dimensional-
ity reduction. Reducing the dimensionality of the
data involves finding a finite set of uncorrelated
variables and establishing a transformation relat-
ing them to a large number of correlated varia-
bles describing the observed data. Feature
extraction is the process of reducing the dimen-
sionality of the data to facilitate subsequent clas-
sification. A standard approach to reduction of
the dimensionality of the data is to use principal
component analysis (PCA) [9] . PCA reduces a large
number of original variables into a smaller num-
ber of uncorrelated components that represent
most of the variance in the original data. At this
stage the vast quantities of data generated are
generally classified. A variety of data analysis
paradigms have been developed for feature
extraction and classification, including multivari-
ate statistical methods and linear discriminant
analysis (LDA) [10] and support vector machines
(SVMs) [11] . Following an assessment of the clas-
sified data, a command signal is generated and a
control function is executed.

to the human body so that the manipulator is
controlled by electromyographically sensed sig-
nals will allow the manipulator to be controlled
by the human brain, as well as provide the ability
to perform tasks requiring large forces.
Alternately, an exoskeleton robot, serving as
an assistive device, may be worn by a human as
an orthotic device. If such a robot is to be prop-
erly matched to assist the human, its joints and
links must correspond to those of the human
body, and its actuators must share a portion of
the external load on the operator. For this rea-
son, it is important to design and build robotic
manipulators that are able to emulate human
limbs. The aim then is to model the human arm
so as to emulate its degrees of freedom, kinemat-
ics, and dynamics.
To model the human forearm, it may be
observed that at three key nodal points—the
shoulder joint, the elbow, and the wrist—the
human arm is characterized by a minimum of
three, two, and three degrees of freedom, respec-
tively. These are adduction/abduction, flexion/
extension, and interior/exterior rotation at the
shoulder; flexion/extension, rotation, and supi-
nation/pronation at the elbow; and flexion/
extension and ulnar/radial deviation at the
wrist and link twist at the wrist end.
To emulate these joints, the human arm is
modeled by four rigid bodies, where the first
one has two degrees of freedom—the joint angle
about a vertical axis and the link twist about a
horizontal axis—and the remaining three links
each have two degrees of freedom characterized
by the joint angles and link twist. The first and
last links are relatively small in length, and the
first two joints along with the first link constitute
the shoulder joint. The joint between the second
and third links constitutes the elbow; the joint
between the third and fourth links and the
fourth link constitute the wrist, which is assumed
to be attached to the hand.
In Section 4.3.1.3 , the performance of the non-
linear controller of such a robot manipulator
( Figure 4.1 ) based on a complete dynamic model

4.2.4 Modeling and Control of
Anthropomorphic Manipulators
Modeling of human limb dynamics by equiva-
lent robot manipulators and establishing a one-
to-one equivalence of links and joints offers
the opportunity to create a new generation of
prosthetic limbs for both healthy and disabled
people.
Humans use naturally developed algorithms
for control of limb movement, which are limited
only by the muscle strength. Furthermore,
whereas robotic manipulators can perform tasks
requiring large forces, the associated control algo-
rithms do not provide the flexibility to perform
tasks in a range of environmental conditions
while preserving the same quality of performance
as humans. Thus, interfacing a robot manipulator