Page 311 - Handbook of Biomechatronics

Upper and Lower Extremity Exoskeletons 303

offer the advantage of unrestricted shape and can therefore be thinner than
the lithium-ion-based design. They provide double the energy density of
lithium-ion batteries.

3.4 Control for Exoskeletons
The exoskeleton control system can be categorized according to the model
system, the physical parameters, the hierarchy, and the usage. These consid-
erations lead to different control schemes (Anam and Al-Jumaily, 2012).
According to the model-based control system, the control strategy
for the skeleton can be divided into two types: the dynamic model and
the muscle model-based control (Anam and Al-Jumaily, 2012). The
dynamic exoskeleton model is derived through modeling the human body
as rigid links joined together by joints (bones). This model is formed from
combination of inertial, gravitational, Coriolis, and centrifugal effects
(Anam and Al-Jumaily, 2012). The dynamic model can be obtained through
three ways: the mathematical model, the system identification, and the
artificial intelligent method (Anam and Al-Jumaily, 2012):
• The mathematical model is obtained by modeling the exoskeleton
theoretically based on physical characteristics of the system (Anam and
Al-Jumaily, 2012).
• The system identification method is based in parameters estimation. In
the BLEEX exoskeleton researchers have implemented the least-squares
method for swing-phase control (Ghan et al., 2006). The least square is
utilized to estimate the parameter of the dynamic model.
• Based on the pairs of input-output data. Aguirre-Ollinger et al. also
employed the recursive least square method to estimate the dynamic
model parameters of one DOF lower exoskeleton (Aguirre-Ollinger
et al., 2007).
• The use of an artificial intelligence method to allow solution many
nonlinear problems has attracted some researchers to employ in the
dynamic model identification. Xiuxia et al. (2008) used the wavelet
neural network to identify the dynamic model of exoskeleton. They
implemented the wavelet neural network in the virtual joint torque
control as inverse dynamic model.
The muscle models have been used in the exoskeleton control schemes.
Unlike the dynamic model, the muscle model predicts the muscle forces
deployed by the muscles of the human limb joint as a function of muscle
neural activities and the joint kinematics (Anam and Al-Jumaily, 2012).