A General Rhythmic Pattern Generation Architecture for Legged Locomotion - Advancing Artificial Intelligence Through Biological Process Applications - page 212

Figure 5. Mimicking CPGs with SER/SMER-driven networks. Activity of flexor (node j) and extensor
(node i) motor neurons during walking of a cockroach and its SER/SMER analogous simulation.

individual OBB modules rather than the integrated
OBB network that is governed by an SMER algo-
rithm. Therefore, the methodology is especially
useful for digital circuit implementation as the
gait model operates asynchronously without the
necessity of a global clock.

ubiquitous in the real world. They are continu-
ous rather than discrete in time domain, better
described by analog circuitry rather than digital
one. Following the digital version we present a
series of novel continuous time OBB structures
which are similar to Hopfield networks but gov-
erned by the SMER algorithm. These structures
can be classified into two major categories, namely
simple OBB and composite OBB, depending on
the network complexity.

Dynamic Properties of Analog Obb
Modules

SER and SMER have the potential to provide the
greatest concurrency among scheduling schemes
on resource-sharing systems (Barbosa & Gafni,
1989; Barbosa et al., 2001). The mutual exclusion
characteristic between any two neighbouring
nodes coupled under SMER makes this schedul-
ing scheme suitably tailored for simulating PIR,
a mechanism widely employed for locomotion
and other rhythmic activities. The dynamics of
SMER depends on the initial allocation of shared
resources, different configurations may lead to
different cycling behaviours and even deadlock
or starvation of the system. This feature leaves
us a space on how to mimic CPGs with SMER
to simulate numerous rhythmic patterns while
avoiding possible wrong design.
Given the ability of SMER to mimic many
gait patterns, this kind of strategy, however, is
essentially intuitive due largely to the discrete
nature of SMER. It is thus more efficient in digi-
tal simulation than analogous applications. The
analogous behaviours, on the other hand, are

Hopfield Neural Network

In 1982 J. J. Hopfield published his seminal
work on the prominent emergent properties of
one kind of neural model which rekindled great
interest of scientists in neural network analysis.
Different from McCulloch-Pitts neural model
(1943) in which neurons are intrinsically Boolean
comparators with limited inputs, Hopfield studied
the biological structure of the large-scale intercon-
nected neural systems and proposed his model
with the emphasis on the whole behaviours and
properties of a neural system. A single neuron is
simplified to a simple electronic device contain-
ing only a resistor, a capacitor and a nonlinear
component stipulating the input-output relation-
ship. The dynamics of an interacting system of
N coupled neurons can be described by a set of
coupled nonlinear differential equations governed
by Kirchhoff's current law,