Biomedical Engineering Reference
In-Depth Information
Immune System Inspired-Algorithms
as behaviour Coordination
(7)
The proposed behaviour coordination has a
dynamic process of decision making defining a
network structure of conditions-action-restric-
tions items. The incoming information provided
by each sensor is considered as an “antigen”,
meanwhile the robot as an “organism”. The robot
(organism) contains “antibodies” in a network
organization to recognize the antigens and to
perform immune responses. Thus, the condition-
action behaviour of the robot could be related with
the antigen-antibody response of the immune
system. The antigen represents the current state
of the environment (Vargas, de Castro, Michelan
& Von Zuben, 2003).
In our recent experiments, we develop an AIS-
based system for behaviour coordination to solve
the problem of tracking a target (i.e, a black line
in the floor or a pipeline in the seabed). In them,
the immune network dynamics was modelled
through Farmer's model (Farmer, Packard &
Perelson, 1986), selecting the most appropriate
antibody to deal with external information. The
concentration of an antibody may stimulate or
suppress other antibodies of the network, and it
may depend on the concentration of other antibod-
ies through network links. Based on the resulting
concentration level, an antibody is selected using
a ranking algorithm, choosing those with higher
concentrations. Only one antibody is chosen at a
time to control the robot.
Antibodies will recognize antigens, and the
action to be taken will be determined by dynamics
of the immune network. The changing process of
antibodies involved in antigenic recognition is
planned to be implemented to perform autono-
mous navigation. Then, the appropriate antibod-
ies recognize the antigen, and the concentration
of each antibody ( a ) involved in the process of
recognition could vary based on the following
equations.
(8)
The terms involved in equations (7)-(8) for the
calculation of the concentration of the i i-th anti-
body at the instant t , are: N , number of antibodies
that compose the network; m i , affinity between
antibody i and a given antigen; m ji , affinity be-
tween antibody j and antibody i , in other words
the degree of stimulation; m ik , affinity between
antibodies k and i , in other words, the degree
of suppression; k i , natural death coefficient of
antibody i . Equation (8) is a squashing function
used to impose boundaries on the concentration
level of each antibody.
As stated, the Khepera © robot has light and
proximity sensors (Figure 4). The antigens refer
to the direction of light sources, direction of the
obstacles, and the proximity of them respect to
the robot body. In this new approach, the sen-
sory system was divided in the zones showed in
Figure 8.
During robot evaluation, the network of anti-
bodies tries to recognise antigens, and the network
dynamics determines the action (behaviour) to
be taken. The behaviour coordination was based
on (Vargas, de Castro, Michelan, Von Zuben,
2003) and (Ishiguro, Kondo, Shirai & Uchikawa,
1996) works. The antigens coding corresponds
to possible situations that activate under certain
conditions (e.g. an obstacle is near, or a light
source is placed on the far-right-front side). The
actions that antibodies represent are, for example,
turn left, turn right, go forward, avoid obstacle,
or do phototaxis.
The experiments carried out contrast with Var-
gas et al.'s work in that an evolutionary adjustment
Search WWH ::




Custom Search