Biomedical Engineering Reference
In-Depth Information
mechanism was not used, in which the immune
“idiotopes” (restrictions to antibody execution)
are evolved using genetic algorithms. This dif-
ference was done on purpose just to find out if it
was possible to evolve using life-time adaptation
without human intervention. Another difference
with Vargas et al.'s work was that behaviours
could relate with evolved neural networks, instead
of simple rules for controlling robots. The use of
evolved neural networks as behaviour enhances
the robot robustness when performing tasks like
path generation in real world, due to its intrinsic
robust characteristics (Nolfi & Floreano, 2000).
In order to evaluate the AIS-based coordi-
nation, a computer simulated environment for
the Khepera
©
robots, the previously introduced
YAKS, was used to cope with the target tracking
problem with obstacles in the environment. In
this case, the target was a black line represent-
ing a pipeline deployment. These preliminary
experiments demonstrated the feasibility of this
approach, even when dealing with perturbations.
The following tests were performed:
test 1
: control
case;
test 2
: sensors light gain ±5%;
test 3
: motor
gain ±5%;
test 4
: random noise in affinities (
m
)
among antibodies ±50%;
test 5
: random noise
in antibody concentrations (
a
) ±50%. In Figure
9 the final trajectory is depicted. Results show
that the AIS-based coordination could deal with
the expected robot performance. The robot was
capable to develop all its dedicated tasks, without
risk of collision.
Figure 8. Definition of zones for antigen recogni-
tion based on waypoints, pipe segments, zones,
and obstacles recognition
Figure 9. Example of tracking trajectory generation using AIS-based behaviour coordination
Search WWH ::
Custom Search