Biomedical Engineering Reference
In-Depth Information
are domain dependent. Universal values for them
are not available.
The introduction of fuzziness within the AIS
properties (see de Castro & Timmis, 2003) in
antigenic recognition, suggests that fuzzy logic
might be appropriate to model several aspects and
mechanisms of the immune system, particularly to
combine the multiple responses of simultaneous
immune network outcome.
In self-organized adaptive systems, some in-
teresting questions still remain open, like how a
bio-inspired system reacts to discover new optima
solutions, or how hybrid solutions can enhance
behavioural coordination by mean of adding
other computational intelligence techniques. For
example, AIS could be combined with GA for
discovering more suitable immune networks for
a specific problem. This could correspond to the
primary response of the AIS and to the conver-
gence phase with an immune network based on
GAs. Therefore, this combination could exploit
further adaptations that are precisely what GAs
is lacking (Gaspar & Collard, 1999).
From a strictly theoretical point of view, the
adaptive nature of behaviours has several con-
sequences that are not well understood yet. For
instance, motor actions partially determine sensor
patterns in an also autonomous sensory-motor
loop. Therefore, on-line coordination between
sensors and actuators can enhance adaptively
the robot ability to achieve its goal, as suggested
in (Nolfi, 2005). This fact resembles the natural
adaptation of animals to interact with their sur-
roundings. This is a cutting-edge research topic
in mobile robots.
a comparison framework for the new trends in
autonomous path and task planning approaches
like the ones in ER. A considerable amount of
research has been done in the last decade in this
area, and a special emphasis was put into autono-
mous robot path generation with environmental
adaptation behaviour, in which authors are cur-
rently researching.
From an engineering viewpoint, a determin-
istic behaviour is still difficult to be designed
straightforward because it requires a special ability
from the designer to infer the rules governing the
interactions between the robot and the environ-
ment that will precisely determine the desired
emergent behaviour. In addition, such a design
methodology is opposite to the inherent philoso-
phy of free evolution. In fact, evolution include
a necessary learning process in which the rules
governing the robot/environment interactions
are progressively assigned, modified through a
process of random variation, and refined during
the lifetime as a part of the adaptive process.
This process allows discovering and retaining
useful properties emerging from such interac-
tions without the need to identify the relation
between rules governing the interactions and
the resulting behaviours. Difficulties arise when
trying to optimize a merit figure, engineers'
obsession, like minimum energy consumption,
shortest path and others. In such cases, the proper
selection of a fitness function may be more an art
than a prescribed method. Another perceptible
drawback of ER is the need of a learning period
previous to the task development with a certain
degree of success. This training can be carried
out off-line or on-line. The on-line learning as
in the case of humans, has involved the risk of
non-stabilities and the impossibility to assure a
complete convergent algorithm.
In spite of these apparent disadvantages of ER
in contrast to other more traditional techniques,
it seems to be the future research in robotics, for
it is being demonstrated that it not only can con-
tribute to solve real engineering problems, but also
CONCLUSION
The problem of path generation in autonomous
robots was addressed from the stand point of
artificial intelligence techniques. Traditional
approaches, in which authors have been work-
ing, were exemplified as case studies. This was
Search WWH ::
Custom Search