Biomedical Engineering Reference
In-Depth Information
ordination depended on the sensing performance
of the obstacle avoidance simple behaviour. Due to
the shape of the lowest part of internal walls, this
obstacle avoidance module could not be able to
sense the obstacles effectively. This coordination
approach could not surpass this inconvenience
giving the control to another behavioural module.
In other words, layered coordination is done over
static modules in a prescribed (by rules) way,
meanwhile AIS coordination is based on self-ad-
justment during task performance. Then, changes
in simple behaviour to be coordinated (like an ad-
aptation of simple behaviours) will cause changes
in the coordination method itself. However, the
observed problems for the AIS coordination were
related to reaching the target through the narrow
corridors (XA and BY ones). This difficulty could
be solved with slower movements.
Comparisons between both approaches of ER
behaviour coordination showed no significant dif-
ferences. Further experiments are planned to test
the long-term tracking problem where adaptation
is essential. In these cases, the rule-based coor-
dination seems to be less adequate than the AIS
one. In addition, an improvement in the emergent
behaviour may also appear if simple behaviours
modules are constructed by learning in a noisy
and uncertain environment just before they are
coordinated.
obstacle avoidance, and others). In this way, this
means that evolutionary algorithms are suitable
for optimizing the solution space in proposed
problems, which besides is not new. The imme-
diate question is about the suitability of AIS as
optimization methods. According to de Castro
& Von Zuben (2002) and Aickelin & Dasgupta
(2005), AIS are probably more suited as an op-
timizer where multiple solutions are of benefit.
AIS can be made into more focused optimizers
by adding hill-climbing or other functions that
exploit local or problem-specific knowledge.
AIS can exploit its potential either when some
concept of “matching” is needed. In some sense,
AIS works as algorithms that adapt their solu-
tion space during test time meaning that they
are suitable for problems that change over time
and need to be several times with some degree
of variability.
When AIS is compared with hand-design
ruled-based and fittest evolutionary coordination
systems, it has the potential benefit of an adaptive
pool of antibodies that can produce adaptive and
robust coordination. Therefore, the benefits of this
computation can be used to tackle the problem of
dealing with changing or noisy environments. In
this robot control context AIS-based behaviour
coordination is useful because it does not optimize
a path, i.e. in finding the best match for a planned
path in a priori unknown environment. Instead,
AIS require a set of antibodies that are a close
match, but which are at the same time distinct from
each other for successful environmental situation
recognition. This is the main interest around most
of AIS implementations in literature and from
our own experience, because the designer can
implement different autonomous methods for path
generation. AIS only require positive examples.
In addition, the patterns that it has learnt can be
explicitly examined (see de Castro & Von Zuben
(2002)). Besides, AIS, as well as the majority of
other heuristics that require parameters to oper-
ate, has the drawback of parameter settings that
FUTURE TRENDS
ER is a key tool to obtain self-organized adaptive
systems, like the ones required in autonomous
robots interacting with the real world. Layered
Evolution and AIS approaches to behaviour coor-
dination share many similarities. In the authors'
opinion, the challenge to obtain a bio-inspired
robot control within ER is to deepen in the way
the emergent behaviour is achieved.
In the previously described case studies of ER,
it was shown that genetics algorithms obtained
“fittest” behaviours for each task (e.g. phototaxis,
Search WWH ::
Custom Search