Biomedical Engineering Reference
In-Depth Information
telligence method for problem solving. We expect
to see in the near future an expansion of research
in this field, providing better understanding of the
underlying principles of self-organization. This,
along with the availability of high-performance
computing machinery, will eventually contribute
to the increase in meaningful, real world applica-
tions in various areas, especially robotics.
potential to automatically recruit additional agents
to complete a task when one agent detects the
necessity of that task.
A second set of experiments, dealing with the
collective transport problem, extends the difficulty
of the first problem not only by adding obstacles,
but also by requiring every object to be trans-
ported by more than one agent. The collectively
moving agents proved able to complete this new
task, confirming that such agents can solve more
challenging problems. Further, these simula-
tions showed that simple collective movements
can produce cooperation between self-organized
sub-teams, as seen for example when a group of
agents coordinated to make a bar-shaped object
turn as it went around a corner.
Our final problem, that of self-assembly, does
not require the agents to transport or manipulate
external objects. However, it presents new chal-
lenges, requiring these agents (which are embod-
ied as blocks) to position themselves in precise
locations in order to form a desired 3D structure,
while subject to constraints which include simu-
lated gravity, in addition to impenetrability. In
spite of this complexity, it was once again pos-
sible to control the spatial self-organization of the
agents via an integration of higher-level, state-
based coordination mechanisms together with
low-level, reflexive behaviors. The latter include
not only collective movements, which proved yet
again to benefit the efficiency of the process, but
also, stigmergic pattern matching, which allows
for an implicit form of communication between
the agents, and complements the explicit message
passing that triggers state changes.
In all of the problems studied, the simula-
tions exhibited group-level decisions not just
about which type of movements to make, but
also about how to more effectively distribute
individual members, and which destinations or
goals to pursue. Our results, as well as related
recent work (Winder, 2004), show that the re-
flexive agents of contemporary particle systems
can readily be extended to successfully support
DISCUSSION
The results discussed in this chapter provide
substantial support for the hypothesis that self-
organizing particle systems can be extended to
exhibit behavior more general than just collec-
tive movements. Specifically, by giving a few
behavioral states to normally purely reflexive
agents often found in particle systems, a simple
finite state transition graph or set of rules that
governs state changes, and a simple memory for
storing spatial information, the resulting agent
team has the ability to collectively solve a variety
of problems, through a process of “guided self-
organization”.
Three specific problems were used to study
the behavior of the resulting agents. In the first
problem, search-and-retrieve, an agent team could
routinely search for, collect, and return discovered
resources to a predetermined home location, all
the while retaining movement as a “flock” of
individuals. Further, it was found in simulations
that a team of agents that moved collectively was
more effective in solving search-and-retrieve
problems than otherwise similar agents that
moved independently. This was because when
one or a few agents on a collectively-moving team
discovered a site with plentiful resources, they
would automatically pull other team members
toward that site, greatly expediting the acquisi-
tion of the discovered resource. Thus, a benefit
of underlying collective movements of particle
systems which, to our knowledge, has not been
appreciated in past work, is that they have the
Search WWH ::
Custom Search