Biomedical Engineering Reference
In-Depth Information
form of self-guided behavior is available to the
particles, producing a guided self-organizing
system. This is achieved by equipping particles
with an internal state, which affects their reac-
tive behaviors. State changes can be triggered
by locally detected events, including particular
observations of states of nearby particles. Once
particle systems have been extended with this
top-down control of the particles, while keeping
the bottom-up approach for the system in general,
particle systems can be investigated as a means
for general problem-solving.
In this chapter we present three examples of
systems designed using this approach to demon-
strate that it can work effectively. The first system
involves foraging in a 2D world by competing
teams of agents that use collective movements
or flocking. This scenario presents teams of
homogeneous agents with the opportunity and
necessity to self-assign to smaller sub-teams that
undertake specialized behavior, while working
for a collective goal and competing for resources
against rival teams. The second system solves the
problem of collection and distribution of objects
by mobile agents, also in a 2D world. This prob-
lem closely resembles the problem of collective
transport, common in the robotics and mobile
multi-agent systems literature due in part to
the fact that it strictly requires cooperation, and
optionally coordination, among the agents. The
third and final system presents the self-assembly
of predetermined 3D structures by blocks that
move in a continuous environment with physical
properties and constraints approximating those of
the real world. Although no single block/agent has
an explicit representation of the target structure
nor has an assigned location within this structure,
the global structure emerges through the systemic
properties of the collection of blocks that com-
municate among themselves through the use of
stigmergy, that is, through modification of the en-
vironment. Basic versions of these three problems,
foraging, collective transport and self-assembly,
are commonly found in nature among social in-
sects, and nature has provided simple, effective
strategies that work despite the relatively simple
cognitive levels of such insects, which have served
as inspiration for human-engineered systems that
solve these and similar problems.
In all cases presented here, the particles or
agents not only collectively move in a continuous
world, but also collectively solve the problem at
hand, leveraging the underlying self-organiz-
ing processes to achieve specific goals. These
systems clearly show that our approach is apt to
produce problem-solving capabilities suitable
for complex, goal-oriented behavior, and that
the approach is general enough to tackle widely
different problems.
bACkGROUND
Interacting particle systems were initially in-
troduced by Reynolds, 1987 as a method for the
behavioral simulation of aggregations of animals,
particularly fish schools and flocking birds. By
modeling each individual animal as a reactive
particle whose movements depend only on inter-
actions with nearby particles, Reynolds achieved
realistic simulation of hundreds of thousands
of individuals that move in a life-like manner,
without scripting the movement of any of them.
This clearly showed that it was possible to achieve
sophisticated, seemingly organized behavior in a
complex system without exerting global control
on its components. Flocking systems using in-
teracting particles have been found to be useful
in the study of crowds (Braun, 2003; Helbing,
2000) and in the control of mobile robots (Balch,
1998; Gaudiano, 2003; Mataric, 1995), especially
when formation is an advantage, as in the case of
cooperative sensing (Parunak, 2002).
After the success of particle systems in the
simulation of biological systems, researchers
found applications for problem solving. One of the
best known examples of the application of swarm
intelligence to engineering problems is particle
Search WWH ::
Custom Search