Alcherio Martinoli.Book
Review: Collective Complexity out of Individual Simplicity, Artificial
Life 7 (2001), 315-319
The concept of Swarm Intelligence (SI) was first introduced by Gerardo
Beni, Suzanne Hackwood, and Jing Wang in 1989 when they were investigating
the properties of simulated, self-organizing agents in the framework
of cellular robotic systems. Beni, G., & Wang, J. (1989). Swarm
Intelligence. In Proceedings of the Seventh Annual Meeting of the Robotics
Society of Japan, Tokyo, Japan, (pp. 425 428).
Eric Bonabeau, Marco Dorigo, and Guy Theraulaz extend the restrictive
context of this early work to include any attempt to design algorithms
or distributed problem-solving devices inspired by the collective behavior
of social insect colonies, such as ants, termites, bees, wasps,
and other animal societies. Swarm Intelligence: From Natural to Arti
cial Systems. Eric Bonabeau, Marco Dorigo, and Guy Theraulaz. 1999,
Oxford University Press.
He gives the three major advantages of SI Robotic approach, since
SI systems have the following properties:
Scalable: The control architecute
of each robot is same, no matter the number of robots.
Flexible: The robots may
be inserted or deleted to/from the environment, no requirement for
any change in the task operation.
Robost: "Not only due to
unit redundancy but also through minimalist unit design".
Critics:
The underlined approach of distributed
control of robots is not thoroughly examined and not mentioned
other works done.
"Second, the authors seem to consider a group of robots more as
a sort of collective em-bedded emulator of natural societies rather
than as another type of system sensing and acting in the physics of
the real world."
At last, Swarm-Robotic approach is not adapted to current
robotic technology and the dissimilarities are not mentioned well.
While the capabilitied of animals in grasping is mush higher than
in robots, robotic systems can comminicate in a better way.
Jan
Wessnitzer,Andrew Adamatzky,and Chris Melhuish:
Towards Self-Organising Structure Formations:
A Decentralized Approach, Proceedings
of ECAL 2001.
Some internal states, local communication among agents and simple
sensory information is proposed to appear as ingredients of the emergence
of macroscopic structures from simple agents with same rules.
Economies, biological cells, stock markets, cellular automata and
collective robotics are the samples of complex
systems.
In "Beni, G., & Wang, J. (1989).
Swarm Intelligence. " : Swarm intelligence is descpribed by "to design a system that while composed of un- intelligent
units,is capable,as a group to perform tasks requiring intelligence
- the so-called swarm intelligence".
Model:
Each unit of the network
can interact with neighbouring units and capable of moving around.
Each processor in the unit, taked some internal state information
from local environment, and employ this data to make decisions.
Additionally, agents possess some sensing units which gives clues
about environment.
Experiments and Tasks: The
task is the formation of two different geometrical structures, namely
line and square. At the end of experiments self-organized
agents succeeded. The algorithm composed of locally interacting agents
with internal states. Initially there is on recruiting agent, which
searches for non-recruited agent in its local radius. When it finds,
there would be two recruited agents and a line with two points. This
process continues until they finish the construction.
Wawerla,
J., Sukhatme, G. & Matari c, M. (2002).Collective
construction with multiple robots.In Proc. 2002 IEEE/RSJ
International Conference on Intelligent Robots and Systems,
Lausanne, Switzerland.
In the beginning the situations where autonomous robots are the
only viable solutions are listed. Collective
construction.
The problem is constructing a simple 2D structure in a planar, bounded
environment.
How should interrobot communication be structured in order to improve
the time effi-ciency of cooperative construction?
The task is the construction of linear barrier out of octagonal
building blocks. They first implement an algorithms with single-robots
and succeeded. Then, they employ a number of robots which communicated
with each other the color of last picked block.
Owen E. Holland and Chris
Melhuish.Stigmergy, self-organisation,
and sorting in collective robotics. Artificial
Life, 5(2):173 202, 1999.
Partially Examined...
While they examine the relation of stigmergy with emerged structures,
they give the signatures of Self-Organization as the creation of spatiotemporal
structures in an initially homogeneous medium, the possible attainability
of different stable states (multistability), and the existence of
parametrically-determined bifurcations.
Gianluca
Baldassarre, Stefano Nolfi, Domenico Parisi.Evolving
Mobile Robots Able to Display Collective Behaviours.Proceedings
of the International Workshop on Self-Organizing and Evolution of Social
Behavior, pages 11-22,
Ascona, Switzerland, September, 2002.
The objective is to evolve a group of robots in order to aggragate
and move together towards a light source. Group act as an individual
and moreover in some situations, some individuals act in different
roles.
"The main advantage of evolution of group of robots is that it is
an ideal framework for synthesizing robots whose behaviour emerge
from a large number of interactions among their constituent parts."
Behaviors are emerged from the interactions of robots and environment.
The requirement to understand the relationship between interactions
and emerge behaviors is essential while designing algorithms by hand.
But artificial evolution process freely exploits the interactions
without the need to understand relation.
Related Work:
"Martinoli (1999) who
used artificial evolution to synthesize the control system of a
group of simulated Khepera robots (Mondada et al., 1993) that were
asked to find food items randomly distributed on an
arena." No striking result except some pair formation while searching
for food.
"Reynolds (1993) evolved the control system of a group of creatures
placed in an environ-ment with static obstacles and a manually programmed
predator for the ability to avoid obstacles and predatation. Despite
the results described in the paper are rather preliminary, some
evidences indicate that coordinated motion strategies begun to emerge.
In the attempt to study the evolutionary origin of herding, Werner
and Dyer (1993) co-evolved two populations of predators and prey
creatures that were selected for the ability to catch prey and to
find food and escape predators respectively."
"In a more recent work, Ward et al. (2001) evolved groups of artificial
fish able to display schooling behaviours."
"Theraulaz and Bonabeau (1995) evolved a population of constructor
agents who collectively build a nest structure by depositing bricks
according to their perception of the local environment and to a
set of behavioural rules (see also Bonabeau et al. 2000).
Critisized by the unrealistic
and free from noise sensory information.
A simulated sound sensor is given with its formula and Attenuation factor. The perceptron controller
has infrared, sound, light
and bias inputs and motor outputs.
At the end, they reach different behaviors. As a result, they propose
artificial evolution is successful, no need to tune the parameters
and/or fitness function...
The evolutionary process to freely exploit interactions without
the need to understand the relation between interactions and emerging
properties as it is necessarily required in other approaches that
rely more on explicit design.
Sampling is used for infrared and ambient light sensors
Simulated sound sensors are employed
population size=100
genotype=8 bit between [-10,10]
in order to move continously, light is turned on and off
Fitness function has 2 components: group's compactness component,
group's speed component
Best 20 genotypes, 5 copies, 2% of bits replaced with random
values
3 qualitatively different classes: Flock, Amobea, and Rose
In order to quantitatively measure performance: Group stability
index(how a formation is stable in time), ,rotational index (rotate
on themselves)
Select each simbot's location, ie. one is very far away from others.