Sunteți pe pagina 1din 4

Swarm Synchronization Control

Douglas D. Hammer
Principal Investigator

clearlineofsight@gmail.com

ABSTRACT 2.CURRENT RESEARCH


In this paper we develop an algorithm for the decentralized Much research has gone into disordered systems governed by
control and synchronization of multi-agent swarms. This work is simple rules of interaction. As found in natural systems many of
based on simulation research in entropy or information capacity the underlying components follow simple rules that determine
regulated phase difference equations. We show how there is a their next state based on the interactions and the forces at work
direct correlation between local signaling entropy and mean in their local environment. Many researchers in this area have
proximity phase rates, this establishes a control function to turned to cellular automata as finite state machines, in hopes of
adjust the rate of swarm synchronization. revealing what indicators cause the formation of emergent order.
In a finite state machines hundreds of interaction components
Categories and Subject Descriptors can represent one of many finite states. Their change in state is
Decentralized Control. governed by global forces, such as gravity and local interaction
of their neighboring components. Through the study of these
General Terms finite state machines complex orders of coherent interaction
Algorithms, Measurement, Control Theory. have been observed. The cellular automation approach takes a
deterministic approach to revealing how these systems are
Keywords composed and what underlying mechanisms are involved to
Information Theory, Statistical Mechanics, Synchronization. bring about higher orders of structure and complexity. Some
models of cellular automata have revealed the formation of
1.INTRODUCTION
rivers, the fractal structure of plants, and the formation of
Synchronization is a phenomenon that occurs throughout both snowflakes.
the natural and mechanical world. On a more prevalent level
ecosystem synchronization can be found in fireflies in the 3.SYNCRONIZATION MODELS
amazon, heart cells, and sleep wake cycles in the human brain. In contrast phase transitions or state transitions also follow the
Some of the deeper hidden aspects of synchronization occur statistical mechanical laws of entropy, the second law of
inside a bee colony, where homeostasis or thermo regulation of thermodynamics, the first being gravity. In a statistical
the hive is kept constant through the interaction of thousands of mechanical model more emphasis is placed on known laws of
bees. This thermo regulation is also found in warm blooded physics that govern interaction than the random generation or
animals, where keeping a constant body temperature is vital to iteration rules found in cellular automata. A considerable
the organism's survival. On a mechanical level, specifically a amount of mathematical insight has been gained over the last
statistical mechanical level, synchronization occurs where tens thirty years to explain why synchronization occurs, specifically
of thousands of atoms reach a state of equilibrium. These in natural systems. From a statistical mechanical viewpoint
equilibrium states can be seen as phase transitions. Such phase these natural ecological systems, can be compared to more
transitions can be found in the function of lasers, were at generalized models of fluid dynamic mechanical systems. One
specific point in time the random interactions of electrons are such example is the cooling of liquid glass. When glass in its
excited into a coherent state, generating beam of light. Other liquid state is cooled at a fast rate the molecules that compose
more generalized examples of state coherency or its structure must anneal to a steady state independent of the
synchronization can be seen as a liquid turns into a solid. These effects of their neighboring interactions. This rapid cooling
phase transitions are different in their mechanical organization causes the final solid to exist in a more disordered final
from living organisms, but share similar properties of global molecular state or in other words the glass has a higher state of
state or phase transition between different orders of structure entropy. Entropy is a measure of the amount of disorder in a
and complexity. system, specifically in our glass model. If our liquid glass is
cooled at a much slower rate the atoms that compose the
structure have more time to settle into an ordered state, where
their orientation to their neighbors is more uniform throughout
Permission to make digital or hard copies of all or part of this work for the liquid as it cools. This slower annealing or cooling decreases
personal or classroom use is granted without fee provided that copies are the amount of entropy and allows the glass model to react on a
not made or distributed for profit or commercial advantage and that local level of molecular interaction in addition to a global level
copies bear this notice and the full citation on the first page. To copy of the forces involved. When statistical mechanical interactions
otherwise, or republish, to post on servers or to redistribute to lists, are incorporated into cellular automata many new forms of
requires prior specific permission and/or a fee.
structural and functional coherence become visible. From the
Copyright © 2007 Douglas D. Hammer…$24.00.
statistical viewpoint emergence found in natural systems doesn't 4.1 Control Formulation
exist, but is a product of the phase transitions governed to a Using standard functions from information theory entropy
greater extent by their mechanical laws. calculations were computed for each iteration or in other words
the density of interaction per iteration. When the entropy for
multiple components was plotted in the standard time series the
4.SYNCRONIZATION CONTROL density or swarm activity aligned with the mean entropy rate
In our studies we first created random topologies on hundreds of prior to synchronization (Fig.1). When plotted with both the
interacting components. Each of the components was governed phase shift per component next to the component's internal
by a simple local rule, adjust its rate of firing based on the mean entropy at specific densities or levels of interaction the phase
or average of its neighbors' phase. Initial firing rates for each rate reached its maximum variance near the maximum entropy
component were regulated by sin waves where the component (Fig.2). Component entropy is a calculation of all the entropy
would fire or signal when the wave reached passed through zero. states at each iteration, this follows a smooth convex curve. In
(1,2). addition the phase rate was approximately twice the entropy
near the maximum. This was convenient due to simplifying a
new synchronization model (6,7,8).

The firing of components never signaled its neighbors, and acted


as a representation of the components internal state. At each
iteration in the simulation the phases of each component's
neighborhood based on a predefined radius were summed and
averaged over this range (3).

This new phase was differenced from its own phase as an update When a component's phase shift or rate would double in size
to each component's phase shift. After a few hundred iterations compared to its internal entropy the component would increase
these random networks of interacting components would all the rate proportional to the difference in entropy and phase,
adjust their phase output, and fire at the same time. This set the driving the system into a state of synchronization (9).
foundation for a basic synchronization model. But the next task
was to understand what factors influenced the rate at which the
networks would synchronize? Multiple plots of phase transition
on local and global network levels were compared to time series
plots of phase rate changes, firing rates, and density levels to
reveal if any indicators that could control the rate of
synchronization. Only until a new model was devised that
calculated the component's entropy as a function of its activity New simulations based on entropy calibrated phase rate
or density was the puzzle solved. Each component would fire adjustments synchronized three to four times faster than our
based on its phase, this was the result of the interaction within original model. This statistical mechanical model helped
its neighborhoods. The firing rate was summed over the each establish a new unique control functional, that the rate of the
iteration step within its number of iterations. This ratio formed network's synchronization was indirectly controlled by the rate
the basis of the probability of a component in a firing state. As of phase change within a local proximity based on the entropy or
the iterations increased, the sum firing rates would also increase. information capacity within each component, and this was the
This creates a density map of probabilities that a specific result of both local and global network dynamics.
component would be in a firing state, and its entropy based in
the density of local interactions (4,5).
5.ACKNOWLEDGMENTS
Our thanks to ACM SIGCHI for allowing us to modify templates
they had developed. In addition to the open source Breve Multi-
Agent Simulation Environment. (Version 2.6 Macintosh OS X)

For example, successive iterations the ratios of firing rates over


iterations, would follow a path (0/1,1/2/,1/3,2/4,...).
Figure 1: Swarm Time Series Phase Model
1.0000
0.9000

0.8000

0.7000

0.6000 Phase
0.5000 Density
Rate
0.4000 Capacity
0.3000

0.2000

0.1000

0.0000

-0.1000
0.0000 10.0000 20.0000 30.0000 40.0000 50.0000 60.0000
Time (Seconds)

The Phase model indicates the change in normalized swarm phase over time. The
maximum phase used in normalization not shown in the model is 3574.9.
Synchronization occurs near thirty seconds in the phase model simulation. Two
indicators are revealed from the phase model. The primary indicator is that
synchronization occurs as the capacity is near the density, further investigated in the
capacity model. The second indicator is the normalized rate converges to a minimum
near the minimum density, as the normalized phase increases to its maximum;
converging to a stable synchronization.

Figure 2: Swarm Information Capacity Model


1.1000
1.0000
0.9000
0.8000
0.7000
0.6000
Capacity
0.5000 Rate
0.4000
0.3000
0.2000
0.1000
0.0000
0.0000 0.2000 0.4000 0.6000 0.8000 1.0000
Density
The Capacity model indicates the information capacity of the swarm density, where the
density is scaled between [0..1]. The rate measured at each of the densities over time is
near the swarm's capacity. In addition when the swarm is out of synchronization the
rate is greater than the capacity. The swarm's maximum capacity is the phase between
chaos and order ( in this case synchronization ). Synchronization is the periodic change
between maximum and minimum densities. At the maximum capacity the greatest
change in rate also occurs that drives the swarm out of synchronization.
6.REFERENCES
[1] S. H. Strogatz 2001. Nonlinear Dynamics and Chaos: with
Applications to Physics, Biology, Chemistry, and
Engineering. Coupled Oscillators and Quasiderpiodicity
273-278.
[2] J. Acebrón and L. Bonilla 2005. The Kuramoto model: A
simple paradigm for synchronization phenomena. Reviews
of Modern Physics, Volume 77, (Jan 2005), The American
Physical Society 137-182.
[3] J. Buchli and L. Righetti 2006. Engineering Entrainment
and Adaptation in Limit Cycle Systems. Biologically
Inspired Robotics Group. School of Computer &
Communication Sciences. Ecole Polytechnique Fédérale de
Lausanne. INN, Station 14, 1015 Lausanne, Switzerland.
[4] J. Klein 2003-2008. Breve, A 3D Simulation Environment
for Multi-Agent Simulations and Artificial Life. v. 2.6,
http://www.spiderland.org

S-ar putea să vă placă și