swarm intelligence: from nature to artificial systems ... · creation of spatiotemporal structures...

Post on 25-Jul-2020

8 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Swarm Intelligence: From Nature to Artificial Systems, Chapter 1

presentation by:Nathan Carlson and Fred Webber

material by:Eric BonabeauMarco Dorigo

Guy Theraulaz

What is Swarm Intelligence‽

The term was originally used to describe cellular robotic systems. ‘SI’ should be used more generallySI is “Any attempt to design algorithms or distributed problem-solving devices inspired by the collective behavior of social insect colonies and other animal societies.”“Swarm is a decentralized problem solving system that solves problems efficiently”

Important Characteristics of SI

RandomnessDecentralizationIndirect interactions among agentsSelf Organization (SO)

Specialization of Labor

Job determined by individual characteristics

Morphology (as in major and minor ants)AgeChance

When one type disappears,others take over

Design

To develop a solution, one must first have knowledge of individual behaviors and interactions are needed to produce such global behaviors

1.2 Modeling Collective Behavior in Social Insects

The four basic ingredients of Self Organization

Pos. Feedback – magnifies a good job (sugar concentration example)Neg. Feedback – counterbalances pos. feedbackAmplification of fluctuations –randomness allows discovery / rediscovery of solutions Multiple Interactions –performing more then one behavior at once

Characteristics of SO phenomena

Creation of spatiotemporal structures (such as nest architectures, foraging trails, etc.)Multiple stable states – local optimaStimuli have ranges where behavior changes greatly (termites building pellet mounds)

Stigmergy

Stigmergy – communicating by modifying the environmentSocial insects require communication amongst each other – direct or indirect.Stigmergy reduces the need for direct communication

Modeling Swarms

Complex actions by individuals are treated as simple tasks on the group level Don’t over complicate the modelAdd more complex assumptions only if the simple model failsIndividuals have limited cognitive abilities

1.3 Modeling as an interface

Using Models

A model…Uses a small number of relevant quantities Has parsimony, coherence, & refutabilityMight have hidden variables

Biological models are great, but tune them

1.4 Robotics

Pros of SI applications to robotics

By removing a central controller…Less communication is neededThe system still won’t fail if the controller breaksIndividual robots can be designed and made cheaper and more simplySystems are more robust

Random fluctuations find new solutions compensate for new problems

Cons of SI applications to robotics:

Global information not available, which can lead to stagnation / deadlockActually programming them and developing the modelLack of miniaturization restrains field because quantity is important

Contributing factors to the current success of collective robotics

Relative failure of classical AI programming in roboticsProgress of hardwareArtificial Life, as in biological systemsPeople like it because of its successes

Obligatory Questions Slide

0,|_|357][0|\|5‽

top related