A robotic implementation of rules for social self-regulation

EmergeNET presentation23rd March 2009Torbjørn S. DahlUniversity of Wales, Newport

Defying the rules: How self-regulatory social systems work

Prof Kim Christensen, Imperial College Dr Ana Sendova-Franks, UWE Dr Angela Espinosa, Hull University Dr Torbjørn Dahl, University of Wales, Newport

Aim: To find a set of generic rules that describe emergent

self-regulation in ant, human and robotic social systems

Ants display specialisation

A small fraction of the ants to a large fraction of a task

Task

Ants

Sufficient and necessary requirements for emergent self-regulation

An example of emergent self-regulation in ant systems is “division of labour”

Presence of at least two tasks Differentiation among agents Concurrency of tasks Flexibility in agents

The attractive field model (AFM) of emergent self-regulation

The AMF model is a bi-partite graph Information flow from tasks

Task attraction

jijj

ji

iji

j rtr

tkS

)(

|)(|

)(tanh

i – ant, j – task r – position, δ – avoids division by zero k – learning factor Φ – Max number of ants on task Ψ – Current number ants on task Θ – temporal availability

Action selection

Discrete time steps Random action selection based on task

stimuli Sensitisation and desensitisation:

FLij

ij Kkk /

Initial results

As individuals specialise, tracks become straighter and spatial distributions more distinct

AFM in (Temnothorax) ants

o – ants x – tasks (brood sorting,

exploring, grooming, random wandering wall construction

AF – stimuli to perform a given task

Simulation reproduces data from observations

Ongoing studies to refine AFM, in particular w.r.t. learning

AFM in human social systems

o – humans x – tasks AF – motivation to undertake task

(understanding, interest, proximity, politics)

The viable systems model (VSM)

Recursive model of tasks required for systems to be viable

1. Tasks2. Coordination3. Monitoring and

regulation4. External factor

monitoring5. Strategic vision

Flow of information in human social systems

Pedro Pablo Cardoso from the University of Hull is currently studying an eco-village in Ireland

The village aims to be self-organised Initially, a number of tasks were ignored The VSM model has been introduced to the

villagers

Flow of information in human social systems

Observing the elements of AFM Differentiation (skill levels, empowerment) Concurrency (activity of multiple S1 and S2 taks) Flexibility (changing tasks)

Recording data about communication Communication channels Type and quality of information exchange Failures in communication

Recording data about tasks covered From meeting minutes Uncovered tasks

AFM in robot societies

x – robots o – (manufacturing) tasks AF – Robot task awareness

Evaluation of AFM in robots

Task selection and learning is programmed into the robots

Task data and attractive fields implemented on a separate server

Three models of information flow will be compared:

1. Global broadcast of tasks2. Local (robot-robot) communication3. Stigmergic communication through artificial

pheromones

Experimental infrastructure

Metrics for self-regulation

Task completion rates Active tasks Spatial distribution

Future tasks

Accommodating emergent task specialisation within a pre-programmed setup

Pre-program tasks using an implementation of the VSM

Introduce mechanisms to identify and accommodate new tasks or specialisations