Stigmergic Cues and their Uses in Coordination: An Evolutionary Approach

In the first part of the chapter, we review current definitions of stigmergy adopted by leading scholars in different fields. Though the core properties of stigmergic behavior have been acknowledged both in biology and computer science, a clear-cut conception is still missing. Occasionally the phenomenon is classified as a kind of coordination, other times as communication. To better clarify what stigmergy is as well as when it is communication and when it is not, we introduce the general theory of behavioral implicit communication and we clarify what is peculiar of the stigmergic case.
In the second part of the chapter, we explore the evolution of stigmergic behavior adopting a simulative approach. In particular, we have simulated a population of artificial agents living in a virtual environment containing safe and poisonous items (fruits): eating safe fruits increases the fitness of an individual, while eating poisonous ones decreases it. The behavior of the agents is governed by artificial neural networks whose free parameters (i.e. the weights of the networks' connections) are encoded in the genome of the agents and evolve through a genetic algorithm. Agents interact with their environment and between each other through the traces that their behaviors leave in such an environment.
Biological plausibility aside, the simulations are designed to provide an operational model of stigmergic cues together with a principled way to understand their possible uses. By making explicit the transition from (1) a multi-agent system in which agents individually look for their resources to one in which (2) each agent indirectly coordinates with what the other agents do (stigmergic self-adjustment) and, finally, (3) to a situation in which each agent acts also to send a message about what kind of resources are available in a risky environment (stigmergic communication), the simulations offer a precise analysis of the difference between traces that are just signs with a behavioral content and traces that are signals with a behavioral message.

Tipo Pubblicazione: 
Contributo in volume
Author or Creator: 
Tummolini, L.
Mirolli, M.
Castelfranchi, C.
CRC Press, Boca Raton, USA
Multi-Agent Systems Simulation and Applications, edited by Danny Weyns; Adelinde M . Uhrmacher, pp. 243–265. Boca Raton: CRC Press, 2009
Resource Identifier:
ISTC Author: 
Ritratto di Luca Tummolini
Real name: 
Ritratto di Marco Mirolli
Real name: