![](bg2.png)
Figure 1: A representation of the adaptation procedure.
GenomeAgents are represented in white. The agent selected
at the end of the procedure is marked with an “X”, while the
CrosserAgent is marked with a “C”.
from the previous generation and some new agents
obtained crossing the agents’ genomes. The num-
ber of individuals per generation is a parameter of
the adaptation process. During the adaptation process
two agents merge their chromosomes according to the
rules of genetic programming that will be discussed
later. The parents transmit their genes to the child
which evolves and gains the capability of reaching
different results. The adaptation process ends when
an agent in the current generation provides a satisfy-
ing service or if an a priori defined number of gener-
ations has been reached. The fittest agent is selected
and becomes a member of the society. The CrosserA-
gent notifies its name to the requesting agent in order
to fulfill the service request. This process will be dis-
cussed more in details in section 3.
2 THE GENOME STRUCTURE
This section introduces the Genome structure shown
in form of an UML class diagram in Fig.2. This fig-
ure highlights the two main parts in which the genome
can be decomposed at a logical level: knowledge and
abilities. The division is pointed out through the in-
clusion of chromosomes in two different packages.
Starting from the higher level, the genome (at the
top of Fig.2) contains all the information needed to
describe the agent; from this information a new agent
can be created.
The genome enables a set of Services which
makes explicit the functions offered by the agent to
the external environment. The genome is composed
of a KnowledgeChromosome and a set of AbilityChro-
mosomes.
The KnowledgeChromosome aggregates genes
which refer to ontological concepts (OntologyGene)
and that are specialized in three categories: (i) Con-
ceptGene: describes an instance of a concept of the
ontology; (ii) ActionGene: describes an instance of an
action of the ontology; (iii) PredicateGene: describes
an instance of a predicate of the ontology.
The AbilityChromosome is composed of a set of
node genes which describe the plan structure (Node-
Gene) and by the contents of these nodes which de-
scribe the action associated to them. Node contents
can be of three different kinds: predicate or action
genes (indicated as PredicateGene and ActionGene in
Fig.2) or other Ability Chromosomes. Plugging in an
Ability Chromosome with a node allows us to asso-
ciate a behavior, described by another plan, to a node,
thus creating a sort of recursive structure. There are
four kinds of nodes: StartNodeGene, EndNodeGene,
ActionNodeGene and IfNodeGene. The node classi-
fication reported here is inspired by (van Der Aalst
et al., 2003).
3 THE AGENT ADAPTATION
PROCESS
The agent adaptation process, led by the CrosserA-
gent, is composed of the following iterative steps:
• definition of the parents’ sub-society: this sub-
society includes all the agents that will be used
in the adaptation process and that contribute with
their ability genes and knowledge genes to the
definition of the resulting agent;
• creation of the new generation by using adaptation
techniques;
• evaluation of the results provided by the new
agents;
• stop of the process if one (or more) agent(s) suc-
cessfully provide the required service.
During the agent adaptation process we create
new generations by using mutation, elitism and cross-
ing techniques. While the first two techniques
are reused from litterature (Banzhaf, 1998)(Mitchell,
1998), crossing is described below.
The agent adaptation process can be divided into
two steps: knowledge crossing and ability crossing.
Knowledge crossing allows to modify the set of
knowledge genes about the environment. Knowl-
edge Chromosomes crossing is inspired by (Noy and
Musen, 1999) and it is performed over each agent’s
knowledge gene by using four techniques:
• Fusion: the two parents’ knowledge genes are
melted in a single gene;
• Selection: one of the parents’ knowledges is cho-
sen while the other one is discarded;
• Union: both of the parents’ knowledges are
copied in the new individual;
A GENOME BASED VISION OF MULTI-AGENT SYSTEMS
407