-A notation for representing agents' organizations to be implemented using Moise+ and Jason is proposed. For this purpose a UML profile was defined for representing the elements of Moise+ organizational model such as role, mission and group. The proposed notation will be fully illustrated and applied to the classical example provided by the J-Moise+ team.
In the context of highly complex, distributed and open
systems, engaged and working in dynamic environments are
widely employed. Such system should include the
capability of continuously reacting, with a re-organization process,
to changes occurring in the environment. Because of their
intrinsic nature, agents have been recognized to be a good
way for solving complex problems both at the design and the
implementation levels [
Organizations [
As regards the agent organization implementation, a robust
approach coming from Hubner et al. [
The Moise+ organizational model is complemented with
the possibilities of quickly and easily programming MAS by
means of J-Moise+ [
This offers a powerful tool to MAS developers, nevertheless
it is not still adequately supported by a well defined
methodological approach. Some researchers in the past developed
methodologies for MASs where some aspects of organization
were modeled. In one of the most known in literature [
The work illustrated in this paper regards the creation of a specific notation for representing the organizational model proposed by Moise+. The advantages of having a graphical notation for representing organizations are evident: first of all, graphical notations are more readable and understandable at a glance than any coding language, secondly it is usually easier to explain a graphical notation to stakeholders involved in the designer (that are not designer) than read the application code with them. The possibility of involving stakeholders like system users enables the adoption of agile or extreme development approaches and improves the flexibility of conventional ones.
The remainder of the paper is organized as follows. In
section II the Moise+ organizational model, J-Moise+ and
Jason are introduced. In section III we explain the proposed
notation by using three kinds of diagram in order to define
graphically the structural, functional and normative
specification of a Moise+ organization. In this section an instance of the
notation in use by using the Moise+ tutorial [
Moise+ [
While the Moise+ implementation is based on two key elements: the Organizational Specification (OS) that is the union of structural, functional and normative specification and the Organizational Entity (OE) that is the instantiation of OS on a set of agents.
different approaches [
Jason [
The Triggering Event describes the situations in which a plan may be applicable for execution. The context can be used to specify the condition to make the plan applicable even if an event has triggered that plan. The body can be considered the consequent of the event linked to the context. Within the body commonly are defined the actions that agent must perform to fulfill its own goals.
tional model. J-Moise+ is based on Jason and consists of both an OrgBox Api and a special agent called OrgManager. Agents use the OrgBox Api to access to the organizational layer. While the OrgManager stores the current state and maintains the consistency of the Organizational Entity during its lifecycle. J-Moise+ basically offers a set of actions to change the state of the organization and produces some events related to organizational changes to which the agent can react.
we here define only the constructs used to model organizations
with Moise+. In the following subsections, we describe three
kinds of diagram applied to the classical example (”Writing
Paper”) reported in the Moise+ Tutorial [
roles, groups, and relations between these within the MAS organization. Using the normative specification we can constrain the agents behavior by specifying what missions an agent ought to follow and what missions an agent is allowed to follow when playing certain roles.
We use a UML class diagram (named Organizational Diagram OD) for representing the structural and normative specification. The Organizational Diagram focuses on Moise+ elements such as Group, Roles, Missions and different kinds of relationships.
The development methodology of an organization with Moise+ is out of the scope of this paper, but in order to understand this diagram we can say that the building of this diagram is subdivided in two phases. During the first phase all elements concerning the Moise+ structural specification are established, while the second phase starts at the end of the definition of the schema structural diagram and it aims to define the norms the agents should obey when they adopt a role. Figure 1 illustrates the graphical representation of organizational diagram elements.
Groups - A group is represented by means of a package with little men icon. It may contain several structural elements (Roles) and other grouping elements (sub-groups). The root group represents the entire organization.
Roles - A role is an UML class depicted as a little man. Its properties are represented in the form of class attributes. Roles can be logically related to one another using associations. An abstract role, instead, is identified using an italic font.
Relationships - Model elements are related each others with dependencies, associations and generalizations. A dependency is a generic relationship, indicating that an element depends in some way on another. A generalization specifies a relationship between roles in which specialized roles inherit features of the general role. An association describes a link between elements of the Moise+ model. We used UML stereotypes in order to attribute a semantic of the association with Moise+ domainspecific concepts.
In particular Moise+ organizational links, defining the way in which social exchanges between agent roles occur, are represented by means of associations between roles labeled with an Authority, Acquaintance or Communication stereotype.
Fig. 1. The Defined Notation Elements Each propriety of a link can be expressed by UML association B. The Scheme Structural Diagram constraints. The Moise+ Functional Specification deals with the con
Moise+ compatibility links, instead, are defined by means cepts of agents’ missions and their global plans. Plan repreof associations between roles labeled with the Compatibility sents the set of goals to be pursued. Plans and missions comstereotype. Two roles connected by a bidirectional compatibil- pose (or are assembled into) the social scheme. We use two ity link define the possibility of an agent to adopt both roles different views ( or models) for representing the elements the at the same time. Each propriety of a link can be expressed functional specification is composed of: the Scheme Structural by UML association constraints. Diagram (SSD) and the Scheme Functional Diagram (SFD).
Figure 2 shows the Organizational Diagram for the Writing The Scheme Structural Diagram allows to model the social Paper organization. In this example, a set of agents aims to schemes of the organization through a UML class diagram. write a paper. For this purpose, the Moise+ authors define an The elements of this diagram are: organization with one group (wpgroup) and two roles (Writer Goal is represented by a class element reporting the name, and Editor). These roles are an extension (represented by the stereotype and the attribute field; each of them corresponds means of UML generalization in the OD) of the abstract to a specific feature of the Moise+ concept of goal: the class role Author. An agent can play several roles only if they name addresses the goal id. The stereotype represents the are compatible. As exemplified in figure 2, an agent playing two types of goal namely achievement and maintenance. The the writer role can play the role editor at the same time and default type for every goal in Moise+ is achievement but in vice-versa because they are linked by an UML compatibility the SSD the goal type has to be stated in any case. As regard association. the attribute compartment, it basically contains the ttf attribute
In the Moise+ model, a role is usually linked by means of value prescribing the time requested for fulfilling the goal. norms (Obligation or Permission) to one or more missions Mission is also represented through a class stereotyped as defined in a particular scheme. mission. Here the attributes’ compartment contains values for
One of the Writer’s mission (see Figure 2) is mbib (i.e. the minimum and the maximum commitments to the mission. getting references for the paper). The norm linking the mission The Social Scheme is modeled by means of a package to the role is an Obligation, that is the agent playing the where classes (i.e. missions) are grouped in order to represent Writer role must commit to this mission. This is shown in the the social organization of goals and missions. There can be Organizational Diagram through the stereotype obligation. The more than one package in a single SSD thus representing the Editor, instead, may commit to the mission mManager because existence of different schemes in the same organization. The the link is a Permission norm. The association between roles is package’s name corresponds to the social scheme id. stereotyped in order to represent organizational links such as As regard relationships among elements, in this diagram Acquaintance, Compatibility, Communication and Authority. we only use two kinds of relationship: the aggregation and the dependency; the latter is used for representing how two different schemes depend on each other, the former is used for relating missions and goals. With respect to Moise rationale, goals are aggregated into missions that can be distributed/ committed to agents.
Figure 3 shows a portion of the SSD for the write paper example1.
In the Scheme Structural Diagram, a Social Scheme is modeled by means of a package containing classes (i.e. missions and goals). Within a package the structural composition of goals and missions is defined. For instance, the SSD for writing paper example is composed by two Social Schemes, writePaperSch and monitoringSch. The portion of writePaperSch scheme reported in figure 3 shows how the mManager mission is a composition of five goals: wp, wtitle, concl, wabs, wsectitles that respectively aim to write the paper, the title, the conclusion, the abstract and the title of each section. While the illustrated portion of monitoringSch scheme shows ms mission formed by only Sanctioning goal. In the SSD is also possible to underline the dependences between different social schema.
As 3 shown, the social scheme writePaperSch is related to the monitoringSch scheme through a “monitoring” dependency relationship.
The Scheme Functional Diagram represents the behavioural view of the Moise+ functional specification, it is realized by means of an UML activity diagram and it aims at representing, through a set of associated activities, how a goal can be decomposed in sub-goals. Each activity represents the work
1Because of space concerns only grams are reported. Complete diagrams http://www.pa.icar.cnr.it/cossentino/moisenotation/ portions can be of found diain
Fig. 4. The Writing Paper Example - writePaperSch SFD done by agent to fulfill the goal. The elements of this diagrams are:
The Goal is then represented by an activity where the name is the goal’s id and the stereotype represents the type of the goal (achievement or maintenance - see the previous subsection) The Moise+ model allows to decompose goal in sub-goals by means of a plan operator. There are three different kinds of plan operator: sequence, parallelism and choice, the first means that a goal gi (having two sub-goals gi,i and gi,i+j ) can be achieved only if the close sequence of gi,i and gi,i+j ). All of them can be easily represented by means of the UML activity diagram syntax, for instance the parallelism is represented through the fork and the choice through the decision diamond. Sequence is represented by a straight arrow line.
As said before, in this paper our concern is about the notation/models to be used for representing MAS organizations. If we would use them during a design process phase we should consider that we can draw more than one SFD, one for each package (i.e. social scheme) of the SSD.
Figure 4 and figure 5 show the Scheme Functional Diagrams (SFD) built for the wp and monitoring goals of the Writing Paper organization which are the root goals of writePaperSch propose notations for agents, roles and the other elements of their proposed metamodel.
Remaining in the context of AUML-related researches, it is
worth to remind the long work done by the FIPA Modeling
Technical Committee and its members at the beginning of
years 2000. In this context several proposals arose. Among
the others, Parunak and Odell presented in [
They introduced swimlanes in class diagrams in order to
partition the diagram in zones representing groups. Within
each organization the diagram may depict roles and the agents
playing them. Another proposed diagram was concerned with
Fig. 5. The Writing Paper Example - MonitorinSch SFD the description of the dynamic behavior of agents/roles in
terms of their interactions. Essentially it is an extension of the
UML 1.0 sequence diagram containing some notation elements
and monitoringSch (defined in the previous section) corre- that have been introduced in following versions of UML.
spondingly. The SFD of the writePaperSch (see figure 4) A more extended notation has been proposed by L. Padgham
explains how to achieve the root goal of the scheme. In detail, et al in [
It is important to highlight there are three different types of details, it has been applied to O-MASE, PASSI, Prometheus
goal execution: sequential, parallel and choice. If two goals are and Tropos (partially). The notation includes graphical icons
related with a sequential relationship then the goal target can for representing almost all the elements of an agent-oriented
be reached only after that the source goal is reached. If two design, organizations included (agent, role, position, goal, . . . ).
goals are related with a parallel relationship then both goals From this point of view, the notation presented in [
Besides it is important to note that (see 4 and figure 5) the of them behavioral. This is a relevant difference with the work
root goal is represented with a box with the goal icon at the presented in this paper. In fact, our notation also includes the
top left corner instead of with an activity, this is due to the Scheme Functional diagram that is a behavioral representation
features of the tool we use for drawing activity diagrams. The of the system.
concept of goal does not depend on the graphical box they are INGENIAS [
A proposal for the introduction of groups in MAS modeling to develop MASs.
has been presented by J. Odell et al. in [
As regards the comparison of this work with the notation Organization is done by dividing the MAS into groups and
we presented in this paper, the authors of [
Each element of the Organization viewpoint has a precise notational counterpart, for instance the goal is represented by a circle and the group in a box with two kinds of head. This allows to model how the organization is divided in groups (each group again can be decomposed in groups) made by agents that play roles. From a modelling point of view it is important to note that the goal is associated to the organization and that in INGENIAS the concept of agent is central and is related to the concept of role whereas in MOISE+ the central element group and the role. This logical difference can be found in the two notations and in the related diagrams we can draw; anyway both of them allow representing the whole portion of metamodel including organizational concept but INGENIAS does not provides means for representing norms.
methodology mainly based on the notion of goal.
The Tropos methodology is articulated in four different phases from the requirements analysis to the agent system implementation. The requirements analysis covers two phases: the Early Requirements Analysis phase, concerning with the studying of the problem, produces an organizational model and the Late Requirements Analysis phase where the system-to-be is described. The agent system implementation is performed through the Architectural Design phase, where the system architecture is defined, and the Detailed Design phase where all system components are specified. During the first two phases an actor and a goal diagram are produced.
cation domain stakeholders, their objectives and dependences. A Goal Diagram is a refinement of the previous diagram underlining the goals of a single actor. These two diagrams show essentially five concepts: Actors, Goals, Resources, Tasks and Dependences. The Actors are the intentional entities such as agents (software or human), roles (abstract representation of behaviors within some specialized domain) or position (set of roles typically played by an agent). Goals are the objectives of an actor, divided into hard goals and soft goals. Tasks are the way to achieve a goal. Resources are means used by agent in order to reach their goals. Finally, Dependences are relationships between actors.
The notation used in the Tropos diagrams in order to represent the above elements is very simple. An actor is depicted by means of a circle, its goals are ovals and its softgoals are clouds shape. The used resources and tasks are represented as rectangles and hexagons respectively. The dependences between actors are arrows with a specific content. This content represents the dependum (i.e. goal, task or resource) that is the element through which two actors depend each other.
tion. Its authors have proposed organizational patterns [
applications led to an increasing number of agents employed in the multi agent system that must expose autonomous and organizing capabilities also for substituting and making decisions on the behalf of user.
One important topic in these kind of systems is how to manage the agents by creating organizations in the same way the human, and more generally, the biological systems do.
The design and the implementation of organization in MASs is related to this topic. Our work concerns the creation of a design process for developing MASs organized in hierarchical structures, such as holons, that can be implemented with JMoise+. In this paper we present the first step of this ongoing work: the UML-based notation to be used for representing organizational elements in the design process work products.
We created a notation enabling us to model organization through three different artefacts where all the elements of Moise+ organizational model are represented. One of the most important results is that we eliminated the difficulty related to the use of the predicative form for representing goal, norms, etc. Moreover we obtained the capability of converting the work product, the diagram, in a xmi file and then through an easy transformation in a xml file thus directly obtaining Moise+ code or better, if necessary, any other kind of code.
FP7-Humanobs and the IMPULSO project funded by the Italian Ministry for Economic Development.
Authors would like to thank Paolo Giorgini for his useful comments and suggestions.