MERCURIO: First Steps in the Development ofan Interaction-oriented Agent Framework

Matteo Baldoni1, Cristina Baroglio1, Federico Bergenti4, Antonio Boccalatte3,Elisa Marengo1, Maurizio Martelli3, Viviana Mascardi3, Emilio Montolivo5,Luca Padovani1, Viviana Patti1, Alessandro Ricci2, Gianfranco Rossi4, and

Andrea Santi2

1 Universita degli Studi di Torino{baldoni,baroglio,emarengo,padovani,patti}@di.unito.it

2 Universita degli Studi di Bologna{a.ricci,a.santi}@unibo.it

3 Universita degli Studi di Genova{martelli,mascardi}@disi.unige.it, nino@dist.unige.it

4 Universita degli Studi di Parma{federico.bergenti,gianfranco.rossi}@unipr.it

5 Elsag Datamat-AMTECemilio.montolivo@elsagdatamat.com

Abstract. This paper discusses the motivations, the starting point andthe first steps of the MERCURIO6 project proposal, submitted to MIURPRIN 20097. The aim of the work is to develop formal models of interac-tion and of the related support infrastructures, that overcome the limitsof the current approaches by explicitly representing not only the agentsbut also the computational environment in terms of rules, conventions,resources, tools, and services, that are functional to the coordination andcooperation of the agents. These models will enable the verification ofthe interaction in the MAS, thanks to the introduction of a novel so-cial semantics of interaction based on commitments and on an explicitaccount of the regulative rules.

1 Introduction

The growing pervasiveness of computer networks and of Internet is an impor-tant catalyst pushing towards the realization of business-to-business and cross-business solutions. Interaction, coordination, and communication, central issuesto any distributed system, acquire in this context a special relevance since theyallow the involved groups, often made of heterogeneous and antecedently exist-ing entities, to integrate their capabilities, to interact according to some agreedcontracts, to share best practices and agreements, to cooperatively exploit re-sources and to facilitate the identification and the development of new products.

6 Italian name of Hermes, the messenger of the gods in Greek mythology.7 Despite the label “2009”, it is the just closed call for Italian National Projects,http://prin.miur.it/index.php?pag=2009.

Multi-Agent Systems (MASs) seem to be the most proper abstraction for devel-oping cross-business systems, since they share with them the same characteristics(autonomy, heterogeneity) and the same issues (interaction, coordination, com-munication). These issues received a lot of attention by the research communityworking on agent models and platforms but, in our opinion, there is no proposalable to integrate them in a seamless and coherent way.

The first limit of existing agent frameworks and platforms concerns the formof communication. Most of them, such as JADE [6, 19, 17, 18] (one of the bestknown MAS infrastructures, sticking out for its wide adoption also in businesscontexts), only supply direct communication. This is a natural consequence ofhaving agents as the only available abstraction, which leads to thinking of com-munication as just consisting in message exchanges. There are models, however,that account for abstractions other than the agent one. These approaches mayallow the MAS designer to cope with a wider range of communication kinds.The Agents & Artifacts model (A&A) [64, 72, 49, 62], for example, provides theexplicit abstractions of environment and artifact that are entities which can beacted upon, observed, perceived, notified, and so forth, thus easily supportingindirect forms of communication. Unfortunately, the A&A misses a semanticsof communication, which would instead be required to achieve the seamless andcoherent integration of interaction, coordination, communication issues that welook forward.

Besides lacking the support of both direct and indirect forms of communica-tion, most existing frameworks also lack a satisfactory semantics of communi-cation. Many of them including the already mentioned JADE, in fact, adopt amentalistic semantics which does not allow the verification of global propertiesof the system: in open MASs, as those we are interested in, agents are unlikely tosucceed in knowing what the other agents know, believe, desire, and this makesthe verification of properties that involve more than one agent at a time im-possible. Viable alternatives to the mentalistic semantics include the social andobservational semantics, that characterize, among other features, commitments[51]. When an agent commits itself towards doing something or achieving somegoal, if it does not satisfy its commitment it becomes liable for a violation whichcan be detected by all the agents in the MAS. Proposals of this kind, that over-come the limitations of a pure mentalistic approach, seem particularly suitableto give to the A&A meta-model the semantics of communication that it stillmisses.

But still, solving the problems concerned with forms and semantics of commu-nication is not enough. In fact, agents belonging to groups that dynamically form,evolve, and split, need to standardize their interaction. In other words, they needinteraction patterns for coordinating. Singh and colleagues define commitment-based interaction protocols that specify the set of social actions, i.e. of thoseactions that affect the social state. When a commitment is added to the socialstate, each agent has an expectation of what the debtor agent will make true,but not of how. In cross-business scenarios, however, expectations about howsome goal will be achieved are at least as important as those on what will be.

This happens because there are laws, rules, habits to take into account, whichhave strong implications on how agents can interact. The need of expressing suchpatterns of interaction and to allow agents to formulate expectations also on the“how” is felt by many researchers, see [15] for a deep discussion.

The last relevant issue in both business-to-business and multi-agent systemswe are interested in, concerns the need to specify interaction in a shared andinspectable way in order to provide the means for verifying, both at design andat execution time, global properties of the interaction, like the interoperabilityof given roles, as well as individual properties, like the conformance of an agentspecification (or of its run-time behavior) to a protocol.

This paper discusses the MERCURIO framework, aimed to overcome thelimitations of existing proposals highlighted in this section. The paper is orga-nized as follows. Section 2 gives an overview of the proposal. Section 3 reports onthe first steps of the project. Section 4 describes the possible application areas,with a particular attention to the scenarios of interest to the Elsag Datamatcompany. An overview of the relevant literature in Section 5 and final remarksin Section 6 conclude the paper.

2 The MERCURIO proposal

The focus of our proposal is to overcome the limitations described in the in-troduction and to specify a unified solution, encompassing formal models ofinteraction and of the related support infrastructures. The solution accounts forboth direct and indirect forms of communication, based on an observational andsocial semantics and explicitly including abstractions for agents, artifacts andthe environment. In this way, it allows for the representation of patterns of inter-action, and which allows for the verification of properties, both from the globalperspective of the system and from the local perspective of single agents.

Figure 1 draws the overall picture of our proposal. We distinguish threelevels: the specification level, the level of the programming abstractions, andthe level of the infrastructure. This setting enables forms of verification thatencompass both global interaction properties and specific agent properties, suchas interoperability and conformance [12]. The approach does not hinder theagents’ autonomy, it guarantees the privacy of the policies implemented by theagents, and consequently favors the composition of heterogeneous agents.

2.1 Specification Level

The model that we propose is specifically thought for open contexts, where itis necessary to coordinate autonomous and heterogeneous agents and it is notpossible to assume mutual trust among them. In these contexts it is necessary tohave an unambiguous semantics allowing the verification of interaction propertiesbefore the interaction takes place [58] or during the interaction [10], preservingat the same time the privacy of the implemented policies.

Fig. 1. The MERCURIO architecture.

The specification level is based on an observational semantics, that can beused to capture direct forms of communication, by implementing communicativeacts, as well as forms of interaction, that are mediated by the environment. Tothis aim, we use the commitment approach [51, 66, 34], which relies on an socialsemantics of the interaction. Even though, in general, the agents’ own behaviorsremain private, agents agree on the meaning of the social actions of the protocol:since interactions are observable and their semantics is shared, each agent is ableto draw conclusions concerning the system as a whole. The support to indirectcommunication is due to the fact that the use of commitments does not requirethe spatio-temporal coupling of the agents. Agents can not only send and receivemessages but they can also act upon or perceive the environment in order tocommunicate. These characteristics offer a degree of flexibility that is adequateto the agents of an open system.

The model includes a notion of programmable environment [63], based onthe A&A meta-model [64, 72, 49, 62], that is specifically thought for supportingthe interaction and coordination among agents. Such computational environmentplays the role of a flexible communication channel, used to represent and manageinteraction protocols based on commitments, patterns of interaction, forms ofdirect and mediated communication and coordination between agents (such asstigmergic coordination). The environment supplies services and artifacts, whichrealize ontologies and ontological mediators, so as to guarantee openness andheterogeneity of MAS. Mediation will occur at two distinct levels: the one relatedto the vocabulary and domain of discourse and the one that characterizes thesocial approach where it is required to bind the semantics of the agent actionswith their meaning in social terms.

2.2 Programming Abstractions Level

This level integrates the abstractions defined in the previous level within pro-gramming languages and frameworks. In particular, we integrate the notions ofagent, of environment, of direct and mediated communication, and of ontologicalmediator. The starting point is given by works that focus on the integrationof agent-oriented programming languages with environments [61] and by theCArtAgO framework [63].

The CArtAgO framework [63] provides the basis for the engineering of MASenvironments on the base of: (i) a proper computational model and (ii) a pro-gramming model for the design and the development of the environments on thebase of the A&A meta-model. In particular, it provides those features that areimportant from a software engineering point of view:

abstraction - it preserves the agent abstraction level, since the main conceptsused to define application environments, i.e. artifacts and workspaces, arefirst-class entities in the agents world, and the interaction with agents isbuilt around the agent-based concepts of action and perception (use andobservation);

modularity and encapsulation - it provides an explicit way to modularizethe environment, where artifacts are components representing units of func-tionality, encapsulating a partially-observable state and operations;

extensibility and adaptation - it provides a direct support for environmentextensibility and adaptation, since artifacts can be dynamically constructed(instantiated), disposed, replaced, and adapted by agents;

reusability - it promotes the definition of types of artifact that can be reusedas tools in different application contexts, such as in the case of coordinationartifacts empowering agent interaction and coordination, such as blackboardsand synchronizers.

These features are advantageous in the realization of commitment stores, com-munication constraints and the interaction mechanisms, w.r.t. approaches like[28], where these elements reside in the middleware, and are therefore shieldedfrom the agents. This latter approach has two disadvantages: the first is thateven though all these elements are accounted for in the high level specification,the lack of a corresponding programming abstraction makes it difficult to verifywhether the system corresponds to the specification; the second is a lack offlexibility, in that it is not possible for the agents to dynamically change therules of interaction or to adopt kinds of communication that are not alreadyimplemented in the middleware.

For what concerns communication, the reference point is the FIPA ACLstandard because widely accepted by the research community and because ofits adoption by agent platforms like JADE, even though FIPA ACL bases on amentalistic approach. For these reasons, as a first step, we are currently study-ing an extension of the language, including primitives for interacting with theenvironment so as to produce also mediated forms of communication.

2.3 Infrastructure Level

The infrastructure level will be developed by taking as references the JaCaMo [5],a platform formulti-agent programming, integrating three orthogonal dimensionsin the development and execution of MAS: the agent level, the environment leveland the organisational level. In particular, JaCaMo is based on Jason [23], as pro-gramming language for developing intelligent BDI-based agents, on CArtAgO [1],for developing and running the artifact-based environments, and on the MOISEframework [39] for modelling organisations and running the organisation man-agement infrastructure.

This makes it possible, on the one side, to improve the separation of concernswhen engineering multi-agent systems, i.e. using the best abstraction level andtools to tackle the specific dimensions, avoiding design pitfalls, such as usingagents to implement either non-autonomous entities (e.g., a blackboard agent)or a collection of autonomous entities (group agent); on the other side, to exploitthe synergy between the dimensions, i.e. exploiting one dimension to effectivelydesign and program also aspects related to the other dimensions. For instance,using the environment to design, implement and represent at runtime the organ-isation infrastructure.

3 A Unified Perspective: First Steps of Realization

Open systems are made of heterogeneously designed and pre-existing partiesthat are assembled with some aim, which none of them can pursue alone. Inorder to allow for a fruitful cooperation, the interaction that each agent carrieson with the others, in the context of the assembled system, must respect somerules. In other words, the regulation of interaction is a decisive factor. The terminteraction protocol refers to a pattern that allows a set of agents to become aMAS, by engaging the expected interaction.

In MERCURIO, we adopt the commitment-based interaction protocols de-scribed in [13]. The main characteristic of this model is a decoupled representa-tion of the constitutive and the regulative specifications of the protocol, whichare both based on commitments. While the constitutive specification definesthe meaning of actions based on their effects on the social state, the regulativespecification reinforces the regulative nature of commitment by adding a set ofbehavioral rules, given in terms of constraints among commitments (and liter-als). These constraints define schemes on how commitments must be satisfied,regulating the evolution of the social state independently from the executed ac-tions: the constitutive specification defines which engagement an agent has tosatisfy, whereas the regulative specification defines how it must achieve whatpromised.

Interaction protocols give the interacting parties the possibility to have ex-pectations on how the other agents will behave. However, the main characteristicof agents is that of being autonomous. This means that they are not forced, butjust requested, to follow the rules imposed by the protocol. The social expec-tations and the taken commitments should guide the agents’ behavior [26, 46].

When agents do not conform to the rules, their violation is to be, somehow, de-tected and managed. These issues are studied in areas like organizational theory[44, 16, 39], normative MAS [75], and electronic institutions [31, 11]. One of thetasks, usually assigned to the environment, is that of medium of communication[72, 49], but A&A environments can be programmed and used at will. In ouropinion, the environment should also have the important role of arbitrator.

In other words, besides supplying the functionalities, needed for communicat-ing, it can also be entrusted to check that the interaction is evolving in obedienceto the rules as well as to reify the social state. A first step in this direction isrepresented by ORA4MAS [38], which proposes the use of artifacts to enablethe access of the agents in the MAS to the organization, providing a workingenvironment that agents can perceive, act upon and adapt. Following the A&Aperspective, artifacts are concrete bricks used to structure the agents’ world: partof this world is represented by the organizational infrastructure, part by artifactsintroduced by specific MAS applications, including entities/services belonging tothe external environment.

The current proposals for electronic institutions do not supply yet all ofthe solutions that we need: either they do not account for indirect forms ofcommunication or they lack mechanisms for allowing the a priori verificationof global properties of the interaction. As [34, 71] witness, there is, instead, anemerging need of defining a more abstract notion of action, which is not limited todirect speech acts. In this case, institutional actions are performed by executinginstrumental actions that are conventionally associated with them. Currently,instrumental actions are limited to speech acts but this representation is notalways natural. For instance, for voting in the human world, people often raisetheir hands rather than saying the name corresponding to their choice. If theenvironment were represented explicitly it would be possible to use a wider rangeof instrumental actions, that can be perceived by the other agents through theenvironment that acts as a medium.

Figure 2 sketches an institution in the setting that we are proposing [42]. Itis composed of three artifacts: one for the social state, one acting as a catalog ofsocial actions, and the last one encoding the constraints upon the interaction.We adopt the definition of artifact in [64, 72, 49, 62]: so, each artifact providesa set of operations, a set of observable properties and a set of links. Using anartifact involves two aspects: (1) being able to execute operations, that are listedin its usage interface, and (2) being able to perceive the artifact observable in-formation. From an agent’s standpoint, artifact operations represent the actions,that are provided by the environment. Moreover, artifacts can be linked togetherso as to enable one artifact to trigger the execution of operations over anotherartifact.

The social state artifact contains the commitments and the facts as envisionedby commitment-based protocols. It is updated after each execution of a socialaction. Agents can observe its contents. The constitutive artifact contains the setof the institutional actions together with their semantics, given in terms of effectson the social state. It must be observable by the agents. The link to the social

Fig. 2. Institution representation by means of agents and artifacts.

state represents the fact that the execution of an institutional action triggers theupdate of the social state. The regulative artifact contains the constraints on theevolution of the social state.

Violations can be detected automatically when the state is updated by check-ing whether some constraint is unattended. The check is triggered by the socialstate itself which, in turn, triggers the regulative artifact. The MERCURIOframework accommodates equally well the solution where checks are performedby an arbiter agent. We propose the one based on artifacts because, as observedin [35], agents are autonomous and, therefore,they are free to choose whetherto sanction a violation or not. Artifacts guarantee that violations are alwaysdetected.

Organizations can be realized in a similar way. The focus, here, is on thestructure of a certain reality, in terms of roles and groups. Each role will havean artifact associated to it, implementing the actions that the role player canexecute on the social state. Organizations can be used to regiment the executionof the protocol [37]. More details in [42].

4 Applicative Scenarios

Aside from the effects on research contexts, we think that MERCURIO willgive significant contributions also to industrial applicative contexts; in particu-lar, to those companies working on software development in large, distributedsystems and in service-oriented architectures. Among the most interesting ex-amples are the integration and the cooperation of e-Government applications

(services) spread over the nation. In this context, the aim is to verify the adher-ence of bureaucratic procedures, of the public administration, to the current laws(see, for instance, [4]). Another interesting application regards (Web) services.Some of fundamental aspects promoted by the SOA model, such as autonomyand decoupling, are addressed in a natural way by the agent-oriented paradigm.The development and analysis of service-oriented systems can benefit from theincreased level of abstraction offered by agents, by reducing the gap betweenthe modeling, design, development, and implementation phases. In this context,it is necessary to deploy complex interactions having those characteristics offlexibility that agents are able to guarantee.

Specific case studies will be defined jointly with Elsag Datamat, the Finmec-canica center of excellence for the design and production of systems, services andsolutions in automation, security, transport, space and information technology.In the following we describe the typical application scenario, underlining theconnections with MERCURIO.

4.1 Elsag Datamat applications

Consider the problem of carrying out an analysis of risks of complex infrastruc-tures such as harbors, airports, stations, commercial centers, industrial sites. Ex-isting tools for risk management typically follow an analytical approach, wherethe risk is a number resulting from applying a complex formula to a huge setof data. Even if this approach is consolidated and currently considered the bestpractice for risk assessment tools, there is room for improvement. In fact, suchtools help identifying the infrastructures vulnerabilities, but in some circum-stances cannot accurately model how such vulnerabilities might combine, leadingto the success of an attack.

A project which faces risk analysis in a completely different way was recentlydeveloped by DISI and Elsag Datamat. The approach consists in simulating at-tacks, performed by malicious entities, by following an agent-oriented approach,and in obtaining empirical evidences of vulnerabilities and strengths of the siteby running the simulations a suitable number of times. In that project, namedSimulRisk, the physical site is described in terms of its topology, existing secu-rity devices (cameras, reinforced doors, anti-intrusion alarms), existing securityprocedures (rules that employees and other persons entering and exiting the sitehave to respect), assets to protect.

Despite of existing protections, a site might be violated because protectionsare too weak with respect to the attacker’s capabilities, or because some securitydevices are out of order, or because procedures are not respected. The SimulRiskmodel has been conceived to cope with all these situations: attackers are softwareagents equipped with capabilities, tools, and a given amount of time, that movein the physical site in order to reach their goal (that might be either to stole ordestroy one set of assets, or to cause as many damages as possible). Obviously,both the SimulRisk model and the two working prototypes (one developed inProlog and another one in Java) are simplified abstractions of the reality. They

proved to be able to provide interesting results but could be further improvedleveraging the capabilities of the MERCURIO framework. Namely:

– Only one role, the “attacker”, is considered. In the prototypes, only oneagent playing that role exists. Other rational entities that inhabit the site(employees, visitors, etc) are not taken under considerations, although theirbehavior is definitely relevant for the outcome of the attack.

– Objects similar to artifacts exist in the SimulRisk environment, but they arenot conceptualized following a rigorous meta-model like the A&A one.

– Since the moves of only one agent are simulated in the prototypes, no needof interaction with other agents arises. However, in the real life attackers ofcomplex infrastructures are usually more than one, and they use to coordi-nate themselves either in an explicit, direct way, or in an implicit, mediatedone (for example, by leaving pre-agreed signs in the environment).

– The malicious behavior of attackers is contrasted by the behavior of thesecurity guards that, again, coordinate themselves and follow well codifiedprotocols. However, this aspect is just briefly mentioned in the SimulRiskmodel.

Both attackers and security guards have expectations on the others’ actions,formulate hypotheses on their behavior, and try to prove in a formal way thatthey can fool their antagonists, whatever their action. The overall goal of thesimulation system, is either to demonstrate (either formally, if a proof mecha-nism is available, or by means of experimental validation) that no set of actionsand interactions of the attackers can cause damage to the site, or to show thesequences of actions and interactions that can lead to the attack success.

The MERCURIO framework, being a unified solution for directed and medi-ated interaction, can cope with all the issues above, thus overcoming the limita-tions of the single-agent SimulRisk model. Because of this and of others concretescenarios where MERCURIO can play a relevant role (including electricity mar-ket, city logistics, resource unavailability management), Elsag Datamat statedits interest in it.

5 Related Works

The issues tackled by MERCURIO, related to the formation of and the inter-action within decentralized structures, are coped with in the literature. Currentproposals, however, are still incomplete in that they supply solutions to singleaspects. For instance, electronic institutions [31, 11, 39, 38] regulate interaction,tackle open environments, and their semantics allows the verification of proper-ties but they only cope with direct communication protocols, based on speechacts, and do not include an explicit notion of environment. Traditional com-mitment protocols [51, 74], effective in open systems and allowing more generalforms of communication, do not supply behavioral patterns, and for this reasonit is impossible to verify properties of the interaction. Eventually, most of the

models and architectures for environments prefigure simple/reactive agent mod-els without defining semantics, that are comparable to the ones for ACL, andwithout explaining how such proposals could be integrated with direct commu-nication models based on speech acts. In the following we comment, without theclaim of being exhaustive, some of the main proposals, classifying them w.r.t.the methodological aspects involved in MERCURIO.

5.1 Formal Models for Regulating the Interaction in MAS

This topic has principally been tackled by modeling interaction protocols. Mostof protocol representations refer to classic models, such as Petri nets, finite statemachines, process algebras, and aim at capturing the expected interaction flow.An advantage of this approach is that it supports the verification of interactionproperties [58, 22, 12], such as: verifying the interoperability of the system andverifying if certain modifications of a system preserve some desired properties (acrucial issue in open domains where agents can enter/leave the system at anytime). Singh and colleagues criticize the use of procedural specifications becausetoo rigid [66, 27, 74]: agents cannot take advantage of opportunities that emergealong the interaction and that are not foreseen by their procedure. Anotherissue is that communication languages use a BDI semantics (FIPA ACL is anexample), where each agent has goals and beliefs of its own. At the systemlevel, however, it is impossible to perform introspection of agents, which are, forthis reason, black boxes. For what concerns the verification of properties thisapproach allows agents to draw conclusions about their own behavior but not toverify global properties of the system [45, 70].

5.2 Environment Models

The notion of environment has always played a key role in the context of MAS;recently, it started to be considered as a first-class abstraction useful for thedesign and the engineering of MAS [72]. A&A follows this perspective, being ameta-model rooted upon Activity Theory and Computer Support CooperativeWork that defines the main abstractions for modeling a MAS, and in particularfor modeling the environment in which a MAS is situated. A&A promotes avision of an endogenous environment, that is a sort of software/computationalenvironment, part of the MAS, that encapsulates the set of tools and resourcesuseful/required by agents during the execution of their activities. A&A intro-duces the notion of artifact as the fundamental abstraction used for modelingthe resources and the tools that populates the MAS environment.

In the state of the art numerous applications of the endogenous environments,i.e. environments used as a computational support for the agents’ activities, havebeen explored, including coordination artifacts [50], artifacts used for realizingargumentation by means of proper coordination mechanisms [48], artifacts usedfor realizing stigmergic coordination mechanisms [60, 54], organizational artifacts[38, 55, 56]. Even if CArtAgO can be considered a framework sufficiently maturefor the concrete developing of software/computational MAS environments it can

not be considered “complete” yet. Indeed at this moment the state of the art andin particular the CArtAgO framework are still lacking: (i) a reference standardon the environment side comparable to the existing standards in the context ofthe agents direct communications (FIPA ACL), (ii) the definition of a rigorousand formal semantics, in particular related to the artifact abstraction, (iii) anintegration with the current communication approaches (FIPA ACL, KQML,etc.), and finally (iv) the support of semantic models and ontologies.

5.3 Multi-agent Organizations and Institutions

The abstract architecture of e-Institutions (e.g. Ameli [31]), places a middlewarecomposed of governors and staff agents between participating agents and anagent communication infrastructure (e.g. JADE). The notion of environment isdialogical: it is not something agents can sense and act upon but a conceptualone that agents, playing within the institution, can interact with by means ofnorms and laws, based on specific ontologies, social structures, and languageconventions. Agents communicate with each other by means of speech acts and,behind the scene, the middleware mediates such communication. The extensionproposed for situated e-Institutions [11] introduces the notion of “World of In-terest” to model the environment, that is external to the MAS but which isrelevant to the MAS application. The infrastructure of the e-Institution, in thiscase, mediates also the interaction of the agents in the MAS with the view ofthe environment that it supplies.

According to [11] there are, however, two significant differences among ar-tifacts and e-Institutions: (i) e-Institutions are tailored to a particular, thoughlarge, family of applications while artifacts are more generic; (ii) e-Institutionsare a well established and proven technology that includes a formal foundation,and advanced engineering and tool support, while for artifacts, these featuresare still in a preliminary phase. MERCURIO gives to artifacts the formal foun-dation in terms of commitments and interaction patterns; engineering tools arecurrently being developed.

5.4 Semantic Mediation in MAS

The problem of semantic mediation at the vocabulary and domain of discourselevels was faced for the first time by the “Ontology Service Specification” [9]issued by FIPA in 2001. According to that specification, an “Ontology Agent”(OA, for short) should be integrated in the MAS in order to provide services suchas translating expressions between different ontologies and/or different contentlanguages and answering queries about relationships between terms or betweenontologies. Although the FIPA Ontology Service Specification represents the firstand only attempt to analyze in a systematic way the services that an OA shouldprovide for ensuring semantic interoperability in an open MAS, it has manylimitations. The main one is the assumption that each ontology integrated inthe MAS adheres to the OKBC model [7]. Currently, in fact, the most widelyaccepted ontology language is OWL [8] which is quite different from OKBC and

cannot be converted to it in an easy and automatic way. Also, agents are allowedto specify only one ontology as reference vocabulary for a given message, whichis a strong limitation since an agent might use terms from different ontologies inthe same message, and hence it might want to refer to more than one ontologyat the same time.

Maybe due to these limitations, there have been really few attempts to de-sign and implement OAs. The first dates back to 2001 [68] and realizes an OAfor the COMTEC platform that implements a subset of the services of a genericFIPA-compliant OA. In 2007 [52] integrated an OA into AgentService, a FIPAcompliant framework based on .NET [69]. Ontologies in AgentService are repre-sented in OKBC, and hence the implementation of their OA is fully compliantwith the FIPA specification, although the offered services are a subset of thepossible ones. The only two attempts of integrating a FIPA-compliant OA intoJADE, we are aware of, are [47], and [25]. Both follow the FIPA specificationbut adapt it to ontologies represented in OWL. The first proposal is aimed atstoring and modifying OWL ontologies: the OA agent exploits the Jena library[40] to this aim. The second proposal, instead, faces the problem of “answeringqueries about relationships between terms or between ontologies”. The solutionproposed by the authors exploits ontology matching techniques [32]. Apart from[25], no other existing proposal faces that problem. Among non FIPA-compliantsolutions, we mention [41], which focuses on the process of mapping and inte-grating ontologies in a MAS thanks to a set of agents which collaborate together,and the proposal in [53], which implements the OA as a web service, in order tooffer its services also over the Internet.

As far as semantic mediation at the social approach level is concerned, we areaware of no proposals in the literature. In order to take the context of count-asrules into account, we plan to face this research issue by exploiting context awaresemantic matching techniques, that extend and improve those described in [43].

5.5 Software Infrastructures for Agents

The tools currently available to agent developers fail in supporting both seman-tic interoperability and goal-directed reasoning. Nowadays, the development ofagents and multi-agent systems is based on two kinds of tools: agent platformsand BDI (or variations) development environments. Agent platforms, such asJADE and FIPA-OS [2] provide only a transport layer and some basic services,but they do not provide any support for goal-directed behavior. Moreover, theylack support for semantic interoperability because they do not take into accountthe semantics of the ACL they adopt. The available BDI development environ-ments, such as Jadex [24] and 2APL [30], support only syntactic interoperabilitybecause they do not exploit their reasoning engines to integrate the semanticsof the adopted ACL.

The research on Agent Communication Languages (ACL) is constantly head-ed towards semantic interoperability [36] because the most common ACLs, e.g.,KQML [33] and FIPA ACL [3], provide each message with a declarative semanticsthat was explicitly designed to support goal-directed reasoning. Unfortunately,

the research on ACLs only marginally investigated the decoupling properties ofthis kind of languages (see, e.g., [20, 21]). To support the practical developmentof software agents, several programming languages have thus been introducedto incorporate some of the concepts from agent logics. Some languages use ac-tions as their starting point to define commitments (Agent-0, [65]), intentions(AgentSpeak(L), [59]) and goals (3APL, [29]).

6 Discussion and Conclusion

The achievements expected from this research are of different natures: scientificresult that will advance the state of the art, software products deriving from thedevelopment of implementations, and upshots in applicative settings.

The formal model developed in MERCURIO exploits commitment protocols,enriched with behavioral rules [15, 14, 13]. This will advance the current state ofthe art with respect to the verification of interaction properties, for which therecurrently exist only preliminary proposals [67]. Another advancement concernsthe declarative specification of protocols and their usage by designers and soft-ware engineers. The proposals coming from MERCURIO conjugate the flexibilityand openness features that are typical of MAS with the needs of modularity andcompositionality that are typical of design and development methodologies. Theadoption of commitment protocols makes it easier and more natural to represent(inter)actions that are not limited to communicative acts but that include in-teractions mediated by the environment, namely actions upon the environmentand the detection of variations of the environment.

For what concerns the coordination infrastructure, a first result is the def-inition of environments based on the A&A meta-model and on the CArtAgOcomputational framework, that implement the formal models and the interac-tion protocols mentioned above. A large number of the environments, describedin the literature and supporting communication and coordination, have beenstated considering purely reactive architectures. In MERCURIO we formulateenvironment models that allow goal/task-oriented agents (those that integratepro-activities and re-activities) the participation to MAS. Among the specificresults related to this, we foresee an advancement of the state of the art withrespect to the definition and the exploitation of forms of stigmergic coordination[60] in the context of intelligent agent systems. A further contribution regardsthe flexible use of artifact-based environments by intelligent agents, and conse-quently the reasoning techniques that such agents may adopt to take advantageof these environments. First steps in this direction, with respect to agents withBDI architectures, have been described in [57, 54].

The MERCURIO project aims at putting forward a proposal for a languagethat uses the FIPA ACL standard as a reference but integrates forms of inter-actions, that are enabled and mediated by the environment, with direct formsof communication. This will lead to an explicit representation of environmentsas first-class entities (in particular endogenous environments based on artifacts)and of the related model of actions/perceptions. Furthermore we will formulate

an improved version of the MAS programming language/framework JaCa, wherewe plan to integrate the agent-oriented programming language Jason, which isbased on a BDI architecture, with the CArtAgO computational framework. Thisresult will extend the work done so far in this direction [61, 63].

As future work, we plan to implement a prototype of the reference infras-tructural model. The prototype will be develop upon and integrate existingopen-source technologies, among which: JADE, the reference FIPA platform,CArtAgO, the reference platform and technology for the programming and ex-ecution of environments, as well as agent-oriented programming languages suchas Jason and 2APL. The software platform will include implementations of the“context sensitive” ontology alignment algorithms developed in MERCURIO.The algorithms will be evaluated against standard benchmarks and also againstthe case studies devised in MERCURIO.

References

1. CARTAGO. http://cartago.sourceforge.net.2. FIPA OS. http://fipa-os.sourceforge.net.3. FIPA specifications. http://www.fipa.org.4. ICT4LAW. http://www.ict4law.org.5. JaCaMo. http://jacamo.sourceforge.net.6. JADE. http://jade.tilab.com/.7. OKBC. http://www.ai.sri.com/ okbc/.8. OWL. http://www.w3.org/TR/owl-features/.9. Fipa architecture board, fipa ontology service specification, 2001.

http://www.fipa.org/specs/fipa00086/.10. M. Alberti, F. Chesani, M. Gavanelli, E. Lamma, P. Mello, and P. Torroni. Verifi-

able agent interaction in abductive logic programming: The sciff framework. ACMTrans. Comput. Log., 9(4), 2008.

11. J. L. Arcos, P. Noriega, J. A. Rodrıguez-Aguilar, and C. Sierra. E4MAS ThroughElectronic Institutions. In Weyns et al. [73], pages 184–202.

12. M. Baldoni, C. Baroglio, A. K. Chopra, N. Desai, V. Patti, and M. P. Singh. Choice,interoperability, and conformance in interaction protocols and service choreogra-phies. In Proc. of the 8th International Conference on Autonomous Agents andMultiagent Systems, AAMAS 2009, pages 843–850. IFAAMAS, 2009.

13. M. Baldoni, C. Baroglio, and E. Marengo. Behavior-oriented Commitment-basedProtocols. In H. Coelho and M. Wooldridge, editors, Proc. of 19th European Con-ference on Artificial Intelligence, ECAI 2010, Lisbon, Portugal, August 2010. Toappear.

14. M. Baldoni, C. Baroglio, and E. Marengo. Commitment-based Protocols withBehavioral Rules and Correctness Properties of MAS. In A. Omicini, S. Sardina,and W. Vasconcelos, editors, Proc. of International Workshop on Declarative AgentLanguages and Technologies, DALT 2010, held in conjuction with AAMAS 2010,pages 66–83, Toronto, Canada, May 2010.

15. M. Baldoni, C. Baroglio, and E. Marengo. Constraints among Commitments: Reg-ulative Specification of Interaction Protocols. In A. Artikis, J. Bentahar, A. Artikis,and F. Dignum, editors, Proc. of International Workshop on Agent Communica-tion, AC 2010, held in conjuction with AAMAS 2010, pages 2–18, Toronto, Canada,May 2010.

16. M. Baldoni, G. Boella, and L. van der Torre. Bridging Agent Theory and Ob-ject Orientation: Agent-like Communication among Objects. In R. H. Bordini,M. Dastani, J. Dix, and A. Seghrouchni, editors, Post-Proc. of the InternationalWorkshop on Programming Multi-Agent Systems, ProMAS 2006, volume 4411 ofLecture Notes in Artificial Intelligence (LNAI), pages 149–164. Springer, 2007.

17. F. Bellifemine, F. Bergenti, G. Caire, and A. Poggi. JADE - A Java Agent De-velopment Framework. In R. H. Bordini, M. Dastani, J. JDix, and A. El Fallah-Seghrouchni, editors, Multi-Agent Programming: Languages, Platforms and Ap-plications, volume 15 of Multiagent Systems, Artificial Societies, and SimulatedOrganizations, pages 125–147. Springer, 2005.

18. F. Bellifemine, G. Caire, A. Poggi, and G. Rimassa. JADE: A software frameworkfor developing multi-agent applications. Lessons learned. Information & SoftwareTechnology, 50(1-2):10–21, 2008.

19. F. Bellifemine, A. Poggi, and G. Rimassa. Developing multi-agent systems with aFIPA-compliant agent framework. Softw., Pract. Exper., 31(2):103–128, 2001.

20. F. Bergenti and F. Ricci. Three Approaches to the Coordination of Multiagent Sys-tems. In Proceedings of the 2002 ACM Symposium on Applied Computing (SAC),pages 367–372, Madrid, Spain, March 2002. ACM.

21. F. Bergenti, G. Rimassa, M. Somacher, and L. M. Botelho. A FIPA Compliant GoalDelegation Protocol. In M.-P. Huget, editor, Communication in Multiagent Sys-tems, Agent Communication Languages and Conversation Polocies, volume 2650of Lecture Notes in Computer Science, pages 223–238. Springer, 2003.

22. L. Bordeaux, G. Salaun, D. Berardi, and M. Mecella. When are Two Web ServicesCompatible? In M.-C. Shan, U. Dayal, and M. Hsu, editors, Technologies for E-Services, 5th International Workshop, TES 2004, volume 3324 of Lecture Notes inComputer Science, pages 15–28, Toronto, Canada, August 2004. Springer.

23. R. H. Bordini, J. F. Hubner, and R. Vieira. Jason and the Golden Fleece of Agent-Oriented Programming. In R. H. Bordini, M. Dastani, J. Dix, and A. El Fallah-Seghrouchni, editors, Multi-Agent Programming: Languages, Platforms and Ap-plications, volume 15 of Multiagent Systems, Artificial Societies, and SimulatedOrganizations, pages 3–37. Springer, 2005.

24. L. Braubach, A. Pokahr, and W. Lamersdorf. Jadex: A bdi agent system combiningmiddleware and reasoning. In Software Agent-Based Applications, Platforms andDevelopment Kits. Birkhauser Book, 2005.

25. D. Briola, A. Locoro, and V. Mascardi. Ontology Agents in FIPA-compliant Plat-forms: a Survey and a New Proposal. In M. Baldoni, M. Cossentino, F. De Paoli,and V. Seidita, editors, Proc. of WOA 2008: Dagli oggetti agli agenti, Evoluzionedell’agent development: metodologie, tool, piattaforme e linguaggi. Seneca Edizioni,2008.

26. C. Castelfranchi. Principles of Individual Social Action. In G. Holmstrom-Hintikkaand R. Tuomela, editors, Contemporary action theory: Social action, volume 2,pages 163–192, Dordrecht, 1997. Kluwer.

27. A. K. Chopra and M. P. Singh. Nonmonotonic Commitment Machines. InF. Dignum, editor, Advances in Agent Communication, International Workshopon Agent Communication Languages, volume 2922 of Lecture Notes in ComputerScience, pages 183–200, Melbourne, Australia, July 2003. Springer.

28. A. K. Chopra and M. P. Singh. An Architecture for Multiagent Systems: AnApproach Based on Commitments. In Proceedings of the AAMAS Workshop onProgramming Multiagent Systems (ProMAS), 2009.

29. M. Dastani, B. M. van Riemsdijk, F. Dignum, and J.-J. Ch. Meyer. A ProgrammingLanguage for Cognitive Agents Goal Directed 3APL. In M. Dastani, J. Dix, andA. El Fallah-Seghrouchni, editors, Programming Multi-Agent Systems, First Inter-national Workshop, PROMAS 2003, volume 3067 of Lecture Notes in ComputerScience, pages 111–130, Melbourne, Australia, July 2003. Springer.

30. Mehdi Dastani. 2apl: a practical agent programming language. Autonomous Agentsand Multi-Agent Systems, 16(3):214–248, 2008.

31. M. Esteva, B. Rosell, J. A. Rodrıguez-Aguilar, and J. L. Arcos. AMELI: An Agent-Based Middleware for Electronic Institutions. In AAMAS, pages 236–243. IEEEComputer Society, 2004.

32. J. Euzenat and P. Shvaiko. Ontology Matching. Springer, 2007.33. T. Finin, Y. Labrou, and J. Mayfield. Kqml as an agent communication language.

In Bradshaw J., editor, Software Agents. MIT Press, Cambridge, 1997.34. N. Fornara, F. Vigano, and M. Colombetti. Agent communication and artificial

institutions. Autonomous Agents and Multi-Agent Systems, 14(2):121–142, 2007.35. N. Fornara, F. Vigano, M. Verdicchio, and M. Colombetti. Artificial institutions:

a model of institutional reality for open multiagent systems. Artif. Intell. Law,16(1):89–105, 2008.

36. S. Heiler. Sematic Interoperability. ACM Computing Surveys, 27(2):271–273, 1995.37. J. F. Hubner, O. Boissier, and R. H. Bordini. From organisation specification to

normative programming in multi-agent organisations. In Computational Logic inMulti-Agent Systems, 11th International Workshop, CLIMA XI, volume 6245 ofLecture Notes in Computer Science, pages 117–134. Springer, 2010.

38. J. F. Hubner, O. Boissier, R. Kitio, and A. Ricci. Instrumenting multi-agent organ-isations with organisational artifacts and agents: “Giving the organisational powerback to the agents”. Autonomous Agents and Multi-Agent Systems, 20, 2009.

39. Jomi Fred Hubner, Jaime Simao Sichman, and Olivier Boissier. Developing organ-ised multiagent systems using the MOISE. IJAOSE, 1(3/4):370–395, 2007.

40. Jena – A Semantic Web Framework for Java. Online, accessed on June, 14th,2010. http://jena.sourceforge.net/.

41. L. Li, B. Wu, and Y. Yang. Agent-based ontology integration for ontology-basedapplication. In Australasian Ontology Workshop, AOW 2005, Proceedings, pages53–59, 2005.

42. Elisa Marengo. Designing and programming commitment-based service-orientedarchitectures on top of agent and environment technologies. Technical Report RT129/2010, Dipartimento di Informatica, Universita degli Studi di Torino, 2010.

43. V. Mascardi, A. Locoro, and F. Larosa. Exploiting Prolog and NLP techniques formatching ontologies and for repairing correspondences. In 24th Convegno Italianodi Logica Computazionale, CILC 2009, Proceedings, 2009.

44. C. Masolo, L. Vieu, E. Bottazzi, C. Catenacci, R. Ferrario, A. Gangemi, andN. Guarino. Social Roles and their Descriptions. In D. Dubois, C. A. Welty,and M.-A. Williams, editors, Principles of Knowledge Representation and Reason-ing: Proceedings of the Ninth International Conference (KR2004), pages 267–277,Whistler, Canada, June 2004. AAAI Press.

45. P. McBurney and S. Parsons. Games That Agents Play: A Formal Framework forDialogues between Autonomous Agents. Journal of Logic, Language and Informa-tion, 11(3):315–334, 2002.

46. T. J. Norman, D. V. Carbogim, E. C. W. Krabbe, and D. N Walton. Argument andMulti-Agent Systems. In C. A. Reed and T. J. Norman, editors, ArgumentationMachines: New Frontiers in Argument and Computation, volume ArgumentationLibrary, pages 15–54. Kluwer Academic Publishers, 2004.

47. M. Obitko and V. Snael. Ontology repository in multi-agent system. In M. H.Hamza, editor, Conference on Artificial Intelligence and Applications, AIA 2004,Proceedings, 2004.

48. E. Oliva, P. McBurney, and A. Omicini. Co-argumentation Artifact for AgentSocieties. In I. Rahwan, S. Parsons, and C. Reed, editors, Argumentation in Multi-Agent Systems, 4th International Workshop, ArgMAS 2007, volume 4946 of LectureNotes in Computer Science, pages 31–46, Honolulu, HI, USA, May 2007. Springer.

49. A. Omicini, A. Ricci, and M. Viroli. Artifacts in the A&A meta-model for multi-agent systems. Autonomous Agents and Multi-Agent Systems, 17(3):432–456, 2008.

50. A. Omicini, A. Ricci, M. Viroli, C. Castelfranchi, and L. Tummolini. CoordinationArtifacts: Environment-Based Coordination for Intelligent Agents. In 3rd Interna-tional Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS2004), pages 286–293, New York, USA, August 2004.

51. Singh M. P. An Ontology for Commitments in Multiagent Systems. Artif. Intell.Law, 7(1):97–113, 1999.

52. A. Passadore, C. Vecchiola, A. Grosso, and A. Boccalatte. Designing agent interac-tions with Pericles. In 2nd International Workshop on Ontology, Conceptualizationand Epistemology for Software and System Engineering, ONTOSE 2007, Proceed-ings, 2007.

53. A. Pena, H. Sossa, and F. Gutierrez. Web-services based ontology agent. In 2ndInternational Conference on Distributed Frameworks for Multimedia Applications,DFMA 2006, Proceedings, pages 1–8, 2006.

54. M. Piunti and A. Ricci. Cognitive Use of Artifacts: Exploiting Relevant InformationResiding in MAS Environments. In J.-J. Ch. Meyer and J. Broersen, editors,Knowledge Representation for Agents and Multi-Agent Systems, First InternationalWorkshop, KRAMAS 2008, volume 5605 of Lecture Notes in Computer Science,pages 114–129, Sydney, Australia, September 2008. Springer.

55. M. Piunti, A. Ricci, O. Boissier, and J. F. Hubner. Embodied Organisations in MASEnvironments. In L. Braubach, W. van der Hoek, P. Petta, and A. Pokahr, editors,Multiagent System Technologies, 7th German Conference, MATES 2009, volume5774 of Lecture Notes in Computer Science, pages 115–127, Hamburg, Germany,September 2009. Springer.

56. M. Piunti, A. Ricci, O. Boissier, and J. F Hubner. Embodying Organisations inMulti-agent Work Environments. In Proceedings of the 2009 IEEE/WIC/ACMInternational Conference on Intelligent Agent Technology, IAT, pages 511–518.IEEE, 2009.

57. M. Piunti, A. Ricci, L. Braubach, and A. Pokahr. Goal-Directed Interactions inArtifact-Based MAS: Jadex Agents Playing in CARTAGO Environments. In IAT,pages 207–213. IEEE, 2008.

58. S. K. Rajamani and J. Rehof. Conformance Checking for Models of AsynchronousMessage Passing Software. In Computer Aided Verification CAV’02, number 2404in LNCS, pages 166–179. Springer, 2002.

59. A. S. Rao. AgentSpeak(L): BDI Agents Speak Out in a Logical Computable Lan-guage. In W. Van de Velde and J. W. Perram, editors, MAAMAW, volume 1038of Lecture Notes in Computer Science, pages 42–55. Springer, 1996.

60. A. Ricci, A. Omicini, M. Viroli, L. Gardelli, and E. Oliva. Cognitive stigmergy:Towards a framework based on agents and artifacts. In Weyns et al. [73], pages124–140.

61. A. Ricci, M. Piunti, D. L. Acay, R. H. Bordini, J. F. Hubner, and M. Dastani.Integrating heterogeneous agent programming platforms within artifact-based en-

vironments. In L. Padgham, D. C. Parkes, J. Muller, and S. Parsons, editors,AAMAS (1), pages 225–232. IFAAMAS, 2008.

62. A. Ricci, M. Piunti, and M. Viroli. Environment Programming in MAS – AnArtifact-Based Perspective. Autonomous Agents and Multi-Agent Systems. Toappear.

63. A. Ricci, M. Piunti, M. Viroli, and A. Omicini. Environment Programming inCArtAgO. In Multi-Agent Programming II: Languages, Platforms and Applica-tions, Multiagent Systems, Artificial Societies, and Simulated Organizations. 2009.

64. A. Ricci, M. Viroli, and A. Omicini. The A&A programming model and technologyfor developing agent environments in mas. In M. Mehdi Dastani, A. El Fallah-Seghrouchni, A. Ricci, and M. Winikoff, editors, PROMAS, volume 4908 of LectureNotes in Computer Science, pages 89–106. Springer, 2007.

65. Y. Shoham. Agent-Oriented Programming. Artif. Intell., 60(1):51–92, 1993.66. M. P. Singh. A social semantics for agent communication languages. In F. Dignum

and M. Greaves, editors, Issues in Agent Communication, volume 1916 of LectureNotes in Computer Science, pages 31–45. Springer, 2000.

67. M. S. Singh and A. K. Chopra. Correctness Properties for Multiagent Systems.In M. Baldoni, J. Bentahar, M. B. van Riemsdijk, and J. Lloyd, editors, DALT,volume 5948 of Lecture Notes in Computer Science, pages 192–207. Springer, 2009.

68. H. Suguri, E. Kodama, M. Miyazaki, H. Nunokawa, and S. Noguchi. Implemen-tation of FIPA Ontology Service. In Workshop on Ontologies in Agent Systems,OAS’01, Proceedings, 2001.

69. C. Vecchiola, A. Grosso, and A. Boccalatte. AgentService: a framework to de-velop distributed multi-agent systems. Int. J. Agent-Oriented Software Engineer-ing, 2(3):290 – 323, 2008.

70. M. Venkatraman and M. P. Singh. Verifying compliance with commitment proto-cols. Autonomous Agents and Multi-Agent Systems, 2(3), 1999.

71. M. Verdicchio and M. Colombetti. Communication Languages for Multiagent Sys-tems. Computational Intelligence, 25(2):136–159, 2009.

72. D. Weyns, A. Omicini, and J. Odell. Environment as a first class abstraction inmultiagent systems. Autonomous Agents and Multi-Agent Systems, 14(1):5–30,2007.

73. D. Weyns, H. Van Dyke Parunak, and F. Michel, editors. Environments for Multi-Agent Systems III, Third International Workshop, E4MAS 2006, Hakodate, Japan,May 8, 2006, Selected Revised and Invited Papers, volume 4389 of Lecture Notesin Computer Science. Springer, 2007.

74. P. Yolum and M. P. Singh. Commitment machines. In Proc. of ATAL, pages235–247, 2001.

75. F. Zambonelli, N. R. Jennings, and M. Wooldridge. Developing multiagent systems:The Gaia methodology. ACM Trans. Softw. Eng. Methodol., 12(3):317–370, 2003.