Auton Agent Multi-Agent Syst (2010) 20:369–400

DOI 10.1007/s10458-009-9084-y

Instrumenting multi-agent organisations

with organisational artifacts and agents
“Giving the organisational power back to the agents”

Jomi F. Hübner · Olivier Boissier · Rosine Kitio ·

Alessandro Ricci

Published online: 5 April 2009

Springer Science+Business Media, LLC 2009

Abstract The social and organisational aspects of agency have led to a good amount of
theoretical work in terms of formal models and theories. However, the conception and engi-
neering of proper organisational infrastructures embodying such models and theories are still
an open issue. The introduction of normative concerns with requirements of openness and
adaptation stresses this issue. The corresponding mechanisms for the current infrastructures
appear to be not appropriate for managing distributed and open normative organisations.
There is still the need of proper abstractions and tools to facilitate application agents taking
part in the monitoring of the organisation on one hand, and in the adaptation and definition of
the organisation in which they are situated on the other hand. In this paper we present and dis-
cuss ORA4MAS (Organisational Artifacts for Multi-Agent Systems), a proposed approach
aiming at these issues. Based on the Agents and Artifacts meta-model (A&A), it introduces
organisational artifacts as first class entities to instrument the organisation for supporting
agents activities within it.

Keywords Multiagent systems · MAS organisations · Artifacts

J. F. Hübner (B) · O. Boissier · R. Kitio

SMA/G2I/ENSM.SE, 158 Cours Fauriel, 42023 Saint-Etienne Cedex, France
e-mail: [email protected]; [email protected]
O. Boissier
e-mail: [email protected]
R. Kitio
e-mail: [email protected]

A. Ricci
DEIS, Università di Bologna, 47023 Cesena (FC), Italy
e-mail: [email protected]

123
370 Auton Agent Multi-Agent Syst (2010) 20:369–400

1 Introduction

Current applications of IT show the interweaving of both human and technological

communities (e.g. pervasive computing and ambient intelligence [23]), resulting in the con-
struction of connected communities (ICities [37]) in which software entities act on behalf
of users, cooperate and co-evolve with infohabitants, taking into account issues like trust,
security, flexibility, adaptation, and openness [23,37]. Current applications have led to an
increase in the number of agents, in the duration and repetitiveness of their activities, and
in the range of decision possibilities for the agents [24]. Moreover the number of agents
designers is also increasing, leading to a huge palette of heterogeneity in these systems.
One relevant topic in such systems is the coordination and control of the autonomous
agents so that coherent functioning is promoted at the system level. The social and organi-
sational aspects of agency are one possible answer to this problem. It has become nowadays
an important focus of interest in the MAS community (e.g. the COIN workshop series [39]).
Recent research in the MAS domain has provided many proposals for organisation modelling
languages used to reify organisations into organisation specifications that are interpreted and
used by application agents. In some sense, agents have become ‘organisation-aware’. In the
context of open systems, organisation-oriented middlewares have been proposed to interpret
and support the management of the organisation according to its specification aiming mainly
at ensuring that agents behave in accordance to the specified organisation.
While no consensual model of organisation emerges from these different works, the design
of organisation-oriented middleware has led to a general architecture composed of services
and agents responsible for organisational concerns such as monitoring, consistency main-
tenance, sanction applications and reorganisation. This middleware is generally introduced
between the application agents and the agent communication platform. In those cases, the
application agents do not have the possibility to take part in the management of the orga-
nisation in which they participate. In some sense, the agents are under the ‘control’ of the
organisational middleware with respect to the management and use of their organisation.
Our motivation is to soften the management of openness promoted by these organisational
middlewares. We want not only to conceive an organisational infrastructure that supports and
controls the agents in their participation to the organisation, but also consider an infrastructure
that the agents can manage and use.
To this end, we consider the organisational infrastructure as part of the computational
environment, engineered by MAS designers, where agents of the MAS are logically situated
and where they can perceive and act upon as a first-class entity of their world. The role
the environment can play in the MAS engineering has been remarked by recent works in
the literature [45], as a first-class abstraction useful to encapsulate and provide functional-
ities for enabling and mediating agent interactions and supporting agent coordination [44].
Moreover, besides being a first-class abstraction for MAS designers at design time, suitable
environment infrastructures make it possible to keep the abstraction alive at runtime, mak-
ing the computational environment a first-class entity that agents can perceive, exploit, and
manipulate to achieve their goals [42]. Among the others, the A&A meta-model introduces
the notion of artifact as the basic abstraction to design and structure complex computa-
tional environments [28,35]: an MAS is conceived as an open collections of agents working
together in workspaces where they dynamically construct, share, and use artifacts modelling
resources and tools populating the agent world. While agents model the goal and task oriented
part of the system, i.e. the locus of decision and control, artifacts model function-oriented
parts, the tools that are used and adapted by agents to achieve their individual and collective
goals.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 371

Following the A&A perspective, in this paper we introduce a model called ORA4MAS
which conceives organisational infrastructures inside MAS composed by both agents respon-
sible of organisation management, referenced as organisational agents, and suitably designed
artifacts, referenced as organisational artifacts. These artifacts are intentionally used on the
one side by agents participating in the organisation to make their activities more effective
and, on the other side, by organisational agents to control, manage, and adapt the organisation
at runtime. This is in analogy with human organisations that are populated by humans (as
participants and part of the organisational machinery), and by a rich set of artifacts and tools
that humans use to support their activities both inside the organisation and in the manage-
ment of the organisation itself. By promoting these mechanisms into first class entities as
artifacts, open normative multi-agent organisations may be conceived and supported at the
same abstraction level as the agents populating them.
In the first part of the paper (Sect. 2), we have a look at the different approaches that have
been developed in the field of multi-agent organisation infrastructures, pointing out some
drawbacks. Section 3 briefly presents the Moise+ and A&A models, which our proposal
is based on. In Sect. 4 we describe the general approach of ORA4MAS: our contribution
towards a solution to some limitations identified in current works. The Sect. 5 and Sect. 6
exemplify the approach for a concrete organisational model and present some implementa-
tion solutions. The contribution is then discussed in Sect. 7. We finally provide concluding
remarks and perspectives on the work in Sect. 8.

2 Organisational infrastructures for multiagent systems

Recent developments in the MAS domain originated many proposals towards the organisation
oriented programming of MAS [3]. In all of them, we distinguish two important components:
a declarative Organisation Modelling Language (OML) (e.g. Moise+ [22], Islander [11])
and an organisation implementation architecture (e.g. S -Moise+ [21], Ameli [12]). The
OML defines the language used to write the specification of the organisation for an MAS. It
is used to express specific constraints and cooperation patterns imposed on the agents by the
designer (or the agents), resulting in an explicit representation that we call an Organisation
Specification (OS). A collective entity, called an Organisation Entity (OE), is created by the
agents when they adopt the roles specified in the OS. A set of computational tools related
to the OML supports these agents to properly ‘play’ their roles. The development of these
tools normally considers both an agent-centred and a system-centred point of view.1 In the
former, the focus lies on organisational agent-level deliberative mechanisms to interpret and
reason about the OS and OE to which the agents participate [5,7]. In some sense, the aim is
to develop organisation-aware agents. In the latter, the main concern is the organisational
infrastructure (OI). In the sequel, we will focus on this second component.
The OI has two purposes: (i) to help and support the agents in the achievement of the
organisation’s global purpose by providing them access to organisational services and (ii)
to control and enforce the satisfaction of the organisational constraints (e.g. agents playing
the right roles, following the specified norms). This second point of view is important in
heterogeneous and open systems where the agents that enter into the organisation may have
unknown architectures and behaviours.
The implementation of an OI normally follows a similar trend to multi-agent communica-
tion platforms [3]. They introduce the organisation support mechanisms into a three layered

1 In some proposals, these points of view are called agent and institutional perspectives [41].

123
372 Auton Agent Multi-Agent Syst (2010) 20:369–400

Application
Domain Agent Agent

Organisational Proxy Organisational Proxy

Organisation
Infrastructure

Organisational state Organisational services

Members Coordination Reorganisation

Specification Role assignment Sanction

Other... Others...

Agent Platform

Fig. 1 Common architecture for organisation infrastructures in open MAS

middleware architecture, as depicted in Fig. 1: (i) domain or application agents that use an
organisational proxy component to interact with the organisation and achieve application
goals, (ii) the organisational infrastructure that is responsible for providing the agents with
organisational services for helping and controlling their actions in the OE according to the OS,
and (iii) the agent platform that connects all components in a distributed and heterogeneous
context (e.g. JADE [2]). Examples of organisational proxies in some OI implementations are
angels in OperA [10], governors in Ameli [12], wrappers in TeamCore [32], and OrgBox
in S -Moise+ [21]. One important point to notice is that all the access to the OI is mediated
by these organisational proxies. This layered architecture results in an engineering approach
where the MAS development is considered to be addressed by three kinds of designers:
domain or application designers (for the agents and the specification of the OS using the
OML), MAS or OI designers (for the organisational infrastructure and OE management),
and agent platforms designers.
In contrast to this general picture, we will turn for a moment to the way that human
organisations are used to facilitate common tasks. They are populated by humans (as par-
ticipants and part of the organisational machinery), and by rich sets of artifacts and tools
that humans use to support their activities inside the organisation. They also are used to
support the organisation itself by encapsulating essential infrastructure services. According
to psycho-sociological frameworks such as Activity Theory and Distributed Cognition [25],
the notion of artifact (and tool, taken here as a synonym) plays a key role for the overall sus-
tainability of an organisation and the effectiveness and efficiency of activities taking place
inside it. In particular, some of these artifacts appear to be vital for supporting the coordi-
nation of organisational processes and management: for instance, by making more effective
the communication among the members of an organisation (e.g. the telephone, instant-mes-
saging services, chat-rooms), by providing information useful for orienting the activities

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 373

of organisation participants (e.g. signs inside a building), by coordinating participants (e.g.

queue systems at the post-office), and by controlling access to resources and enforcing norms
(e.g. the badge used by members in a computer science department to access certain rooms
or use some other artifacts, such as copiers). Human societies and organisations continuously
improve their experience in designing artifacts more and more effectively to support both
organisation participation—helping members to cope with the complexity of social activities
and work—and organisation management—helping managers to monitor and control the
organisation behaviour as a whole.
From this brief introduction, we can point out some features of the current OI that moti-
vate our proposal. It is important to note that the issues stated here do not concern solely the
implementation level, but also the conceptual and theoretical levels.
1. Mismatch of the abstraction level for organisational support. The components of an OI
are considered as services. Even if, in some proposals (e.g. S -Moise+ [21]), they have
an agent status, they are mainly considered as ‘services’ with well-defined operation-
oriented interfaces and contracts, and not really intelligent agents to interact and cooper-
ate with through conversations and articulated ACL-based protocols, possibly providing
unexpected and autonomous behaviours. However, those service-agents are taking deci-
sions on the adaptation of the organisation, on the sanctions, and on the enforcement of
norms which target the agents in the application layer. They are nonetheless not acces-
sible by an application designer when s/he needs to customise some decisions of the
system in the organisational dimension (e.g. a sanction, reorganisation or role assignment
strategies).
2. Organisation and Environment are not integrated. In the current development process
for MAS, the designer (and the agents) have to deal with two kinds of shared medium of
interaction: a virtual organisation (where the agents adopt roles, send messages according
to particular protocols, execute goals according to collective procedures, and participate
in the manipulation, construction and realisation of shared social workflow throughout
which they coordinate themselves) and the real physical environment (where the agents
act, move, coordinate to access shared resources). Although their concerns are similar,
their reification has led to different approaches: dedicated layer in the case of organi-
sation, proper abstractions accessible and usable by the agents in case of environment.
Moreover, even if organisation and environment have each their own specifics, they are
not independent in the sense that organisations are situated in an environment. As shown
by Okuyama et al. [27] some of its constructs (e.g. norms) take a different meaning
according to their anchoring in the environment.
3. Overwhelming power of the OI in the services. According to the general architecture
depicted in Fig. 1, services operating in the OI perform three main types of function:
coordination, detection and decision/judgement on norm enforcement. For example, if
some agent wants to perform some action or send a message that its organisation does
not allow, the infrastructure (by means of the organisational proxy) will detect this viola-
tion attempt and may decide whether to apply a sanction or not without any interference
of the application agents. Based on the way that human organisations work, we con-
sider that it is more suitable that the agents operating on the application layer should get
their organisational power back and take decisions on the sanctions and reorganisation.2
The middleware approach to design the OI precludes the incorporation of strategies for
sanction and reorganisation in the application level.
2 Although different aspects of the state of an organisation can change (e.g. the current agents and groups),
we consider here reorganisation as the process of changing the current OS of the MAS.

123
374 Auton Agent Multi-Agent Syst (2010) 20:369–400

Even if the services are configured to use some application knowledge, the clear separation
into two different abstraction layers hinders agents operating at the application level in
taking part in the regulation or reorganisation of the application. As stated by Castelfran-
chi et al. [6], social order is a necessary mechanism, and this mechanism requires that
the agents themselves become actors of the mechanism.

These global statements lead us to propose a general model for structuring and architecting
organisational support mechanisms by changing their status and promoting them as first class
entities accessible and manageable by the agents. In concordance with this approach, we pro-
pose to move these mechanisms from the organisational middleware layer to the application
layer. A global picture of this proposal is presented in Sect. 4.

3 Foundations

Before presenting ORA4MAS in Sect. 4, we set the context of our work. We briefly describe
the Moise+ organisational model, which is used in the sequel to reify and illustrate
ORA4MAS (a fuller account of it can be found in [22]). We also introduce the basic ideas
provided by the A&A meta-model that are used as guidelines for our approach. A more
detailed description of this meta-model—including aspects not essential for this paper—can
be found in [36].

3.1 The Moise+ model

The Moise+ model proposes an organisational modelling language that explicitly decom-
poses the specification of organisation into structural, functional, and deontic dimensions [22].
The structural dimension specifies the roles, groups, and links of the organisation. The defi-
nition of roles states that when an agent decides to play some role in a group, it is accepting
some behavioural constraints related to this role. The functional dimension specifies how the
global collective goals should be achieved, i.e. how these goals are decomposed (in global
plans), grouped in coherent sets (by missions) to be distributed to the agents. The decom-
position of global goals results in a goal-tree, called scheme, where the leaf-goals can be
achieved individually by the agents. The deontic dimension is added in order to bind the
structural dimension with the functional one by the specification of the roles’ permissions
and obligations for missions.
As an illustrative and simple example of an organisation specified using Moise+ , we
consider agents that aim at writing a paper and therefore have an organisational specification
to help them to collaborate. The structure of this organisation has only one group (wpgroup)
with two roles (editor and writer ) that inherit all properties defined for the role author. The
cardinalities and links of this group are specified, using the Moise+ notation, in Fig. 2a: the
group wpgroup can have from one to five agents playing writer and exactly one playing
editor ; the editor has authority over writer and every agent playing author (and by inheri-
tance everyone playing writer or editor ) has the possibility to communicate with every agent
playing author (there is a communication link from author to author ). In this example, the
editor and the author roles are not compatible. To be compatible, a compatibility relation
must be explicitly added in the specification.
To coordinate the achievement of the goal of writing a paper, a scheme is defined in
the functional specification of the organisation (Fig. 2b). In this scheme, a draft version of
the paper has to be initially defined (identified by the goal fdv in Fig. 2b). This goal is

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 375

(a) (b)

(c) (d)

Fig. 2 Graphical representation of the organisational specification for the writing paper example in the
Moise+ OML. a Structural specification. b Functional specification. c Mission cardinalities (part of FS).
d Deontic specification

decomposed into three sub-goals: write a title, an abstract, and the section titles. The agents
then have to ‘fill’ the paper’s sections to get a submission version of the paper (identified by
the goal sv ). The goals of this scheme are distributed in three missions which have specific
cardinalities (cf. Fig. 2c): the mission m Man is for the general management of the process
(one and only one agent can commit to it), mission mCol is for the collaboration in writing
the paper’s content (from one to five agents can commit to it), and mission m Bib is for
getting the references for the paper (one and only one agent can commit to it). A mission
defines all goals an agent commits to when participating in the execution of a scheme, for
example, a commitment to the mission m Man is indeed a commitment to achieve four goals
of the scheme. Goals without an assigned mission are satisfied by the achievement of their
subgoals. The deontic relation from roles to missions is specified in Fig. 2d. For example,
any agent playing the role editor is permitted to commit to the mission m Man.
The specification of an organisation is written in a suitable language [19], that the agents
and the OI are supposed to interpret. This language is founded on components represented by
predicates and functions. We present here only those components that are used in the sequel
of the paper. Considering an OS, G is the set of all group specifications, R is the set of all
roles, S is the set of all scheme specifications, M is the set of all missions, and
is the set
of all goals.

– compat (g, ρ, C): is a predicate that is true when the role ρ (ρ ∈ R) is compatible with
all roles in the set C (C ⊆ R) when played in the group g (g ∈ G );
– mission_scheme(m, s) is a predicate that is true when the mission m (m ∈ M) belongs
to the scheme s (s ∈ S );
– goal_mission(ϕ, m): is a predicate that is true when the goal ϕ (ϕ ∈
) belongs to
the mission m (m ∈ M);
– obl(ρ, m): is a predicate that is true when the role ρ has an obligation relation to the mission
m;
– per (ρ, m): is a predicate that is true when the role ρ has a permission relation to the
mission m;

123
376 Auton Agent Multi-Agent Syst (2010) 20:369–400

– goal_r ole(ϕ, ρ): is a predicate that is true when the role ρ is obliged to the goal ϕ, this
predicate is defined as follows

goal_r ole(ϕ, ρ) ↔ goal_mission(ϕ, m) ∧ obl(ρ, m)

– deadline: G → Z: is a function that maps each goal to an achievement deadline;

– maxr p: R × G → Z: is a function that maps pairs of roles and groups to the maximum
number of players of that role in the group (upper bound of role cardinality);
– minr p: R × G → Z: is a function that maps pairs of roles and groups to the minimum
number of players of that role necessary for the group to be considered well-formed (lower
bound of role cardinality);
– maxmp : M × S → Z: is a function that maps pairs of missions and schemes to the
maximum number of agents that can commit to that mission in the scheme (upper bound
of mission cardinality);
– minmp: M × S → Z: is a function that maps pairs of missions and schemes to the mini-
mum number of agents that have to commit to that mission for the scheme to be considered
well-formed regarding to that mission (lower bound of mission cardinality).

3.2 The agents and artifacts (A&A) conceptual framework

The notion of MAS environment, as noted in recent literature, has gained a key role in the
recent past, becoming a mediating entity, functioning as an enabler but possibly also as
a manager and constrainer of agent actions, perceptions, and interactions (a comprehensive
survey is presented in the JAAMAS special issue on environments for MAS [45]). According
to such a perspective, the environment is not a merely passive source of agent perceptions
and target of agent actions, which actually is the dominant perspective on agency and in
MAS, but a first-class abstraction that can be suitably designed to encapsulate some funda-
mental functionalities and services, supporting coordination in MAS besides agent mobility,
communications, security, etc.
Among the various approaches, the A&A conceptual framework introduces a notion of
work environment, representing such a part of the MAS explicitly designed on the one hand
by MAS designers to provide various kinds of functionality—including, for instance, MAS
coordination—and perceived on the other hand as a first-class entity by agents of the MAS,
as part of their context that they can reason about, use and possibly adapt to perform their
individual and social tasks [28,36]. A&A work environments are made of artifacts, repre-
senting resources and tools that agents can create and use to perform their individual and
social activities. Artifacts then can be conceived as complementary first-class abstractions to
agents populating an MAS: while agents are goal-oriented pro-active entities, artifacts are
non-autonomous function-oriented entities, i.e. designed by MAS designers to provide some
kind of functionality or service for agents. Among the several sorts of artifacts, coordination
artifacts have been introduced as an important class of artifacts [29], as artifacts mediating
agent interactions and encapsulating some kind of coordinating functionality—whiteboards,
event services, shared task schedulers are examples of such artifacts. Besides the notion of
artifact, A&A introduces the notion of workspace as a logic container of agents and artifacts.
Workspaces can be used to structure the overall sets of entities, defining a topology of the
agent environment. In this view an MAS is conceived as a set of (distributed) workspaces,
containing dynamic sets of agents working together by both directly communicating and
sharing artifacts.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 377

Artifact
observable property

usage observable property

Agent interface
Layer
operation

operation

black dbase Work link operation

board Environment link operation
sche-
duler check Layer link
shared
list interface
KB

workspace w0 workspace w1 manual observable

events

Fig. 3 (Left) Abstract representation of workspaces, populated by agents (represented by circles) and artifacts
(represented by squares). (Right) A representation of the main parts and properties of an artifact, with the
usage interface, the observable properties and the manual

Figure 3 shows an abstract representation of an artifact as defined in the A&A meta-model,

exhibiting analogous parts and properties of artifacts to those found in human society. The
artifact function—and related artifact behaviour—is partitioned into a set of operations, which
agents can trigger by acting on an artifact usage interface. The usage interface provides all
the controls that make it possible for an agent to interact with an artifact, that is to use and
observe it. Agents can use an artifact by triggering the execution of operations through the
usage interface and by perceiving observable events generated by the artifact itself, as a
result of operation execution and evolution of its state. Besides the controls for triggering
the execution of operation, an artifact can have some observable properties, i.e. properties
whose value is made observable to agents, without necessarily executing operations on it. The
interaction between agents and artifacts strictly mimics the way in which humans use their
artifacts. For a simple but effective analogy, consider a coffee machine: the set of buttons of
the coffee machine represents the usage interface, while the displays that are typically used
to show the state of the machine represent artifact observable properties. The signals emitted
by the coffee machine during its usage represent observable events generated by the artifact.
Table 1 shows the basic set of actions the agents can use to work with artifacts in work-
spaces, as supported by computational and programming models implementing A&A (in
particular CArtAgO, described in Sect. 6.2). Among these actions, we remark use and focus.
The former makes it possible to trigger and control the execution of an operation inside the
artifact—by acting on controls of the usage interface. The latter makes it possible for an agent
to be continuously (and automatically) aware of the dynamic value of observable properties
and events generated by the artifact, both modelled as percepts from the environment.
Analogously to the human case, in A&A each artifact type can be equipped by the artifact
designer with a manual, containing a machine-readable description of the artifact’s function
(i.e. its intended purpose), the artifact’s usage interface (i.e. the observable ‘shape’ of the
artifact), and the artifact’s operating instructions (i.e. usage protocols or simply how to cor-
rectly use the artifact so as to take advantage of all its functionalities) [43]. The manual is a
key feature when truly open systems are of concern, with intelligent agents that dynamically
discover and select which kind of artifacts could be useful for their work, and then can use
them effectively even if the artifacts have not been pre-established by MAS designers for the
purpose.
Finally, artifacts can be composed together by means of link interfaces, which are sets
of input/output ports that make it possible to enable inter-artifact interaction and finally link

123
378 Auton Agent Multi-Agent Syst (2010) 20:369–400

Table 1 Abstract description of the core action set provided in CArtAgO to work with artifacts, with action
parameters and action feedback

Action Description

joinWsp(wid ) Joins an existing workspace, identified by wid

quitWsp(wid ) Quits a workspace
use(ar tid , oname, opar s, sid, timeout) Triggers the execution of an operation on the artifact
identified by ar tid , by acting on the usage interface
control oname and passing parameters opar s. A sen-
sor sid can optionally be specified to collect events
generated by the operation execution. If no sensor
is specified, events are made observable to the agent
directly as percepts. A timeout can be specified, to
generate an internal action failure event if the action
is not completed within the specified time
sense(sid, f ilter , timeout) : evt Retrieves (and returns as an action feedback) from the
specified sensor an event matching the filter, suspend-
ing the action until the event is found or a timeout
occurs (and the action fails)
focus(ar tid , sid) Starts observing the artifact identified by ar tid , using
the sensor sid. The values of artifact observable prop-
erties are directly mapped into percepts perceived by
agents
stopFocus(ar tid ) Stops observing an artifact
observeProperty(ar tid , pname) : pvalue Read from artifact ar tid the current value of its pname
observable property
makeArtifact(wid , aname, τ, apar s) : ar tid Creates in workspace wid a new artifact named aneme
from template τ and parameters apar s. The artifact
id ar tid will be returned as action feedback
disposeArtifact(ar tid ) Disposes an existing artifact
lookupArtifact(wid , aname) : ar tid Gets the identifier of artifact aname in workspace wid

together artifact functionalities. Actually, link interfaces serve two purposes: on the one side,
to explicitly define a principle of composability for artifacts, enabling the construction of
articulated and complex artifacts by means of simpler ones; on the other side, to realise dis-
tributed (composed) artifacts by linking together artifacts belonging to different workspaces.

4 ORA4MAS model for organisational infrastructure

In this section we propose an organisational infrastructure called ORA4MAS where the basic
idea is to instrument the multi-agent environment and the organisations living upon it in terms
of agents and artifacts, based on the basic A&A meta-model presented in the previous section.
The overall picture accounts for organisational agents that dynamically articulate, manage
and adapt the organisation by creating, linking and manipulating the organisational artifacts,
which are discovered and used by the member agents to work inside the organisation.

4.1 From organisational model to ORA4MAS

As stated in Sect. 2 and by Dignum and Dignum [9], the organisation of an MAS also serves as
an instrument to control the autonomy of the agents. The success of an organisation depends
on how the behavioural constraints stated in the organisational specification are ensured by

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 379

the organisational infrastructure. These behavioural constraints are entitled by the norms that
are established by the specification of the organisation. In the context of this work, a norm
is an obligation, permission, or interdiction to perform some action or achieve some goal.
A norm also has a condition that states when it is active and a deadline to be fulfilled.3 For
example, whenever an agent plays the role writer in a group and there is a writing paper
scheme with not enough agents committed to the mission mCol (that is the condition part of
the norm), the agent is obliged to commit to the mission mCol (that is the action part of the
norm).
As stressed by Grossi et al. [18], norms without enforcement mechanisms are worthless.
Therefore, these norms, which are normally abstract concepts, should be implemented by
means of mechanisms that instrument them in the organisational entity that is the context in
which they have to be considered. In the ORA4MAS approach, we consider two types of
such mechanisms:4

– Regimentation is a mechanism that simply prevents the agents from performing actions that
are forbidden by a norm. More precisely, we regiment some actions in order to preserve
important features (e.g. wellformedness) of the organisation. For example, the writing
paper group can have at most one agent playing editor; to ensure that this constraint is
respected, the organisational action ‘adopt editor role’ in this group is regimented accord-
ingly.
Since this mechanism has to work in an open system for any kind of agent, it has to be
implemented ‘outside’ the agents in the OI. Therefore, a very special set of actions (those
that are under the control of the OI, such as ‘send a message’ or ‘role adoption’) can be
regimented.
– Enforcement is a mechanism which is applied after the detection of the violation of some
norm. While regimentation is a preventive mechanism, enforcement is a reactive one.
From the point of view of the agents, they may decide to obey the norm or not according
to their local view of the organisation. From a system point of view, the fulfilment of the
norms should be detected, evaluated as a violation or not, and then judged as worthy of
sanction/reward or not.
While detection can be implemented as an automatic process that does not require decision,
the evaluation and the judgement need deliberation and reasoning. Given the difference
among these tasks, the implementation of the enforcement mechanism should be assigned
to different entities of the system. For instance, the detection could be implemented in the
OI, whereas evaluation and judgement are more suitably implemented in agents.

These two mechanisms allow ORA4MAS to tackle the problem of balancing (i) the ensur-
ing of very important properties of the OE by means of regimentation and, by means of
enforcement, (ii) the agents’ autonomy required to keep the possibility to adapt and evolve
the organisation. The norms of the MAS can be instrumented either as regimentations or
enforcement mechanisms depending on which side the designer wants to give more weight.
Briefly, regimentation should be used to fully constrain the actions of the agents and enforce-
ment should be used when some violation is allowed (or even desired).

3 We are aware that the concept of norm is broader and more complex than that used in this paper (e.g. the work
of Tuomela and Bonnevier-Tuomela [40] and the Deontic Logic in Computer Science workshop series [17]).
For the present paper however this simple and informal definition is enough to discuss the proposal.
4 This classification is based on the proposal described in [14,18]. However, we present them in a more specific
context: regimentation is applied only to the interdiction of organisational actions and enforcement is applied
to the other cases.

123
380 Auton Agent Multi-Agent Syst (2010) 20:369–400

Organisational
Regimentation done by
Artifact

mechanisms

used by
Enforcement done by

detection

evaluation
done by
Agent
judgement done by

Fig. 4 General relation between the mechanisms that implement the norms and the organisational agents and
artifacts of ORA4MAS

As illustrated in Fig. 4, two classes of handling mechanisms can be distinguished: man-

agement in terms of interpretation and checking of their satisfaction before executing changes
in the OE; and management in terms of detection of non-fulfilment, statement of violation or
not and sanction execution. From the requirements stated in Sect. 3, these two mechanisms
of the organisation management must be taken into consideration by different entities. In our
proposal, organisational artifacts and agents are those entities.

4.2 Organisational artifacts and agents

Here we use the terms organisational agents and organisational artifacts to identify those
agents and artifacts of the MAS which are part of the organisational infrastructure. They are
responsible for activities and encapsulate functionalities concerning the management and
enactment of the organisation. Different OMLs require a different set of suitable artifacts
and agents. For example, in the AGR organisational model [13], we can conceive artifacts to
manage groups; for Islander [11], the artifacts can be used to manage the scenes. In Sect. 5
a detailed set of artifacts are presented for the Moise+ model.
In our approach (see Fig. 4), regimentation mechanisms are implemented in the artifacts
that the application agents have to use to access the organisation. For instance, to adopt a
role the agent has to use the appropriate artifact. The operation will be successfully executed
only in the case the agent is allowed to adopt the role, otherwise the adoption fails. The agent
autonomy for some organisational actions is therefore limited by the artifact. Organisational
artifacts are therefore those artifacts that agents of an organisation may want or have to use
in order to participate in organisational activities and to access organisation resources. They
encapsulate organisational norms and functionalities, such as enabling, mediating and ruling
on agent interaction, and tracing resource access. Analogously to the human case, artifacts
are used here to reify and modularise the functional part of the organisation’s machinery. It
defines a distributed set of organisational resources and tools that can be perceived and used
by agents as first-class entities and that can be dynamically adapted and possibly replaced (by
agents themselves) during the organisation lifetime. Moreover, these artifacts do not create
a sort of hidden organisational layer where the application is deployed, but rather they are
placed beside the agents as available tools in the environment.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 381

Some norms, however, cannot be implemented in the same way since they may be vio-
lated. In this case, the functionality provided by the artifacts is to detect and show (by means
of observable properties) the non-fulfilment of a norm. We consider that agents should be
informed of the current status of a norm and can evaluate the existence of a violation or not
and take the best decision regarding the application objectives. To stress the embedding of
dedicated reasoning and strategies related to the management of the organisation we call
these agents organisational agents. They can be dedicated agents or agents of the application
embedding special knowledge. They are essentially managers responsible for the creation
and management of the organisational artifacts. Such activities typically include observing
artifacts’ dynamics and possibly intervening, by changing and adapting artifacts or inter-
acting directly with other agents, so as to improve the overall (or specific) organisational
processes or taking some kinds of decisions when detecting violations.5 As an example, in
the context of the Moise+ model, one or multiple scheme manager agents can be intro-
duced, responsible for monitoring the dynamics of the execution of a scheme by observing
a specific artifact. The scheme artifact and scheme manager agents are designed so that the
artifact allows for violation of the deontic rules concerning the commitment of missions by
agents playing some specific roles, and then the decision about what action to take—after
detecting the violation—can be the responsibility of the manager agent.
After sketching the basic concepts underlying the ORA4MAS approach, in the next section
we describe how a fully-fledged organisational model can be implemented with organisa-
tional agents and artifacts.

5 ORA4MAS artifacts for M OISE+

We use here the Moise+ model presented in Sect. 3.1 to identify and shape a basic set of
organisational artifacts that constitute the building blocks of ORA4MAS. They are a sort
of ‘reification’ of the structural specification (SS), functional specification (FS), and deontic
specification (DS). This basic set accounts for: OrgBoard artifacts, used to keep track of the
current state of deployment of the organisational entity in the overall; GroupBoard artifacts,
used to manage the life-cycle of a specific instance of a group of agents; SchemeBoard arti-
facts, used to support and manage the execution of a social scheme; and NormativeBoard
artifacts, used to maintain information concerning the agents compliance or not according to
norms.
An organisational entity is then reified by a set of artifacts: exactly one OrgBoard; one
GroupBoard for each instance of a group of agents; one SchemeBoard for each scheme being
executed by the agents; and one NormativeBoard for each SchemeBoard. These organisational
artifacts are linked together to allow the synchronisation of some operations and to share
information required to maintain a coherent and consistent state of the OE. For example, the
state of the NormativeBoard artifact needs to be updated with the information about the agents
committed to missions so that it can evaluate the fulfilment of norms. The links between the
artifacts are the following: the OrgBoard is linked to all other artifacts of the organisation; the
GroupBoard is linked to all schemes its agents are responsible for; the SchemeBoard is linked
to exactly one NormativeBoard that verifies the status of the norms related to the execution of
the scheme; and finally the NormativeBoard is linked to its SchemeBoard and all GroupBoards
responsible for the corresponding scheme.
5 In this paper we focus on the regimentation management and detection of non-fulfilled norms done by the
organisational artifacts. The use of artifacts as an instrument to help the entire enforcement process is better
discussed in Hübner et al. [20].

123
382 Auton Agent Multi-Agent Syst (2010) 20:369–400

OrgBoard GroupBoard SchemeBoard NormativeBoard

OrgSpecification OrgBoard OrgBoard OrgBoard

Group Boards Type NormativeBoard NormsStatus

Scheme Boards PlayableRoles ResponsibleGroupBoards

Normative Boards PlayersOfRole Type updateAgentStatus

SchemeBoards PlayableMissions

getOrgAgents PlayersOfMissions

getMemberAgents adoptRole
GoalsState
leaveRole
registerOrgArt commitMission

add/removeScheme leaveMission
isMember
setGoalAchieved

Fig. 5 Basic kinds of artifacts in ORA4MAS, with their usage interface, including operations (represented
by circles) and observable properties (represented by rectangles), and the link interface (represented by half
eclipses in the bottom of the artifacts)

In the following we briefly describe these artifacts. Here we consider just a core set of the
characteristics and functioning of the artifacts, skipping most details that would make heavy
the overall understanding of the approach.

5.1 OrgBoard artifact

An abstract representation of the OrgBoard is depicted in Fig. 5. The observable properties

of this artifact are:
OrgSpecification: contains the specification of the OE written in the Moise+ OML. Agents
may use this observable property to get the organisational specification. They can then
reason about it and decide whether they want to enter in the organisation.
GroupBoards, SchemeBoards, and NormativeBoards: contain identifiers of all instances of
GroupBoard, SchemeBoard, NormativeBoard, respectively, within the OE. Generally speak-
ing, these observable properties make it possible for agents observing an OrgBoard to know
the current set of SchemeBoard, GroupBoard, NormativeBoard instrumenting the OE.
The usage interface of the OrgBoard has the following operations:
getOrgAgents(): used to get the set of agents having the status of organisational agent in the
OE (this status is set during the deployment of the system).
getMemberAgents(): used to get the set of all agents playing at least one role within a group
of the organisation entity.6
The main link operation of the OrgBoard is registerOrgArt. It is used in the initialisation process
of each new instance of GroupBoard, SchemeBoard, and NormativeBoard to be registered as an
artifact linked to the OE represented by the OrgBoard.

5.2 GroupBoard artifact

A GroupBoard is an artifact instantiated upon a specific group specification and provides func-
tionalities to manage the corresponding group instance in the OE. It maintains a consistent
6 This operation would be implemented as an observable property; however due to distributed characteristic
of the information (its is managed by all group boards), to maintain the observable property always up to
date may consume a lot of time. Using an operation, the list of member agents can be computed only when
demanded and then cached.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 383

state of that group by regimenting some important norms bearing on structural properties.
For instance, whenever some agent asks for a role adoption in the group managed by the
GroupBoard, the GroupBoard regiments a set of norms that state when a role can be adopted:
(1) the role belongs to its group specification; (2) each role that the agent already plays is
specified as compatible with the new role; and (3) the number of players is less or equal to
the maximum number of players defined in the group’s compositional specification.
As an artifact, the GroupBoard has some observable properties (Fig. 5). Among them, the
most relevant are:
OrgBoard: is the reference to the OrgBoard that represents the OE to which the group belongs.
Type: is the identification of the group specification in the structural specification (an element
of G ).
PlayableRoles: contains all roles that can still be played in this group, i.e. those for which the
number of players is not the maximum yet. This property changes whenever a new agent
enters into the group by adopting a role.
PlayersOfRole: contains the names of all agents belonging to the group and their correspond-
ing roles.
SchemeBoards: contains a set of all schemes the group is responsible for.

Generally speaking, the above defined observable properties of the GroupBoard enable agents
to know which are the available roles and their constraints, which are the participant agents,
and which are the other organisational artifacts linked to the GroupBoard. To perceive the
observable properties of a GroupBoard, agents can use either active perception (polling the
artifact) or the focus mechanism (which continuously send signals to the agent whenever
some observable property change, cf. Table 1).
The usage interface accounts for the following operations:
adoptRole(ρ): used by an agent to adopt a new role in the group, where ρ ∈ R is the identifier
for a role in the SS. Note that an agent enters an organisation entity by adopting a role within
a GroupBoard.
leaveRole(ρ): used by an agent to give up the role ρ that it had adopted previously.

The operations that are regimented, like adoptRole, can fail. For example, when the agent that
triggered it is not allowed to do that. In such cases, instead of receiving a ‘successful’ signal
as the result of the operation, the agent receives a ‘failure’ signal which includes the reasons
for the failure.
The link operations of a GroupBoard manage the coordinations with its linked organisa-
tional artifacts. Among them, we have:
addSchemeBoard(sb): used by a SchemeBoard initialisation process to notify the GroupBoard
that it is responsible for the scheme sb. The GroupBoard updates accordingly its Scheme-
Boards observable property.
removeSchemeBoard(sb): used by a SchemeBoard linked to the GroupBoard, to notify that the
GroupBoard is no longer responsible for the execution of the scheme sb. The GroupBoard
updates accordingly its SchemeBoards observable property.
isMember(α): used by a SchemeBoard to inquire whether an agent α is a member (i.e. is
playing at least one role) of the group managed by the GroupBoard.
Before presenting the norms that a GroupBoard regiments, we introduce some predicates and
functions related to the internal state of this artifact. Let GB, Rg , and A be, respectively, the
set of current group boards, the set all roles that can be played in a group board created from
specification g, and the set of all agents participating in the OE.

123
384 Auton Agent Multi-Agent Syst (2010) 20:369–400

– gr oup_t ype(gb, g): this predicate is true when the group board gb ∈ GB is created based
on the group specification g ∈ G (cf. Sect. 3.1);
– plays(α, ρ, gb): this predicate is true when the agent α ∈ A plays the role ρ in the group
board gb ∈ GB;
– r player s : R × GB → Z with
def
r player s(ρ, gb) = |{α| plays(α, ρ, gb)}| (1)
this function gives the number of current players of the role ρ in the group gb.
Given the above definitions and the functions maxr p and minr p (cf. Sect. 3.1), we are able
to define the wellformedness property of a group
well_ f or med(gb) ← gr oup_t ype(gb, g) ∧
∀ρ∈Rg r player s(ρ, gb) ≥ minr p(ρ, g) ∧
r player s(ρ, gb) ≤ maxr p(ρ, g) (2)
Since role adoption is the very action that may bring a group into an inconsistent state,
two norms bearing on this organisational action are regimented by a GroupBoard: the role
compatibility norm and the role cardinality norm. In the following, these norms are repre-
sented as a pair where the first argument is the condition part stating when the norm is active
and the second argument is the action part stating an obligation, permission, or interdiction.
Role compatibility norm: roles are incompatible unless the contrary is explicitly stated
in the OS. Thus, if it is stated that two roles ρ1 and ρ2 are compatible inside a group g
(compat (g, ρ1 , {ρ2 })), it implies that an agent that plays ρ1 in a group board gb created
from the specification g cannot perform the operation adoptRole(ρi ) for any i
= 2 on the
corresponding artifact. This constraint on role adoption is formalised by the following norm:
( plays(α, ρ, gb) ∧ gr oup_t ype(gb, g) ∧ compat (g, ρ, C),
∀ρi ∈R\C F O R B I D D E Nα adoptRole(ρi )) (3)
The condition of the norm (the first line) is a conjunction of predicates. Its evaluation is
given by the particular circumstance of the group board (that defines whether plays(α, ρ, gb)
and gr oup_t ype(gb, g) hold or not) and the structural specification being used by the artifact
(that defines whether compat (g, ρ, C) holds or not). The action part of the norm (the last
line) states that it is forbidden for agent α to execute the action adoptRole for any role that
does not belong to the set of compatible roles C. Based on this norm, as soon as an agent
adopts a role (activating the norm), the adoption of any other incompatible role is forbidden
for it.
Role cardinality norm: the number of players of a role in a group is limited by the function
maxr p(ρ, g) defined from the SS. The following norm constrains the role adoption based
on the cardinality of the role:
(gr oup_t ype(gb, g) ∧ r player s(ρ, gb) ≥ maxr p(ρ, g),
∀α∈A ∀ρ∈R F O R B I D D E Nα adoptRole(ρ)) (4)
Since these two norms are of the type ‘action interdiction’, they can be easily imple-
mented in the artifact: whenever the adoptRole operation is requested by the agent α, the
conditions of all norms have to be verified using the structural specification and the current
state of the group artifact. If the condition of some of these norms holds, the execution of the
corresponding operation is denied.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 385

5.3 SchemeBoard artifact

The SchemeBoard is an artifact instantiated upon a specific social scheme specification of

the FS. It provides functionalities to manage the execution of a social scheme, coordinating
the commitments to missions and the achievement of goals. It manages the dependencies
between the missions and the goals as described by the social scheme.
The execution of a scheme has three phases: formation, goal achievement, and finishing.
In the formation phase, agents commit to the missions of the scheme. As in the GroupBoard,
a property of wellformedness is defined: roughly a scheme is well-formed if the mission
cardinalities are satisfied, i.e. there are enough agents committed to the missions (the formal
definition is presented below). To be well-formed is the condition to pass the execution to
the second phase. In the second phase, the goals should be fulfilled by the agents. Each agent
has to achieve the goals of the missions they are committed to. When the root goal of the
scheme is satisfied, the third phase starts and the scheme can be finished and removed from
the organisation (i.e. the corresponding artifact is destroyed).
During the execution of a scheme, its goals can be in three different states: waiting,
possible, or achieved. The initial state is waiting. It indicates that the goal cannot be pur-
sued yet because it depends on the achievement of other goals (called pre-conditions for a
goal) or the scheme is not yet well-formed. The set of pre-condition goals are deduced from
the goal decomposition tree of the scheme (as presented in Fig. 2b). For example, the goal
‘to write the conclusion of the paper’ (wcon) can be achieved only after the goal of ‘writing
sections’ (wsec ) was achieved. When all pre-condition goals are verified and the scheme
is well-formed, the state of a goal switches to possible. Then the agent(s) committed to a
mission containing the goal can pursue it. Note that the change from the state waiting to
possible is performed by the SchemeBoard, whereas the change from the state possible to
achieved is performed by the agents.
The observable properties of a SchemeBoard are the following (Fig. 5):
OrgBoard: is the reference to the OrgBoard that represents the OE in which the scheme is
being executed.
NormativeBoard: is the reference to the NormativeBoard of the scheme.7
ResponsibleGroupBoards: contains the references to the groups that are responsible for the
scheme.
Type: is the identification of the scheme specification in the functional specification (an
element of S ).
PlayableMissions: contains all missions that can still be committed to in the scheme.
PlayersOfMission: contains all the agents committed to the scheme and their corresponding
mission.
GoalsState: contains the current state of the goals of the scheme.

By observing a SchemeBoard, an agent can monitor the overall dynamics of the scheme
execution. It can be aware of which missions are assigned to which agents, which goals are
achieved and which can be pursued.
The usage interface provides the following operations:
commitMission(m): used by an agent to commit to a mission m ∈ M;
leaveMission(m): used by an agent to give up a mission m it is committed to;
setGoalAchieved(ϕ): this operation is used to set the state of a goal to achieved.

7 This observable property indirectly links all groups responsible for the scheme to the normative board of
the same scheme.

123
386 Auton Agent Multi-Agent Syst (2010) 20:369–400

Based on the current state of a SchemeBoard, we can define the following predicates and
functions. Let SB and Ms be, respectively, the set of current scheme boards and the set of
all missions that can be played in a scheme board created from specification s.
– scheme_t ype(sb, s): this predicate is true when the scheme board sb ∈ SB is created
based on the scheme specification s ∈ S (the type of a scheme board is defined in its
creation);
– r esp_gr oup(gb, sb): this predicate is true when the group gb ∈ GB is responsible for the
execution of the scheme sb ∈ SB;
– committed(α, m, sb) this predicate is true when the agent α ∈ A is committed to the
mission m ∈ M in the scheme sb ∈ SB;
– achieved(ϕ, sb): this predicate is true when the goal ϕ is already achieved in the scheme
sb;
– possible(ϕ, sb): this predicate is true when the state of the goal ϕ is possible in the scheme
sb; considering
 the set of all goals that are pre-condition of ϕ, this predicate can be
deduced by

possible(ϕ, sb) ← achieved(ϕ  , sb) ∧ well_ f or med(sb) (5)
ϕ  ∈


– mplayer s : M × SB → Z with
def
mplayer s(m, sb) = |{α|committed(α, m, sb)}| (6)
this function gives the number of current players of the mission m in the scheme of sb.
Given the above definitions and the functions maxmp and minmp (cf. Sect. 3.1), we are able
to define the wellformedness property of a scheme:
well_ f or med(sb) ← scheme_t ype(sb, s) ∧
∀m∈Ms mplayer s(m, sb) ≥ minmp(m, s) ∧
mplayer s(m, sb) ≤ maxmp(m, s) (7)
Analogously to the role cardinality norm, we define a norm to forbid an agent to commit to
missions in a scheme.
Mission commitment norm: the number of agents already committed to a mission con-
strains the action commitMission (mission cardinality). Another constraint is that only agents
that play some role in a responsible group of the scheme can commit to a mission in the
scheme. This norm is defined as follows:
((scheme_t ype(sb, s) ∧ mplayer s(m, sb) ≥ maxmp(m, s)) ∨
(r esp_gr oup(gb, sb) ∧ ¬ plays(α, ρ, gb)),
F O R B I D D E Nα commitMission(m)) (8)
The implementation of this norm follows the same algorithm used by the GroupBoard:
whenever an agent attempts to commit to a mission, if the condition of the norm holds, the
operation is denied.
Note that in the current version of ORA4MAS, there is no regimentation on the leav-
ing of a mission or a role. We consider that these organisational actions should give rise to
enforcement and violation. For instance, we could imagine as a violation the fact of leaving
a mission while still having goals of this mission to be achieved. Following the detection of
a violation, sanctions have to be decided by organisational agents.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 387

5.4 NormativeBoard artifact

The NormativeBoard artifact (Fig. 5) embeds the functionalities to manage the specification
concerning permissions and obligations defined between roles and missions. This artifact has
no operation available to the agents since its function is to detect and show as observable
properties the current status of the norms according to the agents’ behaviour related to the
groups and scheme the normative artifact is linked to. There is one link operation (update-
AgentStatus) used by the assigned scheme and groups to trigger an update of the current
status of a particular agent, normally because the agent has performed some operation in the
scheme or group.
The set of norms are defined from the deontic specification and the current state of related
artifacts. As examples, in the sequel some of these norms are presented.
Obligation to commit to a mission: based on the deontic relations obl(ρ, m) included in
the organisation specification (as defined in Sect. 3.1), the roles played by the agents (as
defined in the Sect. 5.2), and the current number of agents committed to a mission (an agent
is not obliged to commit to a mission if the minimum number of players is already achieved),
we can define the following norm:
(obl(ρ, m) ∧ plays(α, ρ, gb) ∧ r esp_gr oup(gb, sb) ∧
scheme_t ype(sb, s) ∧ mission_scheme(m, s) ∧
mplayer s(m, sb) < minmp(s),
O B L I G E Dαt commitMission(m)) (9)
The three first lines of this norm are the condition that states when the norm is active and
the last line represents the obligation for the target agent α to commit to the mission before
the deadline t (where t is a value defined by the application designer).
Permission to commit to a mission: based on deontic relations per (ρ, m) included in the
organisation specification, and the roles played by the agents, we can define the following a
norm:
( per (ρ, m) ∧ plays(α, ρ, gb) ∧ r esp_gr oup(gb, sb) ∧
scheme_t ype(sb, s) ∧ mission_scheme(m, s),
P E R M I T T E Dα commitMission(m))
which can be rewritten as an interdiction as follows
(¬( per (ρ, m) ∧ plays(α, ρ, gb) ∧ r esp_gr oup(gb, sb) ∧
scheme_t ype(sb, s) ∧ mission_scheme(m, s)),
F O R B I D D E Nα commitMission(m)) (10)
Obligation to achieve a goal: once an agent α is committed to a mission m, it is obliged to
fulfil the possible goals of the mission before the deadline of the goals (given by the function
deadline, cf. Sect. 3.1). The norm below specifies that rule.
(committed(α, m, sb) ∧ goal_mission(ϕ, m) ∧ possible(ϕ, sb),
O B L I G E Dαdeadline(ϕ) ϕ) (11)
These norms are not implemented by regimentation, since we would like to allow the agents to
violate them. Their implementation is thus not as simple as the implementation of the norms
of group and scheme artifacts (where only interdictions are considered and regimentation is
used as the mechanism).

123
388 Auton Agent Multi-Agent Syst (2010) 20:369–400

The NormativeBoard manages the state of the norms as follows (more details are described
in algorithm Algorithm 1).8 When an agent starts playing a role in a group and this group
is responsible for some scheme, instances of norms are created for the agent in all related
normative boards. This instantiation process consists of copying the norm definition and
grounding their variables (e.g. replacing ρ of norm (9) by the identification of the role the
agent is playing). The state of a norm instance is initially inactive. It becomes active when
its condition holds. When the agent executes the action as it is stated in the action part of
the norm, the status of the norm becomes fulfilled. In the other case, i.e. the agents does not
behave in time accordingly to the action part of the norm, the status of the norm becomes
unfulfilled.

6 ORA4MAS in action

In this section we illustrate how the above defined artifacts can be used in the deployment of
an example. We first describe how a set of agents can use those artifacts and after how those
agents and artifacts can be implemented. The chosen example is very simple and is used just
to illustrate the functioning of the proposal. Its reduced complexity also makes it entirely
comprehensible.

6.1 The writing paper example

We consider that four agents (Tom, Eva, Bob, and Alice) want to write a paper together using
the proposed architecture described above. Among these agents, Tom is also an organisa-
tional agent and thus it may create the OrgBoard for the system. The creation of the OrgBoard
is based on an organisational specification where the roles, groups, schemes, etc. are defined
as presented in Sect. 3.1. The following steps show one possible sequence of evolution of
this system until their goal of writing a paper is achieved (these steps are also illustrated in
Fig. 6).9
1. Tom creates the GroupBoard based on the specification of Fig. 2a. Once the GroupBoard
is created, the artifact registers itself in the OrgBoard by means of the OrgBoard’s link
interface. To succeed in the creation, the new group should satisfy all constraints defined
in the OS (the group cardinality, for instance).
2. Tom sends then a message to the other agents telling them about the new artifact.10 While
Tom decides to adopt the role editor, Eva, Alice, and Bob decide to adopt the role writer
in this group. To adopt the role, they use the adoptRole operation of the GroupBoard. As
this operation is regimented it may fail in case the agents do not fit the requirements for
the role (cardinality of the role in the group, compatibility of roles, etc., as defined in
norms (3) and (4)). The reasons for the adoption of roles is not covered here, but, for
example, they may decide to become a writer because Tom has invited them to enter into
the group.
8 We are aware that this algorithm is not a general norm interpreter—which would be the ideal solution. Indeed
it works for only the three norms defined for the normative board. Nevertheless, the use of norms in the overall
process of developing this artifact is very useful both as a conceptual abstraction and a development tool.
Although we need to change the algorithm if new norms are included, the changes are quite straightforward.
The development of a more general algorithm with good performance is a future work in the project.
9 All the code used to run this example is available at http://moise.sourceforge.net/demos/jaamas09.
10 Agents can be aware of the current artifacts in the system either by communication or by inspecting a
registry artifact that contains a list of all artifacts in the system.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 389

Algorithm 1 Dynamics of the norms’ state in a normative board

3. Tom creates the SchemeBoard to start the process of writing the paper. This artifact
embeds the specification of Fig. 2b. As for the GroupBoard, scheme artifacts, once cre-
ated, register themselves in the OrgBoard. As every SchemeBoard has one NormativeBoard
and at least one group responsible for it, the initialisation process of SchemeBoard creates
its corresponding normative board and then properly links the artifacts.
4. Once the scheme is created, obligations and interdictions are computed and verified by the
NormativeBoard based on information about players and commitments in the group and
scheme artifacts. The norms defined in ORA4MAS (as those exemplified in norms (9)–
(10)) and algorithm Algorithm 1 are used to perform such computation. The deadline to
fulfil the norms is considered as infinity in this example. The observable properties of
this artifact just after the creation of the scheme are roughly as follows (the value ‘active’
in the fifth column means that the condition of the norm holds and the value ‘fulfilled’
means the agent is respecting the norm):

123
390 Auton Agent Multi-Agent Syst (2010) 20:369–400

OrgBoard (1b) GroupBoard

(1a)

(0) (2) (2) (2)

(2)

Tom Alice Eva Bob

(3a) (4/5) (4/5)

(4/5) (4/5) (3c)

(3b) SchemeBoard (3c) NormativeBoard

Legend
artifact creation

Org Appl. artifact use

Artifact type
Agent Agent
artifact link

Fig. 6 Example of the construction and use of organisational artifacts by the agents

NormativeBoard—missions
agent role missions is committed state of
norm (9) norm (10)
Tom editor m Man no inactive inactive
mCol no inactive active
m Bib no inactive active
Alice writer m Man no inactive active
mCol no active inactive
m Bib no active inactive
Eva writer m Man no inactive active
mCol no active inactive
m Bib no active inactive
Bob writer m Man no inactive active
mCol no active inactive
m Bib no active inactive

By observing the normative board, Alice, that is playing the role writer, realises that
she is obliged to commit to the missions mCol and m Bib (cf. Fig. 2d and norm (9)).
Let us suppose that she is obedient and decided to commit to both. Since the mission
m Bib has a cardinality constraint that sets the minimum and maximum number of com-
mitments to one, the other agents are not obliged to commit to this mission anymore.
After Alice’s commitment to m Bib and mCol, the observable properties of the normative
board include:

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 391

NormativeBoard—missions
agent role missions is committed state of
norm (9) norm (10)
Tom editor m Man no inactive inactive
mCol no inactive active
m Bib no inactive active
Alice writer m Man no inactive active
mCol yes fulfilled inactive
m Bib yes fulfilled inactive
Eva writer m Man no inactive active
mCol no active inactive
m Bib no inactive inactive
Bob writer m Man no inactive active
mCol no active inactive
m Bib no inactive inactive

The other writers in the system commit thus only to the mission mCol. Tom, that plays
editor, commits to m Man. We are assuming here that the agents are obedient and always
commit to their obligations and pursue their organisational goals.
5. Once the scheme is well formed, according to the norm (7), the goals of the scheme can
be pursued. The observable properties related to goals in the SchemeBoard are initially
like the following:

SchemeBoard
mission goal state
m Man wtitle possible
wabs waiting
wsectitle waiting
wcon waiting
mCol wsec waiting
m Bib wref waiting

Initially only the goal wtitle is possible and thus can be pursued (i.e. possible(wtitle)
holds), all the other goals are in the waiting state. The goal wtitle belongs to the m Man
mission, so only Tom has something to do, the others will wait for him to achieve the fdv
sub-goals. To know which goals can be pursued and to set them as achieved, the agents
perceive the SchemeBoard and act on it using the setGoalAchieved operation. The above
observable properties after setGoalAchieved(wtitle), performed by Tom, are:

SchemeBoard
mission goal state
m Man wtitle achieved
wabs possible
wsectitle waiting
wcon waiting
mCol wsec waiting
m Bib wref waiting

By showing the possible goals to the agents, the SchemeBoard works like a coordination
artifact.

123
392 Auton Agent Multi-Agent Syst (2010) 20:369–400

In this example, the NormativeBoard simply shows the state of the norms for each agent,
i.e. whether the norm is inactive, active, fulfilled, or unfulfilled. As an artifact, it is not taking
decisions from this status. An organisational manager agent, like Tom, must perceive this
artifact and decide what to do when some norm violation occurs.
Here we have taken a simple example with only one group board, one scheme board, one
normative board, and one organisational agent just to illustrate the ORA4MAS approach.
We can however suitably design and manage an organisation with several artifacts which are
independent or structured in an hierarchy (in case of an organisation with a group and sub-
groups as can be stated in Moise+ ); where the goals of a scheme board can be distributed
to agents from several responsible group boards; etc. These artifacts can then be distributed
and supervised by several organisational agents.

6.2 A concrete architecture

ORA4MAS is realised on top of the CArtAgO infrastructure [34], embedding algorithms

used in S -Moise+ . CArtAgO (Common Artifact Infrastructure for Agent Open environ-
ment) is a infrastructure based on the A&A meta-model for developing and running arti-
fact-based environments for MAS, integrated with (existing) agent platforms. CArtAgO
technology11 includes (i) a Java-based basic programming model for programming artifacts,
(ii) a suitable API for agents to work with artifacts and workspaces, and (iii) a run-time
platform supporting the existence and dynamic management of artifact-based environments.
A foremost important aspect of CArtAgO is its orthogonality with respect to the specific
agent model and architecture adopted: this promotes the integration between CArtAgO and
(existing) heterogeneous agent platforms, making it possible in principle to develop distrib-
uted, heterogeneous and open MAS, where agents dwelling in different agent platforms can
autonomously join and work together in distributed and shared workspaces. Among those
agent platforms, CArtAgO is integrated with Jason [4], 2apl [8], and jadex [31]—these
integrations are presented in [30,33]. Jason and jadex are used here to exemplify how
the agents can be programmed to use organisational artifacts. The code presented in Fig. 7
contains an excerpt of the program of agent Tom as programmed in Jason and Fig. 8 the
code of agent Eva programmed in jadex. To avoid a more detailed explanation of these two
languages, we added comments in the figures explaining the code.
The final system is therefore formed by an organisation specification written in Moise+
OML, distributed organisational artifacts, and agents programmed in different languages that
are able to cooperate by means of these artifacts.

7 Related works and discussion

Several proposals for organisational infrastructures have been proposed in the literature:
MadKit, based on AGR organisational model [13]; Ameli [12] and Ameli+ [15], based
on Islander [11]; karma, based on TeamCore [32]; OperA [10]; S -Moise+ [21], based
on Moise+ [22]. In the sequel, these works and our proposal are discussed considering
important topics and features of OI as identified in Sect. 2.
Abstraction & encapsulation The current OIs’ components are either agents (Ameli,
S -Moise+ , OperA) or services (MadKit). However, as argued in this paper, the organisa-
tion should be implemented by both agents and services. The approaches that use only services

11 CArtAgO is open-source and available at: http://cartago.sourceforge.net.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 393

Fig. 7 Excerpt of the code of agent Tom written in Jason

are not flexible enough to allow the management of the OI by the agents. The approaches that
use only agents are using them for reactive and task oriented services; some of those agents are
not really pro-active and autonomous entities. By using agents and artifacts to reify both the
organisation and the organisation infrastructure, we raise the level of abstraction with respect
to approaches in which organisation mechanisms are hidden at the implementation level by
means of artifacts. Such mechanisms become parts of the agent world, suitably encapsulated
in proper entities that agents then can inspect, reason and manipulate, by adopting a uniform
approach.

123
394 Auton Agent Multi-Agent Syst (2010) 20:369–400

Fig. 8 Excerpt of the code of agent Eva written in jadex

We will consider the following illustrative example to show one advantage of the pro-
posed approach. In S -Moise+ or in MadKit, the middleware managing the organisation is
deployed and running with a specific strategy which is not defined nor decided by the agents.
In our approach, agents may decide the creation of a particular group type, either Moise+ or
AGR. This creation is simply a matter of creating the corresponding artifact. This operation
is realised and decided by agents. We could imagine, and our approach would facilitate such
a possibility, that agents embed some domain knowledge leading them to reason and decide
on the best machine on which to deploy this artifact, on the best number of artifacts to manage
groups, etc.
Environment It is interesting to compare ORA4MAS and middlewares for e-Institutions,
such as Ameli, for what concerns the notion of environment that both approaches refer to. The

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 395

abstract architecture of e-Institutions (Ameli in particular), places a middleware composed

of governors and staff agents between participating agents and an agent communication infra-
structure (e.g. JADE) [1]. The notion of environment is the one of dialogical environment:
it is not a concrete environment that agents sense and act upon, but a conceptual one that
agents playing within the institution can interact with by means of norms and laws, based on
specific ontologies, social structures, and language conventions. In fact, governors and staff
agents are not meant to be necessarily perceived by agents playing inside the institution as
first-class entities of their world: agents communicate with each other by means of speech
acts and behind the scene the middleware mediate such communication by means of insti-
tutional agents. The extension proposed for situated e-Institutions [1], introduces the notion
of World of Interest (WoI) to model such environment external to the MAS and which is
relevant to the MAS application. Participating agents in this case are able to interact with the
WoI (the entities) not directly but always through the e-Institution, i.e. governors and staff
agents.
In ORA4MAS we take a similar approach, where instead of governors and staff agents we
are using artifacts to enable and mediate the access to the organisation. However, instead of
a dialogical environment we provide a work environment that agents can perceive, act upon,
and adapt as a first-class entity of their world. As first-class entities, organisational artifacts
can be intentionally created and used by the agents—they are not “behind the scenes”. Along
with the other kinds of artifacts, in the A&A perspective, they are concrete bricks used to
structure the agents’ world: part of this world is represented by the organisational infrastruc-
ture, part by artifacts introduced by specific MAS applications, including some representing
(wrapping) entities/services belonging to the external environment, i.e. the WoI of situated
e-Institutions. In this case, participating agents are able to sense and act over the WoI using
such artifacts.
Agent autonomy Although one of the goals of an OI is to manage in a more or less flexible
way the constraints imposed by the organisation on the autonomy of the agents, it should
not extinguish their autonomy. This issue is considered by all the above mentioned OIs. In
Ameli, for instance, the agents are autonomous in the achievement of goals but the commu-
nication is constrained (or regimented) by the OI; in S -Moise+ the agents are autonomous
concerning the communication protocols but constrained (or regimented) in the achievement
and coordination of collective goals. In our proposal, agents are still autonomous with respect
to the decision of using a specific artifact or not—including the organisational artifacts—and
keeping their autonomy—in terms of control of their actions—while using organisational
artifacts. Agents however can depend on the functionalities provided (encapsulated) by arti-
facts, which can concern, for instance, some kind of mediation with respect to the other agents
sharing the same organisational artifact. Then, by enforcing some kind of mediation policy
an artifact can be both an enabler and a constrainer of agent interactions. However, such
a constraining function can take place without compromising the autonomy of the agents
regarding their decisions.
Another important issue of our proposal is to clearly consider two kinds of mechanisms
to implement the norms that are entitled in the OS: regimentation that is implemented in the
artifacts and cannot be violated; and enforcement that is implemented both in the artifacts
(the detection) and in the organisational agents (evaluation and judgement).
Distributed management Some OIs, such as S -Moise+ and MadKit , centralise all the
management of the organisation in one agent or service bringing scalability problems. Dis-
tributing the management of the organisation into different organisational artifacts realises
a distributed coordination (meaning here more particularly synchronisation) of the differ-
ent functions related to the management of the organisation. Completing this distribution

123
396 Auton Agent Multi-Agent Syst (2010) 20:369–400

of the coordination, the reasoning and decision processes which are encapsulated in the
organisational agents may be also distributed among the different agents. Thanks to their
respective autonomy, all the reasoning related to the management of the organisation (mon-
itoring, reorganisation, control) may be decentralised into different loci of decision with a
loosely coupled set of agents.
Openness To be open to the entrance of heterogeneous agents is an important feature
for MAS in general and a reason to establish an organisation for the system. This is thus
also an issue considered by all the above OIs. In most cases (e.g. S -Moise+ , Ameli), the
agents have access to the organisational infrastructure by means of an agent communication
language (KQML, FIPA-ACL) or other open protocols. ORA4MAS does not use a protocol
or communication language; operations are used instead. The interaction between the agents
and the organisation is no longer expressed with a ACL semantics. Besides that, organisa-
tional artifacts, as with any other kind of artifact, can be created and added dynamically as
needed. They have a proper semantic description of both the functionalities and operating
instructions, so conceptually agents can discover at runtime how to use them in the best way.
Still related to openness, the approach promotes heterogeneity of agent societies: arti-
facts can be used by heterogeneous agents, with different kinds of reasoning capabilities.
Extending the idea to multiple organisations, we can have the same agents playing different
roles in different organisations, and then interacting with organisational artifacts belonging
to different organisations.
As shown in the example of Sect. 6.2, the use of artifacts, and particularly the CArtAgO
implementation, allows agents implemented in different languages to use the artifacts and
cooperate using them. Most of the OIs listed above give tools and support only for agents
implemented in a particular language, normally Java—which is not the most appropriate
language to code some types of agents.
‘Organisational power back to agents’ As stressed in Sect. 2, the current implementations
of OIs conceive the organisation as a layer that the application agents rely on to participate
in the organisation’s activities. The agents are not managers of this layer, they are simply
passive users. This conception of OI is captured by the notion of regimentation and organi-
sation artifacts in our proposal. However, our contribution in this context is to allow some
decisions that were embedded in the services to go back to the agents’ layer by means of
organisational agents. In ORA4MAS, artifacts encapsulate the coordination and synchroni-
sation which were implemented in services. Control and judgement procedures are separated
from these aspects and are embedded in organisational agents. Organisational agents can
then use organisational artifacts to help them in deciding and eventually applying sanctions
to other agents.
Besides the above mentioned advantages of the ORA4MAS approach, there are still open
questions to be investigated. First, although artifacts promote an open environment for the
agents, we need to define some mechanics that allows the designer (or the organisational
agents) to specify some sort of control on the access to the artifacts. For example, in the
current version of ORA4MAS every agent can see all the observable properties of the organ-
isational artifacts. Some privacy problems could therefore arrive from that. All the operations
are also available to all agents. Of course the operations may fail in case the agents are either
not allowed to use them or are using them incorrectly. But it could be interesting to have
different ‘interfaces’ for different agents, particularly, different interfaces for member and
organisational agents. An initial work is being developed in the context of CArtAgO platform
to add a kind of RBAC (Role Base Access Control) system so that the roles being played
by the agents constrain their access to the artifacts. Second, an important element for the
interoperability and openness of the approach is the artifact’s manual. The manual is to be

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 397

used by an arriving agent to understand how the artifact can be used and, in our case, how
the organisation can be ‘used’. A good language to write the manual should (1) allow the
developer to specify the organisational actions, (2) be easily readable and processed by the
agents, and (3) have a clear semantics. Other research directions are mentioned as future
work in the next section.

8 Conclusion and perspectives

In this paper, we have followed the A&A approach to give back the power to agents in an
organisational approach. From this perspective, we have defined on the one hand the organi-
sational artifacts which encapsulate the functional aspects of an organisation and organisation
management, and on the other hand the organisational agents, which encapsulate the decision
and reasoning side of the management of organisations.
Although we already have some initial results of the ORA4MAS project, as those pre-
sented in this paper, we have concretely evaluated the proposal for only one OML (the
Moise+ ). The first future work of the project will therefore be an evaluation of its application
for different OMLs such as Islander [11], OperA [10], Moise I nst [16], and AGR [13].
Following this broader application we can then better compare our approach with related
works (e.g. [15]) and even others such as those managing the organisation with communi-
cation acts (e.g. RICA-J [38]) or exploiting the environment to coordinate and constrain the
agents’ behaviour [46].
Other extensions aim at taking advantage of the uniform concepts used to implement
the environment and the organisation abstractions through the concept of artifacts. Such an
homogeneous conceptual point of view will certainly help us to find a suitable solution for
the second problem identified in Sect. 2 binding both concepts together in order to situate
organisations in an environment or to install the access to the environment into organisa-
tional models (in the same direction as proposed by [26]). Other points of investigation are
(1) the study of the reorganisation process of an MAS using the ORA4MAS approach, (2)
the impact of the reorganisation on the organisational artifacts, and (3) the definition of a
meta-organisation for the ORA4MAS, so that we have special roles for organisational agents
that give them access to the organisational artifacts.

Acknowledgements We are grateful to the anonymous reviewers of this paper and the earlier versions
of it published in COIN and EUMAS for their valuable suggestions, contributions, and motivation. We
would like to thank Michele Piunti for the development of the JADEX agents and useful remarks to improve
ORA4MAS. This work is strongly based on previous works developed with Jaime S. Sichman in the con-
text of the project USP-COFECUB 98-04. Jomi F. Hübner is financed by the project ForTrust (Social
Trust Analysis and Formalization) funded by the French Agence nationale de la Recherche (ANR-06-SETI-
006).

References

1. Arcos, J. L., Noriega, P., Rodríguez-Aguilar, J. A., & Sierra, C. (2007). E4mas through electronic insti-
tutions. In D. Weyns, H. V. D. Parunak, & F. Michel (Eds.), Environments for multi-agent systems III,
third international workshop, E4MAS 2006, Hakodate, Japan, May 8, 2006, Selected revised and invited
papers, Vol. 4389 of Lecture Notes in Computer Science (pp. 184–202). Springer.
2. Bellifemine, F. L., Caire, G., & Greenwood, D. (2007). Developing multi-agent systems with JADE. Wiley
Series in Agent Technology. Wiley.
3. Boissier, O., Hübner, J. F., & Sichman, J. S. (2007). Organization oriented programming from closed to
open organizations. In G. O’Hare, M. O’Grady, O. Dikenelli, & A. Ricci (Eds.), Engineering Societies in
the Agents World VII (ESAW 06), Vol. 4457 of LNCS (pp. 86–105). Springer-Verlag.

123
398 Auton Agent Multi-Agent Syst (2010) 20:369–400

4. Bordini, R. H., Hübner, J. F., & Wooldrige, M. (2007). Programming multi-agent systems in agentSpeak
using Jason. Wiley Series in Agent Technology. Wiley.
5. Broersen, J., Dastani, M., Hulstijn, J., Huang, Z., & van der Torre, L. (2001). The BOID architecture:
Conflicts between beliefs, obligations, intentions and desires. In J. P. Müller, E. Andre, S. Sen, & C. Fras-
son (Eds.), Proceedings of the fifth international conference on autonomous agents, Montreal, Canada
(pp. 9–16). ACM Press.
6. Castelfranchi, C. (2000). Engineering social order. In A. Omicini, R. Tolksdorf, & F. Zambonelli (Eds.),
Engineering societies in the agent world, first international workshop, ESAW 2000, Berlin, Germany,
August 21, 2000, revised papers, Vol. 1972 of Lecture Notes in Computer Science (pp. 1–18). Springer.
7. Castelfranchi, C., Dignum, F., Jonker, C. M., & Treur, J. (2000). Deliberate normative agents: Principles
and architecture. In N. R. Jennings & Y. Lespérance (Eds.), Intelligent agents VI, Agent Theories, Archi-
tectures, and Languages (ATAL), 6th international workshop, ATAL ’99, Orlando, Florida, USA, July
15–17, 1999, Proceedings, Vol. 1757 of LNCS (pp. 364–378). Springer.
8. Dastani, M. (2008). 2APL: A practical agent programming language. Autonomous Agent and Multi-Agent
Systems, 16, 241–248.
9. Dignum, V., & Dignum, F. (2001). Modeling agent societies: Co-ordination frameworks and institutions.
In P. Brazdil & A. Jorge (Eds.), Proceedings of the 10th Portuguese conference on artificial intelligence
(EPIA’01), Berlin, LNAI 2258 (pp. 191–204). Springer.
10. Dignum, V., Vazquez-Salceda, J., & Dignum, F. (2004). OMNI: Introducing social structure, norms
and ontologies into agent organizations. In R. H. Bordini, M. Dastani, J. Dix, & A. El Fallah-Seg-
hrouchni (Eds.), Proceedings of the programming multi-agent systems (ProMAS 2004), LNAI 3346,
Berlin. Springer.
11. Esteva, M., de la Cruz, D., & Sierra, C. (2002). ISLANDER: An electronic institutions editor. In C. Cas-
telfranchi & W. L. Johnson (Eds.), Proceedings of the first international joint conference on autonomous
agents and multiAgent systems (AAMAS 2002), LNAI 1191 (pp. 1045–1052). Springer.
12. Esteva, M., Rodríguez-Aguilar, J. A., Rosell, B., & Arcos, J. L. (2004). AMELI: An agent-based mid-
dleware for electronic institutions. In N. R. Jennings, C. Sierra, L. Sonenberg, & M. Tambe (Eds.),
Proceedings of the third international joint conference on autonomous agents and multi-agent systems
(AAMAS’2004), New York (pp. 236–243). ACM.
13. Ferber, J., & Gutknecht, O. (1998). A meta-model for the analysis and design of organizations in multi-
agents systems. In Y. Demazeau (Ed.), Proceedings of the 3rd international conference on multi-agent
systems (ICMAS’98) (pp. 128–135). IEEE Press.
14. Fornara, N., & Colombetti, M. (2006). Specifying and enforcing norms in artificial institutions. In A. Omi-
cini, B. Dunin-Keplicz, & J. Padget (Eds.), Proceedings of the 4th European workshop on multi-agent
systems (EUMAS 06).
15. García-Camino, A., Rodríguez-Aguilar, J., & Vasconcelos, W. W. (2007). A distributed architecture for
norm management in multi-agent systems. In J. Sichman, P. Noriega, J. Padget, & S. Ossowski (Eds.),
Coordination, organizations, institutions, and norms in agent systems III, Vol. 4870 of LNAI (pp. 275–
286). Springer, Revised Selected Papers.
16. Gâteau, B., Boissier, O., Khadraoui, D., & Dubois, E. (2005). MOISEinst: An organizational model for
specifying rights and duties of autonomous agents. In Third European workshop on multi-agent systems
(EUMAS 2005), Brussels, Belgium, December 7–8, pp. 484–485.
17. Goble, L., & Meyer, J.-J. C. (Eds.). (2006). Proceedings of the 8th international workshop on deontic
logic in computer science, DEON 2006, Utrecht, The Netherlands, July 12–14, 2006, Vol. 4048 of Lecture
Notes in Computer Science. Springer.
18. Grossi, D., Aldewered, H., & Dignum, F. (2007). Ubi Lex, Ibi Poena: Designing norm enforcement
in e-institutions. In P. Noriega, J. Vázquez-Salceda, G. Boella, O. Boissier, V. Dignum, N. Fornara, &
E. Matson (Eds.), Coordination, organizations, institutions, and norms in agent systems II, Vol. 4386 of
LNAI (pp. 101–114). Springer, Revised Selected Papers.
19. Hübner, J. F. (2003). Um Modelo de Reorganização de Sistemas Multiagentes. PhD thesis, Universidade
de São Paulo, Escola Politécnica.
20. Hübner, J. F., Boissier, O., & Vercouter, L. (2008). Instrumenting multi-agent organisations with reputa-
tion artifacts. In V. Dignum & E. Matson (Eds.), Proceedings of coordination, organizations, institutions
and norms (COIN@AAAI), held with AAAI 2008, Chicago, EUA (pp. 17–24). AAAI Press.
21. Hübner, J. F., Sichman, J. S., & Boissier, O. (2006). S-MOISE+: A middleware for developing organ-
ised multi-agent systems. In O. Boissier, V. Dignum, E. Matson, & J. S. Sichman (Eds.), Coordination,
organizations, institutions, and norms in multi-agent systems, Vol. 3913 of LNCS (pp. 64–78). Springer.
22. Hübner, J. F., Sichman, J. S., & Boissier, O. (2007). Developing organised multi-agent systems using the
MOISE+ model: Programming issues at the system and agent levels. International Journal of Agent-Ori-
ented Software Engineering, 1(3/4), 370–395.

123
Auton Agent Multi-Agent Syst (2010) 20:369–400 399

23. IST Advisory Group. (2003). Ambient intelligence: From vision to reality. Technical report, Information
Societies Technologies. ftp://ftp.cordis.europa.eu/pub/ist/docs/istag-ist2003_consolidated_report.pdf.
24. Luck, M., McBurney, P., Shehory, O., & Willmott, S. (2005). Agent technology: Computing as interaction
(A roadmap for agent based computing). AgentLink. http://www.agentlink.org/roadmap
25. Nardi, B. A. (1996). Context and consciousness: Activity theory and human–computer interaction. MIT
Press.
26. Okuyama, F. Y., Bordini, R. H., & da Rocha Costa, A. C. (2007). Spatially distributed normative
objects. In P. Noriega, J. Vázquez-Salceda, G. Boella, O. Boissier, V. Dignum, N. Fornara, & E. Mat-
son (Eds.), Coordination, organizations, institutions, and norms in agent systems II, Vol. 4386 of LNAI,
pp. 133–146.
27. Okuyama, F. Y., Bordini, R. H., & da Rocha Costa, A. C. (2008). A distributed normative infrastructure
for situated multi-agent organisations. In L. Padgham, D. C. Parkes, J. Müller, & S. Parsons (Eds.), Pro-
ceedings of 7th international conference on autonomous agents and multiagent systems (AAMAS 2008),
May, 12–16, 2008, Estoril, Portugal, pp. 1501–1504.
28. Omicini, A., Ricci, A., & Viroli, M. (2008). Artifacts in the A&A meta-model for multi-agent systems.
Autonomous Agents and Multi-Agent Systems, 17(3), 432–456.
29. Omicini, A., Ricci, A., Viroli, M., Castelfranchi, C., & Tummolini, L. (2004). Coordination artifacts:
Environment-based coordination for intelligent agents. In AAMAS’04, Vol. 1, New York, USA, 19–23
July 2004 (pp. 286–293). ACM.
30. Piunti, M., Ricci, A., Braubach, L., & Pokahr, A. (2008). Goal-directed interactions in artifact-based mas:
Jadex agents playing in cartago environments. In IEEE/WIC/ACM conferences on web intelligence and
intelligent agent technology (IAT-2008). IEEE/WIC/ACM.
31. Pokahr, A., Braubach, L., & Lamersdorf, W. (2005). Jadex: A BDI reasoning engine. In R. H. Bordini,
M. Dastani, J. Dix, & A. El Fallah Seghrouchni (Eds.), Multi-agent programming: Languages, platforms,
and applications, number 15 in Multiagent Systems, Artificial Societies, and Simulated Organizations,
Chap. 6 (pp. 149–174). Springer.
32. Pynadath, D. V., & Tambe, M. (2003). An automated teamwork infrastructure for heterogeneous software
agents and humans. Autonomous Agents and Multi-Agent Systems, 7(1–2), 71–100.
33. Ricci, A., Piunti, M., Acay, L. D., Bordini, R. H., Hübner, J. F., & Dastani, M. (2008). Integrat-
ing heterogeneous agent programming platforms within artifact-based environments. In L. Padgham,
D. C. Parkes, J. Müller, & S. Parsons (Eds.), 7th International joint conference on autonomous
agents and multiagent systems (AAMAS 2008), Estoril, Portugal, May 12–16, 2008 (pp. 225–232).
IFAAMAS.
34. Ricci, A., Viroli, M., & Omicini, A. (2006). CArtAgO: A framework for prototyping artifact-based envi-
ronments in MAS. In D. Weyns, H. V. D. Parunak, & F. Michel (Eds.), Environments for multiAgent
systems III, Vol. 4389 of LNAI (pp. 67–86). Springer. 3rd International workshop (E4MAS 2006), Hako-
date, Japan. Selected Revised and Invited Papers.
35. Ricci, A., Viroli, M., & Omicini, A. (2007). A general purpose programming model & technology for
developing working environments in MAS. In M. Dastani, A. El Fallah Seghrouchni, A. Ricci, & M. Win-
ikoff (Eds.), 5th International workshop “Programming multi-agent systems” (PROMAS 2007), AAMAS
2007, Honolulu, Hawaii, USA, pp. 54–69.
36. Ricci, A., Viroli, M., & Omicini, A. (2008). The A&A programming model & technology for developing
agent environments in MAS. In M. Dastani, A. E. Fallah-Seghrouchni, A. Ricci, & M. Winikoff (Eds.),
Programming multi-agent systems, 5th international workshop, proMAS 2007, Honolulu, HI, USA, May
15, 2007, Revised and Invited Papers, Vol. 4908 of LNCS, pp. 89–106. Springer.
37. Sairamesh, J., Lee, A., & Anania, L. (2004). Introduction of the special issue on information cities.
Communications of the ACM, 47(2), 28–31.
38. Serrano, J. M., & Ossowski, S. (2007). A compositional framework for the specification of interaction
protocols in multiagent organizations. Web Intelligence and Agent Systems, 5(2), 197–214.
39. Sichman, J., Noriega, P., Padget, J., & Ossowski, S. (Eds.). (2008). Coordination, organizations, institu-
tions, and norms in agent systems III, Vol. 4870 of LNCS. Springer.
40. Tuomela, R., & Bonnevier-Tuomela, M. (1995). Norms and agreement. European Journal of Law, Phi-
losophy and Computer Science, 5, 41–46.
41. Vázquez-Salceda, J., Aldewereld, H., & Dignum, F. (2004). Norms in multiagent systems: Some imple-
mentation guidelines. In Proceedings of the second European workshop on multi-agent systems (EUMAS
2004). http://people.cs.uu.nl/dignum/papers/eumas04.PDF
42. Viroli, M., Holvoet, T., Ricci, A., Schelfthout, K., & Zambonelli, F. (2007). Infrastructures for the envi-
ronment of multiagent systems. Autonomous Agents and Multi-Agent Systems, 14(1), 49–60.
43. Viroli, M., Ricci, A., & Omicini, A. (2006). Operating instructions for intelligent agent coordination. The
Knowledge Engineering Review, 21(1), 49–69.

123
400 Auton Agent Multi-Agent Syst (2010) 20:369–400

44. Weyns, D., Omicini, A., & Odell, J. J. (2007). Environment as a first-class abstraction in multi-agent
systems. Autonomous Agents and Multi-Agent Systems, 14(1), 5–30. Special Issue on Environments for
Multi-agent Systems.
45. Weyns, D., & Parunak, H. V. D. (Eds.). (2007). Journal of Autonomous Agents and Multi-Agent Systems.
Special Issue on Environments for Multi-Agent Systems, 14(1). Netherlands: Springer.
46. Weyns, D., Parunak, H. V. D., & Michel, F. (Eds.). (2006). Environments for multi-agent systems II,
second international workshop, E4MAS 2005, Utrecht, The Netherlands, July 25, 2005, Selected Revised
and Invited Papers, Vol. 3830 of Lecture Notes in Computer Science. Springer.

123