A Task Allocation Middleware for Wireless Sensor Networks in a
Multi-Agent Environment
Luca Caviglione
1
, Mauro Coccoli
2
and Alberto Grosso
2
1
Institute of Intelligent Systems for Automation (ISSIA), National Research Council of Italy (CNR) via de Marini 6, I-16149,
Genova, Italy
2
Department of Informatics, Bioengineering, Robotics and Systems Engineering (DIBRIS), University of Genoa, via Opera
Pia, 13, I-16145, Genova, Italy
Keywords:
Task Allocation, Middleware, Wireless Sensor Network (WSN), Multi-Agent Systems.
Abstract:
This paper presents the design of a task allocation middleware for the coordination of a Wireless Sensor
Network (WSN) of embedded devices. Acquisition and distribution of new tasks are performed via a multi-
agent system, while service oriented principles are used to handle data gathered from the field. Also, an ad-hoc
component has been designed to reduce the impact of heterogeneous networks (e.g., long-haul satellite links).
Lastly, we showcase an experimental setup to prove the effectiveness of the approach when used to enhance a
forest fire prevention application.
1 INTRODUCTION
The division of work in activities, tasks and sub-
tasks implies the availability of task allocation tech-
niques to define specific objectives and responsibili-
ties (Qui˜nonez et al., 2011). By using this design, it is
possible to have parallel/specialized execution flows,
a versatile resource allocation scheme, and the del-
egation of duties to humans. Such features also in-
crease the robustness of the overall framework, which
is critical when deployed into unpredictable or time-
varying environments. To effectively implement such
requirements, a multi agent community is one of the
most suitable solutions (Shehory and Kraus, 1995).
However, some deployments, like those using a Wire-
less Sensor Network (WSN), may lead to very large
decision spaces requiring adjustable data collection
disciplines, or task adaptive algorithms. In this re-
spect, the Service Oriented Architecture (SOA) offers
loose coupling, statelessness, composability and dis-
coverability, thus making the merge between a WSN
and the task allocation middleware more efficient (Ib-
botson et al., 2010). Besides, the SOA enables sensors
to be discovered, accessed, invoked and controlled via
the Web (Delicato et al., 2005) and (Golatowski et al.,
2003). Software agents are another important block
to offer intelligent and flexible entities making possi-
ble the cooperation, competition, and coordinationfor
specific goals. To exploit their social abilities, agents
are often organized in a Multi-Agent System (MAS),
which is an interesting candidate to implement mid-
dleware services (e.g., object request brokers and di-
rectory services) (Omicini and Rimassa, 2004). This
paper presents a middleware based on a cooperative
multi-agent society for managing the allocation of
tasks in a WSN with time-varying topology and a
broad set of functionalities. To provethe effectiveness
of our design, we showcase the experimental testbed
of a forest fire prevention application. The contribu-
tion of this work is the definition of a task allocation
framework based on the SOA paradigm, making the
assignment problem independent both from the evo-
lution of the WSN and the target operations. Addi-
tionally, it enables to dynamically match jobs against
resources, for instance through optimized strategies,
such as the execution time, Quality of Service (QoS)
requirements, and power consumption constraints.
The remainder of the paper is structured as fol-
lows: Section 2 reviews task allocation mechanisms
for WSN, and Section 3 presents the architecture of
the system. Section 4 outlines network design aspects
and Section 5 showcases the task allocation middle-
ware. Section 6 describes the forest fire prevention
application. Lastly, Section 7 concludes the paper.
325
Caviglione L., Coccoli M. and Grosso A..
A Task Allocation Middleware for Wireless Sensor Networks in a Multi-Agent Environment.
DOI: 10.5220/0004634803250330
In Proceedings of the 3rd International Conference on Sensor Networks (SENSORNETS-2014), pages 325-330
ISBN: 978-989-758-001-7
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
2 WSN AND TASK ALLOCATION
Usually, a WSN is composed by a large number of
heterogeneous sensing devices forming highly dy-
namic network topologies due to impairments in the
wireless link, hardware/software outages, external
hazards, or battery drains. Nevertheless, to flatten
costs, a WSN is required to concurrently support dif-
ferent sensing tasks, mainly by sharing a common re-
source (e.g., measurement devices or long-haul com-
munication links), or competing for its exclusive con-
trol. Therefore, according to constraints imposed by
the specific application scenario, each sensor needs
to be associated to the best-served task. This raises
the problem commonly defined as Multi-Sensor Task
Allocation (MSTA). To this aim, in reference (Pizzo-
caro and Preece, 2009) authors discuss a framework
to categorize different instances of the MSTA. They
propose a domain-independenttaxonomy, which con-
siders the number of sensors, the number of tasks,
the time constraints, and the nature of the scenario,
i.e., collaborative versus competitive. As regards the
MAS, the literature already offers a variety of solu-
tions, especially based upon distributed task alloca-
tion techniques (Shehory and Kraus, 1995). Conse-
quently, we decided to exploit such result to coordi-
nate activities belonging to the WSNs, also by taking
into account that auction-based systems are widely
used in MAS (Faratin et al., 1998). Hence, in order
to assign activities, an auction is performed, result-
ing in agents/sensors bidding a value in a shared cur-
rency based on their perceived fitness for that task.
Applying this paradigm to the scenario of interest
leads again to the MSTA problem. Specifically, the
literature proposes several variations of auction and
market-based mechanisms, as well as awarding tasks
according to suited cost functions, e.g., to guarantee
a proper degree of QoS (Pletzer and Rinner, 2010),
and by explicitly considering the energy consumption
(Edalata et al., 2012). Accordingly, our middleware
will model multiple instances of the MSTA problem.
3 THE WSN ARCHITECTURE
The system architecture has been engineered to fit dif-
ferent distributed systems requiring flexibility in the
execution of operations/tasks. Yet, it has to be spe-
cialized for the WSN case. We consider the reference
blueprint as composed by a set of connected entities
whose topology and features can dynamically change.
In order to manage and orchestrate the WSN, the mid-
dleware adopts a MAS deployed among its nodes, and
at least one agent has to be located in each of these
with proper computational and storage capabilities.
Sensors are assumed as attached to an entity offering
a web service interface for data retrieval and manage-
ment. As a result, agents within nodes are the tar-
gets of the task allocation process. To implement this
software design, we use a middleware based on the
AgentService framework (Vecchiola et al., 2008) and
(Grosso et al., 2003) relying upon the Microsoft .NET
Framework and MONO Project. We point out that
AgentService already offers many critical features,
such as, the support of MASs, proper tools for mod-
eling agent societies, standard web service communi-
cation channels, and a workflow engine (Boccalatte
et al., 2005). Furthermore, it can distribute agents via
servers or resource-constrained embedded devices. In
this case, the minimal requirement is a device running
the Microsoft .NET Micro Framework (Fox and Box,
2003).
Figure 1: System architecture based on WSN and MAS.
In our approach, the middleware must be able to
handle two kinds of entities: sensing and coordination
nodes. Sensing nodes are handled in form of services,
thus via a set of operations to retrieve measurements
and information about their state (e.g., the assigned
task). Also, they are provided with basic functions
to exploit task allocation (e.g., for bidding). On the
contrary, a coordination node requires more compu-
tational capabilities, since it contains the core engine
in charge of composing and executing the workflow
of services. The communication infrastructure for the
agent society is based on a SOA, which can be ex-
tended via third party services. In this respect, Fig-
ure 1 depicts the WSN architecture composed by six
sensors connected to three embedded devices form-
ing the sensing nodes. Also, a coordination node
equipped with the task allocation and workflow en-
gine is present. Owing to the generality of the de-
sign, this architecture can be applied to different sce-
narios, ranging from small use cases requiring high
level of reliability in task completion (e.g., to allocate
SENSORNETS2014-InternationalConferenceonSensorNetworks
326
production orders in manufacturing plants), to geo-
graphically distributed systems.
4 NETWORK DESIGN ASPECTS
This section introduces the engineering of a specific
entity to prevent issues arising when in presence of
long-range communication links. The terms node
and sensor will be used interchangeably, except when
doubts arise.
4.1 Long-range Communication
Support
Since nodes could be placed in isolated or rural ar-
eas, public network infrastructures or cellular carri-
ers could be absent. As a consequence, some form of
long-range communicationsmust be provided, e.g., to
deliver measurements to a remote sink. Long-range
links are nowadays widely supported, both from the
viewpoints of hardware and availability as commer-
cial services (Caviglione, 2009a). Though, they can
introduce the following drawbacks:
1. generally, the adopted radio technology has fea-
tures to compensate attenuation/fading, to pro-
vide data confidentiality through proper encryp-
tion algorithms, and to support channel reserva-
tion mechanisms. Alas, they are energy consum-
ing tasks, thus heavily affecting battery-operated
nodes;
2. sensors are usually equipped with L2 air in-
terfaces offering short-range communication ser-
vices. Then, providing medium to long-range data
exchanges could require add-ons, both in terms
of hardware and software. Even if this can be
also guaranteed through L2 bridges, such a choice
increases the number of needed devices, and the
overall network complexity, e.g., it requires con-
figuration/maintenance/powersupply;
3. the standard TCP/IP protocol suite has perfor-
mance issues when in presence of high delay ·
bandwidth or heavily faded channels (e.g., GEO
links). Moreover, intermittent connectivity, as
provided by LEO satellites, could demand for
proper application layer countermeasures. A
common approach uses ad-hoc protocol suites,
e.g., those enabling data custody or deferred trans-
mission. We mention, among the others, the Dis-
ruption Tolerant Networking (DTN) (see, e.g.,
(Mc Donald et al., 2007) for an example of the
usage of DTN in a sensor network for monitoring
a lake). Unfortunately, they require an implemen-
tation available for the target platform;
4. as a partial workaround to 3), many protocol mod-
ifications are available, but usually demand for
“patching low-level software components (e.g.,
the kernel or the drivers), which can be unfeasible
for all the devices, for instance due to source-code
unavailability.
To cope with issues 1) - 4) a more suitable method
is to use proxy-based components, often defined as
Performance Enhancing Proxies (PEPs) or “middle-
boxes”. In this way, it is possible to reduce the impact
of long-range links on the overall network architec-
ture, while at the same time, avoid a major re-engineer
of the WSN. PEPs have been already successfully uti-
lized for Public Safety and Disaster Relief (PSDR)
duties, as described in (Caviglione, 2006).
Figure 2: Reference protocol of the PEP.
Figure 2 depicts the reference protocol architec-
ture of a PEP designed to merge the local WSN
(and/or devices in charge of performing task alloca-
tion), with long-range links. Specifically, it is com-
posed by two protocol stacks: one is devoted to man-
age communications within the WSN (i.e., the left
portion), while the other is used to transmit data over
the long-range link (i.e., the right part). The ISO/OSI
L7 layer is in charge of relaying data, and running
local processes. Thus, it can take in custody data to
cope with intermittent connectivity,or can perform lo-
cal task allocation duties, also by enabling a proper
feedback with the remote facility (in this case, it is
assumed co-located with a coordination node). The
availability of a full-featured application layer can be
also exploited for dynamically turning-on/offsensors,
having only the PEP to constantly communicate with
the proper data collection facility, i.e., it acts as a sink.
The PEP can be also used to perform data process-
ing/compression, to save bandwidth when in pres-
ence of narrow wireless channels, or to use portion
of the transmission resource for fading countermea-
sures, e.g., for Forward Error Correction (FEC). By
decoupling the stacks, each specific protocol can be
tweaked according to well-given design constraints.
ATaskAllocationMiddlewareforWirelessSensorNetworksinaMulti-AgentEnvironment
327
As an example, the L4 of the right stack can be a
TCP-like protocol optimized for satellite communi-
cations, while the L3 layer used in the WSN can have
simplified routing strategies, as well as specific coun-
termeasures to reduce the transmission power and the
impact of local channel interferences.
Furthermore, the PEP entity could be standalone,
or co-located within a coordination node, which can
also be mobile. In this case, let us consider an in-
dividual interacting with nodes by exploiting prox-
imity communications (e.g., the IEEE 802.15 substi-
tutes the generic merge of ISO/OSI L1 and L2 de-
fined as “lower layers sensors dependent” in Figure
2). Upon approaching a node, he/she can perform the
following actions: i) issue/collect task-related infor-
mation and ii) gather local measurements or dissemi-
nate previously collected data. For what concerns i),
the App/Task Logic is in charge of managing and al-
locating tasks. As regards ii), a proper protocol suite
supporting deferred transmissions, data bundling and
intermittent connectivity (e.g., a link is only supposed
to exist between the local node and the operator) has
to be implemented. Owing to the decoupled design,
specific choices do not propagate to the rest of the
framework, especially in the middleware.
5 MIDDLEWARE DESIGN
This section discusses the layered architecture of
the middleware, which is composed by the runtime
framework implementing the task allocation process,
and a set of Application Programming Interfaces
(APIs). To this aim, we specialized the general-
purpose middleware AgentService, where each agent
represents sensor(s). Also, the device on which sen-
sors are attached/deployed is assumed as the geospa-
tial reference.
5.1 Agents and Task Definition
In our design, agents are the only targets of the task
allocation process, and can be classified in three cate-
gories:
sensor agents acting on behalf of a single sensor,
or a device equipped with a set of sensors;
mobile agents deployed on mobile appliances,
interacting with both sensors and coordination
nodes. Their activities can be driven by human
operators, as well as autonomous vehicles duly
equipped;
software agents they only perform pure software
tasks, such as data retrieval and processing.
We point out that using the network component
presented in Section 4.1 prevents the need of a net-
work agent. This is another benefit of having a mid-
dlebox to decouple the WSN from the rest of the net-
work. From the viewpoint of the middleware, each
agent can submit new tasks to the runtime and partic-
ipate to their concurrent execution, according to the
specific capabilities.
To represent a task, the following tuple t, is used:
t =< U, P,C, L >
where, U is the Unique Resource Identifier (URI) as-
sociated to the task, P is a set of input/output pa-
rameters, C is a collection of required capabilities,
and L contains information about the area/node loca-
tion where the task must be executed. If L =
NULL
,
then no specific locations are necessary. Thus, L can
be used to discriminate between localized and non-
localized tasks. A possible example of non-localized
task is a general weather forecast, while for the local-
ized case could be a group of measurements coming
from sensors deployed in a well-defined geographi-
cal area. Besides, if a task does not require a feed-
back, it can be considered terminated once allocated,
otherwise results have to be sent back to the runtime
engine.
5.2 Task Allocation Process
The runtime framework is organized in three layers:
1. Task Decomposition Layer: for each new task, it
generates the relevant workflow of services;
2. Workflow Engine Layer: for each service, it exe-
cutes the service flow and it submits a task request
to the Task Allocation Layer;
3. Task Allocation Layer: it collects task requests
and performs assignments to agents.
Each agent awaits for new tasks by polling the
hosting device. Based on its capability, it may de-
cide to take in charge the execution of a pending task.
In this case, the agent will notify such assignment
both to the device and the coordinator. Owing to
task decomposition, each “simple” task corresponds
to a SOA service, while complex tasks are represented
with a workflow whose actions are modeled as service
invocations. The Workflow Engine Layer creates a
new instance of the given workflow also by supervis-
ing its control flow. Then, upon receiving a new task
request (i.e., a service invocation), it routes proper
commands to the Task Allocation Layer as to com-
plete its assignment to sensor agents. Specifically,
the Task Allocation Layer offers two functionalities:
for localized tasks it arranges their deployment over
SENSORNETS2014-InternationalConferenceonSensorNetworks
328
specific WSN nodes/embedded devices, and it imple-
ments the auction-based strategies.
5.3 Deployment of Localized Tasks
When in presence of localized tasks, the Task Allo-
cation Layer performs a “Call for Task” limited to a
specific area. This reduces to finding a proper match
among areas and nodes, and task tuples are sent to
matching entities. The middleware generates a new
task, which is defined as deployment task. In order
to be accomplished, it is assigned to mobile agents
that are placed on mobile devices. Obviously, the de-
ployment task can be performed by human operators,
or by unmanned vehicles, for instance in presence of
hazardous environments.
6 FOREST FIRE PREVENTION
APPLICATION
To prove the effectiveness of our design, we show-
case a forest fire prevention application. Since areas
to be monitored are often outback, the system should
be disaster-resistant and remotely controllable. Fig-
ure 3 depicts the typical reference blueprint composed
by different sensing nodes. To guarantee the proper
network connectivity, at least one coordination node
must be present.
Figure 3: The configuration of a forest fire prevention sys-
tem including devices and the WSN.
To be successfully adopted, the system uses
standard communication infrastructures, e.g., IEEE
802.11b/g/n for medium range communications,
UMTS or Ka/Ku satellite links for remote data deliv-
ery, and the TCP/IP protocol suite to support Internet
services. Then, as discussed, data-handling services
are exposed through a MAS-coordinated SOA infras-
tructure, which also allocates tasks. The coordination
node collects measurements from the WSN. Such val-
ues can be used to perform analyses and predictions
based on the past history.
The IP enables sending data, images, and video
streams towards the Internet, constrained by the un-
derlying resource availability. The satellite channel
can be considered as a backup, or as the primary long-
haul link, according to specific design constraints
(e.g., the deployment of the frameworkin inaccessible
or rural areas). We highlight that satellite bandwidth
could be expensive or scarce, thus a PEP-like entity
would make possible to locally implement data com-
pression, resource reservation mechanisms and data
shaping techniques. The middlebox also isolates the
WSN, enabling the introduction of specific policies,
such as battery preserving scheme, or communica-
tion technologies (e.g., ZigBee, IEEE 802.11 or Blue-
tooth). This design resembles a “mediated” peer-to-
peer overlay guaranteeing the access to the WSN via
a unique and coherent interface exposed by the single
coordination node (Caviglione, 2009b). Also, recon-
figurations of sensing nodes will not propagate across
the overall system, reducing the complexities, and
easing the management of the resulting data flows.
6.1 Evaluation of the Testbed
To evaluate our design, we developed a small testbed
in a controlled environment. To this aim, we used 3
devices based on the Tahoe-II development platform,
which features the .NET Micro Framework, a Merid-
ian CPU based on a Freescale i.MXS ARM9 proces-
sor with 4 Mbytes of Flash and 8 Mbytes of RAM, a
3.5” touch-screen LCD, wired and wireless L2 inter-
faces, USB ports, and an accelerometer. The choice of
the Tahoe-II platform also guarantees the needed flex-
ibility requirements for implementing adaptive sens-
ing nodes, i.e., nodes that can be replaced to fulfill
specific needs/operations. Also, it enables to simu-
late hazardous conditions such as the presence of fire.
Nevertheless, its on-board display demonstrated how
humans could be effectively placed within the “de-
cision loop”. To implement the coordination node,
we used a standard x86 architecture, while the net-
work relies upon the built-in air interfaces (i.e., IEEE
802.11). Lastly, the satellite link has been emulated
by adding proper transmission delays (i.e., via the
netem tool on a Linux machine).
Trials in such a controlled environment support
that: i) the middleware can match tasks and agents ac-
cording to specific objectives; ii) human operators can
remotely adjust functionalities of the fire prevention
system; iii) owing to the rapid re-configurability of
the system, operatorscan simulate specific alert/alarm
conditions, resulting into an effective tool for the
ATaskAllocationMiddlewareforWirelessSensorNetworksinaMulti-AgentEnvironment
329
training of personnel; iv) the presence of an x86 node
allows to aggregate or pre-process data gathered by
sensors to save transmission resources, or to recover
performance degradations due to excessive transmis-
sion delays (e.g., by properly tuning TCP parameters).
7 CONCLUSIONS
In this paper we presented a task allocation middle-
ware for developing adaptive WSN architectures in a
multi-agentenvironment. To proveits effectiveness, it
has been used to develop a forest fire preventionappli-
cation. Besides, the prototypal implementation also
demonstrated that using devices also offering human-
to-machine interaction enables individuals to easily
access measurements and functionalities of the WSN.
Future work aims at enriching the middleware
with semantic capabilities, for instance by using a set
of tuples defined via the Resource Description Frame-
work (RDF). This allows to develop ontology-based
techniques, enabling to model the application domain,
and to use automatic reasoning for the discovery and
decomposition of complex tasks. Additionally, the
SOA paradigm allows to easily integrate our middle-
ware with a Geographical Information System (GIS),
or to offer the built-in workflow engine as a “service”
to schedule resources and personnel when in presence
of alarms. Lastly, as a part of our ongoing research,
we will evaluate how to merge our solution with a De-
cision Support System (DSS).
REFERENCES
Boccalatte, A., Coccoli, M., Grosso, A. and Vecchiola, C.
(2005). A multiuser groupware calendar system based
on agent tools and technology. In Proc. of the IEEE
2005 Int. Symposium on Collaborative Technologies
and Systems, St. Louis, MO, USA, pp. 144-151.
Caviglione, L. (2006). Introducing Emergent Technologies
in Tactical and Disaster Recovery Networks. Int. J. of
Communication Systems, Wiley, vol. 19, no. 9, 1045-
1062.
Caviglione, L. (2009). Can Satellites Face Trends? The
Case of Web 2.0. In Proc. of Int. Workshop on Satellite
and Space Communications (IWSSC’09), Siena, Italy,
pp. 446-450.
Caviglione, L. (2009) Enabling Cooperation of Consumer
Devices Through Peer-to-Peer Overlays. IEEE Trans.
on Consumer Electronics, vol. 55, no. 2, pp. 414-421.
Delicato, F., Pires, P., Rust, L., Pirmez, L. and de Rezende,
J. (2005). Reflective Middleware for Wireless Sen-
sor Networks. In Proc. of the 20th Annual ACM sym-
posium on applied computing (SAC 2005), Santa Fe,
USA, pp. 730-735.
Edalata, N., Thama, C.-K. and Xiaob, W. (2012). An
Auction-based Strategy for Distributed Task Alloca-
tion in Wireless Sensor Networks. Computer Commu-
nications, vol. 35, no. 8, pp. 916-928.
Faratin, P., Sierra, C. and Jennings, N.R. (1998). Negotia-
tion Decision Functions for Autonomous Agents. Int.
J of Robotics and Autonomous Systems, vol. 24, no.
3-4, pp. 159-182.
Fox, D. and Box, J. (2003). Building Solutions with the
Microsoft.Net Compact Framework: Architecture and
Best Practices for Mobile Development. Addison-
Wesley Longman Publishing Co., Boston, MA, USA.
Golatowski, F., Blumenthal, J., Handy, M. and Haase, M.
(2003). Service Oriented Software Architecture for
Sensor Networks. In Proc. of the Int. Workshop on
Mobile Computing (IMC), Rockstock, Germany, pp.
93-98.
Grosso, A., Gozzi, A., Coccoli, M. and Boccalatte, A.
(2003). An Agent Programming Framework Based on
the C# Language and the CLI. Journal of .NET Tech-
nologies, vol. 1, no. 3, pp.13-20.
Ibbotson, J., Gibson, C., Wright, J, Waggett, P., Zerfos, P.,
Szymanski, B.K. and Thornley D.J. (2010). Sensors
as a Service Oriented Architecture: Middleware for
Sensor Networks. In Proc. of the 6th Int. Conf. on In-
telligent Environments, pp. 209-214.
Mc Donald, P., Geraghty, D., Humphreys, I., Farrell, S. and
Cahill, V. (2007). Sensor Network with Delay Toler-
ance (SeNDT). In Proc. of the 16th Int. Conf. on Com-
puter Communications and Networks, 1333-1338.
Omicini, A. and Rimassa, G. (2004). Towards Seamless
Agent Middleware. In Proc. of the IEEE 13th Int.
Workshops on Enabling Technologies: Infrastructure
for Collaborative Enterprises, Modena, Italy.
Pizzocaro, D. and Preece, A. (2009). Towards a Taxon-
omy of Task Allocation in Sensor Networks. In Proc.
of the 28th IEEE Int. Conf. on Computer Communi-
cations Workshops, IEEE Press Piscataway, NJ, USA,
pp. 413-414.
Pletzer, F. and Rinner, B. (2010). Distributed Task Allo-
cation for Visual Sensor Networks: a Market-based
Approach. In Proc. of the 2010 4th IEEE Int. Conf. on
Self-Adaptive and Self-Organizing Systems Workshop,
Budapest, Hungary.
Qui˜nonez, Y. de Lope, J. and Maravall, D. (2011). Bio-
inspired Decentralized Self-coordination Algorithms
for Multi-heterogeneous Specialized Tasks Distribu-
tion in Multi-Robot Systems. In Proc. of the 4th Int.
Conf. on Interplay between natural and artificial com-
putation, Springer Berlin, Heidelberg, pp.30-39.
Shehory, O. and Kraus, S. (1995). Task Allocation via
Coalition Formation Among Autonomous Agents. In
Proc. of Int. Joint Conf. on AI ’95, Montreal, pp. 655-
661.
Vecchiola, C., Grosso, A. and Boccalatte, A. (2008).
AgentService: a Framework to Develop Distributed
Multi-agent Systems. Int. J. of Agent-Oriented Soft-
ware Engineering, vol. 2, no. 3, pp. 290-323.
SENSORNETS2014-InternationalConferenceonSensorNetworks
330