Towards a Multi-Agent Platform for Cyber-physical Systems
based on Low-power Microcontroller for Automated Intralogistics
A Minimized Embedded Solution for the Internet of Things
in Intralogistical Environments
Arne Stasch
and Axel Hahn
OFFIS Institute for Information Technology, Oldenburg, Germany
Department of Computing Science, Carl von Ossietzky University, Oldenburg, Germany
Keywords: Cyber-physical System, Multi-Agent System, Material Flow, Logistics.
Abstract: Today the fluctuating market requires more flexibility from central controlled material flow systems. A new
approach such as the Internet of Things is able to turn the common structure into a cognitive decentralized
system. This paper addresses a modularized solution for material flow systems in intralogistical
environment. The promising concept of Internet of Things leads to an idea of a cyber-physical system with a
Multi-agent platform which is presented in this paper. In order to develop a system which is capable of
meeting industrial needs, a low-power microcontroller is chosen as basis for the Multi-agent system. The
combination has advantages but also restrictions which are discussed. A demonstrator for the future
implementation of the system and evaluation is introduced.
Flexibility is an increasing key requirement in
intralogistical environments. The material flow
systems (MFS) used in this field have to bear
alternations in the flow rate and flow directions.
Common material flow systems consist of
centralized programmable logic controllers (PLC)
and are designed for a certain capacity. By the time
requirements of the MFS change so that a
retrofitting is necessary or reasonable, then a conflict
occurs: Changes of the MFS are time- and cost-
intensive for logistic providers.
Modular material flow systems are one solution
of this problem (ten Hompel, 2011). Among the
approaches to transfer promising technologies to a
modular MFS, the concept of “Internet of Things
sticks out (Günthner, 2010). In this concept, the
intelligence is distributed over a decentralized
control system. This results in engineering
advantages. The reusability of soft- and hardware
solutions will increase and the systems diversity will
be reduced for example (ten Hompel, 2011). This
effect will reduce development and working costs.
The MFS control architecture was basically
stacked in three layers so far (Günthner, 2010).
These layers are the physical, the PLC and the
coordination layer (see Figure 1). The physical layer
represents mechanical and electrical parts. These are
controlled by the PLC layer which accesses sensors
and actuators. The other layers are coordinated by
the coordination layer which consists of an IT-
Figure 1: Transferring of common MFS layer into the
concept Internet of Things.
In order to modularize the common MFS, every
module must implement this structure. Figure 1
visualizes the change to a decentralized concept.
Stasch A. and Hahn A..
Towards a Multi-Agent Platform for Cyber-physical Systems based on Low-power Microcontroller for Automated Intralogistics - A Minimized Embedded
Solution for the Internet of Things in Intralogistical Environments.
DOI: 10.5220/0004583004290433
In Proceedings of the 10th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2013), pages 429-433
ISBN: 978-989-8565-71-6
2013 SCITEPRESS (Science and Technology Publications, Lda.)
The concept presented in this paper takes this
modularized structure and transfers it into an
embedded solution. Every module has its own
physical sensors and actuators (Physical Layer).
They are controlled by a microcontroller with a
Multi-agent system (MAS) platform (PLC Layer).
Agents coordinate their tasks through a fieldbus
(Coordination Layer).
The “Internet of Things” can be implemented in
different ways such as Near Field Communication
(NFC), Wireless Sensors and Actuators Networks
(WSAN) or IP for Smart Objects (IPSO) (Atzori,
2010). Two appealing ways are described in the
following. Radio Frequency Identification (RFID)
tags allows packages to carry their own information
or code which can be altered. Multi-agent systems
(MAS) are systems which consist of entities which
can collectively solve problems (Wooldridge, 2002).
Those technologies are solutions to distribute the
The RFID tags have to be attached on packages
and need an intelligent infrastructure.
Electromagnetic interference (EMI) and certain
materials (metals, liquids) can restrict the use of it in
industrial environments (White et al., 2007).
A MAS needs a modularized infrastructure to be
meaningful. The task, the MAS has to fulfil, has to
be divisible so that different agents work on the
same task. Further a platform that gives multiple
agents the ability to communicate and interact with
its environment is needed. To approve the reliability
of a system which coordinates through
communication, a pervasive testing is necessary.
Taking the idea of RFID with adaptable
identities (ID) where the packages can represent
themselves in the software and transfer this to an
agent who represents the package in a MAS is on a
par with the idea of ten Hompel (ten Hompel, 2011).
Creating a MAS which can be used in an industrial,
intralogistical environment on a modularized MFS is
the next step.
MAS are used in different applications like
autonomous driving (Beeson, 2008) or decentralized
control of Automatic Guided Vehicles (AGV)
(Weyns, 2008). These approaches use computer
systems to realize the MAS infrastructure. The
approaches of geo-references swarm agents
(Barbera, 2010) or a hardware platform for analogue
power simulator as a test environment for smart
grids (Spencer, 2010) appeal more suited for a
decentralized control system. They use embedded
controllers as basis but one controller represents
only one agent. This is not enough for a platform
that has to handle more than one agent. With the
above mentioned MAS for MFS such an embedded
software platform is needed and is presented in
section 3.2.
Modules which communicate with each other
and interact with the environment physically are
called cyber-physical systems (CPS) (Lee, 2008).
Therefore the aim will be to create a CPS which can
replace common MFS. Microcontrollers can meet all
industrial standards and are widely used. Although
they are not powerful, they meet the requirements
for networked systems which can divide their tasks.
The decision of taking microcontrollers as basis is
discussed in the next section.
The paper is arranged in the following order: The
concept of a new cyber-physical system is presented
in section 2. In section 2.1 MAS frameworks and
their benefits are discussed in the described scenario.
The own idea of a MAS platform is presented in
section 2.2. In the following section 2.3 the
demonstrator which will be used to implement the
developed control system is described and is
followed by the conclusion in section 3.
The goal is to create a CPS which can substitute the
common MFS and deliver new features as loose
coupling. It should be possible to couple or decouple
hardware parts and the software completes the
reconfiguration autonomously. If the MFS will be
split in fine-meshed modules then there are more
options to rearrange the MFS to be on a par with the
changing requirements. Every storage shelf,
transport-vehicle and reasonable smallest part of a
conveyor can be a capsuled module. This means that
in every module an electronic control unit (ECU)
has to control the module and communicate with
other modules.
Communication between these modules is the
key for smooth control and coordination. Every
entity has to fulfil the functions of all layers,
mentioned in section 1, but only between the borders
of the module. The modules must communicate to
achieve the tasks of the whole MFS.
By creating a modular system, the system
structure has to be changed (ten Hompel, 2011).
Figure 2 shows the previous horizontal view and the
modularized vertical view of the systems structure.
In every development step the vertical view must be
These separated modules only need the resources
for a small contingent of the whole system. An
embedded solution sticks out against PCs and PLCs
because a module does not need the processing
power of these. Small, optimised, real-time capable
systems which can be mass-produced for low cost
could be enough for the task.
Figure 2: Changing horizontal to vertical system view.
Hardware-requirements for an ECU are:
1. Enough processing power to control the
mechatronic parts and handle the
communication to other modules.
2. Low-power consumption for modules which
are not connected to a static power supply.
3. Reliable for safety critical applications
4. Interfaces are compatible to industrial
These requirements are not very specific.
However they give a direction to small
microcontrollers as MAS devices. After defining the
hardware, suitable software is needed. This is
discussed in the next section.
2.1 MAS Framework for embedded
A multi-agent system (MAS) delivers handy
functionality for MFS. Agents represent the modules
and packages in the software. This distributes the
responsibility for the package transport to the
modules. Other functionalities like routing can be
implemented as agents, too.
Requirements, like real-time operations,
industrial standards and small footprints sort out
most of the existing MAS frameworks. JADE
(Bellifemine, 1999) is a very common used MAS
framework but doesn’t meet the requirements stated
above. Mobile C uses C/C++ as programming
language and C meets industrial standards (Chou et
al, 2010). However Mobile C needs a General
Purpose Operating System (GPOS) and even the
smallest embedded versions, e.g. Embedded Linux,
need more footprint than 259 KB Flash Memory
and this does not account RAM usage.
linux.php3: conclusion last visited 02.04.2013
In the search of agent friendly architectures
AUTOSAR (AUTomotive Open System
ARchitecture) stands out (Heinecke, 2004). It is a
standardized system which solves software
development issues for automotive ECUs. The
architecture is layered and separated between
applications and basic software which is dependable
on hardware. The basic idea is to reuse written
applications on any device independent from
hardware. This architecture allows the applications
to communicate to other applications on the same
ECU and on other ECUs with the same interface.
The interface is provided by the AUTOSAR runtime
environment (RTE). A patent from the DAIMLER
describes the use of agents with AUTOSAR for
ECU Diagnostics.
Because of the specifications from the
automotive sector, AUTOSAR is not lean. It has a
footprint of 256KB on the leanest AUTOSAR ECU
(Bunzel, 2011). It is not optimised for agents and
their interactions. Exemplary, only one application
has access rights to a specific part of the memory
supplied by the RTE. This application has to manage
and deliver it again to other applications with ports
through the RTE. Therefore the implementation of
the agent communication which needs access to
memory is more complicated.
The listed memory resources do not look much
in the embedded world today. This approach is to
aim for a minimal, reliable and cheap solution for
industrial use. The MAS framework should be lean
as possible because the embedded environment
should be chosen due to the application and not on
the framework. A suitable MAS platform is
presented in the next section.
2.2 Concept of MAS Platform for CPS
Using the software architecture of AUTOSAR gives
a basis for a MAS framework which suits for this
CPS. Figure 3 depicts an adapted AUTOSAR
architecture. On the ground level, the ECU hardware
is represented. The background level (in AUTOSAR
language basic software) represents the Real-Time
Operating System (RTOS), the hardware drivers,
and the driver’s service interfaces which are unified
APIs. Agents reside in the Agent Level were an
Agent RTE connects them with each other and the
Background Level.
Creating the Agent RTE will be the key task in
the future work. It has to support the agents with
Patent Daimler AG: (WO2008095518) USE OF A
Figure 3: Adapted AUTOSAR Architecture.
all necessary information and access to fulfil their
role autonomously. Every Agent needs to run in a
capsuled task so it can act independently. The Agent
RTE has to manage all communication between the
agents and agents on other modules. In Figure 4 the
Routing Agent communicates with a Package
Agent on the same module and with a Routing
Agent on another module. The Agent RTE and the
services below handle the communication. However,
the interface the agents use is independent from their
Figure 4: Agent-to-Agent Communication.
The functionality the Agent RTE has to provide
the agents has to be altered by the knowledge about
the functionality of other MAS frameworks like
Mobile C. Because the limitations of the
microcontroller can hinder the development of some
features, it has to be analysed whether it is a fully-
fledged MAS platform or it has reduced
In the next section the environment where this
MAS framework will be implemented is introduced.
2.3 Implementation in a Material Flow
The research project CogniLog (Overmeyer, 2012)
deals with automated, cognitive logistic-networks. A
demonstrator composed of conveyor modules and
forklifts will be built. It will show the capabilities of
modularized conveyor systems.
The setup consists of conveyers, Industrial PCs
(IPC) with Soft PLCs, frequency convertors,
electrical engines, and sensor barriers. A Profibus
fieldbus is responsible for the communication
between the electronic parts.
Per every “capsuled platform” a microcontroller
will be implemented. A “capsuled platform” is
defined as a conveyor which actuators control solely
itself and no other platforms. The intersections must
have sensor barriers to detect incoming or outgoing
goods. The microcontroller completes the capsuled
platform to a module.
In a full modularized environment the
microcontrollers replace the PLCs. To minimize
development time the PLCs remain as a middleware
for actuators, sensors and bus communication. The
microcontrollers represent the distributed
intelligence. They will do the planning,
coordination, and monitoring.
The microcontrollers which are used are Texas
Instruments (TI) MSP430F6638 with a 16-Bit
Architecture, 256 KB flash memory, 16 KB RAM,
SPI and I²C Interfaces. An EEPROM of 128 KB
non-volatile memory and a Proficonn
Profibusadapter is connected with the SPI Interface.
In figure 5, a draft of the connections between
the different electronic parts is shown. The modules
represent the separated conveyor parts.
Although this setup does not correspond with the
targeted decentralization, it shows most of the
functionality which a fully decentralized system will
The previously described demonstrator setup
uses only established technologies. The MFS can be
logical modularized but not real physical. The
conveyor parts are heavy and not exchangeable.
Heavy electrical cabinets and many electrical cables
do not allow fast reconstruction of the whole setup.
But a conveyor system which can be rearranged is
not the aim of this research. The purpose is to create
a technology that enables the creation of such a
system. In this setup the MAS platform can be tested
on the usability of coordination on communication.
The loose coupling can be shown by connecting and
disconnecting modules to simulate a retrofitting.
Additionally the system has to fulfil the same role as
a common MFS setup. No information should be
lost in the communication and no package routing
problems should occur.
Figure 5: CogniLog demonstrator setup.
This should demonstrate that small embedded
controllers with a proper multi-agent system
framework can be a flexible way to bring the ideas
of “Internet of Things” to material flow systems.
This paper describes the ideas of a modularized
intralogistical control system. After a summary about
suitable technologies a new approach is proposed.
Capsuled modules, which consist of their own sensor
barriers, actuators, intelligence and communication
abilities, form a new MFS. The intelligence is
distributed through a MAS platform for small
embedded controllers. This platform will be
constructed for this purpose. This paper also describes
the implementation environment inside the CogniLog
project. This will be an authentic test environment of
the usefulness of this approach of a CPS.
After the implementation phase the Agent RTE
has to be compared with other MAS frameworks to
evaluate the different features and cuts. When this is
done, the possible features can forecast a profit in
other domains. If a mechatronic task can be split in
many parts which do not need many processing
power, the new MAS platform could bring an
ten Hompel, M., Nettsträter, A., Feldhorst, S. & Schier, A.
2011. `Engineering of Modular Material Flow Systems
in the Internet of Things`, at –
Automatisierungstechnik, vol. 59, no. 4, pp. 248-256.
Oldenbourg Wissenschaftsverlag.
Günthner, W. & ten Hompel, M. 2010. Internet der Dinge
in der Intralogistik – engl.: Internet of Things in
intralogisitics, Springer Verlag. Berlin Heidelberg.
Lee, E. 2008. `Cyber Physical Systems: Design
Challenges`, 11th IEEE Symposium on Object
Oriented Real-Time Distributed Computing (ISORC),
pp. 363-369.
Chou, Y., Ko, D. & Cheng, H. 2010. `An embeddable
mobile agent platform supporting runtime code
mobility, interaction and coordination of mobile
agents and host systems`, Information and Software
Technology, vol. 52, no. 2, pp. 185-196. Elsevier B.V..
Overmeyer, L., Krühn, T., Hahn, A. & Pinkowski, J. 2012.
`CogniLog – Cognitive Logistics for Warehousing`,
6th International Scientific Symposium on Logistics,
pp. 104-122.
Wooldridge, M. 2002. An Introduction to MultiAgent
Systems, John Wiley and Sons Ltd.
Heinecke, H., Schnelle, K. P., Fennel, H., Bortolazzi, J.,
Lundh, L., Leflour, J., & Scharnhorst, T. 2004.
‘AUTomotive Open System ARchitecture-an industry-
wide initiative to manage the complexity of emerging
automotive E/E-architectures`, Paper presented at the
Convergence International Congress & Exposition On
Transportation Electronics, Detroit, Michigan, United
White, G., Gardiner, G., Prabhakar, G. & Razak, A. 2007.
`A Comparison of Barcoding and RFID Technologies
in Practice`, in Journal of Information, Information
Technology, and Organizations, vol. 2, pp. 119-132.
Bellifemine, F., Poggi, A., & Rimassa, G. 1999. `JADE–A
FIPA-compliant agent framework`, In Proceedings of
The Practical Application of intelligent Agents and
Multi-Agent Technology (PAAM), vol. 99, pp. 97-108.
Atzori, L., Iera, A. & Morabito, G. 2010. `The Internet of
Things: A survey’, Computer Networks, vol. 54, no.
15, pp. 2787-2905.
Beeson, P., O`Quin, J., Gillian, B., Nimmagadda, T.,
Ristroph, M., Li, D. & Stone, P. 2008. `Multiagent
Interaction in Urban Driving` in Journal of Physical
Agents: Multi-Robot Systems, vol. 2, no. 1, pp. 15-29.
Weyns, D., Holvoet, T., Schelfthout, K. & Wielemans, J.
2008. `Decentralized Control of Automatic Guided
Vehicles`, Proceedings of the 23rd ACM SIGPLAN
conference on Object-oriented programming systems
languages and applications, pp. 663-674.
Barbera, S., Stallo, C., Savarese, G., Ruggieri, M.,
Cacucci, S. & Fedi, F. 2010. `A Geo-Referenced
Swarm Agents Enabling System for Hazardous
Applications`, Proceedings of the 12th International
Conference on Intelligent Systems, Modelling and
Simulations, pp. 598-603.
Spencer, M., Feliachi, A., Pertl, F., Pertl, E. & Smith, J.
2010. `Hardware Platform for Multi-Agents System
Developement`, International Journal of Latest
Trends in Computing, vol. 1, no. 2, pp. 47-52.