Content uploaded by Pavel Vrba
Author content
All content in this area was uploaded by Pavel Vrba on Nov 08, 2013
Content may be subject to copyright.
Conference Report
Over the past few years, a number of industrial applica-
tions have deployed agents. However, a substantial gap
still exists between the cutting-edge research carried out
mainly in university laboratories and research institutes
and the domain-specific industrial applications that com-
mercial organizations develop.
The articles in this department intend to give some indi-
cation of agent technology’s readiness for commercial de-
ployment, based primarily on the presentations and discus-
sions at the inaugural Industry Track of A
AMAS 2005—the
Fourth International Joint Conference on Autonomous
Agents and Multiagent Systems.
Opportunities for agent deployment
Multiagent systems and autonomous-agent technologies
provide a design-and-implementation paradigm for software
solutions based on collective decision making in a commu-
nity of autonomous, loosely coupled computational entities.
Many agent development environments and agent integra-
tion platforms are available either commercially or in open
source format ready to deploy in commercial applications.
The agent research community has consolidated signifi-
cantly in the last few years. In particular, the formalization
of agent-based computing’s foundations has positioned the
domain in relation to adjacent fields of theoretical research
such as formal logic, game theory, theorem proving and
model checking, distributed and parallel computing, scala-
bility, and complexity theory. The community is also
involved in much research that’s closer to potential business
applications, such as the Semantic Web, open systems, and
ubiquitous computing. The achievements in these fields
form a solid foundation for technology transfer from univer-
sity labs and research institutes to industrial applications.
The agent paradigm and the available agent techniques
perform well in five types of domains. The first is compet-
itive and noncooperative domains, where information-
sharing restrictions prevent a centralized decision-making
architecture—for example, e-commerce applications, sup-
ply-chain management, and e-business. In such domains,
the agent paradigm is employed to design and describe
mainly Web-based systems.
In the second type of domain, the data required for auto-
mated decision making aren’t centrally available. The usual
reasons for this are the geographical distribution of knowl-
edge (for example, logistics, collaborative exploration,
mobile and collective robotics, or pervasive systems) or
environments where communication is partially or tem-
porarily inaccessible. Other reasons include temporal distri-
bution (for example, satellite networks where satellites
have different views of the earth at different times of the
day) and conceptual distribution (for example, in layered
hierarchies, where entities at one layer might have no
knowledge of events or processes at other layers, as in the
Internet or supply chains).
The third type of domain requires survivable time-criti-
cal response and high robustness in distributed scenarios.
Example domains include time-critical manufacturing or
industrial-systems control that requires replanning or fast
local reconfiguration to handle problems instantly.
The fourth type of domain involves simulation and model-
ing. Using agents for simulation has been common. Agents
can be deployed either in simulations requiring easy migra-
tion to the real environment or where traditional simulation
techniques are expensive.
The final type of domain involves open-systems engi-
neering. Early agent deployment projects emphasized such
domains, but the reality of the implementations delivered
so far hasn’t met expectations. Even though using ontolo-
gies and FIPA (Foundation for Intelligent Physical Agents)
standards has addressed many syntax issues, semantic-
integration issues remain problematic. Web services and
Web technologies in general seem to have taken the lead in
applications in this area.
In our experience, industrial organizations frequently
request (and agent technology developers frequently pro-
vide) these functionalities:
A
gent technology provides industrial-applications
developers with new abstractions for distributed-
system development, new methodological tools, and a set of
algorithms for creating autonomous, collaborative systems.
Agents in Industry:
The Best from the A
AMAS 2005
Industry Track
Michal Pˇechouˇcek, Czech Technical University
Simon G. Thompson, BT
86 1541-1672/06/$20.00 © 2006 IEEE IEEE INTELLIGENT SYSTEMS
Published by the IEEE Computer Society
• planning,
• scheduling,
• resource and strategic decision making,
• diagnostics,
• control and real-time replanning,
• software systems integration,
• interoperability,
• knowledge integration,
• ontologies, and
• simulation and modeling.
Despite some successful case studies in
industry, agent technology has suffered from
hype and a loss of momentum. This loss has
seen many unique properties of the early
tools and techniques subverted by changes in
the commercial environment (for example,
the Web’s emergence as the key corporate
application platform, the dot-com crash, and
Microsoft’s emergence as the dominant
provider of desktop-personal-productivity
software) or by the development of rival tech-
nologies such as Enterprise JavaBeans, over-
lay networks, and Web services. Recently,
however, the momentum has reversed course
as academic agent research programs have
borne fruit and particularly as agents have
incorporated advanced techniques from other
AI areas. Key examples of such advances
are the
• utilization of efficient, powerful plan-
ning algorithms,
• development of OWL,
• development of algorithms to reason
about action in teams,
• development of efficient market-clearing
mechanisms, and
• refinement of the general principles and
architectures of agents—specifically,
BDI (belief-desire-intention).
These techniques have enabled the imple-
mentation of applications having a clear
advantage over traditional systems.
In addition, improved computer hardware
and increased availability of open source
components, especially for networking and
software development, have made develop-
ment of effective agent applications cheaper,
faster, more reliable, and, above all, easier.
Another reason for agent technologies’
increased momentum is that the human cap-
ital available to practitioners has developed
rapidly as generations of students who have
been exposed to the principles and possibili-
ties of agents and AI have joined the work-
force. Clearly, without the skills and vision
to implement these techniques into practi-
cal solutions, progress is impossible.
The AAMAS Industry Track
Conferences organized by the agent
research community frequently discuss “blue
sky” research ideas (ideas that aim beyond
immediate application), theoretical- and
empirical-research results, and agent tech-
nology’s potential and actual applicability.
For example, A
AMAS is an annual meeting of
agent technology researchers and practition-
ers that has become the canonical forum for
the presentation of new results in the field.
The conference resulted from the merger of
three successful conferences (the Interna-
tional Conference on Multiagent Systems;
Agent Theories, Architectures, and Lan-
guages; and the International Conference
on Autonomous Agents). In an encouraging
development, for the first time, a special
track at A
AMAS 2005 covered research on
the industrial application of agents.
The Industry Track featured reports on
defense and exploration applications and
reports from commercial business operations.
1
In the aerospace applications session,
NASA presented a monitoring agent for
space shuttle launching criteria, and the Jet
Propulsion Laboratory presented an auton-
omous science agent flying onboard the
Earth Observing One spacecraft. Two pre-
sentations covered defense applications on
autonomous control and teamwork of un-
manned aerial vehicles, and one covered
an agent-based simulation application for
naval training.
In the logistics-and-transport session, the
Catholic University of Lueven presented
a decentralized approach to autonomous-
guided-vehicle control for warehousing.
Whitestein Technologies and Magenta Tech-
nology offered their solutions for transport
optimization in industrial logistics. Univer-
sity Jaume I Castellón researchers described
their concept of agent deployment for traffic
management and control. The ECN (Energy
Research Center of the Netherlands) reported
on the successful application of agent tech-
nology in electricity infrastructure control.
In the manufacturing session, Rockwell
Automation’s presentation on agent-based
industrial control represented a traditional
manufacturing industry. Another presenta-
tion detailed a system that the DFKI (Ger-
man Research Center for AI) developed for
planning and monitoring steel production,
and a Singapore Institute of Manufacturing
Technology presentation dealt with semi-
conductor assembly. Two presentations in
this session focused on general conclusions
instead of specific applications. The Czech
Technical University compared the success
and potential of agent deployment in defense
and manufacturing applications. Tom Wag-
ner, Les Gasser, and Mike Luck gave a panel-
like presentation on agent technology’s po-
tential impact.
The following short articles summarize
what we consider to be the four best contri-
butions to that track. Our goal is to convey
the event’s key contributions to a wider
audience than just the attendees or those
who have time to read the entire proceed-
ings. Unfortunately, we couldn’t include
every significant presentation or mention
every important discussion point, and many
attendees might disagree with our perspec-
tive. However, we hope that this personal
view serves as evidence of agent technol-
ogy’s real and considerable impact.
Reference
1. M. Pˇechouˇcek, D. Stainer, and S. Thompson,
eds., Proc. 4th Int’l Conf. Autonomous Agents
and Multi-Agent Systems—AAMAS 2005 Indus-
try Track, ACM Press, 2005.
Variable-Autonomy Control
of Teams of Uninhabited
Air Vehicles
Jeremy W. Baxter and Graham S. Horn,
QinetiQ
Uninhabited air vehicles are of particu-
lar interest to the defense sector because
they could significantly reduce risk to air-
crews. Current UAV systems typically
require multiple operators to control a sin-
gle platform. QinetiQ has been developing
an approach that lets one operator control
multiple platforms.
The basic concept is a decision-making
partnership between a human operator and
an intelligent uninhabited capability. (A
capability is a collection of platforms, sen-
sors, and weapons.) The human provides
mission-level guidance to the pool of coop-
erating UAVs and takes on a largely super-
visory role. The UAVs self-organize to
achieve the goals the operator sets, such as
to observe an area or to locate and destroy a
high-value mobile target. Owing to regula-
tory or liability issues, a human must make
MARCH/APRIL 2006 www.computer.org/intelligent 87
some critical decisions, such as weapon re-
lease. So, the uninhabited capability must
refer such decisions to the operator. We’ve
implemented this concept using a variable-
autonomy interface onto a multiagent sys-
tem, as part of a larger trials system.
The trials system
We use the trials system (see figure 1) to
evaluate potential concepts of use and tech-
nologies. It includes a synthetic-environment
(SE) simulation that models real-world
dynamic interactions. Human-in-the-loop
trials let us capture the key requirements for
the decision-making partnership. The sys-
tem elements have evolved in response to
feedback from trials (subjective comments
and objective performance measures) and
changes to the concepts of use.
A multiagent system provides a natural,
powerful way to represent multiplatform
tasks and sets of coordinated, cooperating
agents. The trials system contains four types
of agent (see figure 1). The user agent allo-
cates individual UAVs to the tasks that the
operator sets, and provides the operator with
information. Group agents plan and coordi-
nate a task’s execution, sometimes calling
on the capabilities of specialist planning
agents. UAV agents interact directly with
individual platforms, commanding the
autopilot to undertake specific maneuvers
and receiving status and sensor informa-
tion. The user agent routes requests for
critical decisions to the variable-autonomy
interface. Depending on the autonomy
level (set for each request type), the inter-
face will either automatically grant permis-
sion to continue or defer to the operator.
Group agents embody the knowledge of
how to plan and execute coordinated team
tasks using a framework based on joint-
intentions theory.
1
We originally designed
the framework to enable robust execution of
user orders by teams of entities in ground-
based battlefield simulations.
2
It provides a
solid grounding for the communication nec-
essary to keep a team task coordinated.
Originally, it contained only group and vehi-
cle agents and didn’t let an operator issue
new tasks during execution (at start-up, it
provided each team with a single order that
could be decomposed into orders for sub-
groups). Adding the user agent allows for
operator interaction and parallel tasking.
The trials scenario
The scenario for the trials is a time-criti-
cal targeting mission against a high-value
mobile target. The system deploys a pack-
age of four UAVs, containing a variety of
sensors and weapons, to locate and destroy
the target. The operator is the pilot of a
single-seat fighter. The mission consists of
two main phases: search and attack.
One specialist planning agent produces
plans for the search phase. It expands a set of
possible target positions into regions that a
moving target could reach in the next few
minutes. The agent plans routes that let the
UAVs efficiently search these regions with
short-range sensors and take images of poten-
tial targets that the operator will classify.
The attack phase can begin when the op-
erator has classified a ground entity as the
high-value target. Another specialist plan-
ning agent provides access to a dynamic
scheduler
3
that allocates UAVs to the tasks
they must execute during the attack phase:
release the weapon and gather images of the
target to see if the weapon has destroyed it.
We’ve used the multiagent system in
three trials. A single pilot was able to suc-
cessfully control the UAV team to com-
plete the missions. Including the agents in
the trials system has allowed the quicker
completion of more complicated missions,
with reduced operator workload.
Acknowledgments
This research was part of the UK Ministry of
Defence Output 3 research program on behalf of
the Director Equipment Capability—Deep Tar-
get Attack. We gratefully acknowledge their sup-
port. We’re part of a QinetiQ team that’s devel-
oping and implementing the decision-making
partnership concept; we focus on the multiagent-
system element.
References
1. H. Levesque, P. Cohen, and J. Nunes, “On
Acting Together,” Proc. 8th Nat’l Conf. Arti-
ficial Intelligence (AAAI 90), AAAI Press,
1990, pp. 94–99.
2. J.W. Baxter and G.S. Horn, “Executing Group
Tasks despite Losses and Failures,” Proc. 10th
Conf. Computer Generated Forces and
Behavioral Representation, 2001, pp.
205–214.
3. M.J.A. Strens and N. Windelinckx, “Com-
bining Planning with Reinforcement Learn-
ing for Multi-Robot Task Allocation,” Adap-
tive Agents and MAS II, D. Kudenko et al.,
eds., LNAI 3394, Springer, 2005, pp.
260–274.
88 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
User
agent
Operator interfaces
Information
and requests
Commands and
operator approvals
Status and
sensor data
Platform
commands
Group
agent
UAV1
agent
Group
agent
Group
agent
Group
agent
Specialist
planning agent
UAV2
agent
UAV3
agent
UAV4
agent
Multiagent system
UAV2 platform
controller
Synthetic environment
Figure 1. The main components of a trials system for controlling uninhabited air
vehicles. Each UAV agent interacts with a platform controller that’s connected to the
synthetic environment.
The PowerMatcher: Multiagent
Control of Electricity Demand
and Supply
Koen Kok, Cor Warmer, and René Kamphuis,
Energy Research Center of the Netherlands
Distributed generation of electricity is
providing an increasing part of the world-
wide energy supply. DG consists of differ-
ent sources of electric power connected to
the distribution network or to a customer
site. This approach is distinct from the tra-
ditional central-plant model for electricity
generation and delivery. Examples of DG
are photovoltaic solar systems, small and
medium-scale wind turbine farms, and the
combined generation of heat and power
(CHP).
When the share of DG increases in a geo-
graphical area, clustered control of DG by
common ICT (information and communica-
tion technology) systems can add value. As
a result, distribution networks are expected
to evolve from a hierarchically controlled
structure into a network of networks, in
which a vast number of system parts com-
municate with and influence each other.
The number of components actively involved
in coordination will be huge. Centralized
control of such a complex system will reach
the limits of scalability and communication
overhead.
A key technology for solving this prob-
lem is market-based control. In market-
based control, many control agents com-
petitively negotiate and trade on an elec-
tronic market to optimally achieve their
local control action goals. Use of market-
based control in the electricity infrastruc-
ture opens the possibility for distributed
coordination in addition to the existing
central coordination.
The PowerMatcher
The PowerMatcher method provides mar-
ket-based control for supply-and-demand
matching (SDM) in electricity networks
with a high share of DG. It’s based partly on
earlier research by Fredrik Ygge and Hans
Akkermans;
1
Hans Akkermans, Jos Schrein-
emakers, and Koen Kok;
2
and Per Carlsson.
3
In this method, a control agent represents
each device. The agent tries to operate the
device process in an economically optimal
way, within the process’s constraints. The
agents negotiate their electricity consumption
or production on an electronic exchange mar-
ket. The resulting market price determines the
power volume allocated to each device.
From the viewpoint of controllability,
devices that produce or consume electric-
ity fall into six classes, each having a spe-
cific agent strategy. We look at three in this
article. The first class consists of stochas-
tic-operation devices, such as solar and
wind energy systems, where the power
exchanged with the grid behaves stochasti-
cally. The second class is shiftable-opera-
tion devices, which must run for a certain
amount of time regardless of the exact mo-
ment and thus are shiftable in time. An
example of such a device is a ventilation
system in a utility building that needs to
run for 20 minutes each hour. The third
class comprises user action devices, whose
operations result from a user’s direct ac-
tion. Examples include audio and video
devices, lighting, and computers.
Local agents’ self-interested behavior
causes electricity consumption to shift
toward moments of low electricity prices
and causes production to shift toward mo-
ments of high prices. So, SDM emerges on
the global-system level.
A simulation
To investigate distributed SDM’s impact
for a residential area, we simulated a clus-
ter of 40 houses, all connected to the same
segment of a low-voltage distribution net-
work. Heat pumps (electricity consumers)
heated 20 of the dwellings; micro-CHP
units heated the other 20. The simulation
treated washing machines as shiftable-
operation devices with a predefined opera-
tional time window, photovoltaic solar cells
as stochastic-operation devices, and light-
ing as user action devices.
Figure 2 shows the result of a typical
simulation run. In both plots, a single plot-
line indicates the total consumption and
production, and we treat production as
negative consumption. In figure 2a, all
devices are free running; in figure 2b, the
market-based control agents match supply
and demand.
This simulation shows that our method
can exploit flexibility in device operation
through agent bids in an electronic power
market. The peak in electricity demand
is substantially lower in the controlled
case. From the viewpoint of network oper-
ations, this result is important, because the
highest expected peak demand determines
the needed network capacity (transformers
and cables). Reducing this peak reduces
network investments. Furthermore, intro-
ducing SDM results in a flatter, smoother
MARCH/APRIL 2006 www.computer.org/intelligent 89
20
10
0
–10
–20
Load (kW)
0:00 03:00 06:00 09:00
Time
12:00 15:00 18:00 21:00
20
10
0
–10
–20
Load (kW)
0:00 03:00 06:00 09:00
Time
12:00 15:00 18:00 21:00
Total cluster consumption
Total cluster production
Feed in from midvoltage net
Total cluster consumption
Total cluster production
Feed in from midvoltage net
Peak power
30% lower
Flat feed
in profile
(b)
(a)
Figure 2. Results of a simulation of residential electricity distribution: (a) free-running
devices; (b) market-based control agents match supply and demand. Multiagent
control leads to peak load reduction and power profile smoothing.
90 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
profile of the electricity fed in from the
midvoltage network. This result is inter-
esting from the viewpoint of electricity
trading, where increased predictability of
both production and consumption adds
value.
Field testing
We’re investigating the PowerMatcher in
real-life environments for two different busi-
ness cases. One aims to automatically reduce
the imbalance in a commercial trader’s real-
world portfolio by aggregating medium-
sized industrial electricity producing and
consuming installations (see figure 3). In this
experiment, overproduction and underpro-
duction of wind parks induce price changes
on the cluster’s electronic market. The other
devices’control agents react to this with
counteractions, which restore the cluster’s
energy balance. The first test results indicate
a decrease of the total power imbalance by
approximately 25 percent. Reduction of
unpredictability in the trade portfolio reduces
imbalance costs charged to the trader by the
independent network operator.
The other field test, on a cluster of micro-
CHP units operating as a virtual power plant,
demonstrates their ability to contribute to a
common control goal. This experiment uses
15 domestic heating systems at consumer
premises. The virtual power plant can pro-
vide value through electricity trading or local-
grid-operation support.
Acknowledgments
The European Commission partially supported
this research in the context of the EUSUSTDEV Proj-
ect ENK5-CT-2002-00673, called CRISP (Distrib-
uted Intelligence in Critical Infrastructures for
Sustainable Power).
References
1. F. Ygge and J.M. Akkermans, “Resource-
Oriented Multi-Commodity Market Algo-
rithms,” Autonomous Agents and Multiagent
Systems, vol. 3, no. 1, 2000, pp. 53–71.
2. J.M. Akkermans, J.F. Schreinemakers, and
J.K. Kok, “Emergence of Control in a Large-
Scale Society of Economic Physical Agents,”
Proc. 3rd Int’l Joint Conf. Autonomous Agents
and Multiagent Systems (AAMAS 04), IEEE
CS Press, 2004, pp. 1230–1231.
3. P. Carlsson, “Algorithms for Electronic Power
Markets,” PhD thesis, Dept. of Information
Technology, Uppsala Univ., 2004.
Manufacturing Agents
at Rockwell Automation
Vladimír Maˇrík and Pavel Vrba, Rockwell
Automation Research Center, Prague
Kenwood H. Hall and Francisco P.
Maturana, Rockwell Automation Advanced
Technology Laboratory
As the complexity of manufacturing
business environments grows, multiagent-
systems (MAS) technology is becoming
increasingly important for development of
highly distributed, robust, and flexible in-
dustrial-control architectures. From an MAS
viewpoint, the manufacturing system is a
community of highly distributed, autono-
mous, efficiently cooperating, and asyn-
chronously communicating units—agents—
integrated by the plug-and-operate approach.
Rockwell Automation Inc. manufactures
industrial-automation technology, leading
the US market in discrete automation and
control products. In 1995, its first indus-
trial-agent project optimized machine load
balancing and increased reliability of a
steel rod bar mill. It is currently applying
MAS technology to its flagship product,
ControlLogix programmable logic con-
trollers (PLCs), and developing the MAST
(Manufacturing Agent Simulation Tool)
Local agent
Wind turbine park 1
Wind turbine park 2
Residential heat production
Cold storage
Emergency generator
Test dwelling
PowerMatcher aggregator
Local agent
Local agent Local agent
Local agent Local agent
Data
communications
network
Figure 3. Control agents balancing a real-world commercial-trade portfolio.
MARCH/APRIL 2006 www.computer.org/intelligent 91
agent simulation infrastructure to support
design and validation.
Real-time control agents and simulation
Rockwell Automation’s control agent
architecture usually implements each agent
as a module that encapsulates both the real-
time control subsystem and the software
agent (see figure 4). The RT control subsys-
tem directly handles the information from
physical sensors and actuators in real time
and is programmed in a low-level language
(usually the ladder logic programming lan-
guage). The software agent is implemented
in a higher-level programming language
(usually C++ or Java) and handles decision
making and negotiation.
The important part of this solution is an
efficient runtime interface allowing both
information transfer from the RT control
subsystem (I/O and other control or diag-
nostics data) to the software agents and
propagation of the agents’ control actions
to the RT control subsystem. To simplify
this system’s integration with existing in-
dustrial-automation-control architectures
based on PLCs, we gave the agents direct
access to the PLC’s data memory so that
they can observe and influence the RT con-
trol subsystem directly.
When thinking of a real industrial de-
ployment of agents that requires high reli-
ability and strict adherence to real-time
constraints, you must abandon the idea of
hosting the software agents on a PC and
interfacing them to PLCs. Therefore, we
have modified the ControlLogix PLC’s
firmware so that the C++ agents can run
directly inside the PLC in parallel with the
ladder logic code. Rockwell has developed
the Autonomous Cooperative System as a
C++-based agent platform dedicated to
ControlLogix PLCs. The ACS lets us dis-
tribute the agents across several PLCs (one
PLC usually hosts several agents). It also
supports agent management services (reg-
istration, deregistration, services lookup,
and so on) and ensures the transport of
messages conforming to FIPA (Foundation
for Intelligent Physical Agents) standards
among the agents.
The ACS’s first application was the de-
velopment of a reconfigurable control sys-
tem for a US Navy ship’s chilled-water
system
1
that increased the system’s surviv-
ability. An individual agent controls each
element of the physical CWS equipment
(valve, cooling unit, piping section, and so
on). When an agent’s built-in diagnostic
module detects a failure, the agent initiates
negotiation with other agents to reconfig-
ure the CWS—for example, finding an
alternative path for water in the piping sec-
tion to avoid a broken part.
To test and validate the agent-based con-
trol system before deploying it in a manu-
facturing environment, simulation is indis-
pensable. The simulation must emulate the
manufacturing equipment or processes; for
this, we strongly prefer commercially avail-
able simulators such as Matlab or Arena.
Once the simulation proves that the agent-
based control system is mature enough to
deploy, we replace the simulation with the
real physical system. This shift must be as
smooth as possible, preferably without any
modifications to the agent code developed
for the simulation. Because the agents will
interact with the physical system by shar-
ing the control data in the PLC memory, we
use this mechanism also to share data with
the simulation.
For example, for agent-based control of
the CWS, we implemented the simulation
in Matlab and Simulink. After verification
and testing, we successfully deployed the
unchanged agent-based control system to
control the valves, cooling units, and other
actual equipment of a scaled-down physi-
cal model of the ship.
MAST
This simulation environment, which
Rockwell Automation developed and
implemented in Java, serves mainly as an
agent-based demo implementation for
material handling in flexible manufactur-
ing. The developed agent library represents
material-handling systems’ basic compo-
nents such as work cells, conveyor belts,
and switches (diverters). Agent cooperation
focuses on finding the optimal transporta-
tion routes in the system. The proposed
solutions provide fault tolerance and struc-
tural flexibility. You can emulate any com-
ponent failure (for example, any conveyor
belt failure), which causes the agents to
negotiate an alternative route to avoid the
broken component. You can add new com-
ponents (representing new transportation
capabilities) to the system or remove exist-
ing ones on the fly.
Recently, Rockwell Automation has ex-
tended the MAST environment to simulate
the holonic-packing-cell testbed at the Uni-
versity of Cambridge’s Centre for Distrib-
uted Automation and Control (see figure
5). They have extended MAST’s agent li-
brary with a set of agents to represent and
control particular components of the lab’s
equipment such as a Fanuc M6i robot, a
storage unit, a gate in a Montech conveyor
system, a gantry robot, rack storage, and
RFID (radio frequency identification) read-
ers. More important, an agent represents
each manufactured product—in this case, a
customized Gillette gift box. This product
agent autonomously and proactively con-
trols its own production process by negoti-
ating with the other agents. In this case, the
process involves packing a box with differ-
ent grooming items such as gels, deodor-
Agent
(C++ or Java)
Ladder logic
Low-level
control
High-level
control
FIPA
communication
IEC 1131-3
communication
ControlLogix PLC
FIPA
Agent
(C++ or Java)
Data tableTags
Ladder logic
Figure 4. A real-time control agent architecture for the ControlLogix programmable
logic controller. (FIPA stands for the Foundation for Intelligent Physical Agents;
IEC stands for the International Electrotechnical Commission.)
ants, and razors. The agents negotiate over
such issues as which storage location can
provide the requested items and which ro-
bot will pack them.
The industrial case studies in this article
illustrate that you can effectively employ
MAS technology to design the next gener-
ation of large-scale, robust, and flexible
manufacturing control systems. Features
such as fully decentralized decision mak-
ing, dynamic lookup for suitable service
providers, or embedded support for simu-
lations go far beyond the capabilities of
classic centralized and hierarchical indus-
trial-control systems.
Reference
1. F.P. Maturana, R.J. Staron, and K.H. Hall,
“Methodologies and Tools for Intelligent
Agents in Distributed Control,” IEEE Intelli-
gent Systems, vol. 20, no. 1, 2005, pp. 42–49.
Adaptive, Dynamic Transport
Optimization
Klaus Dorer and Monique Calisti,
Whitestein Technologies
Logistic networks’ increasing complex-
ity and dynamic nature motivates a cost-
sensitive rethinking of process and opti-
mization strategies.
1
This goal requires not
only efficient processes but also IT solu-
tions that can deliver the required flexibil-
ity and dynamically respond to change and
customization.
Living Systems Adaptive Transportation
Networks is a comprehensive agent-based
solution for optimization and dispatching
of full and partial truckloads, including track-
ing and real-time event handling. LS/ATN
includes
• a real-time route optimizer,
• an event management system that in-
forms dispatchers about a wide range of
events as they occur (or, proactively, if
expected events don’t occur),
• a tracking facility that provides accurate
data about the progress of orders, and
• a simulation mode that assists in tactical
and strategic decision making.
Agent-based optimization
Finding optimal routes for serving trans-
portation requests from a (usually large) set
of customers is a complex problem. A lim-
ited number of available trucks must pick
up and deliver transportation orders at spe-
cific customer locations. The trucks can be
of different types and capacities and are usu-
ally available at different locations. Truck
drivers must observe drive time restrictions.
Pickup and delivery must occur within spe-
cific time windows, even though time con-
straints can potentially be violated within
some tolerated degree (soft constraints).
The problem is highly dynamic, not only
because transportation requests aren’t all
known in advance but also because various
unpredictable events can affect previously
defined plans. Trucks might be delayed
owing to traffic jams or other unforeseen
problems or even become temporarily
unavailable.
2
You can distribute the solving of this
transport optimization problem among
multiple interacting software agents to
• achieve scalability with growing sizes of
problem instances,
• directly reflect the distributed nature of
transportation organizations and deci-
sion-making centers,
• facilitate the handling of local deviations
without having to propagate local changes
and recompute the whole solution, and
• increase robustness (avoiding a single
point of failure).
In particular, the LS/ATN architecture
reflects how logistics companies manage
this domain’s increasing complexity. A
transportation business is usually divided
into dispatching regions. Transportation
requests arriving at a region are first tenta-
tively allocated and possibly optimized in
that region. If orders’ pickups or deliveries
occur in different regions, these other re-
gions are also informed and asked to han-
dle the request in case they can provide a
cheaper solution to transport the order.
In LS/ATN, distinct software agents rep-
resent different regions. A local AgentRe-
gionManager manages trucks starting in its
region. A centralized AgentDistributor dis-
tributes incoming transport requests ac-
cording to their pickup location. When
receiving a new order, an AgentRegion-
Manager generates a valid solution (that
is, a transportation plan specifying which
orders to combine into which routes, and
which trucks will handle those routes). To
do this, LS/ATN uses a contract net proto-
col to sequentially insert transportation
requests.
3
The system checks all available
trucks in that region both to verify their
capability to transport the order and to de-
termine the cost. However, this approach
could produce suboptimal plans—for ex-
ample, because the “best fitting” truck is
92 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
Figure 5. A simulation of the Cambridge packing cell in the MAST simulation
environment.
MARCH/APRIL 2006 www.computer.org/intelligent 93
already full. So, to improve the solution,
the system schedules cyclic transfers be-
tween trucks.
4,5
A cyclic transfer is an ex-
change of orders between routes—transfer
requests among regions are triggered when
trucks have routes spanning different re-
gions. A simple strategy to select an order
transfer is a hill-climbing approach that
selects the most cost-saving transfers from
a neighborhood of possible transfers. This
hill-climbing process continues with all
changed routes until the system can’t per-
form any more cost-saving exchanges.
The LS/ATN design’s main advantage
stems from its direct mapping to today’s
transport business organizations and its
good scalability. Moreover, its computa-
tional overhead is also lower than a fully
distributed solution using one agent per
truck.
6
Its main drawback is degradation of
the solution, compared to a fully central-
ized approach (from a global-optimum
perspective). However, a fully centralized
solution often wouldn’t be feasible in real-
world scenarios.
Figure 6 illustrates details of a route that
LS/ATN generated.
Results
We ran extensive empirical tests for ABX,
a European logistics company, to determine
what cost savings LS/ATN could provide.
The cost model and constraint checker took
into account real-world costs and constraints,
thereby enabling comparison of the agent-
based solution’s optimization results with real
transport plans that professional dispatchers
created manually. The analyzed data set con-
tained roughly 3,500 real business transporta-
tion requests (orders).
LS/ATN decreased costs 11.7 percent; 4.2
percent of this stemmed from fewer driven
kilometers. Another 2.2 percent of the sav-
ings came from significantly increasing the
number of consecutive routes that are cheaper
to sell on the spot market. The rest of the sav-
ings stemmed from the LS/ATN solution
preferring routes starting in regions where
trucks are cheaper to buy. An additional im-
portant achievement is that the LS/ATN solu-
tion used 25.5 percent fewer trucks than the
manual solution. This is due to higher utiliza-
tion of the trucks and longer utilization, on
average, of each truck. Today, ABX uses
LS/ATN in its day-to-day operations.
LS/ATN draws its strength from a multi-
agent system core. Built on a bottom-up opti-
mization philosophy, goal-directed agents
interact to solve subproblems that, when
consolidated, result in a solution to the over-
all problem. Similar to human decision mak-
ing, solutions to problems arise from the
interaction of individual decision makers
(represented by software agents), each with
its own local knowledge. Traditional IT sys-
tems’ centralized, rule-based nature imposes
intrinsic limits on dealing successfully with
unpredictability. Multiagent systems don’t
have this limitation because collaborating
agents quickly adapt to changing circum-
stances and operational constraints.
References
1. “UK Consumer Products Industry Cites Cost
Reduction as Its Biggest Logistics Challenge,”
Exel News, 10 Sept. 2002, www.exel.com/
exel/home/media/news/newsreleases/
2002/pressreleasecostreduction.htm.
2. M.W.P. Savelsbergh and M. Sol, “The Gen-
eral Pickup and Delivery Problem,” Trans-
portation Science, vol. 29, no. 1, 1995, pp.
17–29.
3. J.-J. Jaw et al., “A Heuristic Algorithm for the
Multi-Vehicle Advance Request Dial-a-Ride
Problem with Time Windows,” Transporta-
tion Research, vol. 20 B, no. 3, 1986, pp.
243–257.
4. P.M. Thompson and H.N. Psaraftis, “Cyclic
Transfer Algorithm for Multivehicle Routing
and Scheduling Problems,” Operations Re-
search, vol. 41, no. 5, 1993, pp. 935–946.
5. S. Mitrovic-Minic, Pickup and Delivery Prob-
lem with Time Windows: A Survey, tech. report
TR 1998-12, School of Computing Science,
Simon Fraser Univ., 1998.
6. K. Fischer, “Cooperative Transportation
Scheduling: An Application Domain for
DAI,” Applied Artificial Intelligence, vol. 10,
1996, pp. 1–34.
Conclusions and
Lessons Learned
Michal Pˇechouˇcek and Simon G. Thompson
The A
AMAS 2005 Industry Track included
several discussions in various formats, rang-
ing from formal debates to philosophical
discussions in the small hours of the morn-
ing. These discussions explored how the
agent community could improve its rele-
vance and impact to build on successes so
far. While a reasonable amount of interac-
tion occurs between agent researchers and
industry, industrial adoption of agent-based
solutions faces these main bottlenecks:
• Limited awareness about the potential of
agent technology in industry. Agents are
used in a few specialized disciplines and
remain unused in others where they might
be appropriate.
• Limited publicity of successful indus-
trial projects with agents.
• Misunderstandings about agent technol-
Figure 6. Details of a route that LS/ATN (Living Systems Adaptive Transportation
Networks) computed, including route and order information, a map of the route, and
schedule information in tabular and graphical form.
ogy’s capabilities, which led to early
industrial adopters’ unrealistic expecta-
tions and subsequent frustration.
Some common unrealistic expectations
and inappropriate uses of agent technology
fall into seven main categories:
1
• Complexity. People often expect that
agent technology can help solve very
complex (perhaps NP-hard) problems.
In our experience, this is obviously incor-
rect, although partitioning problems using
the agent abstraction can often lead to
approximate solutions with lower com-
putational demands.
• Black box. People often view agent
technology as a black-box technology
(like neural networks or genetic algo-
rithms) that you can insert to solve a
particular complex problem. However,
agent technology provides primarily
system concepts and design paradigms
that are useful in well-defined classes
of problems.
• Intelligence. People sometimes think
that agents can directly deal with prob-
lem solving and domain-specific intelli-
gence. However, agent researchers’ prime
concern is the agents’ collective behavior
and decision making, and agent research
often overlooks the technology’s appli-
cation to real-life problems.
• Agentification. People think that you can
fully automate agent integration and leg-
acy system encapsulation. However, no
sophisticated mechanism exists that can
encapsulate any legacy system fully
automatically. Common current solu-
tions involve alternative technologies
(for example, Web services).
• Learning. People frequently overesti-
mate multiagent systems’ potential
for learning. They often think that an
agent should be superadaptable and
able to accommodate to any requested
behavior (this expectation is closely
connected to those of intelligence and
agentification).
• Interoperability. Standards and interop-
erability are computationally expensive.
It isn’t wise to use full FIPA (Foundation
for Intelligent Physical Agents) compli-
ance in systems where full openness
94 www.computer.org/intelligent IEEE INTELLIGENT SYSTEMS
Michal P ˇechouˇcek is
the principal investigator
and head of the Gerstner
Laboratory Agent Tech-
nology Group, an asso-
ciate professor in artifi-
cial intelligence at the
Czech Technical Univer-
sity in Prague, and a
part-time senior consultant for CertiCon. Con-
tact him at pechouc@labe.felk.cvut.cz.
Simon G. Thompson is
a principal research sci-
entist and research group
leader at the Intelligent
Research Center in BT’s
research department.
He’s also a visiting re-
search fellow at the Uni-
versity of Southampton.
Contact him at simon.2.thompson@bt.com.
Jeremy W. Baxter is
a lead researcher at
QinetiQ. Contact him
at jwbaxter@qinetiq.
com.
Graham S. Horn is a
researcher at QinetiQ.
Contact him at ghorn@
qinetiq.com.
Koen Kok is a scientific
researcher in intelligent
energy management at
the Energy Research
Center of the Nether-
lands. Contact him at
j.kok@ecn.nl.
Cor Warmer is a scien-
tific researcher in intelli-
gent energy management
at the Energy Research
Center of the Nether-
lands. Contact him at
warmer@ecn.nl.
René Kamphuis is a
scientific researcher in
intelligent energy man-
agement at the Energy
Research Center of the
Netherlands. Contact him
at kamphuis@ecn.nl.
Vladimír Maˇrík is the
managing director of the
Rockwell Automation
Research Center, Prague,
Czech Republic. Contact
him at vmarik@ra.rock-
well.com.
Pavel Vrba is the lead
of the Agent Technology
Group at the Rockwell
Automation Research
Center, Prague, Czech
Republic. Contact him at
pvrba@ra.rockwell.com.
Kenwood H. Hall is
the vice president for
Architectures and Sys-
tems Technology at
Rockwell Automation.
Contact him at khhall@
ra.rockwell.com.
Francisco P. Maturana
is the Agent Infrastruc-
ture Lead at the Rockwell
Automation Advanced
Technology Laboratory,
Cleveland. Contact him
at fpmaturana@ra.
rockwell.com.
Klaus Dorer is a senior
researcher at Whitestein
Technologies. Contact
him at kdo@whitestein.
com.
Monique Calisti is the
vice president of R&D
at Whitestein Technolo-
gies. Contact her at
mca@whitestein.com.
isn’t necessary (for example, in simula-
tion and modeling).
• Mobility. People often claim that agent
mobility is inevitable and more essential
than is actually the case. Often, migra-
tion of data or simple communication is
sufficient, rather than migration of an
agent’s code and state.
While the Industry Track attendees ap-
preciated the presentations’ high technical
quality, they frequently expressed one spe-
cific concern. The presented applications
demonstrated only a technical, revenue, or
efficiency advantage. These application’s
developers can rightly claim that their
solutions are superior to those that were
the previous state of the art. However, in
business, this isn’t the fundamental test of
value. Businesses assess an advantage by its
return on investment, but the Industry Track
presented no evidence of this. This point (in
a slightly different form) also applies to
fields such as defense and medicine, where a
precise commercial quantification of agent
technology isn’t possible. These fields have
well-known benchmarks and metrics for
evaluating system performance, but agent-
based systems rarely, if ever, prove superior
according to these criteria.
Reference
1. M. Pˇechouˇcek, M. Rehak, and V. Maˇrik,
“Expectations and Deployment of Agent Tech-
nology in Manufacturing and Defence: Case
Studies,” Proc. 4th Int’l Conf. Autonomous
Agents and Multi-Agent Systems—AAMAS 2005
Industry Track, ACM Press, 2005.
MARCH/APRIL 2006 www.computer.org/intelligent 95
Healthcare is a vast open environment characterized by shared
and distributed decision making and management of care. It
requires the communication of complex and diverse forms of
information among a variety of clinical and other settings, as well
as coordination among groups of healthcare professionals with
very different skills and roles. It has been argued in recent years
that intelligent agents’ basic properties (autonomy, proactivity,
and social ability) and multiagent systems’ main features (man-
agement of distributed information, and communication and
coordination between separate autonomous entities) make these
techniques good options for solving problems in this domain. This
special issue aims to provide empirical confirmations of this claim,
by presenting successful applications of agent technology in any
healthcare-related area (for example, diagnosis, monitoring,
scheduling, and decision support systems).
Submitted papers should address at least one of these issues:
• Successful application of agents and multiagent systems in
healthcare (this is the preferred topic).
• Cooperation between intelligent agents to improve patient
management (for example, distributed patient scheduling).
• Agents that deliver remote or elderly care (for example, home
care).
• Agents that provide information about medical services.
• Multiagent systems for patient monitoring and diagnosis.
• Multiagent systems that improve medical training or educa-
tion (for example, tutoring systems).
• Medical agent-based decision support systems.
• Information agents that gather, compile, and organize med-
ical knowledge available on the Internet.
• Solutions to the basic methodological and technological prob-
lems associated with deployment of agent-based healthcare
systems:
– Security and privacy of medical data.
– Social acceptance of agent-based systems.
– Lack of common medical ontologies.
– Lack of centralized control.
– Communication standards.
– Integration with other types of software.
– Legal and ethical issues.
• A survey of the state of the art in healthcare agents.
Important Dates
Submissions due for review: 9 June 2006
Notification of acceptance: 25 Aug. 2006
Final version submitted: 8 Sept. 2006
Issue publication: Nov./Dec. 06
Submission Guidelines
Authors who plan to submit a paper for this special issue are
encouraged to send a one-page text summary before 20 April
to Antonio Moreno, the special issue editor, at antonio.moreno@
urv.net. He will then provide feedback regarding the paper’s
adequacy for the special issue.
Submissions should be 3,000 to 7,500 words (counting a stan-
dard figure or table as 200 words) and should follow the maga-
zine’s style and presentation guidelines (see www.computer.
org/intelligent/author.htm). References should be limited to 10
citations. To submit a manuscript, access the IEEE Computer Soci-
ety Web-based system, Manuscript Central, at http://cs-ieee.
manuscriptcentral.com/index.html.
Questions?
Contact Guest Editor Antonio Moreno, antonio.moreno@urv.net.
Special Issue on Intelligent Agents in Healthcare
Submissions due 9 June 2006
Call for Papers
Call for Papers
IEEE
