Content uploaded by Nane Kratzke
Author content
All content in this area was uploaded by Nane Kratzke on Mar 03, 2016
Content may be subject to copyright.
Collaboration in Network-Centric Warfare – Modeling Joint Fire Support Teams
Christian Gerstner, Robert Siegfried
Universit¨
at der Bundeswehr M¨
unchen
Werner-Heisenberg-Weg 39
85577 Neubiberg, Germany
{christian.gerstner|robert.siegfried}@unibw.de
Nane Kratzke
University of Applied Sciences L¨
ubeck
M¨
onkhofer Weg 239
23562 L¨
ubeck, Germany
kratzke@fh-luebeck.de
Abstract—This paper presents an agent-based model to com-
pare different coordination patterns in joint fire support (JFS)
scenarios. Modern warfighting approaches depend heavily on a
separation of concerns (like reconnaissance, coordination and
engagement) and therefore impose high requirements on the
coordination of all involved parties. Following the General
Reference Model for Agent-Based Modeling and Simulation
(GRAMS), we present an agent-based model of this problem
domain. Our simulations indicate that decentralized JFS coor-
dination leads to smaller average times from identification of
a target to final engagement, while at the same time requiring
extensive resources. Central coordination is more effective in
terms of engaged units and reduced resource requirements, but
tends to take more time.
Keywords-Modeling and simulation; Multi-agent simulation;
GRAMS reference model; Network-centric warfare; joint fire.
I. INTRODUCTION
Joint Fire Support (JFS) is a military term for providing
lethal engagements in an ad-hoc manner in highly dynamic
warfighting scenarios. JFS requests are typically launched in
tactical situations by military ground units confronted with
non-predictable threats which can not be engaged by organic
engagement means of these ground units. JFS is realized by
military engagement, recce and on scene coordination means
provided by army, air force and navy units. These functional
nodes are assigned and combined ad-hoc.
A typical JFS request shall be executed within few min-
utes including the following tasks: determine adequate recce
and engagement assets, check rules of engagement, task
and reposition assets, collect and provide adequate target
data, conduct and assess the (lethal) engagement. A lot
of military command nodes on different command levels
may be involved in processing JFS requests properly and in
accordance with given rules of engagement. As JFS requests
can not be exactly forecasted in time, target location or class
nearly everything has to be coordinated ad-hoc.
A variety of national coordination patterns has evolved in
western countries including israeli armed forces to handle
this JFS problem domain. A coordination pattern is the com-
mand and control communication structure of command-,
engagement- and recce-nodes in order to collectively provide
a JFS service. None of the existing coordination patterns
seems to be adequate in every situation. Each one has advan-
tages as well as disadvantages. An optimal JFS coordination
pattern has to consider the extent and landscape of the
operational area of own forces, the amount of expected JFS
requests, defined areas of responsibilities of command nodes,
the amount of engagement, recce and on scene coordination
means capable to process JFS tasks as well as the applicable
chain of command.
The coordination patterns reach from strictly hierarchical
to completely decentralized (in vision) as well as hybrid
coordination patterns. Especially decentralized patterns are
reflecting modern warfighting approaches like network cen-
tric warfare visions [1], power to the edge approaches [2], in-
formation age combat models [3], [4] and resulting emergent
behaviour models [5] which make agent-based simulation an
obvious analysis approach.
This paper presents the inital version of an agent-based
model for analysing and comparing JFS scenarios as well as
JFS coordination patterns. We present our JFS model (which
is inspired by the information age combat model proposed
by Cares et al. [3], [4]) in section 2 and first simulation
results in section 3. Finally, we close with a conclusion and
an outlook on our ongoing research in section 4.
II. MODELING JOINT FIRE SUPPORT TEAMS
The development of the JFS model [6] is inspired by
the domain specific information age combat model [3],
[4] and follows closely the General Reference Model for
Agent-Based Modeling and Simulation (GRAMS) [7], [8].
Therefore, the description of the model presented here is
structured according to the GRAMS reference model.
To get acquainted to the problem domain, the model
is restricted in many issues. Once the basics of the
problem-domain are well-understood, the restrictions may
be relaxed and a larger parameter space will be covered
(cp. [9]). Currently, three types of agents are distinguished:
the Reconnaissance-Agent, the Coordinator-Agent and the
Engagement-Agent. Each target is modelled as an object,
which means that it is not able to plan or react to his
environment like an agent. An action of a single agent is
triggered by an event, which on the other hand is triggered
!"#"$%&&&'(%)'*)+$%,-./,0-12,03$)2,4./.,5.$2,$(.6$%,-.3317.,5.$0,8$%,-.3317.,-$*7.,-$9.5:,2327;
!"#$%$"&!'$()!)$(*)%+,-&.%%+/+-%)%+0111
230+)%.))%!*40$056.-%)%.-%7
77#
13
12
11
10
9
8
7
6
5
4
3
2
1
R
R
R
R
E
E
E
1 2 3 4 5 6 7 8 9 10 11 1 2 13 14 15 1 6 1 7 18
R
T
Figure 1. Schematics of the model environment (T = Target, R =
Reconnaissance-Agent,E=Engagement-Agent)
by the action of another agent, the environment or the agent
himself.
A. Aims of the model
The intention of the model is to evaluate different co-
ordination patterns. There are two possible directions: one
is to maximize centralization and the other is to minimize
it. Both coordination patterns have their own strengths and
weaknesses. The idea is to find the optimal pattern by
analyzing different parameters. These parameters can be the
time needed by a coordinator for finding and assigning a
required unit or the overall time needed until a specific target
is fought.
B. Macro-Level: Time and Environment
Time is modeled as discrete time steps. The duration of
each time step is not specified any further. This abstraction
seems feasible as the comparison of the coordination patterns
is purely qualitatively at the moment. Nevertheless, future
calibration and validation activities will adress this issue.
As indicated in figure 1, the environment is modeled as a
flat 2-dimensional matrix. Six different types of landscapes
are distinguished, namely forest, mountain, plain, city, sea
and inlandwater. The landscape determines the movement
possibilities of different unit types (army, air force, navy) as
well as specific limitations (e. g. reduced speed in mountain
areas).
C. Micro-Level: Objects and Agents
The Reconnaissance-Agents patrol along their routes,
which are defined by explicit waypoints (indicated by the
colored paths in figure 1). As soon as a target is located,
they stop and report this target to their superior Coordinator-
Agent. These commanders control all their subordinates and
evaluate their suitability of engaging this target. According
Visual perception
Acceptance of order
[Marking necessary]
[Other unit ordered to marking ]
[No marking necessary]
[has superior Coordinator]
[has no superior Coordinator]
Key:
Control-Flow
Data-Flow
Sensor
Send to Broadcastchannel Submit request
Move
Analyse effects of fire
Mark the target
Effector
Figure 2. Behavior of the Reconnaissance-Agent (depicted as sensor-
effector-chains)
to a pre-defined prioritization method they choose one
subordinate Engagement-Agent and order him to fight the
target. If the kind of weapon fire makes marking necessary,
the reporting Reconnaissance-Agent or another available
Engagement-Agent is ordered to serve this marking at the
same timepoint as the weapon fire from the executing
combat unit. The result is controlled by the Reconnaissance-
Agent and reported to the commanding Coordinator-Agent.
The number of Coordinator-Agents as well as the actual
process of coordination is influenced heavily by the number
of coordinators and their hierarchy. For example, if there are
no Coordinator-Agents, the whole process is coordinated by
the Engagement-Agents themselves.
1) Target: At the current state, targets are represented by
immobile objects. Target-Objects appear according to a pre-
defined rate, and disappear according to some distribution
(thereby imitating moving objects which leave the specified
area of operations).
2) Reconnaissance: Reconnaissance units patrol on de-
fined waypoints and keep a given area under surveillance.
They have not the ability to fight, but to analyze and mark
atargetiftheydetectone.Ifthiseventofdetectionhappens
a recce initiates a request for fire support and sends it to the
command and control unit responsible for this part of the
environment (see figure 2).
Later on, after a unit is assigned to the target, the recce
is capable of marking the target, if the combat unit needs
marking (e. g. a fighter bomber using guided bombs). After
the unit has fired, the recce analyzes the impact on the target.
If the target is succesfully destroyed, the recce continues
with his patrol.
3) Command and Control: The command and control
units are represented by Coordinator-Agents which are not
located at any specific position within the given environment.
They control a specific rectangular area. All targets which
are detected within this area are reported to the responsible
Command and Control unit. The Command and Control
77!
Key:
Control-Flow
Data-Flow
Visual perception Acceptance of order
Read Broadcastchannel
Get status
[Suitable]
[Not Suitable]
[has no Superior]
[has Superior]
[Order = Analyse effects of fire] [Order = Mark]
[Order = Fight]
[Platform = Ship]
[Platform = Airplane]
Check operating time
[Platform = Army/
Infantry]
[Not enough Time]
[Enough Time]
Sensor
Submit request Move
Analyse appropriateness
Calculate prioritization
value
Send to
Broadcastchannel
Move
Effector
Analyse effects of fire Fight the target Mark the target
Figure 3. Behavior of the Engagement-Agent (depicted as sensor-effector-
chains)
units have an amount of Engagement-Agents subjected to
them. If a request for fire occurs, they evaluate the situation
and check all subordinates if the request can be fulfilled.
Each subjected Engagement-Agent is listed in a matrix
together with a value representing the suitability for the
reported target. This way the most appropriate Engagement-
Agent is identified and ordered to fight the target.
The Command and Control unit may come to the con-
clusion that no one of his subordinates can fight the target.
In this case, the Coordinator-Agent first tries to pass the
request along to his superior Command and Control unit (if
he has one) or to a neighbouring commander. If all this is
not possible or this agent is on top of the hierarchy, he puts
the request in a queue and checks the feasibility again later.
4) Engagement unit: Figure 3 illustrates the behavior of
the Engagement-Agent. The Engagement-Agent waits at his
starting point until he is ordered to fight a target. As soon
as he gets an order, he starts moving to the target until it
gets in the range of his weapons and attempts to destroy the
target. If the range of his weapons is higher then his line of
sight he has a need for a marker to help him marking the
target. This marker can be any other Engagement-Agent or
the reporting Reconnaissance-Agent if they are capable of
marking. After the target is fought, the agent controls his
operating time if he can be assigned to another mission or
if he has to move back to his starting point.
Together with this agent the problem of decentralization
shall be explained: If there are no superior Command
140
160
100
120
140
160
60
80
100
120
140
160
Timesteps
0
20
40
60
80
100
120
140
160
Timesteps
0
20
40
60
80
100
120
140
160
12345
Timesteps
# Coordination Shortdescri
p
tion
0
20
40
60
80
100
120
140
160
12345
Timesteps
# Coordination Shortdescription
1 centralized Ͳ 3coordinator,3recce,9engagement
2 decentralized Ͳ 3recce,9engagement
3centralized Ͳ3coordinator,6recce,18engagement
dld
0
20
40
60
80
100
120
140
160
12345
Timesteps
# Coordination Shortdescription
1 centralized Ͳ 3coordinator,3recce,9engagement
2 decentralized Ͳ 3recce,9engagement
3centralized Ͳ3coordinator,6recce,18engagement
4decentralized Ͳ6recce,18engagement
5centralized Ͳ5coordination,3recce,9engagement
0
20
40
60
80
100
120
140
160
12345
Timesteps
# Coordination Shortdescription
1 centralized Ͳ 3coordinator,3recce,9engagement
2 decentralized Ͳ 3recce,9engagement
3centralized Ͳ3coordinator,6recce,18engagement
4decentralized Ͳ6recce,18engagement
5centralized Ͳ5coordination,3recce,9engagement
Figure 4. Average time from identification of a target to final engagement
in different scenarios.
and Control units the Engagement-Agents have to manage
among themselves. Because there is no hierachy between
all the Engagement-Agents they have to communicate with
each other to identify the most appropriate agent. To do this,
the reporting agent publishes the request into a broadcast
channel which is accessable by all free unbound agents.
Each agent now evaluates the situation by calculating a value
representing his appropriateness. This value is published by
all agents into the public broadcast channel so that every
agent gets to know the value of all agents. Each receiving
agent now can check if his own value is the highest or if
there is any higher value published by another agent. This
way the most appropriate agent can be clearly identified and
orders himself to fulfill the task of fighting the target.
III. RESULTS AND EXPERIENCES
A. Results
After successful implementation of the model, nine dif-
ferent scenarios were simulated. In these scenarios we tested
three different ways to identify the most appropriate engage-
ment unit while using different approaches of prioritization.
One way of prioritization was to command the unit with the
shortest way to the target, which thus can engage the target
the fastest. A second approach is to identify the engagement
unit, which is just able to fight the target. This way higher-
ranked units are saved for more dangerous targets while
assigning combat units according to their strengths. The third
approach is a mixture of the first two and should combine
the advantages of both.
Each of the nine scenarios was simulated 100 times to
get average values. Figure 4 illustrates our main findings:
Although basically flat hierachies are aimed for, they reach
7(%
their limit quite fast in the constraints of reality. First and
foremost, a huge amount of combat units is needed for
optimal coverage in decentralized coordination. Also, perfect
communication has to be ensured between all units to enable
the necessary interaction and coordination as well as to
avoid multiple fighting of the same target. In summary,
decentralized coordination leads to smaller average times
from identification of a target to final engagement, while
at the same time requiring extensive resources. Central
coordination is more effective in terms of engaged units and
reduced resource requirements, but tends to take more time.
B. Experiences
The most difficult part of the development was to ensure
the correct coordination between the agents. Modeling the
various information exchange relations and the subsequent
activities to be carried out by the agents is a challenging
task. Even though only three different types of agents were
considered, it is difficult to keep track of the intricate
interplay of mutiple agents.
By following the GRAMS reference model to develop
the agent-based model, we could focus purely on domain-
specific issues. In this sense, the GRAMS reference model
served very well as a guideline throughout the development
process. The strict seperation of events and actions defined
by the GRAMS reference model turned out to be helpful
also. This separation allowed us to construct complex event-
action chains where each event could trigger different actions
at the same time, whereas these action could produce events
as well.
While being beneficial, these event-action chains caused
trouble at the same time. In fact, it turned out that they could
hardly be analysed and debugged. This is not necessarily
a drawback of the GRAMS reference model, but has at
least two reasons: First, the tool chain available does not
support specific aspects of the GRAMS reference model very
well and debugging features are far from complete. Second,
and perhaps more notably, this complexity of modeling
coordination patterns may be immanent to these kind of
models.
IV. CONCLUSION AND OUTLOOK
We presented an agent-based model to analyze the influ-
ence of different coordination patterns on so-called joint fire
support teams. In this (restricted) model, three different types
of agents had to coordinate themselves (Reconnaissance-
agents, Coordinator-agents and Engagement-agents). The
coordination patterns investigated in a first stage ranged
from a wholly centralized coordination to completely de-
centralized coordination. A thorough model validation still
outstanding, the first simulation results indicate slight ad-
vantages of decentralized coordination about a centralized
coordination. At the same time, our experience regarding
the modeling of coordination is that the complexity increases
very fast and even smaller scenarios (in our case, with three
different types of agents and less than 50 agents in total) are
quickly hard to overlook.
The trade-offs between centralized and decentralized co-
ordination in combination with the resource needs and
utilization will be in the focus of future work. Furthermore,
a lot of restrictions were made to reduce the complexity of
the first model version. This model was very helpful to get
acquainted to the problem domain. With this background
knowledge the restrictions may be relaxed and a larger
parameter space will be covered (cp. [9]). A first extension
is to implement moving targets which will add a lot of
complexity to the coordination of the agents. Also, improved
route finding algorithms for the recce agents are of interest
(cp. [10]). Finally, we want to calibrate and validate the
model as well as the parameters to move on from qualitative
to quantitative investigations.
REFERENCES
[1] D. S. Alberts, G. J. J., and F. P. Stein, Network Centric
Warfare. DoD Command and Control Research Program
(CCRP), 1999.
[2] D. S. Alberts and R. E. Hayes, Pow e r t o t h e edge . DoD
Command and Control Research Program (CCRP), 2003.
[3] J. Cares, “An information age combat model,” Director, Net
Assessment, Office of the Secretary of Defense, Tech. Rep.,
2004.
[4] ——, Distributed Networked Operations: The Foundations of
Network Centric Warfare. iUniverse, 2006.
[5] M. E. J. Newmann, “The mathematics of networks,” in The
New Palgrave Encyclopedia of Economics, L. E. Blume and
S. N. Durlauf, Eds. Palgrave Macmillan, Basingstoke, 2008.
[6] C. Gerstner, “Erweiterung und Implementierung eines Mod-
ells zur Analyse von Fragestellungen zur Koordination verteil-
ter Organisationsstrukturen,” BSc Thesis, University of the
Federal Armed Forces Munich, December 2009.
[7] R. Siegfried, “A General Reference Model for Agent-Based
Modeling and Simulation,” in EUMAS 2009, December 2009,
7th European Workshop on Multi-Agent Systems.
[8] R. Siegfried, A. Lehmann, R. El Abdouni Khayari, and
T. Kiesling, “A Reference Model for Agent-Based Modeling
and Simulation,” in Proceedings of the Spring Simulation
Multiconference. SCS, March 2009, agent-Directed Sim-
ulation Symposium.
[9] A. Santamaria and W. Warwick, “Sailing to the model’s
edge: Testing the limits of parameter space and scaling,” in
Proceedings of the BRIMS 2010, 2010.
[10] P. Paruchuri, J. Pearce, J. Marecki, M. Tambe, F. Ordonez, and
S. Kraus, “Coordinating randomized policies for increasing
security of agent systems,” Journal of Information Technology
and Management (ITM), vol. 10, no. 1, pp. 67–79, 2009.
7()
