Simulation models represent objects at many different levels of fidelity and as a result, interoperability between models has often required the use of a disaggregation process. This has allowed a pair of simulations to interoperate to a limited degree, based on customized compromises in unit/entity representations and functions. However, as more diverse simulations are added to training suites, the need for consistent disaggregation techniques will increase. It will no longer be sufficient to design a point-to-point solution between two simulations. Rather, we will require a defined method for transforming unit/entity representations that is consistent and repeatable across multiple nodes.
This paper proposes a fundamental structure for organizing the disaggregation process. Specific classes of data transformations are identified and organized into layers for incremental and logical transformation. These layers contain algorithms for transformation based on: Doctrinal Templates, Historic Events, Environment, Large Feature, Small Feature, Internal Operations, External Operations, and Controller Input.
We maintain that disaggregation will always be required, independent of the fate of traditional aggregate level modeling. Projects like the High Level Architecture and the Joint Simulation System aim at consistency through distributed functionality and architecture, both of which will necessitate sharable and reproducible disaggregation methods. Disaggregation layers will also support the continuing expansion of the simulation domain into the areas of command and control computers, combat systems, and tactical networks. These real-world systems are designed to manage and distribute information at levels of aggregation that are useful to the soldiers using them. These levels are seldom identical to each other or to the levels replicated in simulation, therefore, a transformation process will be required to support interoperability in this area as well.
Roger Smith is a Technical Director for Mystech Associates. He is actively involved in designing, developing, and fielding simulations, having contributed to JSIMS, WARSIM, JSIGSIM, BLRSI, ADST, TACSIM, AWSIM, ALBAM, and others. He is very active in the simulation industry serving as the Chairman for the ACM Special Interest Group on Simulation, the General Chairman for the Electronic Conference on Training Simulation, and as a member of the editorial boards of ACM Transactions on Modeling and Computer Simulation and the International Journal of Computer Simulation, Modeling, and Analysis. His research interests include embedded simulations in operational equipment, integrated simulation-C4I systems, and problems in aggregation/disaggregation.
The US military trains soldiers at many different levels - gunnery, team, company, battalion, corps, and echelons above corps to name a few. Each of these is supported by a different set of tools, some of which are simulations. Historically, each area has developed its own tools, but recently the trend has been to use the same tools at multiple levels, or to join multiple levels together and expect them to interoperate consistently. These connections are hindered by the capabilities of the tools and simulations in the field, the fundamental differences in the problem spaces, and scaling issues of the computer systems and operational techniques. This has resulted in the need to design simulations and interfaces that are more scaleable than those of the past.
We believe that interoperable, heterogeneous simulations will require the use of many different levels of aggregation. This is not a compromise of pure entity level, virtual simulation, but rather a recognition of the fact that the perception of the battlefield varies according to the perspective of the human training audience. Commanders at higher levels receive consolidated data in the real world because it fits their decision making processes. It also fits within the limitations of the communications, navigation, and command systems currently deployed on the battlefield.
To support multiple levels of aggregation consistently, a layered solution of the problem is most applicable. This will allow simulations to aggregate and disaggregate to the level of detail appropriate for them, and to remain consistent with other simulations that are operating at higher and lower levels around them. Each layer can be viewed as a plug-in to allow the customer to choose the types of variables that will drive the disaggregation process. These layers are illustrated in Figure 1 and are discussed below.
Figure 1. Disaggregation Layers
Doctrinal Template. The first layer represents the textbook solution for equipment deployment. This is an instantiation of the placement of subordinate units and equipment according to standard military doctrine. It reflects the type, size, and activity of the unit in a manner very similar to the branch-and-node templates described in the previous section. This is the static form from which adjustments are made in the lower layers.
Historic Events. This layer adjusts the content and status of a unit based on past events. These may include attrition, splitting, merging, and task organization operations. These are the fundamental changes to the unit that are based on events rather than external environment. Each unit may have missing or excess equipment and does not perfectly fit the doctrinal template. Basic mechanistic decisions for handling this situation exist in this layer.
Environment. The effects of weather, night, twilight, and chemical environment are included here. Each of these may result in tighter unit formations and movement precautions that are not experienced during good weather, daylight operations.
Large Feature. Following the placement of objects according to doctrine, history, and environment, the object locations are adjusted according to the presence of large environmental features such as oceans, lakes, swamps, rivers, wooded areas, and cities. This layer does not identify the presence of individual trees and buildings within the features, but merely responds to the border, grade, and altitude of the area covered by the large feature. In some cases, the result is a complete avoidance of the area, as with an ocean or lake; in others it results in the adaptation of the formation to operate in cities or wooded areas. The method may be used for naval forces, in which case the presence of land causes avoidance to keep vessels from beaching or foundering on reefs and shoals. For aircraft, the areas to avoid include the incursion of terrain into the sky and the establishment of flight corridors over certain areas. Air Force disaggregation will keep aircraft out of no-fly zones to avoid destruction and the provocation of enemy forces by incursion into their air space.
Small Feature. This layer takes the previous one to more detail. It accounts for the actual contour of the land and the depths and banks of rivers. This is where adjustments are made to place vehicles on roads and to respond to dynamic terrain features such as bomb craters, and tank ditches. Individual trees, buildings, water towers, and power lines are considered here.
Internal Operations. Following all of the above adjustments, consideration must be given to the operational status and environment of the unit. Its total strength will result in deployment modifications to compensate for attrition on one flank. Vehicles will be moved to account for damage and the depletion of supplies. This includes modifications for the presence or loss of night vision equipment, chemical gear, and other small items used to respond to the environment.
Vehicles without fuel, ammunition, or repair parts may be required to fall to the rear or remain behind while the main body advances. In this level we consider the possibility that the unit will not remain cohesive. It is allowed to break into smaller pieces where each has its own objective and limitations.
External Operations. The presence of enemy forces can cause adjustments in vehicle placement and orientation. This will provide a strong defense, prepare for an eminent engagement, or present a pitched battle. Consideration is also given to the presence and density of non-combatants in the area and the placement of known minefields.
Controller. Finally, the placement of objects may be manually adjusted by orders from simulation controllers or the training audience to produce custom designed patterns. These adjustments may also be provided by external simulations that are providing specific direction to individual vehicles. This will support the combination of automated units and vehicles that are controlled by other manned simulators.
This capability may be used to conduct experiments with new pieces of equipment or new concepts in doctrine and tactics. Advanced sensors may require unique placement that is not represented in current doctrinal templating, but may be tested and evaluated in a simulation environment.
A pictorial example of the operations of these layers is shown in Figure 2.
Figure 2. Layered Disaggregation Illustration
When all of the defined layers are being included in a simulation, we have attempted to illustrate them in the order that they may most effectively be used. However, it is unlikely that all simulations will use all layers. In some cases Small Features may be omitted, or Internal Operations, or Controller. Developers may even reorder the layers that are most applicable to their problem domain. For example, the analytical community may require direct control of object placement, choosing to use the Controller layer first, followed by History and Internal Operations. Others may prefer Controller-History-External Operations-Large Feature.
The usefulness of the layer definitions is not limited to a complete implementation. But rather, the conceptualization and organization of the methods such that they may be selected and described consistently across multiple programs is its greatest value.
These layers can serve as a blueprint to be followed as our understanding of the methods to perform these tasks matures. The layering process is intended to isolate specific disaggregation methods so they may be removed and replaced with better algorithms in the future.
It is unlikely that we will create disaggregation algorithms that are superior to the actual inclusion of 2 million soldiers in virtual simulators making dynamic, synergistic decisions on the battlefield. It is also questionable whether a totally automated disaggregation algorithm can make better decisions than a CGF system working in conjunction with a trained military operator. But, a disaggregation algorithm can perform these tasks without the huge investment in computer hardware and man-power required to perform the same tasks under human control.
As the algorithms develop we expect to find that one of the biggest hurdles is the processing power available to execute them. Disaggregation is just one of the operations that must be carried out to conduct simulated combat and it appears that the best place to host these algorithms is on top of one of the existing CGF systems. These contain the type of environmental and force information necessary to perform the calculations and to automate the control of the objects for other combat operations.
It would also be ideal for all of the dependencies between disaggregation variables to exist only within assigned layers. Unfortunately, this does not appear to be possible. Instead, the strongest dependencies are the defining factor in assigning a variable to a layer, but dependencies cross many or all layers. This implies that the omission of a layer will result in a definite skewing of variables in other layers. The minimization of this effect can only be accomplished by the application of and experimentation with the layering method. This will lead to the discovery and quantification of the dependencies. It will also lead to a reorganization of the variables within the layers, and perhaps the addition or deletion of some layers. With an improved understanding, adjustments can be made to compensate for layers omitted if they are deemed significant to the mission of the simulation.
At this point, certain interactions seem to be clear. Environment and External Operations have similar effects on vehicle placement. Internal Operations, such as a shortage of fuel, may drive decisions in the Large Feature layer, such as an attempt to minimize consumption along a route. Analysts may prefer the implementation of the Control layer first to allow them to observe the effects of the other layers on their experimental placements. Finally, it is not clear whether minefields more aptly fit into the External Operations, Environment. or Small Feature layer.
Disaggregation techniques are just one method of information management in simulations. These must be joined with a host of other techniques being developed within simulation projects. The layering of disaggregation operations described above may be useful in performing experiments like the Eagle/BDS-D integration and the AWSIM/ModSAF integration. Both of these attempt to join specific instances of virtual and constructive simulations together, where the need for disaggregation is defined by the legacy models involved. But, future simulations may present a broader view of the battlefield, requiring more dynamic methods for simulation interoperability across levels of aggregation. Systems like the Joint Simulation System, Warfighter Simulation 2000, National Air and Space Warfare Model, and others may represent combat and military entities in different ways. They may also be tasked with missions which require a more analytical focus. In both cases a flexible structure that can be adapted to the mission requirements would be very useful in executing simulation exercises and experiments.
As the field of cognitive modeling matures there will be a need to present information to the simulated decision makers in a form that is similar to that available to their human counterparts. Though cognitive models may be made to operate on vehicle level data, they would be more efficient and representative of human decision making if presented with aggregated data. Layered disaggregation will assist in adjusting the simulated picture to fit the view of many different command decision making simulation models.
In the real world of command and control we are finding a greater emphasis on computer tools that provide decision making support to military officers. Tactical C4I systems must dedicate most of their power to the processing of information concerning current combat activities. The small budget that remains for decision support (a.k.a. simulation execution) may benefit from interoperability with simulations that can provide various levels of detail on unit/object placement and operations. A layered disaggregation technique should be able to support these operations better than currently available static templates and system specific solutions.
Since commanders conceptualize and view the battlefield at the aggregate unit level, simulations supporting or stimulating them must be able to provide information at the same level. The simulations must also be able to accept orders and requests from the commanders at these levels. A consistent aggregation/disaggregation method will be very helpful in bridging this gap between combat and simulation computers, a gap that is widening as simulations move to entity level modeling and greater levels of interoperability.
In conclusion, we wish to reiterate that simulation and combat computers are being integrated on more projects. In the real world, the focus is on providing information to a human in a form that aids quick and accurate decision making. In many cases, commanders (both real and simulated) require a picture that is an aggregation of the real world. We must be able to build systems that can render information at many different levels, rather than the conveniently defined "constructive" and "virtual" levels used in the past.
Stimulating real-world equipment with simulations
results in limitations that are not always present in the simulation world.
Upgrading ten or twenty simulation centers with more powerful computers
and networks to manage a larger exercise is very inexpensive compared to
upgrading hundreds or thousands of tactical computers that may be custom
built for military applications. In addition, the number of operators that
must be re-trained in the simulation world is a fraction of that found
operating tactical computers. Simulations will have to be designed to operate
within the constraints of tactical computers, networks, and operators.
A simulation may drive a Pentium-based computer with a clock speed of 200
MHz, 64 megabytes of RAM, 2 Gigabyte hard drive, and an SVGA monitor, all
connected through a T1 line; and in the same exercise a 80286 running at
25 MHz, with 4 megabytes of RAM, a 40 megabyte hard drive, and an EGA monitor,
connected through a 1200 baud modem device. To support this type of operation
the simulation must be able to adjust the amount of data being transmitted.
These adjustments may be accomplished by compression, reduced transmissions,
variable message formats, and aggregation. To maintain consistent representations
a formal aggregation/disaggregation method must be developed - something
similar to the disaggregation layers presented here.
Calder, Robert B. Peacock, Jeffery C., Panagos, James, and Johnson, Thomas E. 1995. Integration of Constructive, Virtual, Live, and Engineering Simulations in the JPSD CLCGF. Proceedings of the 5th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Clark, Karen J. and Brewer, Dave. 1994. Bridging the Gap Between Aggregate Level and Object Level Exercises. Proceedings of the 4th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Franceschini, Robert W. 1994. Intelligent Placement of Disaggregated Entities. Proceedings of the 4th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Griggs, Larry, et al. 1997. The Technical Approach to Building C4I System Specific Interfaces to the MRCI. 1997 Spring Simulation Interoperability Workshop. Orlando, Florida.
Hartzog, William W. 1996. Land Combat in the 21st Century. US Army Training and Doctrine Command. Fort Monroe, Virginia.
Foss, Billy and Franceschini, Robert. 1996. "Foundations of the New DIS Aggregate Protocol", Proceedings of the 1996 Electronic Conference on Interoperability in Training Simulation. Elecsim96
Joint Simulation System Project Office. 1996. JSIMS Domain Specific Software Architecture. http://www.jsims.mil/lfs_dssa.html.
Karr, Clark R. and Root, Eric. 1994. Integrating Aggregate and Vehicle Level Simulations. Proceedings of the 4th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
McGarry, Stephen. 1995. The DoD High Level Architecture (HLA) Run-Time Infrastructure (RTI) and Its Relationship to Distributed Simulation. Summary Report - 13th Workshop on Standards for the Interoperability of Distributed Simulations. Institute for Simulation and Training. Orlando, FL.
Petty, Mikel D. and Franceschini, Robert W. 1995. Disaggregation Overload and Spreading Disaggregation in Constructive+Virtual Linkages. Proceedings of the 5th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Powell, Edward. 1997. The JSIMS Architecture. 1997 Spring Simulation Interoperability Workshop. Orlando, Florida.
Pratt, David R., Johnson Matthew, and Locke, John. 1994. The Janus/BDS-D Linkage Project: Constructive and Virtual Model Interconnection. Proceedings of the 4th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Root, Eric D. and Karr, Clark R. 1994. Displaying Aggregate Units in a Virtual Environment. Proceedings of the 4th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Salisbury, Marnie R., Booker, Lashon B., Seidel, David W., and Dahmann, Judith S. 1995. Implementation of Command Forces (CFOR) Simulation. Proceedings of the 5th Conference on Computer Generated Forces and Behavioral Representation. Orlando, Florida.
Smith, Roger D. 1995. The Conflict Between Heterogeneous Simulations and Interoperability. Proceedings of the 1995 Interservice / Industry Training Systems and Education Conference. Albuquerque, NM.
TACSIM Project Office. 1991. Rapid Scenario Preparation Unit for Intelligence (RASPUTIN) Detailed Design Document. Manassas, Virginia.
Wilson, Annette L. and Weatherly, Richard M.
1994. The Aggregate Level Simulation Protocol: An Evolving System. Proceedings
of the 1994 Winter Simulation Conference. Orlando, Florida.