|
B. Bodt |
J. Forester |
C. Hansen |
|
E. Heilman |
R. Kaste |
J. O’May |
 |
 |
 |
Command posts in battle are busy hubs for information flow. A commander and his staff must sift through data representing thousands of events to build an accurate picture of a battle. With seconds counting and lives dependent on timely decisions, commanders can use every advantage.
The Simulation Concepts Branch of the U.S. Army Research Laboratory (ARL) is responsible for creation of techniques and systems to assist battlefield commanders in the military decision making process. This process is based in part on Courses of Action (COAs). Our primary focus is COA Analysis (COAA). We are merging operational reality with simulation to benefit the commander and battle staff at Division and lower echelons.
Our project applies military planning and combat simulation software to the evaluation of automated COA generation tools. Envisioned is a testbed that will enable the assessment of COAs in a simulated operational environment. Initially, one COA generation tool, Fox-GA (for Genetic Algorithm) produced by the ARL Federated Laboratory, and one combat simulation, Modular Semi-Automated Forces (ModSAF) produced by Lockheed-Martin, comprise the prototype testbed. The project is exploring statistical analysis and experimental design techniques that will enable simulated exercises to be utilized as a part of COAA.
The project goal is to reveal strengths and weakness inherent in COA generation software. Scientific evaluation of COA attributes will benefit the soldier by providing increased operational possibilities during planning. The complexity of COAA provides fertile ground for development and application of decision aid technology. This work has potential for significantly improving battlefield command and control capabilities.
The COAA testbed contains four process parts with associated computer software, hardware, and data. These parts are 1) Automated COA Generation, 2) Scenario Translation, 3) Experimentation, and 4) Statistical Evaluation.
ARL, in conjunction with theUniversities of Illinois and Minnesota, has created a software tool to generate automated COAs. The program, called Fox-GA, is “an intelligent planning support tool designed to rapidly generate a variety of coarse-grained, high quality, friendly courses of action for military planners.” Conversion of the Fox-GA scenario to a form usable by a combat simulation was accomplished manually. Experimentation was accomplished by executing the translated scenario within ModSAF. Data collected from experimental simulation executions were analyzed using variances and means to gauge overall COA performance.
The Fox-GA scenario portrays combat between a friendly force brigade and an adversary force battalion. The translation captured a complex combat featuring 418 entities. We had difficulty determining the correct amount of computational power necessary to handle this large scenario translation.
Our first attempts to execute the scenario on a Sun Microsystems UltraSPARC60, a system with two 296 MHz UltraSPARC-II processors, were not successful. In fact, the UltraSparc60 could not complete the scenario setup. Additional systems were added to enable simulation execution in a distributed network environment. The UltraSPARC60 was used to display the plan view and did not support scenario entity activities. Using an incremental approach, five more SGI systems were eventually used to share simulation entities: three SGI O2s with one175 MHz (IP32) R10000 processor each, one SGI Maximum Impact with one 195 MHz (IP29) processor, and one SGI Infinite Reality Onyx with four 194 MHz (IP25) processors. The modified network and hardware configuration was still insufficient to gather data of completed scenarios; all attempts to execute the translated Fox-GA scenario on local computers failed.
We used a different ModSAF combat formation to reduce the translated scenario to 280 entities. A lower number of entities resulted in the generation of fewer network packets. Yet the networked solution was still insufficient for ModSAF data collection. A network packet bundling technique was used to diminish the number of network transmissions, but also did not provide enough network capacity to alleviate the lack of computing power.
We decided that super computers would be needed to enable data collection from combat simulation runs. The entire process, from obtaining accounts on ARL’s Major Shared Resource Center’s high performance computers to actual scenario execution, was accomplished in the space of five days. We installed the ModSAF software on four SGI Origin 2000 systems with little difficulty. Each system has at least 32 processors (250 MHz IP27 R10000) and 32 gigabytes of main memory. These features enabled a single SGI Origin 2000 system to support successful scenario completion and data collection, while eliminating the requirement for network transmissions.
Without the availability of super computers, the resources available for this project would have been insufficient to provide results. The simulation data compiled using one Origin 2000 system over the space of two weeks show that ModSAF and Fox-GA are consistent, as friendly forces were winning most of the time in each. Fox’s internal wargamer may be good enough to provide battle results for competing COAs; however further experimentation will be necessary to gauge the accuracy and degree of these results. Battlefield commanders will benefit from laboratory-tested, expedient decision aids, and we will continue to use the power of super computers to improve the Army’s command and control abilities.
|
B. Bodt |
J. Forester |
C. Hansen |
|
E. Heilman |
R. Kaste |
J. O’May |
|
|
|
|