Modeling, Simulation and Training -- Projects

Automated Discovery of Innovative Tactics and Behaviors

Capturing Behavioral Influences in Synthetic C2

Computational Embedded Training Strategies

Diagnostic and Analysis Tools for Agent-based Combat Simulation Models

Distance Learning with Intelligent Agents

En Route Airspace Modeling

Realistic Schedules for Future Aircraft Traffic

TFM Post-Event Analysis

Modeling, Simulation and Training

Modeling, Simulation and Training focuses on information technology to support training, and the technology and innovative application of modeling and simulation. The information revolution is fueling changes in the workplace at an unprecedented rate, and these changes are threatening to overwhelm conventional education and training approaches. Fortunately, advanced instructional technologies like embedded training and collaborative learning environments can help warfighters and intelligence analysts adapt to these changes. Advances in simulation infrastructure, interoperability architectures, and modeling paradigms have simplified the application of simulation, demonstrated the feasibility of building simulations from reusable components, and otherwise facilitated a revolution in simulation application.

Automated Discovery of Innovative Tactics and Behaviors

Lashon B. Booker, Principal Investigator

Washington only

Problem
Modeling and simulation play a key role in the design, analysis, and implementation of new military concepts and systems. Effective modeling in this context requires the capability to quickly generate innovative twists on operational concepts, tactics, and possible threat responses. Currently, the only possibilities examined are those few that happen to come to mind for the human designers and analysts.

Objectives
Any technique that enables the human to systematically examine a broader range of options, or suggests alternatives the human may not have considered, would greatly increase the effectiveness of these simulation-based activities. We will develop new machine learning techniques to address this need. Our hypothesis is that innovative tactics and behaviors can be learned automatically from experience in a simulation.

Activities
The research will first develop techniques that can learn rule-based reactive behaviors given feedback about outcomes. That approach will be extended to learn more structured, distributed behaviors (e.g., requiring teamwork). The final improvement will address behaviors requiring primitive problem-solving capabilities. The techniques will be tested on benchmark machine learning problems, then applied to discover tactics in some combat simulation.

Impacts
This research will develop new capabilities that will enhance the effectiveness of simulation technology in critical applications like Simulation Based Acquisition and Joint Experimentation. If successful, these developments will also advance the state of the art in machine learning and produce several refereed publications.

Capturing Behavioral Influences in Synthetic C2

Douglas Flournoy, Principal Investigator

Bedford only

Problem
According to a recent National Research Council (NRC) study, "...an enormous gap exists between the current state of the art in human and organizational modeling technology on the one hand and the military needs on the other." A key area that needs further effort is the application of this technology to time-critical command and control decision-making processes.

Objectives
This project will investigate the potential for emerging human behavioral representation (HBR) technologies to deliver better C2 by strengthening analysis, decision aids, training, and experimentation. We will evaluate HBR modeling frameworks for their applicability to C2, then develop and experiment with a behavioral model of a Joint Surveillance Target Attack Radar System (JSTARS) operator.

Activities
The project will configure a testbed facility and install/evaluate battle simulations and HBR tools in the test-bed facility. We will develop a JSTARS Surrogate Human (JOSH); perform cognitive task analysis of a representative human operator; select HBR tool(s) and "draft" JOSH; perform iterative experimentation in which we compare JOSH's behavior to that of a human operator and adjust JOSH accordingly; and assess the strengths/limitations of JOSH.

Impacts
This effort improves the representation of human behavior in simulated C2. This facilitates more effective C2 analyses and decision-aid applications. Projects of this sort also build a better understanding of operator decision performance and strengths/limitations. The development of software surrogates that can perform C2 operator tasks extends/enhances man-in-the-loop experimentation efforts and eases staffing requirements for large-scale battlestaff training exercises.

Computational Embedded Training Strategies

Brant Cheikes, Principal Investigator

Bedford and Washington

Problem
Today's military and intelligence organizations depend on complex software applications to support all aspects of their operations. Training a cadre of software operators and maintaining their readiness in the face of frequently changing assignments, rapid deployments, and personnel turnover continues to be a critical task. Effective training requires periodic hands-on practice sessions focused on relevant mission functions.

Objectives
The general objective of this research is to apply artificial intelligence methods to the development of embedded tutors. Our specific objective is to develop a dialogue-oriented instructional agent that pursues explicit instructional strategies aimed at teaching individual users how to operate a complex software application in support of mission objectives.

Activities
We will initially implement a training agent that can assign an exercise and provide simple within-exercise coaching guided by an overarching instructional strategy. The agent will be enhanced to include planning a small curriculum, modeling and adapting to student behavior, and exhibiting a range of strategy-driven coaching and feedback. We will evaluate the agent's instructional effectiveness in a realistic training context.

Impacts
Our embedded training agent will be able to provide intelligent instruction driven by explicit instructional strategies. Through the design process, we are learning how to construct intelligent software applications to support education and training. Our evaluation studies should indicate whether such computer-assisted learning technologies represent a promising advance over conventional methods.

Diagnostic and Analysis Tools for Agent-based Combat Simulation Models

Garry Jacyna, Principal Investigator

Washington only

Problem
An emerging area of interest is agent-based simulation, where the process unfolds in a highly unpredictable manner, i.e., small changes in one part of the battle space produce profound effects in another. It is extremely difficult to quantitatively evaluate the outcomes or, for that matter, to determine the relevance or novelty of what we see in simulation. Stated more simplistically, where do we look for interesting behavior that can be exploited to understand the process of warfare?

Objectives
The intent of this project is to focus on developing complexity-based analysis tools that address three key problems in agent-based simulation: data interpretation, "novelty" filtering, and statistical characterization of outcomes.

Activities
The first phase addresses the development of novelty detectors using multifractal techniques and a GUI-based complexity toolbox for the rapid analysis and editing of combat simulations. The second phase examines the statistical characterization of the combat surface using multifractal methods.

Impacts
The major impact of this work would be to assist the Marine Corps officer in developing realistic and useful warfighting principles and doctrine. Much of the technical development effort involves creating ways to test various hypotheses, as well as interpreting the prodigious amount of data generated through simulation. Moreover, it is often difficult to determine where to look for interesting behavior and how to diagnose what to do in a particular situation. Our project would have an important impact in these areas. It would also impact issues related to the statistical distillation of data suitable as input to more traditional simulation-based models.

Distance Learning with Intelligent Agents

Brad Goodman, Principal Investigator

Bedford and Washington

Problem
Classroom learning improves significantly when students participate in learning activities in small groups of peers. As the U.S. military moves from schoolhouse instruction to Web-based distance learning, students risk losing this important opportunity to collaborate with other students. Adding conventional groupware tools, such as chat and email, is a start, but these tools do not necessarily remove the deficiencies.

Objectives
This project will develop and insert a learning agent into a collaborative distance-learning environment to promote interaction amongst students and help warriors become better thinkers. Collaboration tools allow multiple students to participate together from a distance, but they cannot guarantee quality interaction. We will develop a learning agent capable of acting as a peer with the students to enhance learning.

Activities
A learning agent will be developed that plays different instructional roles. The agent will observe and manipulate the environment, as well as communicate directly with students. Research in multi-agent planning and studies on paradigms for instructional support in collaborative learning groups will be conducted to determine the proper roles of learning agents. Finally, empirical evaluations of the learning agent will be performed.

Impacts
The proposed research will provide a new and more effective foundation for the Web-based distance learning programs underway in the military. Our intelligent system and collaborative learning research has already spawned a new Army program in companion-based learning and has been applied to a number of research prototypes.

En Route Airspace Modeling

Michael J. White, Principal Investigator

Washington only

Problem
Present models of en route airspace used in MITRE and other simulation models do not accurately reflect the operation of airspace in the real world. Also, they are not sensitive to many of the changes brought about by new technologies and procedures under consideration. This severely limits the utility of existing models, and the accuracy, precision and completeness of simulation studies.

Objectives
This project will develop a more complete, accurate and comprehensive method of representing en route airspace in simulation models. The new model will more accurately represent the functions of real airspace, and will be sensitive to changes in factors influenced by new technologies and procedures under study now and in the future.

Activities
We will research existing en route models in MITRE and elsewhere in the industry, develop new en route models with the desired properties, and experiment with and demonstrate these models in prototypes created in a fast prototyping language. Individual sectors and areas of about six sectors will be prototyped. We will select and demonstrate the preferred model to the MITRE modeling community.

Impacts
The new en route airspace model will enable simulation of the impacts of new technologies and procedures much more comprehensively than existing models permit. It may permit enhancement of existing models, or support the creation of new models. The prototype models created this year may themselves be useful for certain types of analysis.

Realistic Schedules for Future Air Traffic

Lisa Schaefer, Principal Investigator

Washington only

Problem
MITRE does not have a tool that creates future aircraft schedules that account for new kinds of demand and new air travel paradigms, such as those that would exploit small low-priced jets to give better service to passengers. In addition, the current tool for generating future traffic does not give aviation analysts the flexibility they need to examine the sensitivity of their results to different assumptions about how air traffic will grow in the future.

Objectives
This project will develop algorithms that create several schedules that represent a reasonable range of characteristics of future demand.

Activities
The activities will concentrate on variations of determining flight times, origin-destination pairs, and flight-leg linking strategies. The final product will create schedules with ranges of demand scenarios for the NAS Operational Evolution Plan studies and scenarios that represent changes in fleet mix and origin-destination strategies.

Impacts
MITRE will be better able to model a variety of air traffic demand scenarios to help assess a broader range of future scenarios. These scenarios would include changes to the current distribution of demand by time of day, new air transportation hubs, and more realistic itineraries linking successive flight of each aircraft. In addition, it would help MITRE to model new types of air travel paradigms, such as more flexible service to passengers facilitated by small, low-priced jets.

TFM Post-Event Analysis

Leonard A. Wojcik, Principal Investigator

Washington only

Problem
At present, decision-making by the FAA at the Air Traffic Control System Command Center (ATCSCC) does not adequately account for information uncertainty, especially regarding weather impact predictions. Often, decisions are based on what seemed to work or did not work in recent operational experience, rather than a probabilistic understanding of weather predictions. Of particular concern are relatively low-probability severe weather events that are predicted hours in the future.

Objectives
The initial objective is to develop a model to aid TFM post-event analysis with uncertain weather forecasts, based heavily upon actual experience with weather events in the past. The perspective of decision analysis will be taken to account for uncertain information, although formal decision analysis (which includes utility assessment) will not be undertaken. The final objective is to change decision-making policies to reflect information uncertainty.

Activities
The project will define the decision analysis perspective and apply it to past TFM decision-making events. We will develop a conceptual model accounting for information uncertainty in TFM decision making for the selected scenario type, populate a prototype model for post-event analysis with data from actual TFM events, complete an initial prototype model with expert opinion, and define requirements, if any, for simulation modeling of TFM events for post-event analysis for the selected scenario type. We will test the prototype model against an initial set of TFM events, and generate a set of requirements for the post-event analysis tool based on experience with the prototype tool.

Impacts
If successful, the tool has the potential to fundamentally change decision-making strategy at the ATCSCC, and to benefit the flying public during aviation schedule disruptions. We also expect to present an early application of the decision analysis approach to TFM at a major Air Traffic Management R&D symposium.