Virtual Combat Training Center
A simulation without a tutor is like a CTC without an O/C.
Realistic tactical simulations are beginning to proliferate, but the developers focus on simulation fidelity not on training. If used in a training context, the simulations are employed as a substitute for the live 'sandbox'. Basic data is collected and diagnostic evaluation is conducted by human observers. On the other hand, the training community has been developing intelligent tutoring systems (ITS) that can perform the key functions of a live tutor/coach. An excellent opportunity exists to couple an intelligent tutor with realistic simulations to provide supplemental training to the CTC experience.
The tutor is hooked into an existing simulation, using software connectors, rather than extensively modifying simulations or building new ones from scratch. This approach will be demonstrated with an existing high-fidelity tactical simulation of combined-arms warfare at the battalion and company level (Armored Task Force). The tutor system is a modular to allow re-use of key components for other applications. In particular, the separation of tutor and simulation, the use of software connectors, and a software architecture approach where the tutor knowledge bases are built from reusable ontologies should all enhance portability, extensibility, and reusability.
Due to these limited resources, it is critical that the training experiences at these CTCs be optimized. The benefits of these live training experiences can be extended by providing low-cost, readily available, realistic, and relevant PC-based training prior to CTC rotations to better use the time there, and subsequent to rotations, to enhance retention and allow for in-unit follow-on training that builds on what has just been learned. This supplemental training also allows much greater time on task, allowing trainees to spend more time developing a wider range and more in-depth tactical skills and increased automaticity in applying those skills.
Simulations and games that apply to Defense needs are beginning to proliferate, some from the DoD and from entertainment. These developers are usually domain experts who do not have the interest or resources to add significant training (if any) to their simulations. If the simulations are used in a training context, they are typically employed in a similar manner as the live training centers. Basic aggregate data (e.g., number of kills) is collected and diagnostic evaluation is conducted by human observers -- a reduction in equipment cost of CTCs, but still requiring observer personnel. High-end workstation simulations such as JANUS, may also require additional personnel to operate the interfaces for trainees.
A segment of the technical training community focuses on development of intelligent tutoring systems (ITS) that can perform the key functions of a live tutor/coach. These tutoring systems tend to be handcrafted for each application and are expensive and time consuming to build when so developed. The members of this community are not ususally subject matter experts in military subject matter so any simulations developed by them tend to be rather low-fidelity by comparison to those developed by military subject matter experts.
We see an excellent opportunity to couple intelligent tutoring systems technology with existing subject-matter developed high-fidelity simulations to provide supplemental training to the CTC experience.
This approach will be demonstrated with an existing high-fidelity tactical simulation of combined-arms warfare at the battalion and company level, called Armored Task Force (ATF). The predecessor to ATF, Brigade Combat Team (BCT), has been used at the Joint Readiness Training Center (JRTC) Leader Training Program (LTP). The tutoring system will be developed so that the main components (student model, domain knowledge, and tutor strategies) are reusable for other simulations. The tutor component is also intentionally designed as a separate component to existing simulations to promote its reusability.
Tutoring System. Most computer-based tutoring systems build student models based on recognition-based exercises, such as multiple-choice, drag-and-drop, and fill-in-the-blank exercises. While these types of measures are easily collected, they do not provide a full assessment of student state. Not only is it important to have knowledge about a domain, but one must also be able to apply those skills necessary to perform the tasks and do so with confidence. This tutoring system will build a skills and knowledge student model within the context of realistic scenarios and simulations. The V-CTC builds a skills and knowledge student model within the context of realistic scenarios, such as CTC exercises, and in the context of task performance in these high-fidelity simulations.
The tutor provides a real-time assessment of student state that is richer than current approaches. It includes performance-based measures of actions and choices during a realistic simulation, as well as knowledge-based measures of student plans, perceptions (e.g., of enemy threat), and explanations for actions taken or not taken. A dialog capability between the trainee and simulated instructor also provides a rich source of user modeling information, in addition to allowing the trainee to directly ask and answer questions in a natural way. The data collected includes latency and self-assessment measures that provide information for a model of confidence. A Bayesian analysis takes these various measures and forms a student state model consisting of knowledge, skills, and confidence.
Evaluation of student actions in the simulation is performed by deductive reasoning. This knowledge-based reasoning is supported by a domain knowledge representation which is a domain specific knowledge base built over domain-specific ontologies ultimately backed by a standard upper ontology. The ontologies provide a high-level organization of the knowledge, and furthers the user model's extensibility and reusability. The domain-specific knowledge base provide the tutor with an expert level of active knowledge of domain concepts and rules and solutions. The student state model is continually updated, and influences the tutor's strategies so that it customizes interactions to the individual student.
The tutor system will be hooked into an existing high-fidelity tactical simulation, using software connectors, rather than either extensively modifying simulations or building new ones from scratch. These hooks into the simulations allow the tutor to control basic operations of the simulation, such as starting, freezing, or replaying particular scenarios at specified points in time. They will allow the tutor opportunities to gather input from the student and provide feedback and explanations.
The tutoring system will be designed with a generic, modular architecture to enable re-use of the key components for other applications. Re-usable components include the student state model, tutoring strategies, domain knowledge representation, and the methodology for connecting the tutor and simulation. We also build the tutor separate from the simulations to allow reuse of each in multiple applications.
Simulation. The Virtual Combat Training Center concept will be demonstrated with a high-fidelity tactical simulation of combined-arms warfare at the battalion and company level, called Armored Task Force (ATF).
ATF is the recently released successor to a previous simulation called Brigade Combat Team (BCT). BCT was an innovation in that it provided most of the fidelity of JANUS (a simulation used extensively at the Command and General Staff College) but eliminated the need for high-end workstations or controllers to interpret commands. BCT included detailed scenarios from NTC training rotations, and combat situations from the first Gulf War to hypothetical engagements in Kuwait, North Korea, and Cuba. Brigade Combat Team has been used for training at the Joint Readiness Training Center (JRTC), Leader Training Program. ATF and BCT were developed by an active duty Army artillery officer, who is an observer/controller at NTC, and was previously stationed at JRTC.
The ATF game pits a friendly force of up to battalion/company size against an enemy force of up to brigade/regimental size against each other in simulated combat. ATF allows a user to take the role of the friendly forces while it plays the opposing force (OPFOR). It randomly selects from multiple enemy course of actions (COAs) stored with each scenario. The user manipulates NATO-standard icons that represent companies, platoons, or sections. Commands can be given from the company-level on down to the platoon-level and specify paths and orders for individual vehicles. Just as in modern land warfare, the user fights with and against units consisting of a wide variety of assets. These include armor, infantry, artillery, engineers, air defense, and aircraft. These units must be synchronized and massed at the key point on the battlefield to win. The cybernetic battlefield is a digitized elevation map of actual terrain and uses UTM coordinates. Actual National Training Center (NTC) maps (e.g., of Crash Hill) are used in the NTC scenarios.
ATF includes scenarios from National Training Center, the Fulda Gap in Europe, the first Gulf War and a hypothetical second Gulf War. It improves on BCT by providing a better simulation of military command since missions can now be assigned to companies and platoons and they will carry out their orders independently. ATF also includes more accurate vehicle and turret modeling, better modeling of weather effects, the addition of civilians, the addition of vehicle smoke capabilities, and an improved user interface and graphics. The maps are not hexes, but continuous terrain features including trees, buildings, and roads in contour-map representations.
BCT and ATF are real-time simulations (1X, 2X, 4X, or 8X of battle real-time) of combined arms warfare. Note that this real-time aspect is very important in helping trainees acquire an intuitive feel of how fast the battlefield changes and in learning how to synchronize different battle operating systems such as artillery and armor.
Operational concept. The Virtual Combat Training Center (V-CTC) could be used for individual development in the unit prior/after CTC rotations. The tutoring component emulates an Observer/Controller (O/C) at NTC. The virtual coach 'pops-up' and points out poor tactical decisions as they are being made, teaching Army doctrine at that time. Later, in the After-Action Reviews, the virtual coach summarizes what the commander did wrong and what he should have done.
The V-CTC and simulation can be used in the classroom to illustrate tactical concepts. It can be used to train commanders in different echelons and for networked team training for different roles such as FSO, S-2, and S-3.
Intelligent tutoring systems (ITS) coupled with high fidelity simulations can provide this supplemental training. ITS provide individualized instruction with the potential 2 sigma (standard deviation) improvement that good human tutors can accomplish.
We expect that the proposed Virtual Combat Training Center will provide the following advantages and benefits:
Improved quality of training
Reduced cost of training
Low-cost training available anywhere and anytime
A high-fidelity simulated task environment coupled with an intelligent training system with a rich student state model is a vast improvement over current computer-based training programs. This new approach can transform military training by providing continuously available, on-demand mission-level training for all forces at all echelons.