(SAF); computer-controlled hostiles, intelligent forces (IFOR); and command forces (CFOR) models. Both IFOR and CFOR models have been developed in the context of the synthetic theater of war (STOW), an advanced-concept technology demonstration jointly sponsored by the U.S. Atlantic Command and DARPA. A key element of STOW is the representation of both fighting forces and their commanders in software.
ModSAF is an open architecture for modeling semiautomated forces, developed by the Army's Simulation, Training, and Instrumentation Command (STRICOM). It provides a set of software modules for constructing advanced distributed simulation and computer-generated force applications. Semiautomated forces are virtual simulations of multiple objects that are under the supervisory control of a single operator. According to Downes-Martin (1995), the human behaviors represented in ModSAF include move, shoot, sense, communicate, tactics, and situation awareness. The authoritative sources of these behaviors are subject matter experts and doctrine provided by the Army Training and Doctrine Command (TRADOC). Task-based explicit behaviors are enumerated by a finite-state machine that represents all the behavior and functionality of a process for a limited number of states. The finite-state machine includes a list of states, a list of commands that can be accepted while in each state, a list of actions for each command, and a list of state conditions required for an action to be triggered.
In ModSAF there is no underlying model of human behavior, so that any behavior representation must be coded into the finite-state machine. As a result, it is impractical to use ModSAF to construct general-purpose behavioral or learning models. Typically, ModSAF models are employed to represent individual soldiers or vehicles and their coordination into orderly-moving squads and platoons, but their tactical actions as units are planned and executed by a human controller. Ceranowicz (1994) states that because the human agents in ModSAF are not intelligent enough to respond to commands, it is necessary for the human in command to know everything about the units under his/her control. This would not be the case in a real-world situation. ModSAF supports the building of models of simple behaviors at the company level and below.
The initial ModSAF (developed by the Army) has been adopted by the other services—Navy, Marine Corps, and Air Force. Each of these services now has its own ModSAF source and capability tailored to its own needs. In a recent STOW exercise, four types of ModSAFs were used. The scenarios for this exercise were preset and the role of the human controller minimized.