The availability of intelligent adversaries in a training simulator environment can clearly enhance the training experience for students. However, implementation of this capability into simulators has been slow as well as difficult. The semi-automated forces presently available for SIMNET, although quite sophisticated, still represents a partial solution, as the name itself indicates.

Representation of tactical expertise in rules gives rise to the problem of encapsulating every possible scenario within simple rules. This could lead to the need for a very large number of rules which, not only would have to be developed, but would also have to be efficiently executed in a real-time environment. This represents an unacceptable situation.

Improvements could be made by grouping rules according to the mission being simulated, but the number of rules required would still be large, and there would be no benefit of reusing situational knowledge commonly required in different missions.

The approach described in this paper is to develop a hierarchical ordering of rules which, at the highest levels, can be used to recognize the general situation being faced by the adversary. Examples of these situations are when an adversary needs to remain hidden from the student, or when it is appropriate to attack, the student. Recognition of this high level situation will activate a lower level set of rules which will attempt to implement the prescribed course of action within the context of the situation. These will, in turn, activate another set of rules which will carry out the low level implementation details of the action within the simulation software.