Military Operations Planning in SIPE-2


As part of the ARPA/Rome Laboratory Planning Initiative (ARPI), SIPE-2 was applied to joint military operations planning. It successfully generated employment plans and expanded deployment plans for getting the relevant combat forces, supporting forces, and their equipment and supplies to their destinations in time for the successful completion of their mission. Input to the system includes threat assessments, terrain analysis, apportioned forces, transport capabilities, planning goals, key assumptions, and operational constraints. Most of the above came from miltary databases; around 100 plan operators in SIPE-2's input language were developed to describe military operations.


This application of SIPE-2 was done as part of SOCAP -- System for Operations Crisis Action Planning, and was the core of the second Integrated Feasibility Demonstration of the ARPI.

SOCAP includes an application of SIPE-2 to military operations planning together with a user interface tailored to this domain using a situation-map display system. The following inputs are fed into the Socap database from available military databases:

Other inputs come from the mission commander or his/her joint staff: Most of the above information is inherently dynamic, and is best represented in SIPE-2 as first-order predicates. There are over 2100 predicate statements in the Socap knowledge base. However, a great deal of the available data are static (i.e., they do not change over time as actions are executed), and for efficiency reasons are represented in the sort hierarchy and not as predicate statements.

A large set of SIPE-2 operators describes describe military operations that can achieve specific employment or deployment goals. There are operators at every level of abstraction. SIPE-2's levels of abstraction naturally map onto the following levels of the operations planning process:


SOCAP successfully demonstrated that AI planning techniques can be used for the generation of large-scale military operations plans. It provides the first steps toward an operational prototype. SIPE-2 supported efficient plan generation for the scenario used. The plans generated have approximately 200 executable actions in the final plan. (The entire plan has many times this number of nodes, since some nodes are conditions which help record the rationale behind the plan and information for repairing it.) The sort hierarchy has around 250 classes and 500 objects with 15-20 properties per object. The initial world contains approximately 2100 predicate instances. These plans are generated in 1 to 2 minutes on a Sun Sparcstation 20. For a large-scale military operation, such as in the Gulf crisis, the size of both the plans and knowledge bases would probably be at least an order of magnitude larger.

The effectiveness and efficiency stemmed from SIPE-2's hierarchical planning process (which naturally fit the hierarchical structure of the domain), extensive use of its sort hierarchy for encoding military databases, the ability to place constraints on planning variables, its powerful graphical interface for viewing plans and data, and the interactive planning capabilities.


D. E. Wilkins and R. V. Desimone, "Applying an AI planner to military operations planning," in Intelligent Scheduling (M. Fox and M. Zweben, eds.), pp. 685--709, Morgan Kaufmann Publishers Inc., San Mateo, CA, 1994.


This application was funded by the ARPA/Rome Laboratory Planning Initiative and done by a team lead by Dr. Marie Bienkowski.

Dr. David E. Wilkins
Last modified: Mar 7 16:19:09 2007