Soar IFOR Project

The Soar Intelligent FORces project is investigating the application of Soar to the modeling of real-time agent behavior. The ultimate intent behind this effort is to develop automated pilots whose behavior in simulated battlefields is nearly indistinguishable from that of human pilots (and to go beyond this to develop generic agents that are readily specializable for this and other domains). If such agents can be created, they should provide close to ideal force supplements for many of the applications anticipated for distributed interactive battlefield simulation.

As of Dec 1994, prototype agents have been constructed that can, in real-time, interact with humans in several realistic Beyond Visual Range tactical air scenarios. These agents can fly solely or within a section against another single aircraft or section. Agent behaviors include the ability to patrol in a racetrack pattern; select radar modes, detect opponents on radar, perform search and acquire activities when opponents drop off of radar, and maneuver so as to confuse the opponent's search and acquire activities; determine and attempt to achieve appropriate intercept geometries and launch-acceptability regions (LARs); select, fire, and support missiles; and detect and evade enemy missiles.

These prototype pilot agents have participated in a variety of demonstration engagements against expert human pilots. The most recent demonstration was part of a much larger engagement called Simulated Theater of War, Europe (or STOW-E), which was held November 4-7, 1994. STOW-E was an operational (and international) military exercise -- in fact the largest distributed simulation exercise to date -- involving approximately 2000 entities originating from nineteen sites across the United States and Europe. These entities represented everything from aircraft, ships, and ground vehicles to individual soldiers and missiles. Some of these entities represented manned simulators. However, most of the entities corresponded to computer generated forces. Our role in STOW-E was to provide pilot agents for a range of air-to-air, air-to-ground and air-to-surface missions. This was a real learning experience for us in terms of understanding what happens in such complex situations, both in terms of the stress it puts on the low-level technology as well as the many surprises it provides for the agents (such as their being shot down by surface-to-air missile batteries that we had no idea existed, or the unexpected forms of behavior exhibited by unconstrained human pilots). Nonetheless we were able to successfully participate in a range of missions across all four days of the exercise, and on more than one occasion shot down human pilots flying in simulators. Our understanding is that this makes TacAir-Soar the first AI system to have participated directly in an operational military exercise.

If you would like to find out more about the Soar IFOR project there are numerous publications and project documents available.


The Soar IFOR effort is a collaborative effort between researchers at the Carnegie Mellon University School of Computer Science, the University of Michigan's Artificial Intelligence Laboratory, and the University of Southern California's Information Sciences Institute. Project personnel at USC/ISI include: This research is supported under contract N00014-92-K-2015 from the Advanced Systems Technology Office of the Advanced Research Projects Agency (Peggy Feldmann, Program Manager) and the Naval Research Laboratory. The simulation system used in this project is ModSAF, a product of Loral Advanced Distributed Simulation (LADS).
Milind Tambe
Karl B. Schwamb