Site Map

TOH Neural Network

Situated Agents

Information Retrieval

Expert System

AI Backgammon

Technology Transfer Between Government Applications of Computer Generated Agents and Commercial Entertainment

Rich Warren

GreyStone Technology, Inc.
10766 Goldentop Road
San Diego CA 92127

1. Abstract

The transfer of Computer Generated Forces (CGF) technology between government simulation and commercial entertainment communities, facilitates the development of more evolved and cost effective Autonomous Intelligent Adversaries (AIA). As commercial AIA requirements begin to also meet government CGF requirements, breakthroughs in intelligent adversary technology are incorporated into Commercial Off The Shelf (COTS) and Value Added software. The commercial reusable software tools are in turn made available to government CGF managers who realize immediate reductions in development costs and program maintainability.

This paper will describe early applications of CGF technology to both government simulation and commercial gaming environments. More recent applications of the technology will also be discussed to show that the fidelity requirements of AIA in simulation and gaming are, by todayís standards, nearly identical. Motivation to reduce the development, acquisition and operations cost of CGF and AIA software tools that increase the fidelity, performance and portability of behavior models is also offered.

2. Common Vision

The Distributed Interactive Simulation (DIS) glossary defines CGF and Semi-automated Forces (SAFOR) as the ìSimulation of friendly, enemy and neutral entities on the virtual battlefield in which the individual platforms are operated by computer simulation of the crew and command hierarchy.î The Virtual or Electronic Battlefield is likewise defined as the ìIllusion resulting from simulating the actual battlefield (IST, 1994).î

The application of CGF technology in government military Test, Training and Analysis exercises satisfies the governmentís need to reduce cost and logistics support while maintaining the density, depth and diversity of forces required to accomplish the exercise objectives. Though the human or live forces in the exercise remain the focal point, the use of CGF provides an economic solution that stimulates interactions between ìplayersî on the virtual battlefield.

The commercial entertainment industry, like the military, has similar needs for an economic solution that stimulates live (i.e., cash paying) customers. The profit for commercial entertainment is derived from enticing the customer to participate in and repeatedly return to the gaming environments and location based experiences. Commercial Virtual Reality opportunities are growing through the application of technology that offers a solution.

The DIS glossary defines VR as the ìeffect created by generating an environment that does not exist in the real world. Usually, a stereoscopic display and computer-generated three-dimensional environment giving the immersion effect. The environment is interactive, allowing the participant to look and navigate about the environment, enhancing the immersion effect. Virtual environment and virtual world are synonyms for virtual reality (IST, 1994).î

Notice that a ìvirtual battlefieldî is one representation or application of ìvirtual reality.î GreyStone Technologyís commercial virtual reality entertainment systems combine multi-sensory human-computer interfaces with real-time simulations and dynamic models that display intelligent and interactive behaviors

As outlined above, government simulation and commercial entertainment managers now share a common goal to reduce the cost of development, acquisition and operation of CGF technology. In the sections that follow, early applications of CGF technology to both government simulation and commercial gaming environments will be presented to show that CGF requirements across the two environments were at one time distinct. More recent applications of CGF technology, however, will also be discussed to show that the fidelity requirements of Intelligent Adversaries in simulation and gaming are by todayís standards nearly identical. Closing sections of this paper will offer motivation to reduce the development, acquisition and operations cost of CGF and AIA software tools that increase the fidelity, performance and portability of behavior models.

3. Early Applications of CGF Technology

Until recently, user requirements across government simulation and commercial gaming environments have placed differing emphasis on the fidelity of the CGF and AIA. While the government simulation environments required high fidelity CGF , the commercial gaming environments required high fidelity presentation systems. In the sections that follow, several exemplar applications of early work from both government simulation and commercial gaming will be presented to show that computational resources were not sufficient to simultaneously host both CGF and display technology.

3.1 Early Government Simulation

The government has shown an interest in modeling adversarial forces since WWII. Much of modern CGF technology can in fact be traced back to the work of John Von Neumann and his theories of game play (Von Neumann, 1944). This section will trace early work in game theory and Operations Research to motivate the discussion of applications that show governmentís early emphasis on CGF.

Adversarial Agent Modeling and Computer Generated Vehicle Commanders are applications that are also described to show an early government emphasis on CGF technology rather than on presentation technology. Following, GreyStone Technologyís Advanced Maneuvering Logic - 90 (AML-90) (GreyStone, 1994) will be presented to illustrate a government interest in CGF technology hosted in a computational environment sufficient enough to also provide two dimensional bitmapped graphics.

3.1.1 Operations Research

Operations Research is an activity with a long history that dates back to World War II. Methods of Operations Research, including statistical analysis, theory of probability and gaming theory, have been applied to tactical analysis and operational experiments with equipment and procedure for over half a century. (Morse and Kimball, 1951).

Tactical analysis became necessary as the onset of WWII introduced many new tactics and equipment types for which effective measure-counter measures were needed. A counter measure to minimize the threat of the Japanese Kamikaze, for example, was found through statistical solutions that considered damage due to suicide planes, the effects of maneuvering and the effect of angle of approach. The results of the Kamikaze study produced a number of suggested tactics which resemble, in content, the consequent of modern day expert system rules. Below is a summary table of some of the tactics learned through the statistical analysis of suicide planes.

Rule NoTactic Learned
1All ships should attempt to present their beams to high-diving planes.
2All ships should attempt to turn their beams away from low-diving planes.
3Battleships, cruisers, and carriers should employ radical changes of course
4Destroyers and smaller fleet units should turn slowly to present the proper aspect to the diving plane.
5Destroyers and smaller fleet units should not turn rapidly enough to affect the accuracy of their AA.

Table 1: Tactics Learned

In addition to statistical analysis, search and game theories were developed to provide more analytical solutions to tactical analysis. Search theories, for example, can state the probability of making contact with a target placed at random within some given area. The probability of hit (Pk) is likewise computed using statistical theory.

Game theories were also developed as problem solving methodologies to tactical analysis. Specifically, the analysis of countermeasure action is accomplished using principals established by Von Neumann. These principals are particularly effective under situations where battlefield intelligence is complete and the opposing forces are reasonably familiar with measures and countermeasures that apply to the tactical situation. The driving principal under such situations, know as the minmax principal, works to maximize gain while minimizing loss.

3.1.2 Enemy Platforms

Much of the Operations Research described above was conducted at the very dawn of the computer revolution. Since then, a number of research efforts have contributed towards the development of computerized tactical decision aids which incorporate many of the principals and strategies developed through Operations Research.

The Naval Air Development Center, for example, has sponsored research efforts to model plan recognition agents that operate within adversarial domains. For program development, verification and validation purposes, Computer Adversarial Agents that model enemy platforms (e.g. aircraft and ships) are generated. These computer adversaries pose a threat to Naval aircraft carrier Task Forces and are capable of interaction in a dynamically changing world.

The intelligently guided operators of Azarewiczís plan recognition systems, project plan hypothesis forward in time much like implementations of the minmax game principal. Due to the dynamics of the battlefield, however, Azarewiczís models use differential gaming principals that better model domains where joint moves by both opposing forces might simultaneously occur (Azarewicz, 1987).

3.1.3 Combat Commanders

The preceding section introduced the application of game theory to computer generated autonomous adversaries. This section will introduce the use of expert system technology as it is applied to model a combat vehicle commander.

Gibson describes an expert system used to model a combat vehicle commanderís thought or combat decision-making process (Gibson, 1989). Additionally, Gibsons system applies MYCIN certainty factor methodology to model uncertainty common to many battlefield situations.

3.1.4 Adaptive Maneuvering Logic - 90

While the previous section introduced an expert system based vehicle commander, this section will describe a fully autonomous rule based air combat adversary that GreyStone Technology has commercialized.

Adaptive Maneuvering Logic - 90 (AML-90) is an advanced, synthetic adversary control model that allows for real-time, interactive air combat with a six degree-of-freedom air combat simulation for one-versus-one and two-versus-one engagements (GreyStone, 1994). The decision-making process is implemented using a rule-based system that contains a set of air combat rules and associated target behavior modes that consider multiple phases of a fighter combat mission including: Beyond Visual Range (BVR), Intercept, Close-in-Combat (CIC), and Bugout. The adversaries execute in real-time to provide realistic air target simulation for air engagements.

The GreyStone Adaptive Maneuvering Logic - 90 (AML-90) software provides several user selectable aerodynamic models of fighter aircraft platform and allows for 4 computer generated pairs (ownship and wingman), 8 aircraft total, to be simulated simultaneously during a session. The AML-90 pairs can be controlled as adversarial forces within a simulation exercise and may be directed by the user to engage other aircraft entities in either a 1-v-1 or a 2-v-1 engagement.

The Relative Reference Display (RRD) allows the user to control and monitor the AML-90 simulation environment. The RRD provides a two dimensional gods-eye-view of both the simulated arena and the CGF. A Graphical User Interface (GUI) is provided by the RRD which allows the user to set-up the initial positions of AML-90 entities and to establish routes of flight. The GUI also allows the user to specifically control the targets that the AML aircraft will intercept or engage.

3.2 Early Entertainment Environments

The previous section (3.1) offers examples of early applications of CGF technology to government simulation and decision aiding environments. The early government applications clearly demonstrate that limited computer processing power forced a development emphasis on high fidelity Intelligent Adversaries. At the same time, however, commercial gaming applications were developed with an emphasis on high fidelity presentation because the limited computer processing power allowed for only rudimentary or brute force computer adversaries in the gaming environments.

The following sections offer examples of early applications of CGF technology to commercial gaming environments. GreyStoneís Purple Heart Corner is exemplar of the presentation fidelity common to high end adversarial gaming environments. GreyStoneís Pteranodon Experience is also an early example of the high fidelity presentation system common to many commercial gaming environments.

3.2.1 Purple Heart Corner

Purple Heart Corner is a commercially available entertainment game that combines state-of-the-art in virtual reality computer graphics with a detailed mock-up of a WWII bomber gun station. The shell of the gaming simulator closely represents the interior of the B-17 bomber at the waist gunner position, complete with stringers and bulkhead rings. An accurately-sized window opening in the fuselage side holds a copy of the 50 caliber Browning machine gun, and is flanked by a standard issue ammunition box (GreyStone, 1992).

The replica 50-cal Browning machine gun faithfully reproduces the heft and feel of the original gun, while a built-in pneumatic actuator recreates the hammering recoil of the big weapon. The sights are also accurately replicated, allowing the user to glimpse the skill required to aim the Browning accurately in the heat of the battle.

A description of the action of air combat experience shows that a heavier emphasis is placed on presentation rather than on the intelligence of the computer adversaries.

Me109ís and FW190ís, silhouetted against the sky, drop in from the south west, ahead of the formation. The relentless drone of the engines is interrupted, first by the top turret gunner, as he sends orange-red tracers out to greet the incoming fighters. Wave after wave of fighters dive through the formation. Tracers arc across the smoky sky. The 50-caliber Browning clatters on its mount as the fighters loom in the sights, with muzzles flashing. A hit! White flames claw the fighter to pieces as it spins downward, raining embers and trailing smoke. No time to exult; you turn to face the next attacker.

The above excerpt illustrates that a heavier emphasis is placed on graphics and sound technology. Though the experience provides computer generated targets as well as adversaries, they are controlled using scripted programming techniques.

3.2.2 Pteranodon

The Pteranodon experience, first developed as a showcase for Silicon Graphicsí powerful Onyx image generator, represents the state of the art in premium virtual reality. The Pteranodon experience offers 180 degree visibility afforded by three large screens, and thousands of fully-textured, anti-aliased polygons refreshing the screens at thirty times a second. The detailed, colorful textures and realistic movement of objects in the simulation are complemented by the rich, booming, natural sounds of the environment (Crowe, B., 1994).

Although the Pteranodon is programmed to follow the commands of a rider, an intelligent obstacle avoidance system will execute a course deviation when necessary to avoid a collision with obstacles. The Pteranodon is also programmed to search for and follow other creatures with its gaze.

A description of the action of the Pteranodon experience shows that a heavier emphasis is placed on display and presentation.

Echoes of a primal screech rumble through the canyon, announcing the arrival of the raptor. Wings sweep the sky, mocking the winds. A rider guides the beast around cascades of water which plummet from dizzy heights to the river below ... vampire bats flutter above the next bend and monstrous wasps dive and swoop over a whirlpool that devours the river.

As the master of the Pteranodon, you guide it with the reins and by leaning in the saddle. It obeys your every command, but as you cruise through the canyons of this fantasy world, it skillfully avoids obstacles on its own.

The above excerpt should illustrate that a heavier emphasis is placed on graphics and sound and other presentation technology.

4. Recent Simulation and Gaming

The previous section (3.0) describes early applications of CGF Technology and shows that limited computational resources forced government and commercial CGF applications to place differing emphasis on the fidelity of the Intelligent Adversary. This section will demonstrate that increased processing power has made it possible to incorporate both high fidelity CGF and high fidelity virtual reality in todayís simulation and gaming environments.

4.1 Recent Government Simulation

A review of more recent application of CGF technology to government simulation environments will show a continued emphasis on high fidelity Intelligent Adversary technology but will also show a move towards the coupling of the technology with a high fidelity real-time virtual reality graphics environment.

The government simulation community has realized the full potential of todayís computer technology and has coupled Computer Generated Forces technology with high fidelity real-time Virtual Reality. The Naval Postgraduate Schoolís continued development of NPSNET, internationally known for its networked virtual environments technology, has incorporated Autonomous Players into their virtual battlefield. GreyStone Technologyís AML-D RAGE , like NPSNET, is a networked application that combines the latest in real-time VR technology with high fidelity Computer Generated Forces technology. Both the NPSNET and AML-D RAGE environments will be discussed in detail.

4.1.1 NPSNET Autonomous Players

The Navel Postgraduate School has included Autonomous Players in the NPSNET simulation environment to ìprovide interactive players when live players are not available or affordable (Zyda 1994)î. The NPS Intelligent behaviors are modeled using expert system rule based technology capable of commanding unmanned vehicles in the simulation environment.

Among the autonomous NPS players are the autonomous tank forces players that provide intelligent behavior models which have the capability to work cooperatively. If numerous entities, for example, approach the autonomous tank force from several directions, then the NPS autonomous computer generated force can distribute their fire such that some tanks fire one direction and others in another direction. Should the autonomous player become outnumbered, it has the additional capability to call for reinforcements.

The NPSNET autonomous players are also using elevation data to reason about terrain. If, for example, a forward observation vehicle has Line of Site (LOS) with an enemy vehicle, the Autonomous Player can relay the coordinates to one of several howitzers. The threat is fired upon if it is in range of the howitzer.

Additionally, the autonomous players are equipped with intelligence reports that provide them with knowledge of the battlefield. Given whether a vehicle is friendly or not, the autonomous players can prioritize targets so that those targets with a higher priority are fired upon before lower priority vehicles.

4.1.2 RAGE AML-D

GreyStoneís Real-time Advanced Graphics Environment (RAGE) coupled with AML-D is a showcase of both high fidelity graphics and intelligent adversaries running seemlessly together.

RAGE™ is a 3D visualization product designed for US and foreign government agencies and military services, and any members of the US or international defense industries who simulate operational scenarios, avionics and weapons systems, airframes, and mission planning/preview/rehearsal/training systems or conduct range operations for test or training, or manage C4I systems. It is particularly designed for organizations that have a need for advanced visualization but do not have the time, resources, or expertise to buy and build their own visualization products.

RAGE™ provides a 3D visualization component for avionics, weapons and aircraft system simulations, constructive and virtual mission simulations, mission preview, rehearsal and training systems/simulators, live training missions and actual combat missions. It can receive object (entity) state and event data from multiple intelligence and instrumentation sources and present a near real-time representation of live or simulated events at a 30Hz refresh rate. In addition to three standard viewpoint options: Stealth, Out-the-Window, and Tethered, RAGE™ features two simultaneous viewpoints (i.e., Stealth on one monitor, out-the-window on another), sensor-related control functions including scan, slave, and zoom, and 3D visualization of non-visible phenomena such as weapons envelopes, platform signatures, EW signals, sensor beams and scan volumes, and operational area boundaries. RAGE™ can interface with large multi-screen displays, single screen systems, standard system monitors, helmet mounted displays and mini-domes (Shillicutt, 1994).

Depending on user requirements, RAGE™ can be integrated with multiple simulations, various user input/output interfaces, and display alternatives. Specific models and environment renderings can be produced. Non-visible phenomena functionality can be made dynamic such that they respond to the physical criteria which influence their behavior. Examples include dynamic SAM envelopes based on target altitude, and velocity vector and radar detection volumes based on pulse repetition interval or radar cross section.

Distributed Interactive Simulation provides a specialized method for integrating simulations into your visualization environment. RAGE™ is certified to receive DIS entity state and event PDUs. If a user needs to add a 3D visualization product to any entity state source (constructive or virtual simulation, live range data or live combat data), entity state source can be translated into DIS protocols.

AML-D is a (DIS) compliant version of the Adaptive Maneuvering Logic - 90 software (detailed in section 3.1.4 and GreyStone ë93) that translates entity state data to DIS PDU which are forwarded to the RAGE application. This combination of high fidelity VR and CGF technology allows for interaction between dynamic AML-90 aircraft and large multi-player exercises (GreyStone, 1994).

4.2 Commercial Entertainment

The following sections will show that commercial entertainment has realized the full potential of todayís computer technology and has incorporated intelligent adversary technology and high fidelity real-time virtual reality technology in a single synthetic environment.

The ThunderBolt commercial gaming environment, developed at GreyStone for a state-of-the art amusement park ride, is a synthetic environment that satisfies many requirements also common to government simulation. Thunderbolt has successfully combined high fidelity adversary technology with state of art presentation technology.

4.2.1 The Thunderbolt Experience

The computer generated forces developed for the ThunderBolt application are designed to provide a constant level of action for the human players who participate in the gaming environment. The underlying goal is to keep a continuous flow of adversary aircraft (both target and threat) in the field of view of each of the human players.

The technologies used to develop the ThunderBolt intelligent adversary fundamental behaviors are based on the modeling techniques utilized within the AML-90 adversary software. Although the number of actual CGF players required for the experience is significant, the constraints of the ThunderBolt compute environment allowed a CGF design based on a derivative of AML-90 behavioral model.

Like AML-90 adversaries, the ThunderBolt CGF derive relative threat geometry and apply a set of logic in order to assess an appropriate action. The logic is both phase and goal-based in that geometrical parameters such as range and target aspect are determined. The four CGF phases are Intercept, Engage, CIC, and Bugout. The phase is used to determine the set of tactical logic to apply to the situation and ultimately determines the CGF's flight behavior and actions. While the AML-90 CGF includes a robust set of tactical logic, including cooperative logic with a wingman, the ThunderBolt CGF operate independently of one another.

5. Conclusion

GreyStone Technology believes that many of the needs of both government simulation and commercial entertainment communities can be satisfied through Virtual Environments technology. Furthermore, these virtual environments are combinations of Multi-Sensory Human-Computer interfaces with real-time simulations that are populated by dynamic, intelligent and interactive behavioral models (CGF). In the final solution, the distinction between government CGF and commercial Intelligent Adversaries, is defined by the userís needs and the personality or behavior of the application. The underlying software structures and technologies should be common and reusable. As the user determines the fidelity of both the adversaries and the interfaces needed, a compromise must be made on the requirements of costs and operational logistics.

Figure 1: VR Application Axioms

As a single solution for all applications will unlikely satisfy all users, we propose that a common foundation class of object oriented CGF libraries can be cost effectively shared across both commercial entertainment and government simulation applications. With the axioms shown in figure 1 above, the end user can determine the optimal operations point and the developer can determine which of the libraries are needed to ensure the requirements of a specific exercise or experience are met with optimal efficiency.

6. References

Azarewicz, J., et. al. (1987) Multi-Agent Plan Recognition in an Adversarial Domain, Third Annual Expert Systems in Government Proceedings, pages 188-193, Washington, DC, October, 1987.

Crowe, B., Pteranodon Sighting at SIGGRAPH '93, Virtual Reality World, November 1994

Crowe, M., Virtual Environments at GreyStone, technical presentation, GreyStone Technology

Gibson, T. J., Modelling a Combat Vehicle Commander with an Expert System, DTIC, AD-A208 533, 1989.

GreyStone (1994), AML-D Userís Manual , GSD-AMLD-UM110, GreyStone Technology.

GreyStone (1992), Purple Heart Corner, Tech Memo, GreyStone Technology.

GreyStone (1993), ThunderBolt, Tech Memo, GreyStone Technology

IST, A Glossary of Modeling and Simulation Terms for Distributed Interactive Simulation, 11th DIS Workshop on Standards for the Interoperability of Distributed Simulation, Vol. 1, 1994

Morse, P.M., Kimball, G. E. (1951) Methods of Operations Research, First Edition, Peninsula Publishing.

Shillicutt, D., On the Cover ..., Simulation, Vol. 63, No. 5, 1994

Von Neumann, J. and Morgenstern, O. (1944) Theory of Games and Economic Behavior, Princeton University Press.

Zyda et. al., The Software Required for the Computer Generation of Virtual Environments, Presence, Vol. 2, No. 2.

7. Authorí Biography

Rich Warren is a Staff Engineer and Intelligent Systems Technology Group Leader at GreyStone. Mr. Warren holds a Bachelors degree in Computer/Cognitive Science. His research interest is in Artificial Intelligence and Autonomous Intelligent Adversaries.