You are on page 1of 6

The ARTEMIS Air-to-Air Combat Model

Clinton Heinze; Brian Hanlon1; Michael Turner; Kelvin Bramley; John Rigopoulos; David Marlow; Kurt Bieri2
Air Operations Research Branch Air Operations Division Defence Science and Technology Organisation firstname.lastname@dsto.defence.gov.au Abstract. Operations Research is an important element supporting the decision process in major Air Force acquisition programs and must often address itself to a range of issues of varying complexity. An issue of particular complexity is air-to-air combat. Air-to-Air combat is a highly dynamic interaction between, possibly several, high performance aircraft employing sophisticated weapon and sensor systems and applying novel tactics designed to yield an advantage against the opponent. The combination of numerous entities, multiple system capabilities and complex tactics makes the air-to-air combat problem an especially challenging area of military operations research. Currently, Australia is evaluating the capabilities of the F-35 Joint Strike Fighter (JSF) towards possible acquisition of this aircraft to replace Australia's ageing F/A-18 and F-111 fleets. Evaluation of the air-to-air combat performance of the F-35 JSF is a central element of this assessment. Building on the body of analytical studies completed in previous study phases, this evaluation will employ a new air-to-air combat simulation model currently under development, the AiR Tactical Engagement MIssion Simulator (ARTEMIS) to explore issues specific to the F-35 JSF. The requirement for such a model follows an assessment of available air combat models in Australia and overseas. ARTEMIS is based on a flexible simulation architecture and employs teamed intelligent agent technology to model complex multi-aircraft tactics and Command & Control structures. This paper will describe the ARTEMIS model and the context of its development. The paper will then be extended to a consideration on how ARTEMIS will be employed in operations research and the issues that will be addressed. 1. THE HUNT CONTINUES partly a project management story; and points to a change in the ways that AOR does business that have improved verification and validation and taken AOR in a direction toward experimentation and the study of future warfighting concepts (such as network centric warfare). This paper sets out to tell this story in the context of ARTEMIS, the latest and (arguably) most sophisticated simulation system as yet developed by AOR. ARTEMIS Parentage PACAUS (1988-1998): sophisticated air combat model that proved highly effective for modelling small numbers of aircraft in BVR and WVR combat. Coded almost entirely in FORTRAN with very little thought given to architectural design, as it grew it became increasingly more difficult to modify and maintenance was all but impossible. SWARMM (1995-2002): saw the introduction of intelligent agents and the commencement of an effort to improve software engineering, V&V and build a closer relationship with subject matter experts. Gains included the flexible representation of tactics; quicker development times; and a

ARTEMIS is the latest in a series of simulators developed by Air Operations Research Branch (AOR) of the Defence Science and Technology Organisation (DSTO). Rather than report only on the technology behind ARTEMIS (Section 2), this paper will also provide: insights into the development of a simulation where there is a large and diverse stakeholder base (Section 3) including a discussion of some of the project management challenges; the purpose to which the simulator will be put and a sample scenario (Section 4); an indication of future directions for ARTEMIS and related technology (Section 5). 1.1 Four and Counting In the lineage outlined below ARTEMIS (and other simulation systems currently under development at AOR are representative of fourth generation systems. Over the course of a decade the capability of AOD to develop and deploy large simulations has improved. The maturation of this capability is a software engineering story, at least
1 2

Brian Hanlon has since left DSTO and now works for the Department of the Treasury of the Australian Government. Kurt Bieri is a KESEM International employee engaged by DSTO as the Software Project Manager for the ARTEMIS project.

significantly more modular and simpler to maintain simulation. AEW&C BattleModel (1998- ): the requirements of the Project Wedgetail tender evaluation process3 resulted in AOD developing BattleModel. A considerable improvement in robustness resulted from the improvement in the software engineering practices that surrounded the project. The gradual maturing of software engineering was pushed by the tight schedule and hard deadlines of a major Air Force acquisition project that allowed zero tolerance for development delays. Unsurprisingly there were a number of side benefits. 1.2 Modelling for Operations Research Simulations developed by AOR are for operations research: advice to the acquisition process; development of operational concepts and tactical procedures; and the general gaining of knowledge about current and likely future air operations. These simulations often facilitate the exploration of large parameter spaces and assist in developing a statistical understanding of the nature of the operations being studied. Performance Typically many thousands of simulated missions are flown in any given study and, as the simulated timescales can be several hours, performance often becomes a significant issue. Within the typically available time and computational resource limits, simulation performance of much faster than real time is often a requirement. Level of Representation In any modelling and simulation endeavour there are decisions made about the level of representation that is required. For operations research simulation, and consequently ARTEMIS, there is a trade-off to be made between level of detail, computational performance and cost. It is important to note here that any given scenario might involve as many as sixteen aircraft. Furthermore, the scenarios of interest to ARTEMIS do not necessarily require the highest levels of representation in the models. Sensitivity Analysis Sensitivity analyses can reveal those areas of the simulation that will have the greatest impact on the outcome. A simple example will illustrate this point. A study might be considering the effectiveness of a new short-range air-to-air missile. This missile will be launched well inside the normal detection range of the radar and often after extensive combat manoeuvring and whilst
3

within visual range. The effectiveness of the missile under these conditions will be sensitive to the missile performance, the target and launch aircraft manoeuvring and any countermeasures. A highly sophisticated radar model will add nothing to results and will impact on the computational performance. It is necessary to select the level of representation appropriate for the studies being undertaken, sensitivity checks can ensure this and give indications of error bounds over any reported measures of effectiveness. Requirements Volatility Within these requirements that drive simulations like ARTEMIS there are always compromises that must be made due to unavailability of data, uncertainty in requirements, or the use of legacy systems. The representation of an aircraft whose systems are still under development requires flexibility in the representation so that as the design firms up the models can be tailored to match. This is also true of the models of the tactical employment of those systems. Third Party Components The core-business of AOR is to provide advice across a broad spectrum of matters related to air operations. Maintaining the level of technical knowledge required for all of the modelling and simulation activities is beyond the capabilities of the relatively small team. Other branches of Defence assist with the provision of technical advice, data, and operational knowledge about specific hardware. Provided knowledge might be limited to background technical briefings or as thorough as completely documented software models. In this way AOR outsources some of the simulation development activities to appropriately skilled and resourced third parties. Validation of the model can then be vested with those third parties. A consequence of this approach is that AOR requires an approach to simulation development that is flexible enough to accommodate third party software components as they become available. 1.3 Risk ARTEMIS combines a suite of technologies and architectural design choices that have resulted in the most sophisticated air combat OR simulation yet deployed by AOR. Despite AORs capability maturity, the innovative nature of some of the design choices has resulted in a system that is as technologically sophisticated as any deployed by AOR but has the added pressures of a large and diverse development team, tight budgets and schedules and a base set of requirements that continue to display some uncertainty. The following section outlines the technologies in use

Project Wedgetail was the acquisition and introduction into service of an Airborne Early Warning and Control Aircraft.

in ARTEMIS and very brief descriptions of the relevant architecture. 2. TECHNOLOGY

This section outlines the core technologies. These technologies are related to the kernel or simulation architecture; the tactical reasoning component; the physical systems modules; and the interfaces. 2.1 Simulation Architecture The simulation architecture in use in ARTEMIS is BattleModel. BattleModel was designed to provide a capability to incorporate third-party software components into a larger simulation. Models sourced from elsewhere within DSTO or from other organisations can be obtained and bundled together with the existing library of models to create a complete simulation. Faster than real time performance Unlike human in the loop training simulators OR requires faster than real time performance often much faster than real time performance in order that they produce the many thousands of runs necessary for a detailed statistical analysis. Repeatability The simulations that are described here generate knowledge: understanding of tactics; knowledge of the manner in which systems interact; and measures that indicate the relative performance of various systems. As such they provide answers to questions and, fundamentally, those answers must be explainable and justified. Inevitably this leads to a stronger sense of validation than might be required for training simulators where the behaviour need only be perceived as correct to be beneficial. OR simulations must be explainable, visible, repeatable and trusted in ways that other systems might not need to be. Modularity Modularity, a generally desirable property of software, influences OR simulations quite strongly. Making choices in the performance/level of representation/cost trade-off will commonly result in requirements for models of varying complexity. A simulation architecture is required that supports the substitution of models at different points on the performance/level of representation trade-off spectrum. 2.2 Tactical Reasoning The tactical reasoning in ARTEMIS makes use of JACK intelligent agent technology. JACK

JACK4 is an intelligent agent programming language, based on the BDI agent model of Rao and Georgeff [4]. In many respects it is successor to PRS and dMARS, but is implemented in JAVA providing greater portability. Observe Orient Decide Act (OODA) ARTEMIS uses a design pattern that has proved useful in several simulations [2]. During the development of SWARMM it proved useful to partition the tactical reasoning in a way that simplified the conceptualisation its functioning in keeping with nature of the domain. Boyds OODA loop model of military decision making was preferred for several reasons: it was familiar to Air Force fighter pilots and matched closely with their introspective account of tactical decision making; it can be implemented within languages like JACK in a manner that is consistent with time-stepped simulation; and it provides a simplifying layer on top of the complex semantics of the BDI model in a manner that can enforce repeatability. The particular implementation of the OODA loop model serialises the tactical decision making on a time step by time step basis into four components. The first O is a currently sparsely populated component responsible for handling situation awareness. This is really little more than a mapping from the data made available by the rest of the simulation environment into the form required by the agent. The second O (orient) includes most of the processing of the current state to provide situation and threat assessment. Decide deals mainly with the agents reasoning about a course of action to select and finally the Action provides the implementation of the standard operating procedures and tactical plans. TEAMS A significant challenge in modelling air combat is simulating the complex team behaviours associated with pairs and fourship tactics. JACK provides some basic functionality for modelling teams and recent additions have added sophistication. This has changed the manner in which team tactic modelling is undertaken. Previously, hand coded infrastructure was necessary for modelling teams [3]. Tidhar [5] defined the semantics of a model that provides representation of team tactics and views the team as an explicit and separate entity. In ARTEMIS a pair is represented by three agents, one for each of the individual aircraft and one that represents the pair. This pair entity encapsulates any aspects of the reasoning that are associated
4

JACK is a product of Agent Oriented Software, further information is available from www.agentsoftware.com

with coordinated activity. This is often conceptually difficult but offers knowledge and software engineering advantages. A variant of the OODA loop that caters for team decision-making has been developed specifically for ARTEMIS. 2.3 Physics Models The latest fighter aircraft the JSF being a prototypical example are fitted with a number of new systems or technologies. The radar models are an interesting case in point. The JSF and most latest generation fighter aircraft are fitted with electronically scanned radars. This represented a new challenge for AOR branch. All prior modelling of electronically scanned radars had been confined to early warning and surveillance aircraft. A range of similar technical modelling challenges was met for the electronic support measures, the aerodynamics, radios and data-links, weapons and countermeasures. 2.4 Interfaces ARTEMIS provides the user with a number of important interfaces. One of these is a traditional graphical user interfaces for visualising air combat, one allows the visualisation of the tactical reasoning component during the execution of the scenario and the last, but most important, is a data collection module that allows the analyst to gather relevant data from the simulation. MOE Module Unlike many simulators, BattleModels primary interface is not graphical. The measures-ofeffectiveness (MOE) module is programmed to provide basic data gathering and first cut analysis from a batch set of runs. Scenario Initialisation and Visualisation The scenario initialisation and visualisation guis available as a bundled part of BattleModel are included as a part of ARTEMIS. (see screenshots)

range of information are available throught the GUI.

Figure 2: XCombat is a three dimensional visualisation tool in wide use in AOR. The package provides a relatively light-weight and standard GUI for visualising air combat in three dimensions. Out-of the cockpit and gods eye modes are available. It is likely to be incorporated into ARTEMIS in a future release. Tactical Reasoning The standard JACK GUIs are part of ARTEMIS. They are mentioned here because AOR branch has invested in the development and improvement of the guis with features that have been custom-built (see screenshots). The ability to program the tactical reasoning in a graphical manner has in the past provided advantages for intelligent agent development and there is long term expected benefits for ARTEMIS: validation and verification; and building trust with subject matter experts and operational military personnel.

Figure 1: BattleModel provides the BattleVision 2D interface for visualising scenarios. Control of the simulation, playback and record, and a wide

Figure 3: A screen shot of the JACK Development Environment. By utilising a programming language that expresses the tactics in a graphical form with natural language descriptions it is possible to ease the validation process and improve the understanding of the code.

3.

PROJECT MANAGEMENT

4.

STUDIES

The ARTEMIS project is sponsored through the ADO by the NACC project office who are the ultimate customer for the results of any studies. Responsibility for the development of the simulation lies with AOR as does the ultimate sign-off on the validity of the any results generated and hence the validity of the simulation. AOR have contracted project management support for the project and, furthermore, employ contractors for the development and support of simulation architecture (KESEM) and for the development of large components of the tactical reasoning (Agent Oriented Software). 3.1 Stakeholders, obligations responsibilities and

With the understanding that important but classified study details must remain unreported in this open forum the following are indicative of ARTEMIS questions and scenarios. The New Air Combat Capability (NACC) project In June 2002, Australia entered the System Development and Demonstration (SDD) phase of the JSF project with the proposed intent of acquiring the aircraft as a replacement of the capability currently provided by the F/A-18 and F111. As a result, the New Air Combat Capability (NACC) project was created to determine how the JSF would be integrated into the Australian Defence Force (ADF). Consequently AOR has been tasked to address the following key questions: Can the JSF meet the Australian requirements for air combat and strike capability? How many JSF aircraft are required by the ADF? What and how many adjunct systems are required to complement the JSF (Tankers, AEW&C etc.)? The initial application of ARTEMIS will be to evaluate the effectiveness of the JSF in air combat missions, which will ultimately contribute to addressing the key questions of the NACC project. Such evaluation studies will explore variations on a number of parameters to determine the circumstances under which operational effectiveness hold. These parameters are primarily: Air combat missions, such as Defensive Counter Air (DCA) and Offensive Counter Air (OCA) Team sizes, such as 2-ship and 4-ship formations Operational concepts, ranging from current RAAF tactical procedures to novel concepts. Novel concepts will take advantage of the revolutionary systems technology inherent on the JSF. For example, stealth design, active and passive sensors, and data-links enabling Network-Enabled Concepts with teams of JSF and with AEW&C as an adjunct system. Threat aircraft types projected for the timeframe in which the JSF would be in service. Example Study Question In air combat, the fighter that is able to detect, target, and launch a Beyond Visual Range (BVR) weapon against the opposing fighter earliest has an

AOR has approximately 9 Commonwealth groups providing input into ARTEMIS development and two contractor organisations (AOS and KESEM) which are undertaking the development. In addition, as previously, mentioned ARTEMIS is also faced with challenging schedule and cost constraints. These diverse project interfaces and constraints require amore formal engineering processes than have previously been applied. The processes include such things as the formalisation of all model requirements, configuration management, quality assurance and appropriate levels of software engineering rigour; and combine to provide confidence that the project constraints can be met. 3.2 V&V System level validation and verification is ultimately the responsibility of AOR. Some models, particularly those models that represent the performance of aircraft sub-systems are validated at the component level by DSTO technical divisions with expertise in relevant technologies. The validation of the entire simulation system is undertaken by AOR in a variety of ways: system-level regression testing against previously developed and validated air combat models; model-level validation against empirical data obtained from a variety of sources; sanity checks against the expertise of AOR branch; future plans for experimentation and consequent validation by Air Force personnel in the style of AORs previous validation activities [1].

advantage in winning the engagement. Therefore, an important question is: How do the JSF systems capabilities and the way they are employed contribute to the ability to achieve first-look and first kill? Key drivers to this question are the ability of the JSF to have Situation Awareness of the threat and deny Situation Awareness to the threat during the intercept. The former depends on how the sensor systems are employed, whether active or passive, on-board JSF, off-board JSF, or adjunct AEW&C. The latter depends on the nature of the emitted and reflected signatures the JSF presents to the threat sensors. Sample Scenario An example study is a 2v2 scenario of offensive counter air (OCA) versus defensive counter air (DCA). The JSF and threat fly to a common waypoint and may employ similar offensive and defensive tactics. Various cases are investigated which parameterise the JSF and threat system performance and tactics. Each case involves Monte-Carlo runs, each with a randomised initial threat flight paths. The statistical engagement outcomes for each case are then evaluated and analysed. Reporting and Advice The result of these studies will contribute to a report providing Science and Technology (S&T) advice to support Defence and Government approval of the JSF acquisition. 5. WHEN THE WAR IS OVER

AOS and AOR during the course of the last five years. 5.2 Lessons Learned There will always be a need to trade-off computational performance, level of representation and cost. This is perhaps even more critical in OR simulations than in real-time training systems. The requirements of the studies (and indicative scenarios) should drive the development of the models. This is particularly true of the tactical decision making models. If the requirements are not firmly pinned down there is little choice but to build for flexibility but there is a price to pay. Even though this is a fourth generation system and AOR have had extensive development experience with similar systems, the new technology (JACK, BattleModel) and a changed management structure involving greater contribution from contractors have combined to create relatively high levels of risk. ARTEMIS will be used now and into the future to meet the operations research simulation studies requirements of the ADF. Development is ongoing to meet the tight schedules involved. ARTEMIS has the flexibility to meet the needs of future air combat studies in AOR and will underpin the major new Aerospace Battlelab Facility. REFERENCES
[1] Clint Heinze, Martin Cross, Simon Goss, Torgny Josefsson, Ian Lloyd, Graeme Murray, Michael Papasimeon, and Michael Turner. Agents of change: The impact of intelligent agent technology on the analysis of air operations In L. Jain, N. Ichalkaranje, and G. Tonfoni, editors, Advances in Intelligent Systems for Defence, volume 2 Series on Innovative Intelligence, chapter 6, pages 229--264. World Scientific, River Edge, New Jersey, USA, 1 edition, December 2002. Heinze, C. et al., (2002) Interchanging Agents and Humans in Military Simulation, AI Magazine, 23:2, pp37-47 G. Tidhar, C. Heinze, and M. Selvestrel. Flying Together: Modelling Air Mission Teams. Applied Intelligence, vol. 8, pp. 195-218, 1998. A.S. Rao and M.P. Georgeff. Modeling rational agents within a BDI-architecture. In J. Allen, R. Fikes, and E. Sandewall, editors, Proceedings of the Second International Conference on Principles of Knowledge Representation and Reasoning (KR'91), pages 473--484. Morgan Kaufmann, 1991. Tidhar, G. Team Oriented Programming: Theory and Practice. PhD Thesis, University of Melbourne, 1999.

This section documents the current capability of ARTEMIS, some of the more important lessons learned in the development to date, and briefly maps the future development. 5.1 Current Capability ARTEMIS is part way through its development process and development is currently ongoing. The software is at a stage where studies to commence and useful results obtained. The first round of studies will be reported by the middle of this year. Current implementation includes a variety of simulations of physical systems modelled at the appropriate level. The models of fighter pilots are implemented in JACK and these include a formalising of many of the ad-hoc design approaches used in earlier air combat models. An example of this is the four box model, a computational version of the OODA loop model of reasoning underpins the decision cycle of the intelligent agents. ARTEMIS is the first air combat model developed by AOR to take advantage of the teaming extensions that have been developed by

[2]

[3]

[4]

[5]

You might also like