ISAAC

Citation

Adamatzky, A. (2002), "ISAAC", Kybernetes, Vol. 31 No. 3/4. https://doi.org/10.1108/k.2002.06731cad.001

ISAAC

ISAAC:Irreducible Semi-Autonomous Adaptive CombatDeveloper: Andrew Ilachinski Center for Naval Analyses, Alexandria VA22302, U.S.A.

Keywords: Combat simulation, Automata collectives, Social dynamic, Pattern formation, Complex systems, Artificial life, Distributed intelligence

AbstractHow does behaviour of individual combatants determine outcomes of the battle? Why lattice gas can be applied to simulation of a battlefield? What are most common patterns of a battle? One can find answers on these, and many other, questions if familiarise himself with the ISAAC simulation software. The software is briefly tackled in the review.

A discrete simulation of crowds, collectives and societies is one of the most advanced disciplines in the field of artificial life and complex systems (Bonabeau et al, 1999; Helbing, 1992; Helbing and Molnar, 1995; Epstein and Axtell, 1996; Axelrod, 1997). Typically, mass pools of digital entities obey simple rules of local interaction and inhabit a two-dimensional discrete space. They interact with one another and form complex patterns during their evolution. A correspondence between local interaction rules and patterns they implicitly produce is usually a subject of investigation.

An evolution of discrete collective is called a population dynamic when entities consume each other. This is a battlefield when they destroy each other in order to fulfil tasks different from simply supporting their existence.

Majority of the software related to a battlefield simulation is based mainly on a virtual reality and employs "see-the-battlefield" paradigm. No attention is usually paid to development of tactical thinking on a large-scale. No resources at all are allocated to simulation of minimalist agents engaged in a combat operation. Noticeably, more spatially oriented war gaming software is well spread in the computer games world. However, again it is rather a superficial imitation than true simulation of complex combatant systems.

Recent development of simulation tools for battlefields perfectly mirrors the evolution of simulation in natural sciences. It gradually shifts from differential equations, which do not account for spatial dynamics of simulated systems, toward cellular automata, where behaviour of any particular entity influences system's progression.

A first step from Lanchester equations to discrete models has been already done by Woodcock et al. (1988), who designed quite simple cellular automata rules to represent development of automata configurations in a sense of a battlefield. Less than a decade later Ilachinski made the second step (Ilachinski, 1996a, 1996b; 1997a; 1997b). He designed the ISAAC, a simulator of multiple-agent collectives, parameters of which are interpreted in terms of combat, and space-time dynamic of which simulates development of a battlefield. When designing ISAAC, Ilachinski aimed to "use complex systems theoretic inspired basic research to develop fundamentally new conceptualisations of combat" (Ilachinski, 1997a; 1997b).

ISAAC operates models based on a paradigm of discrete entities moving on a discrete lattice and interacting locally to one another. Lattice gases (Frisch et al., 1987) and lattice chemistries (Dab et al., 1990) are best examples. Sometimes, these systems are ignorantly named "mobile cellular automata" (see e.g Miramontes et al, 1993).

Why combat? Here is the ironic answer. Because "... individual behaviour under the combat rule is different from individual behaviour in the non-combat case..." (Epstein and Axtell, 1996).

In one of his talks Andrew Ilachinski quotes Carl von Clausewitz: "War is ... always the collision of two living forces". From mass flow models of forces' collisions to emerging patterns of colliding combatants – is a main idea of the simulator.

Basics of simulation

The ISAAC is a basic core simulation engine, working in DOS. EINSTein forms a Windows interface to the simulation engine. This is an "... interactive toolbox to ... for the general exploration of combat as a complex adaptive system" (Ilachinski, 1997). When discussing particulars of the simulation we keep in mind both versions of the software.

An action takes place on a two-dimensional lattice, where two forces are engaged in a battle. Each force has its own flag and battles to capture enemy's flag or to destroy enemies.

Discrete mobile automata – combatants – are elementary entities of the model. They could, for example, represent infantrymen or tanks. The combatants obey simple rules: they act depending on values of their sensors and local states of their environment.

Each combatant is supplied with a personality vector. The vector consists of six components, corresponding to alive friendly agent, alive enemy agent, injured friendly agent, injured enemy agent, own flag, and enemy flag. Values of the vector's components may be set up in a range of 25-50 units from defensive (lowest) to offensive (highest) values (Figure 1).

Figure 1. The EINSTein simulation screen with combatants' data displayed

Combatant's neighbourhood consists of five nested sub-neighbourhoods. They are, from internal to external, as follows: movement range, constraint range, fire range, sensor range and communication range. When the combatant decides which neighbouring lattice node to move in he virtually calculates an image of his neighbourhood inhabitants weighted by offensive/defensive values of its personality vector, e.g. moves toward or away an enemy or a friend.

Both rules and personality vectors of combatants may be either constant or dynamically changing during battlefield development.

Additional features of the model include local and global commanders, reconstitution fratricide (which reflects a probability to be inured by friendly fire) and attrition.

There are several run modes in the simulator, including the following:

• •

  Interactive (one can change combatants' parameters on flight);

• •

  Multiple series (multiple trials of the same combat but with different initial configurations of combatants);

• •

  Fitness landscape;

• •

  Genetic algorithm (searches in specified space for the combatants' parameters optimal for a defined mission).

Space-time configurations of combatants are probably most important data, collectable in result of simulation. They are invaluable. Additionally, one can gather integral data of combatants dynamic. These include:

• •

  Force strengths (measured in fractions of alive, injured or killed combatants);

• •

  Statistics of combatants with specified weapons;

• •

  Distributions of combatants' populations;

• •

  Distances to enemy's flag;

• •

  Spatial entropy and territorial possession; and

• •

  Mission fitness landscape.

It does not take to much time to become acquainted with the simulation engine. Almost immediately one can start to explore a rich space-time dynamic of combatants. Their behaviour varies from simple forward advance and frontal attack, to more complicate penetration or retreat, to even more sophisticated flanking manoeuvers and encirclement. An example of battlefield development is shown in Figure 2.

Figure 2. Snapshots of a simple battlefield simulated in ISAAC. The combatants start at opposite sides of the battlefield. They move toward each other and eventually engage. The simulation stops when a combatant reaches the flag of the opposite side

Applications

It is difficult to find for sure whether military actively employs the discussed simulator or not. Few results are published anyway. Thus, e.g., Horne and Lauren from the United States Marine Corps and New Zealand Defence Force, respectively, utilised ISAAC in their Operational Synthesis project (Horne and Lauren, 1999; Lauren, 1999) to explore non-linearity, intangibles, and co- evolving landscapes of modern battlefields, and to create a model useful for field commanders. These problems are particularly well tackled by ISAAC.

Outcomes of non-linearity are intrinsic for locally communicating large-scale automata collectives. It is quite typical for any lattice swarm to drastically change appearance of their global patterns when "naively small" changes in automaton local transition function are made, see e.g. (Adamatzky, 2000).

Such person-oriented parameters as a degree of training, state of a morale, ethos, leadership (Horne and Lauren, 1999; Lauren, 1999), despite their high importance, not often been incorporated in simulation systems. One can fully exploit these features in ISAAC via proper definition of agents' personality vectors.

Co-evolving landscapes and adaptability of combatants is also very important issues. In ISAAC the agents' rules may be either fixed or evolving. When the rules evolve during development of the battlefield the combatants adapt not only to specific of their environment but to tactics adopted by enemy agents as well.

Spatial characterisation of simulated combats allows one also to answer the question "how manoeuvre affects warfare" (Lauren, 1999). Thus, as reported in (Horne, 1999), the ISAAC is put to use in the Marine Corps' Project Albert, dealing with space-time relations between manoeuvres, attrition and engagement the enemy. Computer simulation shows that a degree of manoeuvrability clearly determines a success of goal achievements as well as amount of potential casualties.

Further development

The simulator is clearly a beta-version. There are a lot of bugs and the program may behave maliciously. Also it is not unusual for the simulator to simply crash. However, this does not spoil a general charm of the software. The simulator is perfect as a "proof-of-concept".

An integration of the package with other tools for army training and intelligent simulation of battlefields (see e.g. Stone et al, 1996) could be a great idea. Civil applications, e.g. social dynamic, of the simulator can be also quite productive. The ISAAC bears a huge potential for modelling pedestrian flows (see e.g. Helbing and Molnar, 1995) and collective outbursts like panic, craze and hostility (Smelser, 1962).

What is about parallelism? Behaviour of real and simulated combatants is intrinsically parallel. Thus, implementation of ISAAC on a massively parallel hardware could be quite logical. There are examples of successful parallel simulation of battlefields. The MasPaWS, a massively parallel war simulator (Sekharan et al, 1996), is one of them.

The MasPaWS is written on Maspar Programming Language and is implemented on a massively parallel SIMB machine with several thousands elementary processors. The elementary processors communicate either locally, as in typical mesh processors, or at long distances via specific routing procedures. Tanks are basic elements of the model. The tanks move on two- dimensional lattice. Each tank has a certain radius of perception, obeys simple rules of engagement its enemy, and follows intuitive migration protocols, including reduced path planning utilities. Each processor of the parallel hardware is responsible for path of the simulated battlefield. Such a set-up proves to have reasonable computational costs because the natural parallelism of the battlefield dynamic is exploited. Merging of ISAAC and MasPaWS in a single product could be advantageous.

As far as we know (Brandestein et al, 1998) Maui High Performance Computing Center has already ported the ISAAC to the massively parallel processor IBM SP to simulate and investigate models of multiple combats.

Distribution

The ISCAAC and EINSTein, together with documentation and sample combat scenarios, are downloadable free of charge on its home page at the Center of Naval Research site http://www.cna.org/isaac/ .

Andrew AdamatzkyIntelligent Autonomous Systems LabUniversity of the West of EnglandAndrew.Adamatzky@uwe.ac.uk