ISAAC & EINSTein

Report
Marcin Waniek


Towards a Science of Experimental
Complexity: An Artificial-Life Approach to
Modeling Warfare
Andy Ilachinski, Center for Naval Analyses




Homogeneous forces that are
continually engaged in combat
Soldiers always aware of the
position and condition of all
opposing units
Appropriate for static trench
warfare or artillery duels
Rather unrealistic for modern
(and also much older) battlefield

"War is ... not the action of a
living force upon lifeless mass ...
but always the collision of two
living forces.„
- Carl von Clausewitz

“The fight is chaotic yet one is
not subject to chaos.”
– Sun Tzu




Dynamical system composed of many
nonlinearly interacting adaptive agents.
Local action, which often appears disordered,
induces long range order.
No master “voice” that dictates the actions of
each and every combatant.
Military forces must continually adapt to a
changing combat environment.




Irreducible Semi-Autonomous Adaptive
Combat
Bottom-up, synthesist approach to the
modeling of combat.
„Conceptual playground" to explore highlevel emergent behaviors arising from various
low-level interaction rules.
Model patterned after mobile cellular
automata rules.




Doctrine: a default local-rule set specifying
how to act in a generic environment
Mission: goals directing behavior
Situational Awareness: sensors generating an
internal map of environment
Adaptability: an internal mechanism to alter
behavior and/or rules




Agent belongs to one of two armies – Red or
Blue
Agent exists in one of three states – alive,
injured or dead
Each agent has defined sensor and weapon
range
Each agent is equipped with personality
defined by vector ω = (ω1, ω2, ..., ω6) where
-1 ≤ ωi ≤ 1 and |ω1| + ... + |ω6| = 1.






ω1
ω2
ω3
ω4
ω5
ω6
- the number of alive friendly agents
- the number of alive enemy agents
- the number of injured friendly agents
- the number of injured enemy agents
– the distance from friendly flag
– the distance from enemy flag


ω = (1/20, 5/20, 0, 9/20, 0, 5/20)
five times more interested in moving toward
alive enemies than alive friendlies, even more
interested in moving toward injured enemies
ω = (-1/6,-1/6,-1/6,-1/6,-1/6,-1/6)
wants to move away from, rather than
toward, every other agent and both flags, i.e.
it wants to avoid action of any kind.


Rules telling how to alter agents personality
according to environmental conditions.
Basic meta-rule classes: advance toward
enemy flag, cluster with friendly forces,
engage the enemy in combat

Examples of other meta-rules: retreat,
pursuit, support, hold position.


Red effectively
encircles Blue
forces
Fixed Blue
personalities
unable to find
countermeasures


Example of nonmonotonic behavior
Enlarging Red forces
sensor range leads
to a worse outcome


Red forces bred using
genetic algorithm,
Blue forces fixed
Red able to weaken
the center of Blue
line, and then attack
the weak spot with all
forces




Enhanced ISAAC Neural Simulation Toolkit
Context-dependent and user-defined agent
behaviors (i.e. personality scripts)
On-line genetic algorithm, neural-net,
reinforcement-learning, and pattern
recognition toolkits
Agents fighting as a part of small units

similar documents