Robotics on the Battlefield Part II The Coming Swarm

OCTOBER

2014

Robotics on the Battleeld Part II

The Coming Swarm

By Paul Scharre

Cover Image

A ock of auklet birds exhibit swarm behavior.

(U.S. FISH AND WILDLIFE SERVICE)

Also in this series:

“20YY: Preparing for War in the Robotic Age” by Robert O. Work and Shawn Brimley

“Robotics on the Battleeld Part I: Range, Persistence and Daring” by Paul Scharre

Acknowledgements

I would like to thank the numerous colleagues at CNAS and elsewhere who assisted in the development of this report. Liz

Fontaine led the production and design of this report. Adam Elkus provided invaluable assistance in understanding animal

and robotic swarms. CNAS research interns James Marshall and Matthew Seeley assisted in background research. In addi-

tion, a number of outside experts provided valuable insights on a range of issues. I would like to thank Russell Rumbaugh

of the Stimson Center, Michael Sulmeyer of the Oce of the Secretary of Defense, Dean Wilkening of Lawrence Livermore

National Laboratory, David Scheidt of Johns Hopkins University Applied Physics Laboratory, Sachin Jain of Aurora Flight

Sciences, Timothy Chung of the Naval Postgraduate School and Andrew Herr of Helicase for their assistance, corrections

and insights. I would also like to thank the numerous colleagues at CNAS and other institutions who provided feedback on

draft versions of this report.

Any errors of analysis, fact or omission are mine alone. CNAS does not take institutional positions.

Robotics on the Battleeld Part II

The Coming Swarm

By Paul Scharre

OCTOBER 2014

VII. Enemy Swarms and Countermeasures 42

VIII. Conclusion: Building the Swarm 44

IX. Recommendations 50

TABLE OF CONTENTS

I. Executive Summary 5

II. Introduction: 10

The Reconnaissance-Strike Swarm

III. Mass 13

IV. Coordination and Intelligence 24

V. Speed 33

VI. Commanding the Swarm 35

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

About the Author

Paul Scharre is a fellow and director of the 20YY Warfare Initiative at the Center for a New

American Security.

ROBOTICS ON THE BATTLEFIELD PART II:

THE COMING SWARM

By Paul Scharre

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

By Paul Scharre

I. EXECUTIVE SUMMARY

e unfolding robotics revolution is transform-

ing a range of industries, from manufacturing to

transportation, warehouse management, household

appliances, toys, elder care and more. Similarly,

it will lead to signicant and perhaps surprising

changes in warfare. Uninhabited vehicles, like the

Predator aircra or the Packbot ground robot,

have already proven invaluable in today’s conicts.

As uninhabited vehicles incorporate increasing

automation and become true robotic systems,

they will have tremendous value in future military

operations. Individually, they will allow military

forces to extend their reach into the battlespace,

operating with greater range and persistence than

would be possible with human-inhabited systems.

With no human on board they can be sent on dan-

gerous or even suicidal missions, allowing more

daring concepts of operation. Individually, robotic

systems can provide warghters signicant advan-

tages in a range of missions. Collectively, swarms

of robotic systems have the potential for even more

dramatic, disruptive change to military operations.

Swarms of robotic systems can bring greater mass,

coordination, intelligence and speed to the battle-

eld, enhancing the ability of warghters to gain a

decisive advantage over their adversaries.

Secretary of Defense Chuck Hagel has called for

a renewed eort to sustain American military

technological dominance, and uninhabited and

autonomous systems are an important component

of such a strategy.

Today the U.S. military faces

a pernicious cycle of ever rising platform costs

and shrinking quantities. As a result, the number

of combat ships and aircra in the U.S. inven-

tory has steadily declined, even during periods of

signicant growth in defense spending. Today’s

acute scal pressures only exacerbate these trends,

forcing a crisis not only in military moderniza-

tion and readiness, but also in the ability to eld

sucient quantities to be relevant in future ghts.

As precision-guided munitions proliferate to other

adversaries – both state and non-state actors – the

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

shrinking numbers of U.S. combat assets becomes

a major strategic liability. Adversaries can con-

centrate their weapons, which are becoming

increasingly accurate and capable at ever-longer

ranges, on the relatively small number of U.S.

ships and bases, overwhelming their defenses. e

current trend of attempting to compensate for

ever-shrinking numbers of capital assets through

increasingly exquisite systems is not sustain-

able. Clinging to greater quantities by eschewing

modernization, however, is not a recipe for success

either. A new paradigm is needed, one that sustains

the qualitative superiority of U.S. forces in aggre-

gate, but that disperses combat power among a

greater number of platforms, increasing resiliency

and diversity and imposing costs on adversaries.

Uninhabited systems can help bring mass back to

the ght by augmenting human-inhabited combat

systems with large numbers of lower cost unin-

habited systems to expand the number of sensors

and shooters in the ght. Because they can take

more risk without a human onboard, uninhabited

systems can balance survivability against cost,

aording the ability to procure larger numbers

of systems. Greater numbers of systems compli-

cates an adversary’s targeting problem and allows

graceful degradation of combat power as assets are

attrited. e disaggregation of combat power into a

larger number of less exquisite systems also allows

the ability to eld a family-of-systems approach,

increasing diversity and reducing technology risk,

driving down cost. Uninhabited systems need not

be exquisite multi-mission systems, but rather can

be purpose-built for specic missions at lower cost.

For example, uninhabited missile barges, under-

sea payload modules, airborne “missile trucks”

and robotic appliqué kits for ground vehicles can

supplement the striking power of existing manned

platforms at relatively low cost. e result can be

greater combat power on the battleeld, at the

same cost. By embracing uninhabited and autono-

mous systems, the United States can disperse its

combat capabilities, increasing resiliency, and

expand its oensive striking capacity, all within

realistic budget constraints.

e power of swarming lies in more than just

greater numbers, however. Today’s modern mili-

tary forces ght as a network, with interconnected

human-inhabited platforms passing surveillance

and targeting data across great distances. Future

military forces will ght as a swarm, with greater

coordination, intelligence and speed. Autonomous

and uninhabited systems will be networked and

cooperative with the ability to autonomously coor-

dinate their actions in response to events on the

ground. Swarming, coordinated action can enable

synchronized attack or defense, more ecient allo-

cation of assets over an area, self-healing networks

that respond to enemy actions or widely distrib-

uted assets that cooperate for sensing, deception

and attack. Harnessing the power of swarming

will require new command-and-control models

for human supervision of large swarms. is will

mean moving beyond existing paradigms where

humans directly control a vehicle’s movements to

one where human controllers supervise the mission

at the command level and uninhabited systems

maneuver and perform various tasks on their own.

Increased automation also has the potential to

speed up the pace of warfare by helping to shorten

decision cycles and, in some cases, remove humans

from them entirely. Increased automation can

allow humans to process large amounts of data

quickly, allowing warghters to react to chang-

ing events on the ground faster than the enemy.

In some cases, the fastest reactions might come

from removing humans from some tasks entirely,

as is already done for some defensive actions like

dispensing ares or other countermeasures. While

increased automation may have tactical benets

in allowing faster reaction times to enemy actions,

it could also have strategic consequences if the

speed of action on the battleeld eclipses the speed

of decision-making for policymakers. Increased

autonomy in the use of force raises the dangerous

specter of “ash wars” initiated by autonomous

systems interacting on the battleeld in ways that

may be unpredictable. While militaries will need

to embrace automation for some purposes, humans

must also be kept in the loop on the most critical

decisions, particularly those that involve the use of

force or movements and actions that could poten-

tially be escalatory in a crisis.

Increasingly sophisticated autonomous systems

will still fall short of human intelligence in many

respects, and uninhabited systems will not be

useful or appropriate for all missions. A human-

machine teaming approach will be needed to nd

the optimal mix of human-inhabited and uninhab-

ited platforms and human and machine cognition

for various tasks. As one example, the Army has

adopted an approach of teaming human-inhabited

Apache helicopters with uninhabited Gray Eagle

aircra to perform armed aerial reconnaissance.

Developing the doctrine, training, concepts of

operation and organization to enable eective

human-machine teaming will be critical to lever-

aging the unique advantages of uninhabited and

autonomous systems in a wide range of mission

areas.

e introduction of greater numbers of uninhab-

ited and autonomous systems on the battleeld will

not lead to bloodless wars of robots ghting robots,

but could make more warfare more deadly and

dangerous for human combatants. Humans will

still ght wars, but new technology will give com-

batants, as it always has, greater stando from the

enemy, survivability or lethality. Exploiting those

advantages will depend principally on the ability to

uncover the most innovative applications of robotic

swarms, which will require not only increased

resources but also an aggressive campaign of

experimentation and technology development.

Many of the underlying technologies behind

increased autonomy are driven by commercial

sector innovation, and as a result will be available

to a wide range of state and non-state actors. In

a world where some of the most-game changing

technologies will be available to everyone, uncov-

ering the best uses of that technology – and doing

so urgently – will be vital to sustaining American

military dominance.

Humans will still ght wars,

but new technology will give

combatants, as it always has,

greater stando from the enemy,

survivability or lethality.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

KEY RECOMMENDATIONS

THE OFFICE OF THE SECRETARY

OF DEFENSE SHOULD:

• Undertake a study on swarming plat-

forms to examine the potential for

low-cost uninhabited systems to impose

costs on adversaries.

• Fund a multi-year series of experiments

in cooperative multi-vehicle control and

swarming.

• Establish a Defense Robotics Systems

Oce, directly reporting to the Deputy

Secretary of Defense, to coordinate

ongoing eorts on uninhabited systems

across the Department.

THE JOINT STAFF SHOULD:

• Ensure that lessons learned from ex-

periments regarding uninhabited and

autonomous systems are centrally col-

lected and widely shared throughout the

Department.

THE NAVY SHOULD:

• Build an experimental prototype of an

uninhabited missile barge that can dem-

onstrate the ability to remotely control

and launch missiles from a large unin-

habited vessel.

• Build a proof-of-concept demonstra-

tion of an undersea payload module to

exploit U.S. sanctuary undersea.

• Move aggressively to eld autonomous

swarming defensive boats to protect U.S.

ships from enemy fast attack craft. This

should include further experimentation

to rene concepts of operation, a rapid

elding initiative to equip combatants in

high-risk areas like the Straits of Hormuz

and a program of record for outtting all

Navy surface combatants with optional-

ly-manned small boats that can operate

as a defensive swarm.

THE AIR FORCE SHOULD:

• Investigate the potential for low-cost

swarming uninhabited air vehicles, in-

cluding expendable or non-recoverable

systems such as missiles or decoys, to

conduct a variety of missions including

suppression/destruction of enemy air

defenses, reconnaissance, battle damage

assessment and electronic warfare.

• Conduct an analysis of alternatives of

lower-cost uninhabited aircraft to sup-

plement existing manned aircraft with

additional sensors and missiles, such as

an uninhabited “missile truck.”

• Conduct a series of experiments in

human control over large numbers of

swarming air vehicles.

THE ARMY AND MARINE CORPS

SHOULD:

• Develop a concept of operations for

using appliqué kits for ground convoy

operations and an associated program of

record.

• Conduct a series of modern day “Louisi-

ana Maneuver” experiments on “robotic

wingman” ground robots for long-range

scouting and maneuver operations,

in order to inform further technology

development and requirements for an

eventual program of record.

• Conduct a series of experiments on

swarming uninhabited air vehicles for

persistent surveillance, close air support,

aerial resupply and communications re-

lay to support ground maneuver forces.

THE MARINE CORPS SHOULD:

• Conduct experiments on amphibious

swarming robots for reconnaissance

and counter-mine operations to clear

beaches ahead of an amphibious assault.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

II. INTRODUCTION: THE

RECONNAISSANCESTRIKE SWARM

From Fighting as a Network to Fighting

as a Swarm

Advances in information technology achieved

in the twentieth century allow modern military

forces to ght as a network. Sensors can detect

enemy forces and pass targeting data through

communications links to shooters, who engage

enemy targets with precision-guided weapons.

e U.S. military was the rst to harness the

advantages of the information revolution to build

a networked force, but other nations are following

suit. Adversaries are building reconnaissance-

strike networks that can detect U.S. forces at long

range and strike them with precision-guided

weapons. ese developments, oen captured

under the label of anti-access, area denial (A2/

AD) challenges, threaten many traditional U.S.

modes of power projection, such as ships, carriers

and air bases that can be targeted with long-

range weapons.

As detailed in CNAS’s recent

report, “Robotics on the Battleeld Part I: Range,

Persistence and Daring,” uninhabited systems

can help U.S. forces to counter this threat because

of their increased range, persistence and ability

to take greater risks, enabling new concepts of

operation.

But these developments are merely the precursor to

a larger shi in warfare.

Emerging robotic technologies will allow tomor-

row’s forces to ght as a swarm, with greater mass,

coordination, intelligence and speed than today’s

networked forces. Low-cost uninhabited systems

can be built in large numbers, “ooding the zone”

and overwhelming enemy defenses by their sheer

numbers. Networked, cooperative autonomous

systems will be capable of true swarming – coop-

erative behavior among distributed elements that

gives rise to a coherent, intelligent whole. And

automation will enable greater speed in warfare,

with humans struggling to keep pace with the

faster reaction times of machines. e result will

be a paradigm shi in warfare where mass once

again becomes a decisive factor on the battleeld,

where having the most intelligent algorithms may

be more important than having the best hardware,

and where the quickening pace of battle threatens

to take control increasingly out of the hands of

humans.

Keeping Pace with the Unfolding Robotics

Revolution

ese developments will pose profound opera-

tional and policy challenges. Adapting to these

challenges will require the development of new

capabilities, experimentation with new concepts

and development of new doctrine and organi-

zational structures. Despite the U.S. military’s

dominance today, other nations may be better

prepared to capitalize on these coming changes.

e U.S. military is heavily invested – both

nancially and bureaucratically – in today’s

technologies and methods of ghting. While

uninhabited systems have been embraced for

some missions like reconnaissance and bomb

disposal, across the force they largely remain

relegated to niche roles. Only one out of every

20 Department of Defense (DOD) research,

development and procurement dollars goes to

uninhabited systems.

Furthermore, elements of

the U.S. military continue to resist technologies

Emerging robotic technologies

will allow tomorrow’s forces

to ght as a swarm, with

greater mass, coordination,

intelligence and speed than

today’s networked forces.

that disrupt familiar operational paradigms, such

as automation that would change the paradigm of

control for human pilots over aircra.

For many

missions, uninhabited and autonomous systems

are seen as an unproven technology and even

potentially threatening when human jobs may be

eliminated. In the face of this discomfort, a “go

slow” approach might be tempting.

e problem is that the enemy gets a vote.

By 2018, global spending on military robotics

is estimated to reach $7.5 billion per year. At

the same time, global spending on commercial

and industrial robotics is estimated to top $43

billion a year.

As a result, many of the under-

lying advances in robotics will come from the

commercial sector and will be widely available.

e U.S. military is used to competing in a world

where some of the most game-changing innova-

tions – such as stealth, GPS and precision-guided

weapons – come from the U.S. defense sector. It is

ill-prepared for a world where such technologies

are widely available to all.

If the U.S. military is to keep pace with the

unfolding robotics revolution, it will need to

adopt an aggressive strategy of targeted research

and development, experimentation and concept

and doctrine development. is will require not

only increased resources, but also better institu-

tional processes. Existing acquisition processes

are too sluggish to keep pace with rapid tech-

nological change and pose a strategic risk to the

United States. If they cannot be reformed, then

DOD leaders will increasingly have to operate

outside the traditional processes in order to rap-

idly adapt to emerging needs, as they repeatedly

did when adapting to urgent needs for Iraq and

Afghanistan.

Most importantly, when new technologies upend

existing operational paradigms, the alternative

Emerging robotic technologies

will allow tomorrow’s forces

to ght as a swarm, with

greater mass, coordination,

intelligence and speed than

today’s networked forces.

What is a Robot?

Robotic systems combine two key attributes: (1)

uninhabited, or unmanned, platforms or vehicles;

and (2) autonomous or semi-autonomous opera-

tions. While a true “robot” incorporates both attri-

butes, they can be separated. Some uninhabited

platforms or vehicles are remote-controlled, and

autonomous features can and often are incorpo-

rated onto human-inhabited platforms.

Removing the human from a vehicle can have

several advantages. Vehicles that are uncon-

strained by human physical limitations can have

increased range, endurance, maneuverability,

persistence, speed or stealth. Without a human

onboard, commanders can also use the vehicle

to undertake more hazardous missions without

risking a human life.

Autonomy is the ability of a machine to perform a

task without human input. Increased automation

or autonomy can have many advantages, includ-

ing increased safety and reliability, improved

reaction time and performance, reduced person-

nel burden with associated cost savings and the

ability to continue operations in communications-

degraded or -denied environments.

Elements of the U.S. military

continue to resist technologies

that disrupt familiar

operational paradigms, such as

automation that would change

the paradigm of control for

human pilots over aircra.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

In the 1980s, technological devel-

opments in sensors, command-

and-control networks and preci-

sion-guided munitions enabled

the possibility of real-time preci-

sion targeting of ground forces,

with the potential for strategic

eects that were not previously

possible without resort to nuclear

weapons. Soviet military theorists

were the rst to recognize the

game-changing potential of these

technologies and coined the term

“reconnaissance-strike complex”

to describe the synergistic com-

bination of sensors, networks and

precision-guided munitions work-

ing together.

The rst battle networks actually

originated in World War II. During

the Battle of Britain, the United

Kingdom used a network of radars

and spotters, connected with radio

and telephone cables, to direct

British ghters toward incoming

German bombers. Actual engage-

ments were still conducted with

unguided weapons, however.

During the next several decades,

precision-guided munitions

increased in accuracy while sen-

sors and network technology also

improved. By the early 1990s they

had reached a culminating point,

and the overwhelming U.S. victory

in the Persian Gulf War validated

Soviet theories about the value of

information technology-enabled

reconnaissance-strike networks.

Today, sophisticated nation-states

operate reconnaissance-strike

battle networks comprised of

sensors, command-and-control net-

works and precision-guided weap-

ons. The combination of these

elements allows forces to ght

as a networked whole capable of

long-range precision strikes. These

technologies are not only prolifer-

ating to other states over time, but

many low-cost versions are within

the reach of non-state actors. The

United States should expect future

adversaries, state and non-state

alike, to be able to operate battle

networks capable of targeting U.S.

forces with great precision.

Uninhabited and autonomous

systems will enable the next evolu-

tion, as forces shift from ghting as

a network to ghting as a swarm,

with large numbers of highly

autonomous uninhabited systems

coordinating their actions on the

battleeld. This will enable greater

mass, coordination, intelligence

and speed than would be possible

with networks of human-inhab-

ited or even remotely controlled

uninhabited systems. Human

judgment will still be essential for

many decisions, but automation

will help humans to process large

amounts of data rapidly, control

large numbers of vehicles simul-

taneously and shorten decision

cycles, accelerating the tempo of

operations.

What is a Reconnaissance-Strike Network?

concepts they enable should be embraced through

experimentation and innovation. e history of

revolutions in warfare has shown they are won

by those who uncover the most eective ways of

using new technologies, not necessarily those who

invent the technology rst or even have the best

technology. is report is an attempt to chart out

what those new uses might be, and how they could

change operations on the battleeld.

III. MASS

e United States outproduced its enemies in

World War II. By 1944, the United States and its

Allies were producing over 51,000 tanks a year to

Germany’s 17,800 and over 167,000 planes a year

to the combined Axis total of just under 68,000.

Even though many of Germany’s tanks and aircra

were of superior quality to those of the Allies, they

were unable to compensate for the unstoppable

onslaught of Allied iron.

Paul Kennedy writes in

e Rise and Fall of Great Powers:

… by 1943-1944 the United States alone was

producing one ship a day and one aircra every

ve minutes! … No matter how cleverly the

Wehrmacht mounted its tactical counterattacks

on both the western and eastern fronts until

almost the last months of the war, it was to be

ultimately overwhelmed by the sheer mass of

Allied repower.

e Cold War saw a shi in strategy, with the

United States instead initially relying on nuclear

weapons to counter the growing Soviet conven-

tional arsenal in Europe. By the 1970s, the Soviets

had achieved a three-to-one overmatch against

NATO in conventional forces and a rough parity in

strategic nuclear forces. In response to this chal-

lenge, the U.S. military adopted an “oset strategy”

to counter Soviet numerical advantages with quali-

tatively superior U.S. weapons. e result of this

approach was the invention of stealth technology,

advanced sensors, command and control networks,

and precision-guided weapons.

e full eect of these weapons was seen in 1991,

when the United States took on Saddam Hussein’s

Soviet-equipped army. Casualty ratios in the Gulf

War ran an extremely lopsided 30-to-1.

Iraqi

forces were so helpless against American precision

airpower that the White House eventually termi-

nated the war earlier than planned because media

images of the so-called “highway of death” made

American forces seem as if they were “cruelly and

unusually punishing our already whipped foes,”

in the words of Gulf War air commander General

Chuck Horner.

Precision-guided weapons,

coupled with sensors to nd targets and networks

to connect sensors and shooters, allowed the

information-enabled U.S. military to crush Iraqi

forces ghting with unguided munitions.

e proliferation of precision-guided weapons to

other adversaries is shiing the scales, however,

bringing mass once again back into the equa-

tion. e United States military can expect to face

threats from adversary precision-guided munitions

in future ghts.

At the same time, ever rising plat-

form costs are pushing U.S. quantities lower and

lower, presenting adversaries with fewer targets on

which to concentrate their missiles. U.S. platforms

may be qualitatively superior, but they are not

invulnerable. Salvos of enemy missiles threaten to

overwhelm the defenses of U.S. ships and air bases.

Even if missile defenses can, in principle, inter-

cept incoming missiles, the cost-exchange ratio of

attacking missiles to defending interceptors favors

the attacker, meaning U.S. adversaries need only

purchase more missiles to saturate U.S. defenses.

Uninhabited systems oer an alternative model,

with the potential to disaggregate expensive multi-

mission systems into a larger number of smaller,

lower cost distributed platforms. Because they can

take greater risk and therefore be made low-cost

and attritable – or willing to accept some attri-

tion – uninhabited systems can be built in large

Quantity has a quality all

of its own.

 

  

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

numbers. Combined with mission-level autonomy

and multi-vehicle control, large numbers of low-

cost attritable robotics can be controlled en masse

by a relatively small number of human controllers.

Large numbers of uninhabited vehicles have sev-

eral potential advantages:

• Combat power can be dispersed, giving the

enemy more targets, forcing the adversary to

expend more munitions.

• Platform survivability is replaced with a concept

of swarm resiliency. Individual platforms need

not be survivable if there are sucient numbers

of them such that the whole is resilient against

attack.

• Mass allows the graceful degradation of combat

power as individual platforms are attrited, as

opposed to a sharp loss in combat power if a

single, more exquisite platform is lost.

• Oensive salvos can saturate enemy defenses.

Most defenses can only handle so many threats

at one time. Missile batteries can be exhausted.

Guns can only shoot in one direction at a time.

Even low cost-per-shot continuous or near-con-

tinuous re weapons like high energy lasers can

only engage one target at a time and generally

require several seconds of engagement to defeat

a target. Salvos of guided munitions or uninhab-

ited vehicles can overwhelm enemy defenses such

that “leakers” get through, taking out the target.

Some examples of ways in which these advantages

could translate to new, innovative approaches for

using uninhabited systems are below.

ATTRITABLE UNINHABITED COMBAT AIRCRAFT

In 2008, a now-infamous study by RAND Project

Air Force examined a potential future air-to-

air exchange between the U.S. and China over

Taiwan.

Because U.S. ghters had to y from pro-

tected air bases located in Guam, U.S. numbers in

the ght were signicantly reduced. Even with the

entire U.S. F-22 inventory located at Guam, range

and endurance constraints meant that only six

F-22s could be maintained over Taiwan continu-

ously. By contrast, because of both greater numbers

Artist depiction of Chinese DF-21D anti-ship missile widely circulated on Chinese defense-related web forums.

of ghters and larger, closer air bases, China was

able to surge seventy-two aircra to the ght.

e analysis assumed that every single air-to-air

missile that came o of a U.S. F-22 hit a Chinese

ghter (probability of kill = 1.0) and that zero

Chinese missiles hit any U.S. F-22s. Even still,

China won the air-to-air engagement because U.S.

ghters ran out of missiles. (F-22s can carry eight

air-to-air missiles each.) Once the F-22s had run

out of missiles, or “gone winchester,” the remaining

Chinese ghters were free to attack vulnerable U.S.

tankers and surveillance aircra.

An uninhabited “missile truck” that brought addi-

tional air-to-air missiles to the ght to supplement

human-inhabited F-22s could tip the scales back in

the United States’ favor. Such an aircra need not

have the full performance characteristics of a 5

generation ghter aircra. It would only need to

have sucient stealth to get close enough to launch

its missiles against Chinese ghters. If it then

perished in the engagement, that would be accept-

able provided it took a sucient number of enemy

ghters with it. It would still have accomplished

the mission. e uninhabited aircra would not

need advanced autonomy, merely enough to y

in a straight line under a human’s control and

suciently robust communications links for the

human-inhabited F-22s to pass targeting data. All

targeting and ring decisions would be made by

the F-22 pilots. If such an aircra could be built at

relatively low cost, this uninhabited “loyal wing-

man” could be a tremendous force multiplier for

U.S. human-inhabited ghters.

Such a concept is not far from the original vision for

the joint unmanned combat air system (J-UCAS), a

relatively low-cost “aircra in a box.”

e Air Force

should begin an analysis of alternatives to deter-

mine whether such an uninhabited aircra could be

built that would have sucient stealth and payload

capacity to augment the missile capacity of existing

manned aircra at relatively low cost.

SMALL UNINHABITED AIR VEHICLES AND

AIRMOBILE ROBOTS

e miniature air-launched decoy (MALD) and

miniature air-launched decoy – jammer (MALD-J)

– loitering air vehicles that are not quite muni-

tions and are not aircra – hint to the potential

of small, loitering uninhabited air vehicles and

air-mobile robots. e MALD functions as an

aerial decoy to deceive enemy radars, while the

MALD-J jams enemy radars.

Similar future unin-

habited air vehicles, launched from aircra, ships

or submarines, could saturate enemy territory with

overwhelming numbers of low-cost, expendable

systems.

Like D-Day’s “little groups of paratroop-

ers” dropped behind enemy lines, they could sow

confusion and wreak havoc on an enemy.

Loitering electronic attack weapons could cre-

ate an electronic storm of jamming, decoys and

high-powered microwaves. Small air vehicles could

autonomously y down roads searching for mobile

missiles and, once found, relay their coordinates

back to human controllers for attack.

Large numbers of cheap, expendable systems

could be used to deny an enemy use of an air-

eld, “mining” the airspace above it by swarming

overhead like locusts, risking collisions if enemy

aircra tried to takeo or land. Air mobile systems

could conserve power by landing near an aireld

and attacking only periodically, either based on

acoustic signatures of landing aircra or randomly

timed sorties, disrupting air trac for days.

Such aircra would be small and would require a

means of getting to the ght. is could include

submarines parked o an enemy’s coast, uninhab-

ited missile boats that race to the enemy’s coastline

before launching their payloads into the air, large

bomber or cargo aircra or even uninhabited

undersea pods like DARPA’s Hydra program.

e Air Force has recently initiated development of

a new “ight plan” for small uninhabited air vehi-

cles. As it begins to scope out the potential for such

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

RISING COSTS: AUGUSTINE’S

LAW

In 1984, Norm Augustine observed

as one of “Augustine’s Laws” that the

cost of military aircraft was growing

exponentially, while the defense

budget was only growing linearly.

He humorously noted:

In the year 2054, the entire de-

fense budget will purchase just

one tactical aircraft. This aircraft

will have to be shared by the Air

Force and Navy 3½ days each per

week except for leap year, when

it will be made available to the

Marines for the extra day.

Of course, such a trend becomes a

problem long before the Depart-

ment of Defense gets down to only

one aircraft. That time is now.

RISING COSTS, SHRINKING

QUANTITIES

Rising costs have pushed down

procurement quantities for not only

aircraft but also ships. Furthermore,

shrinking procurement quantities

have the pernicious cyclical eect of

further driving up per-unit procure-

ment costs, as developmental costs

are spread over fewer and fewer

units. This can lead to more cuts in

production numbers.

From 2001 to 2008, the base

(non-war) budgets of the Navy

and Air Force grew 22% and 27%,

respectively, adjusted for ination.

Meanwhile, the number of combat

ships and aircraft in the U.S. inven-

tory declined by 10% for ships and

nearly 20% for aircraft over the

same period.

A number of fac-

tors contributed to this decrease in

numbers despite an overall budget

rise, and this was in part due to a

deliberate choice by the Navy and

Air Force to emphasize quality over

quantity.

But better quality can

only compensate so much.

THE QUALITY OF QUANTITY

AUGUSTINE’S LAW: RISING AIRCRAFT COSTS

OVER TIME

Sources: Marcelle Knaack, Encyclopedia of USAF Aircraft & Missile

Systems; Congressional Budget Oce, Total Quantities and Costs of

Major Weapon Systems Procured, 1974-1993; and DOD: F/A-18E/F SAR

(2012), Air Force FY 2011 Budget Estimate and F-35 SAR (2013).

Per Unit Costs of U.S. Combat Aircraft

(2014 USD in millions)

Year

F-100

F-4

F-15

F-16

F/A-18E/F

$20

$40

$60

$80

$100

$120

$140

$160

$180

$200

1940 1960 1980 2000 2020

F35



F-22

DECLINING AIR COMBAT POWER FROM 20012014

Source: International Institute for Strategic Studies. Includes aircraft in store. Budget data from U.S. Department of Defense.

$20,000

$40,000

$60,000

$80,000

$100,000

$120,000

$140,000

$160,000

$180,000

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

Air Force Base (Non-War) Budget

(2014 USD in millions)

Bombers

Fighters

Air Force base (non-

war) budget

Number of Combat Aircraft

DECLINING NAVAL COMBAT POWER FROM 20012014

Source: International Institute for Strategic Studies. Includes ships in reserve. Budget data from U.S. Department of Defense.

Number of Combat Ships

Navy Base (Non-War) Budget

(2014 USD in millions)

$20,000

$40,000

$60,000

$80,000

$100,000

$120,000

$140,000

$160,000

$180,000

100

150

200

250

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

2011

2012

2013

2014

Carriers

Cruisers

Destroyers

Frigates

Submarines

Amphibious Ships

Navy base (non-war) budget

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

NUMBERS MATTER:

LANCHESTER’S LAW

A standard rule-of-thumb for the

advantage of quantity vs. qual-

ity in military engagements is

Lanchester’s Square Law. Lanches-

ter’s Square Law states that,

all things being equal, having

twice as many units in the fight

actually translates to a fourfold

increase in combat power for

units with aimed-fire weapons.

This is because the numerically

superior force can double up on

attacking enemy units, while the

numerically inferior force can only

attack half of the opposing force

at one time. This is in contrast

to hand-to-hand combat, where

combatants can only attack one

person at a time, and a twofold

increase in numbers translates to

only a twofold increase in combat

effectiveness.

A numerically inferior force can

compensate with greater qualitative

superiority, but a force that is out-

numbered by its opponent 2-to-1

must therefore be four times better

in quality in order to simply match

its opponent. There is, in essence, a

limit to how much qualitative supe-

riority can compensate for smaller

numbers.

As one example, a 2009 RAND study

of a hypothetical U.S.-China air war

over Taiwan highlighted the value

of numbers and the limits of qualita-

tive superiority alone. Analysts con-

ducted a detailed model of aircraft

engagements, accounting for quali-

tative and quantitative advantages.

Even though U.S. ghters were as-

sessed to be far more capable than

Chinese ghters – 27 times better in

the case of the U.S. F-22 – China was

able to launch nearly 800 sorties in

the rst day of ghting and won the

battle.

LANCHESTER’S SQUARE LAW

5 vs. 5

5 vs. 10

Relative combat power is proportional to the square of

the relative sizes of opposing forces. A two-fold numerical

advantage leads to a four-fold increase in combat power.

systems, it should be careful to include loitering air

vehicles like MALDs, not just recoverable aircra.

e Air Force should also be sure to examine the

full range of possible missions including strike and

electronic attack, not merely tactical surveillance,

as small uninhabited aircra are used for today.

SQUADORGANIC CLOSE AIR SUPPORT

One particularly notable use for small uninhabited

air vehicles is the ability to put organic close air

support directly in the hands of ground troops. e

Switchblade is a back-packable, loitering precision-

strike weapon. Weighing only 5.5 pounds, it can be

issued directly to ground troops to carry on patrol.

If engaged, they can launch the Switchblade, use its

onboard sensors to nd enemies, and then destroy

them with its low collateral damage warhead.

A small, loitering precision-strike air vehicle like

the Switchblade is a “reght ending weapon” that

places close air support directly in the hands of

ground troops and makes it instantly available.

Because of its low cost and high value, it can be

purchased in large numbers and issued to ground

troops to provide squad-organic close air support.

Switchblade has been deployed to Afghanistan and

the Army and Marine Corps are procuring addi-

tional systems.

GROUND VEHICLES  ROBOTIC APPLIQUÉ KITS

e Army has thousands of fully functional

ground vehicles such as HMMWVs and M113

armored personnel carriers that will not be used in

future conicts because they lack sucient armor

to protect human occupants. At very low cost,

however, on the order of tens of thousands of dol-

lars apiece, these vehicles could be converted into

robotic systems. With no human on board, their

lack of heavy armor would not be a problem.

is could be done at low cost using robotic appli-

qué kits – sensors and command systems that are

applied to existing vehicles to convert them for

remote or autonomous operation. Robotic appliqué

kits have already been used to convert construction

vehicles into remotely operated Bobcats and bull-

dozers to counter improvised explosive devices.

Applied to existing vehicles, robotic appliqué kits

could give the Army a massive robot ground force

at extremely low cost. e sheer mass of such a

force, and the ability to apply it in sacricial or

suicidal missions, could change how the Army

approaches maneuver warfare.

Uninhabited ground vehicles could be the van-

guard of an advance, allowing robots to be the

“contact” part of a “movement to contact.” Robotic

vehicles could be used to ush out the enemy, ank

or surround them or launch feinting maneuvers.

Uninhabited vehicles could be air-dropped behind

enemy lines on suicide missions. Scouting for tar-

gets, they could be used by human controllers for

direct engagements or could send back coordinates

for indirect re or aerial attacks.

e Army is investigating appliqué kits for cargo

resupply, which will have signicant cost saving

advantages, but not yet for maneuver warfare.

e Army should begin a series of experiments

with uninhabited ground vehicles, akin to the

1940 Louisiana Maneuvers that accompanied the

adoption of the tank, to better understand the

role uninhabited vehicles could play in maneuver

operations.

UNINHABITED ARSENAL SHIPS TO EXPAND

MAGAZINE CAPACITY

A U.S. Aegis ballistic missile defense destroyer

is an amazing piece of machinery. A exible and

mobile oensive and defensive weapons platform, it

can engage and destroy enemy short- and interme-

diate-range ballistic missiles and can strike targets

deep on land with Tomahawk cruise missiles. e

Aegis destroyer’s sole shortcoming is its limited

magazine capacity. A Flight II Arleigh Burke-class

destroyer has 96 vertical launching system (VLS)

cells, a modular system that can be used to carry

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

a mix of oensive and defensive missiles.

VLS

space is limited, and once its missiles are exhausted

the Aegis’s oensive and defensive capabilities are

signicantly reduced.

Uninhabited arsenal ships could be used to

expand destroyer magazine capacity, dramatically

enhancing the oensive and defensive capacity of

existing destroyers. ese could take the form of

uninhabited surface barges or undersea pods. e

commercial shipping industry is already explor-

ing the possibility of uninhabited ships at sea to

reduce costs.

By leveraging this technology, the

Navy could eld additional missile capacity at rel-

atively low cost. Uninhabited arsenal ships could

be built much cheaper than additional destroyers

because they would not be warships. ey would

not have radars, guns and other combat capabili-

ties. ey would simply need large numbers of

VLS cells and reliable communications link to

human-inhabited ships, both for targeting and

safety purposes. Any additional survivability

features would need to be balanced against cost,

because such vessels could be attritable in a ght.

In fact, all things being equal, a large number of

lower-cost missile barges would be preferable to a

single high-cost one, since having to target more

barges would be cost-imposing to an adversary.

Undersea payload modules stocked with missiles

would be even better, since they could be parked

very close to an enemy’s coastline and would be

extremely dicult to detect. On warning, they

could launch missiles or even uninhabited aircra

into the ght.

e Navy has already tested the basic concept

of uninhabited missile boats, demonstrating the

ability of a small uninhabited surface vessel to

launch missiles to intercept enemy swarming small

boats in 2012.

A missile barge loaded with VLS

cells would be a scaled up version of the same

concept. e missile barge is very similar to the

concept of a minimally-manned low-cost “arse-

nal ship,” which was oated in the mid-1990s.

e Navy should begin experiments to scale up

already-demonstrated uninhabited missile boats to

larger, VLS-capable surface and subsurface ves-

sels and begin a study of possible designs. As the

commercial shipping industry begins to explore

uninhabited barges, the Navy may be able to lever-

age commercial-o-the-shelf technologies to lower

cost.

“BILLIONS OF DRONES”

Ultra-cheap 3D-printed mini-drones could allow

the United States to eld billions – yes, billions – of

tiny, insect-like drones. Researchers at Harvard

have developed a technique for 3D-printing

drones cheaply and eectively, without errors, by

the sheet. e “Mobee” drone is manufactured

by a 3D printer in a two-dimensional sheet, and

then pops out of the sheet and folds into a tiny,

bug-sized drone.

e Mobee is tiny and lacks a

power source or sophisticated programming, but

computer processing power is becoming smaller

and faster at an exponential rate. Just as swarms

of insects, which individually are not particularly

intelligent, can nevertheless collectively perform

complicated tasks, a cloud of tiny drones could

similarly be used in novel and inventive ways.

“Smart clouds” of 3D printed drones could ood a

building, locating and identifying enemy combat-

ants and civilians, or could even be airdropped

over a wide area to nd enemy personnel and

materiel. If a useful tiny drone could be manufac-

tured using 3D printing techniques for less than a

dollar apiece, procuring a billion is not out of the

Department of Defense’s reach.

Cost-Exchange Ratio

e concept of deploying large number of unin-

habited systems on the battlespace hinges squarely

on the issue of cost. If such systems cannot be

made cheaply, they cannot be made in large

numbers. But “cheap” is a relative term, as is

“expendable.” How cheap do systems need to be in

order to be useful?

e key metric is not the cost of uninhabited

platforms themselves, but rather the cost-exchange

ratio between adversaries. is ratio is tradition-

ally used in the context of ballistic missile defense,

measuring the aggressor’s marginal cost of over-

whelming enemy defenses as compared to the

defender’s marginal cost of countering the attack.

In general, the cost-exchange ratio can be thought

of as the ratio of the cost of an approach compared

to the cost of its countermeasure. e U.S. military

should consider its investments within the context

of cost-exchange ratios and seek favorable or at

least minimally disadvantageous cost-exchange

ratios. In some cases, an innovation that oper-

ates at an unfavorable cost-exchange ratio but

is less unfavorable than before may still be an

improvement.

Non-material costs and relative cost to an adver-

sary should also be considered. An innovation that

can be countered cheaply may still be advanta-

geous if it forces the enemy to expose himself in a

dangerous way or consumes other scarce resources

for the enemy, such as time or personnel. Similarly,

costs should be considered within the context of

an adversary’s resources. An even cost-exchange or

even a slightly unfavorable one may be a perfectly

acceptable approach if one has deeper pockets than

one’s enemy and is willing to outspend them.

A New Paradigm for Assessing Qualitative

Advantage

e point of building large numbers of lower cost

systems is not to eld forces on the battleeld that

are qualitatively inferior to the enemy. Rather, it

is to change the notion of qualitative superiority

from an attribute of the platform to an attribute

of the swarm. e swarm, as a whole, should be

more capable than an adversary’s military forces.

at is, aer all, the purpose of combat: to defeat

the enemy. What uninhabited systems enable is

a disaggregation of that combat capability into

larger numbers of less exquisite systems which,

individually, may be less capable but in aggregate

are superior to the enemy’s forces.

Disaggregating combat power will not be possible

in all cases, and large (and expensive) vehicles

will still be needed for many purposes. Expensive,

exquisite systems will inevitably be purchased in

small numbers, however, and so where possible

they should be supplemented by larger numbers

of lower-cost systems in a high-low mix. Neither

a cheap-and-numerous nor an expensive-and-few

approach will work in every instance, and U.S.

forces will need to eld a mix of high and low-cost

assets to bring the right capabilities to bear – and

in the right numbers – in future conicts.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

The chief value of mass is that it can

be used to impose costs on adversar-

ies because it forces one’s adversary

to counter large numbers of sys-

tems. The value of mass in a cost-

imposing strategy can be illustrated

with a simple vignette:

Two adversaries, Red and Blue, are

engaged in a technological com-

petition of innovation and counter-

measures. In an unguided muni-

tions regime, both seek to maximize

the rate of re of their weapons and

the density of the barrage landing

on enemy positions. Because the

unguided munitions are inaccurate,

large numbers are needed to maxi-

mize the odds of a successful hit

on an enemy target. Blue develops

guided munitions rst, however,

and has a game-changing advan-

tage.

With guided munitions, Blue is able

to trade large numbers of unguided

weapons for smaller numbers of

higher-cost guided munitions.

This tradeo is a winning strategy

because the guided munitions

have a high probability of kill (Pk),

meaning they have a high probabil-

ity of hitting and destroying their

target. Rather than pour thousands

of unguided munitions at Red, Blue

can invest scarce defense dollars in

a mere handful of guided munitions

that home in on Red and strike their

target. Red is powerless against this

approach.

This works great for Blue until Red

develops guided munitions as well.

Now Red can counter both Blue’s

munitions and, more importantly,

Blue’s power projection platforms

that launch the munitions with

Red’s own guided munitions. This is

a winning strategy for Red because

– in this vignette – there is a funda-

mental asymmetry between Red’s

and Blue’s strategic approaches.

Blue is a naval power attempting

to project power around the globe

far from its home, while Red is a

land power with a large land mass

within which Red can hide mobile

missile launchers and build scores

of dispersed airelds. Blue has many

targets to engage and Red has few.

Blue now has two approaches to

counter this new challenge from

Red. Blue can continue to invest

in fewer numbers of ever-higher

quality assets or buy larger num-

bers of lower-cost and therefore

lower-quality assets. Blue has a xed

amount of money, so every dollar

spent on one type of platform or

munition is a dollar robbed from

another. Let’s consider the value of

each approach:

Higher-quality: Let’s assume

that for its money Blue can buy 1

high-cost, high-quality asset for

every given Red target , with a

Pk of 0.9.

That is, this asset has a

90% probability of achieving a kill

against a Red target.

High-quantity: Alternatively,

for the same amount of money

Blue can invest in 20 lower-cost

and lower-quality assets, each

with a Pk of 0.11. Each individual

weapon has only an 11% chance

of killing a Red target, but the

aggregate odds of one of them

killing a target if twenty are red

is 90%.

From Blue’s perspective, both strate-

gies are equal. They cost the same

and achieve the same eect. In one,

all of Blue’s investment dollars go

to a single high-quality asset. In the

other, Blue resources are spread

over a larger number of lower-cost

assets which, in aggregate, achieve

the same capability. Both are viable

strategies for Blue, but how do they

look from Red’s perspective?

For Red, countering Blue’s high-

quantity approach is much more

dicult if Red’s preferred method of

countermeasure is hit-to-kill inter-

ception. If Blue adopts a high-quali-

ty approach, Red’s only challenge is

to nd a way to hit Blue’s single as-

set. If Blue adopts a higher-quantity

approach, on the other hand, then

Red needs to hit and kill all of Blue’s

assets, even though most of them

will not actually succeed in killing

the Red target.

From Red’s per-

spective, this is a nightmare. Even

though most of Blue’s assets are

not a threat, Red can’t know which

will miss and which will hit, making

the problem of intercepting Blue’s

assets twenty times harder.

Blue’s

choice to disperse combat power

among a large number of assets is

very cost-imposing to Red, since

all of Blue’s assets eectively act as

decoys for the few that get through.

What if Red tried a dierent ap-

proach?

THE COSTIMPOSING VALUE OF MASS

The challenges involved in getting

a kinetic hit-to-kill on all of Blue’s

assets might drive Red to focus

instead on reducing the Pk of all

of Blue’s assets across the board

through the use of decoys or some

kind of wide-area spoong attack.

What is the eect of such an ap-

proach on Blue?

Let’s assume that Red adopts a

countermeasure that reduces the

eectiveness of Blue’s assets by

50%.

For the high-quantity approach,

Blue’s Pk drops from 0.11 to 0.055

for each asset.

In order to get

back up to a 90% probability of a

hit, Blue must launch 41 assets at

Red instead of twenty, or just over

double what was required before

Red degraded Blue’s munitions’

eectiveness.

For the high-quality approach,

Blue’s Pk drops from 0.9 to 0.45

for each asset. In order to get the

total probability of a hit back up

to 90%, Blue must now eld 4

assets instead of one, an increase

of fourfold over its original ap-

proach.

Again, numbers matter. Red’s

countermeasure that reduces the

Pk of any given Blue asset by 50% is

much more cost-imposing to Blue

when Blue relies on high-quality

assets. This makes sense, because

when Blue was using a low-cost,

high-quantity approach, Blue wasn’t

1. See Appendix for a table of Pk values.

relying heavily on the quality of its

assets individually for securing a kill

anyway. Blue’s approach favored

mass and Blue can counter degrad-

ed quality by simply throwing more

mass at the problem.

Are larger numbers of low-cost

assets always the answer? Of

course not. The merits of any given

approach in a specic exchange

depend heavily on the particular

assets at play, their cost, actual Pk,

the cost of countermeasures and

counter-countermeasures and the

cost of any platforms to get them

into the ght. The example of trad-

ing twenty lower-cost assets with

a Pk of 0.11 for one high-cost asset

with a Pk of 0.9 is notional and used

only to illustrate the value of mass.

It is not necessarily indicative of any

specic cost-quantity-Pk tradeo.

All things being equal, however, dis-

persing one’s combat power impos-

es signicant costs on the enemy by

forcing the enemy to counter many

threats, even if individually each of

those threats is less capable.

Even if the cost-quantity-Pk tradeo

for a particular asset favors mass,

the ability to get additional mass to

the ght is essential to the success

of this strategy. If Blue relies on $1

billion power-projection platforms

that can carry only four missiles

each, then Blue ought to go with

higher-quality munitions. If Blue can

eld large numbers of low-cost ar-

senal ships and missile trucks to get

more assets into the ght … Well,

then it is another matter entirely.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

IV. COORDINATION

AND INTELLIGENCE

A large number of uncoordinated uninhabited

systems is not a “swarm;” it is a deluge. A swarm

consists of disparate elements that coordinate and

adapt their movements in order to give rise to an

emergent, coherent whole. A wolf pack is some-

thing quite dierent from a group of wolves.

Ant

colonies can build structures and wage wars, but

a large number of uncoordinated ants can accom-

plish neither. Harnessing the full potential of the

robotics revolution will require building robotic

systems that are able to coordinate their behaviors,

both with each other and with human controllers,

in order to give rise to coordinated re and maneu-

ver on the battleeld.

Swarms in Nature

Swarms in nature are wholly emergent entities that

arise from simple rules. Bees, ants, and termites

are not individually intelligent, yet their colonies

can exhibit extraordinarily complex behavior.

Collectively, they are able to eciently and eec-

tively search for food and determine the optimal

routes for bringing it back to their nests. Bees can

“vote” on new nesting sites, collectively decid-

ing the optimal locations. Ants can kill and move

very large prey by cooperating together. Termites

can build massive structures, and ants can build

bridges or oat-like structures over water using

their own bodies.

COLLECTIVE INTELLIGENCE IS AN EMERGENT

PHENOMENON

ese collective behaviors emerge because of

simple rules at the individual level that lead to

complex aggregate behavior. A colony of ants will,

over time, converge on an optimal route back from

a food source because each individual ant leaves

a trail of pheromones behind it as it heads back

to the nest. More ants will arrive back at the nest

sooner via the faster route, leading to a stronger

pheromone trail, which will then cause more ants

to use that trail. No individual ant “knows” which

trail is fastest, but collectively the colony nonethe-

less converges on the optimal route.

SWARMS USE IMPLICIT AND EXPLICIT

COMMUNICATION

Animals communicate through a variety of

methods, both explicit and implicit. Bees com-

municate the degree of interest over their

potential food site through a “waggle” dance,

while wolves use body language and barks to

communicate within the pack.

Implicit communication also plays a signicant

role. Flocks of birds, schools of sh and herds of

animals do not stay together because of explicit

communication signals between individual ani-

mals, but because each animal keys its movements

o of those around it. Once on the attack, a wolf

pack operates as a synchronized whole because

individual members adapt their behavior based

on that of other wolves.

is is not dissimilar

from military small unit tactics, where “battle

drills” allow a well-trained re team to execute

coordinated maneuvers with little or no explicit

communication among them, once the decision

has been made to execute a particular drill.

A novel and signicant method of communica-

tion between animals is stigmergy, where animals

e battleeld is a scene

of constant chaos. e winner

will be the one who controls

that chaos, both his own

and the enemy’s.

N B

alter their environment and, in so doing, leave

signals for other members of the swarm. An ant’s

pheromone trail is an example of stigmergy, as are

the implicit signals termites leave each other in the

environment as they construct nests.

SWARMS COMMUNICATIONS CAN BE EXPLOITED

BY OTHER ANIMALS

ere are many examples in nature of animals

exploiting swarm communication signals to

deceive members of a swarm, either to hide within

the swarm or to hijack it for their own purposes.

Several animals exploit swarm communication

signals to hide within a swarm, freeloading on

the benets of security that the swarm brings. e

silvershmalayatelura ponerphila lives among nor-

mally aggressive army ants by rubbing itself on ant

larvae and pupae, absorbing their scent. e West

African Rubber Frog, on the other hand, directly

secretes a pheromone that prevents the normally

aggressive stinging ant paltothyreus tarsatus from

attacking it. e frog then lives inside the colony

during the dry season, reaping the benets of the

nest’s humidity and protection from prey.

“Slave-making” ants, by contrast, invert this trick

by fooling other ants into working for their colony.

Slave-making ants raid other rival ant colonies and

steal their larvae, taking them back to the slave-

making ants’ nest and raising them to take care

of the slave-making ants’ workers. Raised their

entire lives in a rival colony, the captured ants are

unaware that they have been hijacked by a rival

species.

e slave-making ant polyergus breviceps takes this

method a step further and, in addition to raiding

larvae, can hijack an entire colony. A polyergus

queen can inltrate a rival colony, kill the queen,

and assume control of the colony as its new queen.

Her ospring are then raised by the hijacked

colony and its workers.

us, a polyergus queen is

able to take control of the entire swarm and use it

for her purposes by lling one key role.

ese examples of animals exploiting commu-

nication signals among members of a swarm are

analogous to spoong and cyber attacks in the

military domain. Swarm security – ensuring that

other members of a swarm can be “trusted,” in

particular any element that serves as a leader – will

be especially important for military swarms.

Robot Swarms Dier from Animal Swarms

in Important Ways

Like ants, termites and bees, simple rules govern-

ing the behavior of robots can lead to aggregate

swarming behavior for cooperative scouting, forag-

ing, ocking, construction and other tasks. Robot

Ants work together to build a bridge with their bodies.

(SHUTTERSTOCK)

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

swarms can dier from those found in nature in

several interesting and signicant ways. Robot

swarms can leverage a mix of direct and implicit

communication methods, including sending

complex signals over long distances. Robot swarms

may consist of heterogeneous agents – a mix of

robots of dierent types or robots working together

to perform a task. For example, the “swarmanoid”

is a heterogeneous swarm of “eye-bots, hand-bots,

and foot-bots” that collectively work together to

solve problems.

Swarm security is an even larger concern for robot

swarms than for animals. Robot swarms have the

potential to fall victim not only to spoong attacks

like those of the West African Rubber Frog, but

also direct cyber attacks that usurp control of an

uninhabited system.

In December 2012, a hacker

demonstrated the ability to take control of a widely

used commercially-available drone by hacking

its unencrypted wi-.

“Swarm intelligence” can

help individual members be resilient against some

forms of cyber attacks. “Voting” mechanisms can

allow members to communicate to one when it has

fallen victim to a spoong attack. For example,

swarm elements could share position information,

allowing some measure of resiliency against GPS

spoong.

At the same time, whole swarms could

potentially fall victim to hijacking if an enemy

is able to spoof the entire swarm as a whole or

assume the role of a central node. Cyber vulner-

abilities are not unique to uninhabited systems, but

the lack of a human on board does introduce addi-

tional vulnerabilities. Human-inhabited systems

can, in principle, be equipped with physical over-

rides to be used in the event of a cyber attack, and

human “common sense” may aord a measure of

resiliency against some forms of spoong attacks.

e most important dierence between animal and

robot swarms is that robot swarms are designed

while swarm behavior in nature is evolved. Swarms

in nature have no central controller or “common

operating picture.” Robot swarms, on the other

hand, ultimately operate at the direction of a

human being to perform a specic task.

More research is needed to leverage the potential

for emergent swarm phenomena. Researchers have

only just begun to understand how simple rules

give rise to complex behavior. Simple robot swarms

have been demonstrated in laboratory settings,

but scientists do not yet have a universal model for

understanding what emergent behaviors will arise

from simple rules.

While swarms in nature rely on emergent behav-

ior for complex tasks, such as those performed by

insects, this may not be necessary for robot swarms

or even desirable if it makes swarm behavior less

predictable as a result. Instead, robot swarms could

leverage cooperative behavior for relatively simple

advantages, some of which are explained below.

Concepts for Military Swarming are Largely

Unexplored

Military applications for swarming are intriguing

but largely unexplored. Examples of ghters employ-

ing swarming tactics date back to Genghis Khan,

but have oen played a less-than-central role in

military conict. In their ground-breaking mono-

graph Swarming and the Future of Conict, John

Arquilla and David Ronfeldt articulate an evolution

of four doctrinal forms of conict across history:

• Melee – Chaotic combat among groups with

individuals ghting non-cohesively

• Massing – Large formations of individuals ght-

ing together in ranks and les, such as the Greek

phalanx

• Maneuver – Multiple formations ghting

together, like the Blitzkrieg, coordinating re

and movement across distances to achieve a

coherent aim across the battleeld

• Swarming – Large numbers of dispersed indi-

viduals or small groups coordinating together

and ghting as a coherent whole

MELEE

MELEE VS. MASS

MASS

In melee ghting, combatants ght as individuals, uncoordinated. Massed formations have the advantage of synchro-

nizing the actions of combatants, allowing them to support one another in combat. Massing requires greater organi-

zation, however, as well as the ability for individuals to communicate to one another in order to act as a whole.

MANEUVER

MASS

MASS VS. MANEUVER

Maneuver warfare combines the advantages of mass with increased mobility. In maneuver warfare, mutually sup-

porting separate massed formations move as independent elements to outank the enemy and force the enemy

into a disadvantageous ghting position. Maneuver warfare requires greater mobility than massing as well as the

ability to communicate eectively between separated ghting elements.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

MANEUVER VS. SWARM

SWARM

MANEUVER

Swarm warfare combines the highly decentralized nature of melee combat with the mobility of maneuver and a

high degree of organization and cohesion, allowing a large number of individual elements to ght collectively.

Swarming has much higher organization and communication requirements than maneuver warfare, since the

number of simultaneously maneuvering and ghting individual elements is signicantly larger.

ese types of warfare require increasingly sophis-

ticated levels of command-and-control structures

and social and information organization. Examples

of all four forms, including swarming, can found

dating to antiquity, but widespread use of higher

forms of warfare did not occur until social and

information innovations, such as written orders,

signal ags or radio communication, enabled

coherent massing and maneuver.

While low-cost uninhabited systems allow a

military to eld large numbers of forces, informa-

tion technology allows them to ght as something

more than an uncoordinated deluge. Instead,

networking and automation let systems cooperate

to ght together as a coherent entity, even while

dispersed at great distance. Swarming as a doctri-

nal concept has the advantage over maneuver in

that it distributes one’s forces across the battleeld,

while still harnessing them to ght as a coherent

whole. While this places greater pressures on one’s

forces for communication and coordination, it also

presents the enemy with a much more formidable

challenge. Rather than ghting against a forma-

tion, the enemy faces an insuppressible collection

of targets that are, seemingly, everywhere and

nowhere at once.

Intelligent Swarms Have Several Advantages

on the Battleeld

Arquilla and Ronfeldt recommend a tactic of

swarming as “sustainable pulsing” where forces

mass res, then disperse and come together again

to attack.

Sean Edwards, in a comprehensive

review of swarming tactics, denes swarming as

“when several units conduct a convergent attack on

a target from multiple axes.”

ese are tactical denitions, however, relating to

but one possible advantage stemming from large

numbers of networked, autonomous elements on

the battleeld coordinating to achieve a coherent

whole. For our purposes, we will dene an intel-

ligent robotic swarm more broadly as a network

of uninhabited vehicles that autonomously coor-

dinate their actions to accomplish a task under

some degree of mission-level human direction. e

ability to make networked, autonomous systems

cooperate has many advantages, including but

not limited to coordinated re and maneuver. Just

a few of these potential advantages are explored

below.

DYNAMIC SELFHEALING NETWORKS

Autonomous uninhabited vehicles can coordi-

nate their actions to form self-healing, responsive

networks for reconnaissance, communications

relay and other activities. DARPA’s Heterogeneous

Airborne Reconnaissance Team (HART) suc-

cessfully demonstrated the ability of a network

of cooperating uninhabited aircra to cover an

area for reconnaissance purposes, autonomously

re-tasking assets to cover areas of interest based

on warghter input.

Similar self-coordinating

networks could be used for ground and maritime

reconnaissance. Uninhabited undersea vehicles,

working together in concert with undersea pods,

could form a reconnaissance network to coopera-

tively identify and track submarines.

Uninhabited

vehicles could autonomously de-mine coastal

areas and beaches and, using simple “anti-social”

communication means to avoid other uninhabited

vehicles, could ensure adequate coverage over a

given area. Intelligent mineelds, conversely, could

shi to re-cover areas that have been de-mined,

presenting enemies with an adaptive threat.

Intelligent networks of uninhabited vehicles with

communications relay payloads could adapt to

network disruptions and shi to cover moving

forces or areas of high bandwidth. When con-

fronted with jamming, networks could shi and

adjust their topology to optimal advantage to avoid

interference.

Networks of loitering uninhabited

aircra could provide ground forces not only with

reconnaissance and communications, but also

Swarming as a doctrinal

concept has the advantage

over maneuver in that it

distributes one’s forces across

the battleeld, while still

harnessing them to ght as a

coherent whole.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

air-mobile resupply, jamming and close air support

all responsive to the needs of warghters on the

ground.

COORDINATED ATTACK AND DEFENSE

Just like wolves in a pack present their enemy with

an ever-shiing blur of threats from all directions,

uninhabited vehicles that can coordinate maneuver

and attack could be signicantly more eective

than uncoordinated systems operating en masse.

Waves of kinetic attack weapons could synchro-

nize their strikes to occur simultaneously or from

multiple directions to saturate and overwhelm

defenses. Decoys, jamming and electronic attack

weapons could autonomously coordinate their

eorts and time them to be optimally advanta-

geous to support kinetic strikes.

For defensive purposes, clouds of uninhabited

vehicles could protect ships or ground troops from

attack, autonomously shiing to interdict potential

threats. e Navy recently demonstrated the ability

of a swarm of autonomous uninhabited surface

vessels to intercept and surround an unknown and

potentially hostile vessel.

Swarms of uninhabited

surface, air and undersea vehicles could provide

early detection and interdiction of threats to U.S.

ships, including from enemy swarming small

boats, and the Navy should rapidly move to inte-

grate this capability into a full program of record.

Coordinating attack or defense could allow more

intelligent target selection, leading to more tar-

gets being eectively engaged. One way this could

occur is if munitions had onboard sensors to

conduct battle damage assessment before attack.

In such a case, they could conrm whether a given

target had already been destroyed by any previous

munitions and, if so, could proceed on to second-

ary targets. Communication between munitions

would thus occur indirectly through a form of

stigmergy. is could expand the total number of

targets that a sortie could successfully take out.

For example, if a missile with a Pk = 0.5 were

launched against a target, four missiles would be

needed to get a 90% assured probability of kill. If,

on the other hand, each missile independently had

the ability to look rst before deciding whether to

strike or move on to secondary targets, then four

missiles would still need to be launched, but 50%

of the time the rst missile would strike the target

successfully and the remainder would move on to

secondary targets. 25% of the time, the rst mis-

sile would successfully strike the primary target,

the second missile would successfully strike the

secondary target, and the remaining two missiles

would move onto tertiary targets.

Sometimes the

swarm would take out three or even four targets.

Sometimes all of the missiles would be unsuc-

cessful. On average, though, the swarm would

successfully take out two targets rather than just

one. us, in this example, the ability for the mis-

siles to operate cooperatively through stigmergy

would double striking capacity.

e reverse is also true. If a swarm of munitions

were sent aer a group of targets and each muni-

tion had a very high probability of kill such that

only one was needed per target, the munitions

could directly communicate in ight to deconict

their targets in order to ensure that multiple muni-

tions were not wasted on the same target.

DISAGGREGATED FUNCTIONALITY FOR LOWER

COST, MORE DIVERSE SOLUTIONS

A cooperative swarm of uninhabited systems can

distribute its functions across a mix of platforms,

allowing more numerous lower-cost systems.

Rather than a single exquisite multi-mission

platform, a swarm can employ a suite of many

low-cost single-mission platforms that are able to

work cooperatively to accomplish the same mis-

sion. In addition to complicating the enemy’s

targeting problem, employing a diverse mix of

heterogeneous assets has the advantage of forc-

ing the adversary to respond to multiple dierent

kinds of threats. Such an approach can also lower

costs by using a “high-low mix” of assets, with a

small number of more expensive assets working in

concert with a large number of lower-cost assets to

solve a problem.

For example, a reconnaissance network might

employ a high-low mix of a few expensive, high-

quality sensors combined with large numbers of

lower-quality sensors. e lower-cost sensors could

be distributed over a wide area to nd targets and

then, upon discovering them, cue a platform with a

more expensive sensor to come conrm the target’s

identity. is could enable wider and more ecient

coverage at lower cost.

Similar approaches could be used for inertial

navigation systems and communications relay.

Inertial navigation systems are used for GPS-

independent navigation, an essential capability in

anti-access environments. e estimated posi-

tion of an inertial navigation unit “dris” over

time, however, leading to position inaccuracy that

grows with time. More expensive, higher-quality

systems can compensate for this problem, as can

regular precise position updates from an outside

source, such as an element outside a GPS-denied

area. A high-low mix of inertial navigation sys-

tems among a network of uninhabited vehicles

can allow one to act as a “navigator” for others,

with other platforms requiring only a capable-

enough system to get them through until the next

position update. Similarly, higher-cost long-range

protected communications systems can be located

on a “communicator” within the swarm, with

other systems passing their communications

through this node.

e capabilities of a single exquisite multi-mission

system can be entirely disaggregated among a

swarm of low-cost vehicles performing the same

functions, but merely as a distributed swarm.

is could include surveillance and reconnais-

sance, electronic attack, decoys and deception,

battle damage assessment and strike. By leveraging

A swarm of uninhabited surface vessels patrols in formation in a demonstration on the James River.

U.S. NAVY

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

existing platforms with modular payloads, such

a swarm could be built and elded at relatively

low cost. Each individual system need not be and

should not be a new, exquisite system in its own

right. Instead, each element of the swarm should

leverage modular design and existing platforms

as much as possible, striving for a simple, low-cost

solution to solve one problem. By being simpler,

they can be built on shorter timelines with lower

technology risk and fewer requirements creep, get-

ting at the underlying conditions behind platform

cost growth.

An example of such an approach can be found in

the MALD and MALD-J air-launched decoy and

jammer. While these platforms do not compete

with the functionality of a multi-mission aircra,

they are able to perform their tailored missions –

deception and jamming – at low cost. e same

concept can be leveraged for an expanded array of

missions, and indeed has already been proposed. A

proposed MALD-V variant would use a modular

design to incorporate a variety of single-mission

payloads, from target acquisition to battle damage

assessment to strike.

Disaggregating functionality from an exquisite

multi-mission system to a family of heteroge-

neous simpler, special-purpose systems also has

other benets. In addition to forcing the adversary

to respond to a range of threats, it can enhance

resiliency against enemy innovations and coun-

termeasures. If a weakness is shown in any one

system, only that discrete element needs to be

replaced or modied, rather than attempting

modications to a large multi-mission system.

Monocultures have inherent risk. Malfunctions,

like the oxygen system problems that grounded

the F-22 eet, or enemy countermeasures or cyber

attacks can have catastrophic consequences in

a monoculture.

at vulnerability may in fact

incentivize adversaries to concentrate their eorts

on attacking that single system.

DISTRIBUTED SENSING AND ATTACK

e ability to distribute sensors over a wide area

has tremendous advantages for both active and

passive sensing and precision geolocation. Multiple

cooperative vehicles can accomplish passive preci-

sion geolocation by triangulating targets. For

example, multiple widely distributed sensors can

locate emitters by comparing the dierences in

time of arrival and frequency due to Doppler shi

from relative movement.

For active detection,

distributed sensors can function like a multi-static

radar, with one sensor emitting a radar pulse and

multiple sensors detecting the reection, allowing

stealthier and higher-quality radar detection.

e same physical principles can also be used for

distributed attack. An array of electromagnetic emit-

ters can deliver focused electronic attack over long

distances. is approach depends on precise syn-

chronization of the relative location and timing of the

emitters, such that the electromagnetic waves arriving

at the target overlap in time and space. If achievable,

however, a distributed approach can deliver more pre-

cise electronic attack at lower power and with fewer

interference problems than blanketing a wide area.

DECEPTION

Swarms of vehicles could be used to deceive an

enemy into thinking a larger vehicle or even an

entire formation is moving through an area.

is could be done through a variety of means,

including generating false acoustic, visual or elec-

tromagnetic signatures. Cooperative decoys could

even be used to generate precisely-timed false radar

returns that create the illusion of a radar track

moving through integrated air defenses.

SWARM INTELLIGENCE

Distributed voting by numerous systems could

also be used to solve problems. Swarm intelligence

could be used to improve target identication, with

multiple sensors providing estimates of a target’s

probable ID and weighing their “votes” based on

their estimated condence.

V. SPEED

Machines cannot yet match human intelligence

in solving ambiguous or novel problems, but they

excel at speed. Automation, particularly among

networked autonomous systems, enables a much

faster pace of military operations. Automation

can be used to assist in processing large amounts

of information quickly, compressing the decision

cycle for human operators. is can result in faster

operations, helping commanders to understand

the battlespace, then adapt and react before their

adversaries. In addition, swarming can delegate

control to autonomous elements reacting immedi-

ately to events on the battleeld, allowing tactical

decision-making closer to the edge of battle.

Finally, removing humans entirely from the sense-

decide-act loop can result in faster reactions that

may be impossible for humans to match.

The Ever-Shifting Swarm

e result could be an accelerated pace and speed

of operations that would be impossible for adver-

saries to match unaided by automation. e sheer

volume of information in the future battlespace

may be impossible for commanders to understand

without automated information processing tools.

Moreover, swarming can present enemies with an

ever-shiing, constantly-adapting challenge. Just

as an adversary is beginning to understand the

threat environment, it changes once again.

In his

explanation of the importance of “getting inside”

an adversary’s observe-orient-decide-act loop, John

Boyd denes the objective of an accelerated action-

reaction cycle:

Goal: Collapse adversary’s system into confu-

sion and disorder by causing him to over and

under react to activity that appears simultane-

ously menacing as well as ambiguous, chaotic or

misleading.

Automated decision-making may not always be as

good as human decision-making, but it need not be

if it is faster, and if that speed leads to a sucient

advantage on the battleeld.

“Flash Wars” and Fragile Stability

An accelerated tempo of operations has the poten-

tial for signicant military advantages, but also

raises serious concerns. Just as the introduction

of automated trading algorithms has led to “ash

crashes,” automation in military crises could

introduce instabilities. e lure of quicker reac-

tion times or merely the fear that other nations

might develop automated weapons systems could

spark an automation arms race. is potential

“gunslinger” quality of automation is exceptionally

dangerous and destabilizing, particularly in cyber-

space where operations move at “net speed.”

ere is a tension between the speed of operations

and the speed of decisions. Automation that might

make sense tactically would be disastrous strategi-

cally if it led to “ash wars.” While militaries will

need to embrace automation for some purposes,

humans must also be kept in the loop on the most

critical decisions, particularly those that involve

the use of force or movements and actions that

could potentially be escalatory in a crisis.

During the Cold War, defense planners faced a

similar problem of “fragile stability,” whereby

vulnerable nuclear arsenals incentivized an enemy

to strike rst. In response, strategists developed

a doctrine of an assured second-strike capac-

ity in order to reduce incentives for a rst strike.

Similarly, strategists today must focus on resiliency

in order to be able to absorb a sudden destabilizing

attack and buy time for decision-makers to under-

stand a crisis before deciding how to respond.

While automation will be essential for some

Speed is the essence of war.

 ,    

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

purposes, it should not take the place of humans

for decisions about when and how to use force.

Humans in the Loop

e potential for autonomous systems to increase

the speed of operations raises challenges for both

policy and strategy. When humans are the weak

link in an accelerated decision cycle, there are

very real operational incentives to delegate actions

directly to machines. Delegation of certain deci-

sions, however, particularly regarding the use of

force, raises challenging issues.

Two particularly notable policy concerns are

the role of autonomy in decisions regarding the

use of force and the role of human cognitive

enhancement technologies to assist humans in

keeping pace with an accelerating battleeld. e

Department of Defense has a policy on the role

of autonomy in the use of force, DOD Directive

3000.09, “Autonomy in Weapon Systems,” signed

in 2012.

State parties to the United Nations

Convention on Certain Conventional Weapons,

which has dealt with “special” weapons in the past

like blinding lasers, took up the issue of increased

autonomy in future weapons for discussion in the

spring of 2014. Discussions on this important topic

should continue, and the United States should con-

tinue to lead in this area.

e need to keep pace with an accelerated tempo

of war also raises dicult issues regarding human

cognitive performance enhancement. Human

performance modication technologies, includ-

ing pharmaceuticals such as Ritalin, Adderall

and other “study drugs,” or other modication

techniques, such as transcranial direct current

stimulation (tDCS), could allow humans to pay

attention, process information and react faster

than would otherwise be possible without these

aids. ese technologies raise dicult legal, policy,

ethical and social issues that must be addressed.

DOD has a responsibility to take care of its service-

members, both by ensuring that they are treated in

an ethical and responsible manner, and also by giv-

ing them access to the best life-saving capabilities.

DOD currently lacks a Department-wide policy on

human cognitive performance enhancement, how-

ever. DOD should take steps to address the policy

issues associated with these technologies in order

to guide research in this sensitive area.

VI. COMMANDING THE SWARM

Swarming models directly imported from nature

are not likely to be ideal for military settings,

since animal swarms lack a central commander.

At the same time, human controllers will not

want to be in a position to micromanage each

element of a swarm, nor will they have the band-

width to do so even if they desired. Human

commanders will need to control swarms at the

mission level, giving overarching guidance, but

delegating a wide range of tasks to autonomous

systems. In the near term, this will entail a shi to

mission-level autonomy and multi-vehicle con-

trol. In the long term, new command-and-control

models are needed to allow humans to employ

large swarms eectively.

Simple Multi-Vehicle Control is Possible with

Even Limited Autonomy

e U.S. military has demonstrated and used

multi-vehicle control, where one human controls

two or more uninhabited vehicles at the same time,

in experiments and limited operational settings.

e Air Force has experimented with rudimentary

control of multiple uninhabited aircra while in

transit, including in limited real-world operations,

although it is not routinely used.

e biggest challenge in adopting multi-vehicle

control is not technical, but rather understand-

ing the cognitive demands placed on the human

operator and how many vehicles can be eec-

tively controlled. ere is no easy answer, and

how many vehicles a single person can manage

depends on the task at hand, the human-machine

interfaces and the level of vehicle autonomy.

Early Air Force concepts for multi-vehicle control

suggested that a pilot could control up to four

vehicles at a time.

is is possible for simple

missions, like stationary surveillance, jamming or

point-to-point transit, even with relatively limited

autonomy. In fact, with greater autonomy, human

operators might be able to control far more than

four aircra at a time, so long as the demands

on the human operator per aircra are relatively

minimal.

If human operators are required to respond to

unanticipated events and make decisions, however,

then there are limits to how many vehicles a person

can eectively control. Switching between tasks

introduces ineciencies, as operators reorient their

situational awareness to each new task. e more

situation-specic the cognitive tasks are, and the

more they dier from each other, the greater the

time lag will be. Multiple overlapping events can

lead to wait times between the emergence of a need

for an operator and his or her ability to respond.

Whether or not these wait times are acceptable will

depend on the mission. For emergencies, human

attention may be needed urgently, but in other

situations a vehicle may be able to loiter until the

operator is able to attend to it. In addition, some

tasks may require an operator to focus his or her

attention solely on a single vehicle for some period

of time, such as following an emerging target

or taking control of a vehicle in an emergency,

while others will more readily lend themselves to

multi-tasking.

Many of these issues can be addressed through

better technology, concepts of operation or train-

ing. Multi-vehicle control architectures should be

networked, so that primary operators can pass o

control of vehicles to other operators in the event

that they need to focus their attention on one

vehicle. From a concept of operations perspective,

No plan survives rst contact

with the enemy.

  

()



Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

there should be a sucient number of operators in

aggregate to allow for slack in the system to adapt

to unanticipated events, so that all operators are

not maxed to their cognitive load in steady-state

operations. In addition, human-machine interfaces

should be designed to assist operators in priori-

tizing multiple competing tasks. For example,

blinking lights might be used to draw attention

to an urgent task, while non-urgent tasks could

be placed in a queue so as not to distract opera-

tors from the task they are presently performing.

Improved automation can also help to reduce the

human task loading. Finally, training, experience

and even psychological orientation may play a

signicant role in operators’ ability to handle mul-

tiple, competing cognitive tasks.

Cooperative Multi-Vehicle Control Enables

More Complex Tasks

Cooperative multi-vehicle control takes this

concept to the next level, with a person tasking a

group of vehicles that then coordinate amongst

themselves to accomplish the task as a swarm.

For example, a human might task a swarm of

missiles with a set of targets, but let the missiles

coordinate among themselves to determine which

missile will hit which target. Or a human might

task a group of vehicles to maintain coverage over

an area, whether for surveillance, communica-

tions relay, electronic warfare or establishing

a defensive perimeter, and the vehicles might

coordinate to determine how best to cover the

area. ese vehicles could exist across multiple

An autonomous rigid hull inatable boat (RIB) participates in a demonstration on the James River as part of a swarm of thirteen uninhabited

autonomous boats.

U.S. NAVY

domains, such as air, sea surface and undersea

vehicles operating collectively with one person

controlling the group.

Concepts for cooperative multi-vehicle control

have been demonstrated in simulations and some

real-world experiments, and many applications

are technically feasible today.

In the summer

of 2014, the U.S. Navy demonstrated the ability

for one operator to control a swarm of thirteen

autonomous uninhabited small boats escorting

a high-value ship through a mock strait transit.

When a potentially hostile vessel was spotted, the

operator tasked the swarm to interdict and sur-

round the vessel, which it executed autonomously.

According to naval researchers in charge of the

experiment, such a concept could be scaled up to

one person controlling twenty or thirty boats at

a time.

e associated manpower savings and

reduction in risk to personnel are tremendous. An

interdiction operation that normally would have

included forty to y sailors closing with poten-

tially hostile actors can be executed by a single

sailor safely removed from harm’s way. Similar

concepts could be used in other domains and for

other missions, such as a swarm of air and ground

vehicles searching over a wide area to nd and

positively identify targets. While the Navy is mov-

ing out in this area for swarming boats, cultural

resistance to multi-aircra control in the Air Force

has hindered progress for air vehicles.

Multi-Vehicle Control Faces Cultural Barriers

to Adoption

Early Air Force experiments with multi-aircra

control led to dissatisfaction with human machine

interfaces and human task loading, including the

inability of pilots to hand over control of their

aircra to other pilots if they needed to focus atten-

tion on a single airplane.

Rather than improve

the technology to allow for networked control and

better interfaces, however, multi-aircra control

was deemed an “unfunded requirement.”

In 2010, then-Defense Secretary Robert Gates

directed the Air Force to develop improved multi-

aircra control interfaces to overcome concerns

about the technology as it existed at the time. His

direction included nearly $50 million in funding.

e Air Force never developed the technology,

however, instead arguing that the multi-aircra

control concept needed to be developed further

rst.

e Air Force is no further along in develop-

ing multi-aircra control today. e Air Force’s

recently released Remotely Piloted Aircra Vector

discusses multi-aircra control, but it is not funded

in the DOD budget.

Privately, Air Force ocials

claim that multi-aircra control is a “decade aer

next” technology. e reality is that the technology

exists today and has been demonstrated in its basic

form by many companies. What will take a decade

or longer on the current trajectory is cultural

acceptance of a model where pilots are not in direct

physical control of only one aircra at a time.

e belief that a human must control only one

aircra at a time comes from applying an exist-

ing paradigm – human-piloted aircra – to a new

technology. Viewing uninhabited air vehicles

through this lens is a choice, however. Military

forces already routinely employ uninhabited air

vehicles that are not directly controlled by human

operators. ey just aren’t called “unmanned

aircra.” ey are called missiles or decoys. Some,

like cruise missiles, y pre-programmed routes.

Others, like homing missiles, are highly autono-

mous and maneuver to targets on their own. And

many can receive new targeting data in ight and

respond to human taskings.

As uninhabited

vehicle technology matures, the lines between

uninhabited aircra, missiles and decoys will

continue to blur. If the U.S. military is to fully

capitalize on the potential of uninhabited systems,

it will need to be willing to change the operational

paradigm and embrace new concepts of operation.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

Large Swarms Require New Command-and-

Control Paradigms

Scaling multi-vehicle control up to large swarms

will require even more fundamental shis in

the command-and-control paradigm. e Naval

Postgraduate School is working on a 50-on-50

swarm vs. swarm aerial dogght, and researchers

at Harvard have built a swarm of over a thou-

sand simple robots working together to create

simple formations.

As the number of elements

in a swarm increases, human control must shi

increasingly to the swarm as a whole, rather than

micromanaging individual elements.

How to exercise eective command-and-control

over a swarm is an area of nascent research.

Possible command and control models, ordered

from more centralized to increasingly decentral-

ized control, include:

• Centralized control, where swarm elements feed

information back to a central planner that then

tasks each element individually.

• Hierarchical control, where individual swarm

elements are controlled by “squad” level agents,

which are in turn controlled by higher-level con-

trollers, and so on.

• Coordination by consensus, where swarm ele-

ments communicate to one another and converge

on a solution through voting or auction-based

methods.

• Emergent coordination, where coordination

arises naturally by individual swarm elements

reacting to others, like in animal swarms.

Each of these models has dierent advantages,

and may be preferred depending on the situation.

While completely decentralized swarms are able to

nd optimal solutions to complex problems, like

how ant colonies converge on the shortest route

for carrying food back to the base, converging on

the optimal solution may take multiple iterations,

and therefore time.

Centralized or hierarchical

planning may allow swarms to converge on opti-

mal, or at least “good enough,” solutions more

quickly, but requires higher bandwidth to transmit

data to a central source that then sends instruc-

tions back out to the swarm. Action by consensus,

through voting or auction mechanisms, could

be used when low bandwidth communications

exist between swarm elements.

When no direct

communication is possible, swarm elements could

still rely on indirect communication to arrive at

emergent coordination, however. is could occur

by co-observation, like how animals ock or herd,

or stigmergic communication by altering the

environment.

DECENTRALIZED SWARMS ARE INHERENTLY

ROBUST AND ADAPTIVE

Centralized control is not always optimal even if

robust communications exist, since detailed plans

can prove brittle amid fast-paced changes to a

battleeld environment. Decentralized control,

either through localized “squad commanders,”

voting-based consensus mechanisms or emer-

gent coordination has the advantage of pushing

decision-making closer to the battleeld’s edge.

is can both accelerate the speed of immedi-

ate reaction and make a swarm more robust to

communications disruptions. Swarms of indi-

vidual elements reacting to their surroundings in

accordance with higher-level commander’s intent

represent the ultimate in decentralized execution.

With no central controller to rely upon, the swarm

cannot be crippled or hijacked in toto, although

elements of it could be. What a decentralized

swarm might sacrice in terms of optimality, it

could buy back in faster speed of reaction. And

swarms that communicate indirectly through stig-

mergy or co-observation, like ocks or herds, are

immune to direct communication jamming.

Hordes of simple, autonomous agents operating

cooperatively under a centralized commander’s

intent but decentralized execution can be devil-

ishly hard to defeat. e scattered airdrop of

SWARM COMMANDANDCONTROL MODELS

Centralized Coordination

Swarm elements communicate with a centralized

planner which coordinates all tasks.

Hierarchical Coordination

Swarm elements are controlled by “squad” level

agents, who are in turn controlled by higher-level

controllers.

Coordination by Consensus

All swarm elements communicate to one another

and use “voting” or auction-based methods to con-

verge on a solution.

Emergent Coordination

Coordination arises naturally by individual swarm

elements reacting to one another, like in animal

swarms.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

paratroopers over Normandy during the D-Day

invasion wrecked detailed Allied plans, but had

the unintended eect of making it nearly impos-

sible for Germans to counter the “little groups of

paratroopers” dispersed around, behind and inside

their formations. Simple guidance like “run to the

sounds of gunre and shoot anyone not dressed

like you” can be eective methods of conveying

commander’s intent, while leaving the door open

to adaptive solutions based on situations on the

ground. e downside to an entirely decentralized

swarm is that it could be more dicult to control,

since specic actions would not necessarily be

predictable in advance.

COMMANDANDCONTROL MODELS MUST

BALANCE COMPETING OBJECTIVES

Choices about command-and-control models for

swarms may therefore depend upon the balance

of competing desired attributes, such as speed of

reaction, optimality, predictability, robustness to

disruption and communications vulnerability.

e optimal command-and-control model for any

given situation will depend on a variety of factors,

including:

• Level of intelligence of swarm elements relative

to complexity of the tasks being performed;

• Amount of information known about the task

and environment before the mission begins;

• Degree to which the environment changes

during the mission, or even the mission itself

changes;

• Speed of reaction required to adapt to changing

events or threats;

• Extent to which cooperation among swarm ele-

ments is required in order to accomplish the task;

• Connectivity, both among swarm elements and

between the swarm and human controllers, in

terms of bandwidth, latency and reliability; and

• Risk, in terms of both probability and conse-

quences, of suboptimal solutions or outright failure.

e best swarm would be able to adapt its

command-and-control paradigm to chang-

ing circumstances on the ground, such as using

bandwidth when it is available but adapting to

decentralized decision-making when it is not. In

addition, the command-and-control model could

change during dierent phases of an operation,

and dierent models could be used for certain

types of decisions.

HUMAN CONTROL CAN TAKE MANY FORMS

Human control over a swarm can take many

forms. Human commanders might develop a

detailed plan and then put a swarm into action,

allowing it to adapt to changing circumstances

on the ground. Alternatively, human command-

ers might establish only higher-level tasks, such

as “nd enemy targets,” and allow the swarm to

determine the optimal solution through central-

ized or decentralized coordination. Or human

controllers might simply change swarm goals or

agent preferences to induce certain behaviors. If

the cognitive load of controlling a swarm exceeds

that of one person, human tasks could be split up

by breaking a swarm into smaller elements or by

dividing tasks based on function. For example,

one human controller could monitor the health of

vehicles, with another setting high-level goals and

yet another approving specic high-risk actions,

like use of force.

Ultimately, a mix of control mechanisms may be

desirable, with dierent models used for dier-

ent tasks or situations. For example, researchers

exploring the use of intelligent agents for real-time

strategy games developed a hierarchical model of

multiple centralized control agents. Squad-based

agents controlled tactics and coordination between

individual elements. Operational-level agents

controlled the maneuver and tasking of multiple

squads. And strategy-level agents controlled over-

arching game planning, such as when to attack.

In principle, cooperation at each of these levels

could be performed via dierent models in terms

of centralized vs. decentralized decision-making or

human vs. machine control. For example, tacti-

cal coordination could be performed through

emergent coordination; centralized agents could

perform operational-level coordination; and

human controllers could make higher-level strate-

gic decisions.

In order to optimize their use of swarms, human

controllers will need training to understand the

behavior and limits of swarm automation in real-

world environments, particularly if the swarm

exhibits emergent behaviors. Human controllers

will need to know when to intervene to correct

autonomous systems, and when such intervention

will introduce suboptimal outcomes.

Basic research on robotic swarms is underway

in academia, government and industry. In addi-

tion to better understanding swarming behavior

itself, more research is needed on human-machine

integration with swarms. How does one convey to

human operators the state of a swarm simply and

without cognitive overload? What information is

critical for human operators and what is irrelevant?

What are the controls or orders humans might

give to a swarm? For example, a human controller

might direct a swarm to disperse, coalesce, encir-

cle, attack, evade, etc.

Or a human might control

a swarm simply by using simulated “pheromones”

on the battleeld, for example by making targets

attractive and threats repellent.

To harness the

power of swarms, militaries will not only need to

experiment and develop new technology, but also

ultimately modify training, doctrine and organi-

zational structures to adapt to a new technological

paradigm.

To harness the power of

swarms, militaries will not

only need to experiment

and develop new technology,

but also ultimately modify

training, doctrine and

organizational structures to

adapt to a new technological

paradigm.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

VII. ENEMY SWARMS

AND COUNTERMEASURES

Many of the game-changing innovations that

enable swarming – low-cost uninhabited systems,

autonomy and networking – are driven by com-

mercial sector, not military, innovation. ey will

be widely available to a range of actors, and many

states and non-state groups may be more eager

to embrace them than the U.S. military, which is

invested heavily in current operational paradigms.

Strategists should not be deceived by the apparent

lack of sophistication of the cheap drones cur-

rently in the hands of non-state groups like Hamas,

Hezbollah or the Islamic State of Iraq and Syria.

Fully autonomous GPS-programmable drones can

be purchased online today for only a few hundred

dollars. Large numbers of them could be used to

eld an autonomous, jam-proof swarm carrying

explosives or even crude chemical or biological

agents. Just as cheap improvised explosive devices

wreaked havoc on U.S. forces operating in Iraq and

Afghanistan, low-cost drones could similarly be

extremely disruptive and cost-imposing. U.S. forces

must begin to think now about how to counter

these threats and in cost-eective ways.

Reversing the Cost Equation

It is not enough merely to nd a way to destroy an

enemy’s drone; it must be done in a cost-eective

manner. If taking out a $1,000 enemy drone

requires a $1 million missile, then every drone shot

down is a win for the enemy, because it imposes

tremendous costs on the defender.

LOW COSTPERSHOT WEAPONS

Developing low cost-per-shot weapons will be criti-

cal to countering enemy swarms. ese consist of

exotic technologies like lasers and electromagnetic

rail guns as well as more traditional technolo-

gies like machine guns. e Navy is currently

developing laser weapons and rail guns , and will

test a laser weapon at sea in 2014 and a rail gun at

sea in 2016.

Lasers and rail guns are appealing

counter-swarm weapons because they are electri-

cally powered and therefore have relatively low

costs for each shot – signicantly lower than a

missile – assuming the power sources are avail-

able. e Navy has already demonstrated the

ability of a laser to shoot down an enemy drone,

although defeating an entire swarm of drones is

a more signicant challenge. Machine guns, like

the sea-based Phalanx and land-based counter-

rocket, artillery and mortar (C-RAM) system are

also eective at defeating incoming projectiles

or drones at low cost. High-energy lasers, if they

could be operationalized successfully, would have

even longer range.

COUNTERSWARM

One method of taking out a swarm could be

with another swarm. As long as the counter-

swarm was cheaper and/or more eective than

the enemy swarm, it could be a relatively low-cost

way to defend against enemy swarm attacks. e

Naval Postgraduate School is currently research-

ing swarm-on-swarm warfare tactics, with the

intent of testing a 50-on-50 aerial swarm ght.

Basic research in swarming tactics will be critical,

as winning in swarm combat may depend upon

having the best algorithms to enable better coor-

dination and faster reaction times, rather than

simply the best platforms.

ELECTRONIC ATTACKS

Electronic warfare countermeasures, such as spoof-

ing, jamming, cyber-weapons and high-powered

microwaves are particularly attractive for counter-

ing swarms since they can, in theory, be applied

over a wide area without requiring one to target

individual swarm elements. If a swarm relies on

communications for its coordination, then jam-

ming or cyber attacks could be quite eective

if they disrupted communications and forced

swarm elements to ght individually, devolving

a swarm ght into a melee. Spoong attacks that

trick a swarm with false data or cyber attacks

that potentially even take control of a swarm are

also appealing ways to counter large numbers of

systems.

Uninhabited systems are not uniquely susceptible

to electronic warfare, cyber attacks and spoof-

ing, but the consequences of some types of attacks

could be greater in some cases.

Because machine

cognition lacks the “common sense” capabilities

of a human, there is a higher risk that the system

might fail badly if manipulated with false data.

Whereas humans can adjust and adapt to new and

unanticipated situations, machines can be “brittle”

if presented with situations for which they were not

programmed, leading to unpredictable outcomes.

Counter-countermeasures

e dicultly in tricking a person with false data

points to one possible safeguard against spoong

or cyber attacks.

Maintaining a human “in the

loop” – not for every swarm action but for particu-

larly high-risk ones, such as target authorization

– is one potential fail-safe. By building in “human

rebreaks” within a swarm’s decision-cycle,

militaries can ensure that even if an adversary

were to be able to insert false data, there would be

limits to what the swarm could do without human

approval.

While humans are not incapable of

mistakes, a human rewall would ensure that there

was at least a common sense check before authoriz-

ing high-risk swarm actions.

At the very least, a

human rewall would ensure that in order for an

adversary to take control of a swarm, the enemy

would have to actually exercise some direct human

control and replicate the human control interfaces,

not merely insert malicious code.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

VIII. CONCLUSION: BUILDING

THE SWARM

A New Paradigm for Technological

Superiority

We need not wait until 2054 when the Department

of Defense will only have sucient dollars to fund

one aircra, split between the Air Force, Navy

and Marine Corps, for Augustine’s Law of rising

costs to take its toll. e crisis in rising costs and

shrinking quantities that Norm Augustine warned

about is here today. Quantities for next-generation

systems are so low that even dramatic qualitative

advantages over adversaries, on the order of 27-to-

1, are not sucient to overcome the challenges

inherent in projecting power at long range and

with limited numbers of assets. e solution is not

to stop modernizing, however, or to retain legacy

force structure at the expense of higher-quality sys-

tems. Instead, we need a new paradigm that allows

the United States to eld high quality systems and

in large quantities. And this strategy must work

in the midst of a constrained scal environment.

Accommodating rising costs through massive

growth in the defense budget in perpetuity is nei-

ther feasible nor responsible.

Distributing functionality from single, exquisite

multi-mission systems to large numbers of lower

cost, less complex systems is one approach to help

address this problem. Because uninhabited sys-

tems have no person on board, survivability can

be balanced against cost, replacing the concept of

platform survivability with swarm resiliency. Large

numbers of systems can be built to be attritable.

If some are destroyed, the remainder can carry

on the mission, allowing graceful degradation of

combat capability rather than risk the catastrophic

loss of a single expensive platform. A large number

of diverse assets also imposes costs on adversaries,

dramatically expanding the number and diversity

of the targets they must strike, and enhances resil-

iency by avoiding monocultures.

Perhaps most importantly, because they would be

less complex than exquisite multi-mission sys-

tems, adopting a distributed approach could begin

to reign in rising costs. Increasing complexity of

systems and, in particular, shiing requirements is

a major factor behind ever-rising platform costs.

is need not be the case, however. Many com-

mercial technologies, including computers and

automobiles, are as complex as modern weapon

systems in terms of numbers of parts and lines of

code, and yet do not face the same challenge of

exponentially rising costs.

100

Economies of scale

are a factor – another advantage of adopting a

cheap-and-many approach – but development

timelines are also a major contributor to the

problem. DOD frequently develops major weapon

systems on twenty or thirty year timelines, which

creates perverse incentives to generate unrealistic

requirements based on unproven technology. Long

development timelines also virtually guarantee

that requirements will change over time as adver-

sary capabilities evolve, which further increases

costs. In the worst cases, even with shiing

requirements, platforms are unable to keep pace

with adversary or commercial sector innovation

and risk being obsolete before they are even elded.

A smarter approach would be to break down

exquisite systems into smaller components. One

typical way in which this is done is to disaggre-

gate modernization across time, building modular

platforms with incremental improvements in each

procurement “block” over time.

101

With only mar-

ginal changes between each “block,” this approach

reduces technology risk and, as a consequence,

cost. Another approach would be to disaggregate

a system spatially into many components, adopt-

ing a family-of-systems approach. is would

consist of a number of single-mission systems

working together to accomplish a task, rather than

a single exquisite multi-mission system. Because

single-mission systems would be less complex than

multi-mission systems, they could be produced

with lower technology risk and at lower cost. In

addition, provided that network architectures are

designed with sucient interoperability up front,

such an approach is inherently modular. Concerns

about size, weight and power that traditionally

bedevil modular design approaches no longer

matter when combat functions are disaggregated

spatially among many platforms. Provided they

can plug into the network, new systems are inher-

ently “plug and play.”

Disaggregating complex multi-mission systems

into a family of lower-cost single-mission sys-

tems has not been particularly appealing to date

since, without automation, human operators were

ultimately needed to control them, either physi-

cally onboard the platform or remotely. People

cost money, and rising personnel costs have placed

steady downward pressure on end-strength for all

of the military Services. In a world where Services

envision having fewer aircra, ships and ground

vehicles because they have fewer people to control

them, highly capable multi-mission systems make

sense.

Autonomous uninhabited systems oer the poten-

tial for a dierent approach. ey can be used

to augment existing human-inhabited systems,

putting additional sensors and missiles into the

ght at relatively low cost. e onboard automation

need not be intelligent enough to replace human

operators entirely, but need merely be sucient

to reduce the cognitive load for a human supervi-

sor such that he or she can control many vehicles

at one time. is breaks the current relationship

between people and platforms, and allows a force

small in personnel to eld and control a very large

force in platforms. Some human-inhabited plat-

forms may be needed forward in the battlespace for

various functions, including to “quarterback” the

ght. ere also may be little benet to removing

humans from very large and expensive systems like

ships or bombers. But augmenting these human-

inhabited assets with larger numbers of lower cost

uninhabited systems may dramatically increase

the ability of those humans to accomplish their

mission.

Human-Machine Teaming

e Army’s manned-unmanned teaming concept

for its aviation assets is an instructive example

of this approach. e Army’s recent decision to

retire its aging Kiowa armed reconnaissance

helicopter without a replacement allowed the

Army to adopt an alternative approach: teaming

the human-inhabited AH-64 Apache helicopters

with its uninhabited MQ-1C Gray Eagle aircra.

e uninhabited Gray Eagles do not on their own

replace every task the Kiowa once performed,

but in aggregate the Gray Eagles and Apaches

do. Not only is this approach less expensive, it

also enables new concepts of operation, since

Gray Eagles can be sent forward to undertake

more dangerous missions without risking human

lives. Gray Eagles also incorporate a high degree

of automation, which reduces the human task

loading for simple missions. As a result, the Gray

Eagle can even be controlled directly from the

Apache’s cockpit.

In this model, human-inhabited and uninhab-

ited systems are both leveraged for their relative

strengths, as are human and machine cognition

and automation. is is not to say that unin-

habited and autonomous systems are without

signicant limitations. An uninhabited vehicle is

a poor choice for demonstrating political will to

an adversary, when one wishes to show the resolve

to suer and die for a piece of terrain. Removing

a person from a vehicle also means removing the

most advanced information processing system

on the planet – the human brain – and placing it

at another location. Cognition for uninhabited

systems depends instead on some combination of

onboard autonomy and a communications link to

ooard human controllers, both of which have

limitations. Communications can be disrupted

and degraded, and even under the best conditions

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

bandwidth today is insucient to convey all of

the sensory information a person can take in

when physically present. Machine intelligence has

limitations as well. While machines exceed human

cognitive capacities in some areas, particularly

speed, they lack robust general intelligence that is

exible across a range of situations. Some deci-

sions, particularly those requiring judgment or

creativity, will be inappropriate for autonomous

systems. ose who can eld a mix of human and

machine abilities, maximizing the advantages of

both, will be best suited to capitalize on the poten-

tial of the robotics revolution.

As uninhabited systems become increasingly

autonomous, this balance of tasks will shi over

time. In some cases, trusting automation will

be dicult. Humans may be unwilling to cede

control for some tasks over to machines. Debates

over autonomous cars are an instructive example.

Human beings are horrible drivers, killing more

than thirty thousand people a year in the United

States alone, or roughly the equivalent of a 9/11

attack every month. Self-driving cars, on the other

hand, have already driven nearly three quarters of

a million miles, including in crowded city streets,

without a single accident.

102

Autonomous cars have

the potential to save literally tens of thousands of

lives every year, yet rather than rushing to put self-

driving cars on the streets as quickly as possible,

adoption is moving forward cautiously.

103

At the

state of the technology today, even if autonomous

cars are far better than human drivers overall,

there would inevitably be situations where the

autonomy fails and humans, who are better at

adapting to novel and ambiguous circumstances,

would have done better in that instance.

104

Even

if, in aggregate, thousands of lives could be saved

with more autonomy, humans tend to focus on

the few instances where the autonomy could fail

and humans would have performed better. Ceding

human control to automation requires trust, which

is not easily given.

Increased autonomy can run into similar obstacles

in military contexts, especially when cultural

issues related to identity compound the issue.

While Army uninhabited aircra incorporate a

high degree of automation, equivalent Air Force

aircra do not, even though they are built by the

same contractor. In fact, Air Force MQ-9 Reapers

do not include automated takeo and landing

functionality, even though the vast majority of

MQ-9 accidents occur on takeo and landing.

105

Automating takeo and landing would require

ceding control, however, changing the relation-

ship of the human controller to the aircra. For

Army soldiers, who see uninhabited aircra as just

another piece of equipment, ceding control is not

problematic. In the Air Force, however, piloting is

central to the collective sense of identity. Tellingly,

the Air Force insists upon identifying uninhabited

aircra as “remotely piloted,” even applying this

term to future aircra which, in principle, ought

to have a high degree of automation.

106

Conversely,

not only does the Army refer to its uninhabited

aircra as “unmanned aircra systems,” the

people controlling them are called “operators,”

not pilots.

107

Terminology aside, the paradigm

that equates “piloting” to direct physical control of

the aircra, rather than human supervision and

mission command, leads to resistance to automa-

tion that could prevent accidents and increase

eciencies.

Culture matters, both to individuals and orga-

nizations. It cannot be blithely swept aside, but

nor can it be accepted when it hinders necessary

change. When existing cultural paradigms prevent

the adoption of new approaches that could have

game-changing results on the battleeld, change is

required. Cultures must adapt. e Army cavalry

is a ne example in this regard. While tanks and

motorized ghting vehicles have long replaced

horses in modern combat, the cavalry ethos lives

on in Army “cavalry” units today. Cavalry sol-

diers honor their heritage with Stetsons and tall

boots for dress uniforms. New cavalry soldiers

must “earn their spurs.” But the denition of

what it means to be a cavalry soldier has evolved.

Similarly, the denition of what it means to be a

“pilot” will change over time. e signicance of

physically controlling an aircra by stick and rud-

der will fade, to be replaced with greater emphasis

on areas where human cognition is still needed,

such as mission-level command and decisions over

the use of force.

While pilots may be the rst to grapple with this

paradigm shi, autonomous systems will raise the

same issues across many military positions, from

truck drivers to tank commanders. As uninhabited

and autonomous systems are increasingly intro-

duced into the force, the skills that we require of

military personnel may change. Physical prowess

for some tasks, like piloting an aircra, driving a

vehicle or ring a rie will be less important in a

world where aircra y themselves, vehicles drive

on their own and smart ries correct for wind,

humidity, elevation and the shooter’s movements

all on their own. e cognitive skills that are most

in demand for humans will change as well, per-

haps in surprising ways. As machine intelligence

becomes increasingly capable, the tasks that are

required of humans will change, to include not

only the tasks that machines do poorly, but also

the supervision of complex autonomous systems in

combat.

108

is places new burdens on the selec-

tion, training and education of military personnel.

The Rapid Pace of Technological

Advancement

DOD will need to move swily to harness the

advantages of swarming if it is to retain its current

fragile lead in military robotics.

109

e technology

that enables swarming is generally not new plat-

forms, vehicles or munitions, but rather improved

autonomy for existing hardware. For some appli-

cations, such as swarming boats, the degree of

autonomy required to enable swarming exists

today. For others, improved autonomy is needed,

and the gulf between what is possible today and

what is needed for some applications can be quite

large. However, the underlying technologies that

enable more advanced autonomy, such as improved

computer processing power, are advancing at an

exponential rate. As a result, many information-

based technologies that may have seemed like

science ction only a few years ago, like “smart”

glasses and watches, self-driving cars or bionic

exoskeletons, exist today.

110

e rate of their dif-

fusion into our highways, homes and businesses

is a product of price, business models, market

availability and legal and policy issues, but the

capabilities exist today.

Likewise, the ability to eld fully autonomous,

cooperative vehicles and munitions may be real-

ized sooner than anticipated. Many swarming

applications have already been demonstrated in

simple form. Beyond the military domain, there

are strong commercial incentives for invest-

ments in cooperative robotic systems, given their

potential to develop solutions for area coverage,

self-healing networks, optimal transport and other

tasks. DOD may have to import into the defense

sector swarm algorithms rst developed for non-

defense applications, reversing the traditional

paradigm of DOD exporting advanced technology

into the commercial space.

111

Whether the U.S. military successfully capitalizes

on swarming’s potential will depend upon bureau-

cracy and culture. To tap into commercial sector

innovation in robotics, DOD will need to lower the

barriers to entry that currently exist for non-tradi-

tional defense companies and make itself a more

attractive customer, or risk freezing itself out of an

important market. DOD’s sluggish requirements

and acquisition process is also a strategic risk.

If DOD continues to develop “next-generation”

weapon systems on thirty-year timelines, it will be

hard-pressed to maintain the most cutting-edge

computer hardware and soware.

112

Shorter acqui-

sition timelines and more modular system designs

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

are needed. e cultural lens through which tech-

nology is viewed also matters. When acquisition

challenges are combined with a desire to “go slow”

in areas where automation raises uncomfortable

cultural issues, the U.S. military may nd itself at

risk of falling behind.

The Human Element

Cultural resistance to robotic systems oen stems

from a perception that they are replacing humans,

and terminology that refers to robotic systems as

“unmanned” can feed this perception. e reality,

however, is a future of human-machine teaming.

Many of the tasks humans perform in warfare will

change, but humans will remain central to war,

for good or ill. e introduction of increasingly

capable uninhabited and autonomous systems on

the battleeld will not lead to bloodless wars of

robots ghting robots, with humans sitting safely

on the sidelines. Death and violence will remain

an inescapable component of war, if for no other

reason than that it will require real human costs

for wars to come to an end. Nor will humans be

removed from the battleeld entirely, telecom-

muting to combat from thousands of miles away.

Remote operations will have a role, as they already

do in uninhabited aircra operations today, but

humans will be needed forward in the battlespace,

particularly for command-and-control when long-

range communications are degraded.

Even as uninhabited and autonomous systems

play an increasing role on the battleeld, it is still

humans who will ght wars, only with dierent

weapons. Combatants are people, not machines.

Technology will aid humans in ghting, as it has

since the invention of the sling, the spear and

the bow and arrow. Better technology can give

combatants an edge in terms of stando, surviv-

ability or lethality, advantages that combatants

have sought since the rst time a human picked

up a club to extend his reach against an enemy.

But technology alone is nothing without insight

into the new uses it unlocks. e tank, radio and

airplane were critical components of the blitz-

krieg, but the blitzkrieg also required doctrine,

organization, concepts of operation, experimenta-

tion and training to be developed successfully. It

was people who developed those concepts, who

draed requirements for the technology, who

restructured organizations and rewrote doctrine

and who ultimately fought. In the future, it will

be no dierent.

War will remain a clash of wills. To the extent that

uninhabited systems allow an actor to reduce the

costs of war, they can be a major advantage. ose

who master a new technology and its associated

concepts of operation rst can gain game-changing

advantages on the battleeld, allowing decisive

victory over those who lag behind. But techno-

logical innovation in war can be a double-edged

sword. If this advantage erodes a nation’s willing-

ness to face squarely face the burden of war, it can

be a detriment. e illusion that such advantages

can lead to quick, easy wars can be seductive,

however, and those who succumb to it may nd

their illusions shattered by the unpleasant and

bloody realities of war.

113

Uninhabited systems

can lead to greater stando from the enemy, but

the millennia-long evolution of weapons and

countermeasures suggests that such weapons will

proliferate: no innovation leaves its user invulner-

able for very long. Similarly, automation has the

potential to accelerate the pace of warfare, but not

necessarily in ways that are conducive to the cause

of peace. An accelerated tempo of operations may

lead to combat that is more chaotic, but not more

It is still humans who will

ght wars, only with dierent

weapons.

controllable. Wars that start quickly may not end

quickly.

Uninhabited and autonomous systems raise chal-

lenging operational, strategic and policy issues, the

full scope of which cannot yet be seen. e nations

and militaries that see furthest into the future to

anticipate these challenges and prepare for them

now will be best poised to succeed in the warght-

ing regime to come.

Conclusion

e past decade of conict has seen the intro-

duction of uninhabited systems in warfare in

important ways, saving lives and money, but their

use has to-date been conned largely to niche

roles. is is merely the precursor to a larger shi

in warfare where large numbers of autonomous

uninhabited systems play signicant roles on the

battleeld.

Autonomous and uninhabited systems have

the potential to give tremendous advantages to

actors who gure out how best to employ them.

As detailed in “Robotics on the Battleeld Part

I: Range, Persistence and Daring,” uninhabited

systems can operate with longer endurance and

therefore greater range and persistence. is is

particularly important in countering anti-access

threats where long-range missiles threaten U.S.

ships and bases. Uninhabited systems can also

enable more daring concepts of operation, allowing

commanders to take risks with uninhabited assets

that they would not with human-inhabited ones.

Large numbers of uninhabited systems can

bring greater mass onto the battleeld, and with

it greater resiliency and diversity. Cooperative,

autonomous systems can operate as self-healing

networks and self-coordinate to adapt to events as

they unfold. And automation can accelerate the

pace of battle, compressing decision cycles and

constantly altering the adversary’s threat picture

before he can respond.

For actors who are able to harness the advantages

of uninhabited and autonomous systems, their

forces will be able to operate with greater:

• Range and persistence

• Daring

• Mass

• Coordination and intelligence

• Speed

In aggregate, these advantages will lead to the evo-

lution from today’s reconnaissance-strike networks

to tomorrow’s reconnaissance-strike swarm.

Perhaps most signicantly, the underlying technol-

ogy that will enable these innovations does not

stem from secret U.S. defense labs, but in many

cases will be widely available. Moreover, much

of the technology that will enable autonomous

operations and swarming is better algorithms and

soware, not necessarily new platforms. ere is

an urgent need to innovate faster than adversaries,

and to discover the best ways of employing swarms

rst. To do so, the United States must invest in a

robust plan of experimentation, prototyping, and

iterative concept and technology development.

Where these technologies raise challenging policy

concerns, such as increased automation in the use

of force or human cognitive performance enhance-

ment to keep pace with faster machines, the United

States should grapple forthrightly with the issues

and cra sensible policy guidelines to guide tech-

nological development.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

IX. RECOMMENDATIONS

Elements of the Department of Defense are

conducting experiments on swarms and other

applications of uninhabited and autonomous

systems, but DOD currently lacks a comprehensive

plan to take full advantage of their potential.

Developing the technology alone is not sucient,

as the truly game-changing innovations come from

the ways in which a new technology is ultimately

used. e best path forward for developing these

applications is an iterative process of experimenta-

tion with new technologies and concepts, which

then informs further technology development.

Simulations can be useful, but concepts of opera-

tion ultimately must be tested in the eld. Actual

experimentation with real users also can solicit

new ideas, as well as feedback on what avenues

for research are promising, or where a concept or

technology does not work well.

e chief stumbling block in DOD is the “val-

ley of death” between cutting edge research and

development and formal programs of record.

rough organizations like the Defense Advanced

Research Projects Agency (DARPA), Oce of

Naval Research (ONR) and other labs, DOD does

an excellent job of undertaking high-risk / high-

reward proof of concept research. Transitioning

new, promising concepts to actual funded

DOD programs, however, is uneven at best.

Experimentation and technology development can

help bridge the gap by clarifying what is possible,

what is promising and what is not.

To sustain the U.S. military’s current, but fragile,

lead in robotics:

The Oce of the Secretary of Defense

should:

• Undertake a study on total lifecycle costs for

uninhabited systems, including the potential for

automation to reduce costs by reducing the need

to train operators to physically control vehicles.

e study should focus particular attention on

aircra, where pilot training costs run high.

• Undertake a study on swarming platforms to

examine the potential for low-cost uninhabited

systems to impose costs on adversaries. e study

should include an evaluation of platform surviv-

ability, total cost, amount of vehicles elded per

dollar and costs to adversaries to respond.

• Investigate the potential for uninhabited systems

to increase resiliency and reduce costs by disag-

gregating complex systems into a larger number

of smaller, simpler systems.

• Ensure future military systems are built with

modular designs and open architectures to allow

upgrades and plug-and-play interoperability into

a family of systems.

• Fund a multi-year series of experiments in coop-

erative multi-vehicle control and swarming.

• Establish a Defense Robotics Systems Oce,

directly reporting to the Deputy Secretary of

Defense, to coordinate ongoing eorts on unin-

habited systems across the Department.

114

• Undertake a comprehensive policy review of

human cognitive performance enhancement

technologies.

• Continue to lead in international discussions on

autonomy in weapon systems.

The Joint Sta should:

• Ensure requirements for all new programs are

written so as not to exclude uninhabited or

autonomous solutions or partial solutions as part

of a family of systems.

• Include cost – and not only platform costs but

also total lifecycle costs – as a factor in balancing

new program requirements.

• Ensure that lessons learned from experiments

regarding uninhabited and autonomous sys-

tems are centrally collected and widely shared

throughout the Department.

The Navy should:

• Build an experimental prototype of an unin-

habited missile barge that can demonstrate the

ability to remotely control and launch missiles

from a large uninhabited vessel.

• Build a proof-of-concept demonstration of an

undersea payload module to exploit U.S. sanctu-

ary undersea.

• Move aggressively to eld autonomous swarming

defensive boats to protect U.S. ships from enemy

fast attack cra. is should include further

experimentation to rene concepts of operation,

a rapid elding initiative to equip combatants in

high-risk areas like the Straits of Hormuz and a

program of record for outtting all Navy surface

combatants with optionally-manned small boats

that can operate as a defensive swarm.

• Conduct a series of further experiments in multi-

domain swarms of air, surface and subsurface

vehicles for a variety of missions.

• Sustain development of low cost-per-shot coun-

ter-swarm weapons such as high-energy lasers

and electromagnetic rail guns.

The Air Force should:

• Investigate the potential for low-cost swarming

uninhabited air vehicles, including expendable

or non-recoverable systems such as missiles

or decoys, to conduct a variety of missions

including suppression/destruction of enemy air

defenses, reconnaissance, battle damage assess-

ment and electronic warfare.

• Conduct an analysis of alternatives of lower-

cost uninhabited aircra to supplement existing

manned aircra with additional sensors and

missiles, such as an uninhabited “missile truck.”

• Fund development of improved multi-aircra

control interfaces for existing uninhabited

aircra.

• Conduct a series of experiments in human con-

trol over large numbers of swarming air vehicles.

The Army and Marine Corps should:

• Develop a concept of operations for using appli-

qué kits for ground convoy operations and an

associated program of record.

• Conduct a series of modern day “Louisiana

Maneuver” experiments on “robotic wingman”

ground robots for long-range scouting and

maneuver operations, in order to inform further

technology development and requirements for an

eventual program of record.

• Conduct a series of experiments on swarm-

ing uninhabited air vehicles for persistent

surveillance, close air support, aerial resupply

and communications relay to support ground

maneuver forces.

• Include ground robotics as part of the set of pos-

sible solutions as part of a family of systems for

all future programs, such as a light airborne tank

or new ground combat vehicle.

The Marine Corps should:

• Conduct experiments on amphibious swarm-

ing robots for reconnaissance and counter-mine

operations to clear beaches ahead of an amphibi-

ous assault.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

ENDNOTES

1. Chuck Hagel, “Defense Innovation Days Opening Keynote,” (Southeastern

New England Defense Industry Alliance, Newport, September 3, 2014),

http://www.defense.gov/Speeches/Speech.aspx?SpeechID=1877. Deputy

Secretary of Defense Bob Work has also expounded on this issue: Bob Work,

“National Defense University Convocation,” (National Defense University,

Washington, August 5, 2014), http://www.defense.gov/speeches/speech.

aspx?speechid=1873.

2. T.X. Hammes has made a similar, and compelling, argument: T.X. Hammes,

“The Future of Warfare: Small, Many, Smart vs. Few and Exquisite?”

Warontherocks.com, July 16, 2014, http://warontherocks.com/2014/07/

the-future-of-warfare-small-many-smart-vs-few-exquisite/#_).

3. Department of Defense, “Air-Sea Battle: Service Collaboration to Address

Anti-Access and Area Denial Challenges,” (Air-Sea Battle Oce, May 2013).

4. Paul Scharre, “Robotics on the Battleeld – Part One: Range,

Persistence and Daring,” (Center for a New American Security, May

2013), http://www.cnas.org/sites/default/les/publications-pdf/CNAS_

RoboticsOnTheBattleeld_Scharre.pdf.

5. David Klein, “US Department of Defense 2015 Budget Analysis,” Auvsi.org,

May 2, 2014, http://www.auvsi.org/Mississippi/blogs/david-klein/2014/05/02/

us-department-of-defense-2015-budget-analysis.

6. Paul Scharre, “How to Lose the Robotics Revolution,” Warontherocks.com,

July 29, 2014, http://warontherocks.com/2014/07/how-to-lose-the-robotics-

revolution.

7. Michael C. Horowitz, “The Looming Robotics Gap,” ForeignPolicy.

com, May 5, 2014, http://www.foreignpolicy.com/articles/2014/05/05/

the_looming_robotics_gap_us_military_technology_dominance.

8. Robert O. Work and Shawn Brimley, “20YY: Preparing for War in the Robotic

Age” (Center for a New American Security, January 2014), 10-19, http://www.

cnas.org/20YY-Preparing-War-in-Robotic-Age; Barry Watts, “The Evolution

of Precision Strike” (Center for Strategic and Budgetary Assessments, August

2013), http://www.csbaonline.org/publications/2013/08/the-evolution-of-

precision-strike/; Barry Watts, “Six Decades of Guided Munitions and Battle

Networks: Progress and Prospects” (Center for Strategic and Budgetary

Assessments, March 2007), http://www.csbaonline.org/publications/2007/03/

six-decades-of-guided-munitions-and-battle-networks-progress-and-

prospects/; and Wayne P. Hughes Jr., Fleet Tactics and Coastal Combat

(Annapolis, MD: Naval Institute Press, 2000), 285.

9. Paul Kennedy, The Rise and Fall of Great Powers (New York: Random House,

1987), 353-354.

10. The standard metric for weighing the value of quantitative advantages of

aimed-re weapons is Lanchester’s Square Law, which states that the military

advantage of increased numbers increases with the square of the combat

ratio. So, for example, a two-fold superiority in numbers actually translates

to a four-fold military advantage. This is because aimed-re weapons can

focus their attacks, bringing all repower to bear at the same time. This is in

contrast with the linear-scaling advantage in additional numbers in an era

of hand-to-hand combat, where phalanxes of ghters could only engage one

person at a time. Under Lanchester’s Square Law, German tanks would have

had to have been nine times better than Allied tanks to compensate for their

three-fold numerical disadvantage.

11. Kennedy, The Rise and Fall of Great Powers, 356.

12. William J. Perry, “Technology and National Security: Risks and

Responsibilities,” April 7-8, 2003, http://stanford.edu/dept/france-stanford/

Conferences/Risk/Perry.pdf.

13. The United States suered 148 battle deaths during the war with 210

coalition partners killed. See Patrick Cooper, “Coalition deaths fewer than

in 1991,” CNN.com, June 25, 2003, http://www.cnn.com/2003/WORLD/

meast/04/17/sprj.irq.casualties/. Estimates of Iraqi military casualties vary

wildly, from roughly 1,000 to over 100,000. For a brief overview of the range

of estimates and associated debate, see Jack Kelly, “Estimates of deaths in rst

war still in dispute,” Post-Gazette, February 16, 2003, http://old.post-gazette.

com/nation/20030216casualty0216p5.asp. For the purposes of calculating

casualty ratios, we estimate 12,000 Iraqi military killed based on the Gulf War

Air Power Survey, yielding a ratio of approximately 33:1. Thomas A. Keaney

and Eliot A. Cohen, “Gulf War Air Power Survey Summary Report,” Washington,

DC 1993, 249, http://www.afhso.af.mil/shared/media/document/AFD-100927-

061.pdf.

14. Tom Clancy and Chuck Horner, Every Man a Tiger (New York: Berkley Books,

1999), 499-500.

15. Department of Defense, “Air-Sea Battle: Service Collaboration to Address

Anti-Access and Area Denial Challenges.”

16. John Stillion and Scott Perdue, Air Combat Past, Present, and Future,

RAND Corporation, August 2008, http://www.docstoc.com/docs/42891479/

Air-Combat-Past-Present-and-Future. The study generated a signicant deal

of controversy, not necessarily for the numerical analysis outlined here but

because of derogatory statements in the brief about the performance of the

F-35 Joint Strike Fighter. Adding additional color, the study became associated

with the phrase that U.S. forces were “clubbed like baby seals.” Eventually,

RAND had to issue a clarication. Graham Warwick, “UPDATED: F-35 Criticisms

– RAND claries,” Aviation Week, September 25, 2008.

17. This scenario obviously includes a number of assumptions, some of which

are questionable and some of which are clearly unrealistic, but used to

simplify the analysis. It assumes that Chinese ghters do not turn and run,

even after suering heavy losses. It ignores any possible dogghting kills by

F-22s using guns against Chinese ghters (the F-22 has superior stealth and

maneuverability). And it assumes that U.S. tankers and surveillance aircraft

do not begin to ee immediately when Chinese ghters are seen. The analysis

also assumes, however, that U.S. missiles have a probability of kill (Pk) of 1.0

and Chinese missiles have a Pk of 0. More reasonable Pk assumptions would

tilt the simulated ght further in China’s favor.

18. DARPA, “Joint Unmanned Combat Air Systems,” http://archive.darpa.

mil/j-ucas/index.htm.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

19. Norman R. Augustine, Augustine’s Laws (American Institute of Aeronautics,

1984).

20. U.S. Department of Defense, National Defense Budget Estimates for FY

2015, Table 2-1, http://comptroller.defense.gov/budgetmaterials.aspx.

21. International Institute for Strategic Studies, The Military Balance (2001)

and International Institute for Strategic Studies, The Military Balance (2008).

22. In the latter half of the twentieth century, shipbuilding costs rose on an

average annual basis by 7.4% for nuclear aircraft carriers, 9.8% for attack

submarines, 10.7% for surface combatants, and 10.8% for amphibious ships.

Mark V. Arena et al, Why Has the Cost of Navy Ships Risen? (Washington: The

RAND Corporation), 2006. For more insight on the sources behind cost growth

for aircraft, see Mark V. Arena et al, Why Has the Cost of Fixed-Wing Aircraft

Risen? (Washington: The RAND Corporation), 2008.

23. Lanchester’s Law is a very rough rule of thumb given for illustrative

purposes. In a precision-guided weapons exchange, models focusing on the

probability of kill are more accurate. Nevertheless, the overall point about

a limit on how much better quality can compensate for reduced quantity

remains valid.

24. David A. Shlapak et al., A Question of Balance (Washington: The RAND

Corporation), 2009, 65-67, t http://www.rand.org/pubs/monographs/MG888.html.

25. “Miniature Air Launched Decoy (MALD),” Raytheon.com, http://www.

raytheon.com/capabilities/products/mald/.

26. The MALD is air-deployable from a C-130 cargo plane. See “Miniature Air

Launched Decoy (MALD),” youtube.com, September 21, 2012, http://www.

youtube.com/watch?v=OmG5Q4i5R3s.

27. DARPA, “Network of unmanned undersea platforms would assist

manned vessels,” August 22, 2013, http://www.darpa.mil/NewsEvents/

Releases/2013/08/22a.aspx.

28. AeroVironment, “Switchblade,” https://www.avinc.com/uas/adc/

switchblade/.

29. Andrew Tarantola, “America’s kamikaze drone makes the skies

way less friendly,” Gizmodo, September 5, 2013, http://gizmodo.com/

americas-kamikaze-drone-makes-the-skies-way-less-frien-1227821895.

30. “US military bringing a switchblade to a gun ght,” Defense

Industry Daily, August 29, 2013, http://www.defenseindustrydaily.com/

us-army-brings-a-switchblade-to-a-gun-ght-07071/.

31. For example, see QinetiQ, “Robotic appliqué kit,” https://www.qinetiq-na.

com/products/unmanned-systems/robotic-controller-kit/.

32. FAS Military Analysis Network, “DDG-51 Arleigh Burke – class,” fas.org,

http://fas.org/man/DOD-101/sys/ship/ddg-51.htm.

33. Adam Leach, “Are Unmanned Cargo Ships on the Horizon?” Ship-

tTechnology.com, May 13, 2014, http://www.ship-technology.com/features/

featureare-unmanned-cargo-ships-on-the-horizon-4262804/.

34. Tamir Eshel, “US Navy Tests Rafael Spike Missiles on Unmanned

Vessels,” Defense-Update.com, October 31, 2012, http://defense-update.

com/20121031_us-navy-tests-rafael-spike-missiles-on-unmanned-vessels.

html.

35. “Missile barge program could swamp carriers,” Baltimore Sun,

3 September 1995, http://articles.baltimoresun.com/1995-09-03/

news/1995246035_1_cruise-missiles-carrier-warships. “Revisiting the arsenal

ship,” New Wars, October 6, 2007, http://newwars.blogspot.com/2007/10/

revisiting-arsenal-ship.html.

36. “Printing drones by the sheet (or how we get to tens of billions of drones

by 2020),” Global Guerrillas, February 16, 2012, http://globalguerrillas.

typepad.com/globalguerrillas/2012/02/printing-drones-by-the-sheet.html.

37. For reference, historical Pk values for various air-to-air missiles: AIM-7

(estimated pre-Vietnam) = 0.7; AIM-7 (actual in Vietnam) = 0.08; AIM-9

(estimated pre-Vietnam) = ~0.65; AIM-9 (actual in Vietnam) = 0.15; AIM-9L

(actual in Falklands) = 0.73 (19 kills for 26 missiles red); AIM-9M (actual in

Desert Storm) = 0.23 (11 kills for 48 missiles red); AIM-120 (actual to-date) =

~0.59 (10 kills for 17 missiles red in combat). Source: John Stillion and Scott

Perdue, Air Combat Past, Present, and Future, slides 19-28.

38. Total probability of a hit for a salvo is given by the following formula:

Probability of hit = 1 – (1 – Pk)^N where Pk is the probability of kill for a

single munition and N is the number of munitions red in a salvo.

39. Of course, reducing the number of Blue’s assets has some value for Red,

even if Red does not eliminate all of them. Partially reducing Blue’s salvo

reduces the likelihood of Red’s target being taken out, even if it does not

eliminate it entirely.

40. This is not the case for unguided munitions, for example in the case of

Hamas’ rocket attacks against Israel. Because Hamas uses unguided rockets,

Israel can calculate the rockets’ trajectory and determine which are likely

to strike populated areas, and therefore only target those that are threats.

Guided munitions that home in on targets move, however, considerably

complicating the problem of distinguishing which will miss.

41. This does not at all consider the critical question of “shot doctrine,” or how

many munitions Blue should shoot before stopping to assess whether the

target has been taken out and whether or not to shoot more munitions. In

an ideal scenario, one would have both the time and the means to “shoot-

look-shoot,” that is, re one munition, observe whether it has taken out the

target, and only re a second munition if it has not. This would maximize the

eectiveness of each munition and ensure that none were wasted against

already destroyed targets. In practice, time is a factor, as is the practical

matter of actually conducting a real-time battle damage assessment. In some

situations it is more advantageous to shoot two munitions, then observe, then

shoot again if necessary (shoot-shoot-look-shoot). Other combinations are

also possible. Suce to say, shot doctrine plays a signicant role.

42. Some researchers distinguish between “swarming” and “teaming,” only

using the term swarming to refer to cooperative behavior with large numbers

of agents and where the agents themselves are not particularly intelligent.

In this report, we use the term “swarming” to cover cooperative behavior

regardless of the number of agents or their degree of individual intelligence.

Rather, the term swarm is used to refer to cooperative behavior of a number

of individual agents where their cooperation leads to a whole greater than the

sum of its individual parts.

43. For a vivid illustration, see this video of a pack of wolves ghting a grizzly

bear over a kill, “Wild Kingdom – Wolves vs. Grizzly,” https://www.youtube.

com/watch?v=jY7Xmt4HzV0.

44. For an excellent overview of animal swarming, see Eric Bonabeau, Guy

Theraulaz, Marco Dorigo, Swarm Intelligence: From Natural to Articial

Systems (New York: Oxford University Press, 1999).

45. Matt Soniak, “3 sneaky chemical tricks used by animals,”

Mental_Floss, March 3, 2014, http://mentaloss.com/

article/55203/3-sneaky-chemical-tricks-used-animals.

46. Rumsaïs Blatrix, Claire Sermage, “Role of Early Experience in Ant

Enslavement: a Comparative Analysis of a Host and a Non-Host Species,”

FrontiersinZoology.com, August 2, 2005, http://www.frontiersinzoology.com/

content/2/1/13.

47. Howard Topo, Ellen Zimmerli, “Colony Takeover by a Socially Parasitic

Ant, Polyergus Breviceps: the Role of Chemicals Obtained During Host-Queen

Killing,” The British Journal of Animal Behaviour, 46 no. 3 (September 1993),

479-486.

48. “Swarmanoid: towards humanoid robotic swarms,” http://www.

swarmanoid.org/.

49. The natural world does have examples of direct cognitive hijacking, akin

to cyber attacks. The protozoan parasite toxoplasma gondii, or “toxo”, alters

the behavior of mice it infects by making them less afraid of cats. Once the

mice are eaten, toxo reproduces in the cat’s intestines and infects other mice

through cat feces. The “zombie ant” fungus ophiocordyceps unilateralis is

even more inventive. It infects ants’ brains and directs them to very specic

spots in the rain forest that are the optimal humidity and temperature for

fungal growth. The infected ants then climbs to the underside of a leaf of a

particular height o the ground on the northern side of the plant and bites

down. The ant’s jaws lock in place in a “death grip” and the ant dies. Then the

fungus grows stalks from the ant’s body, which it uses to rain down spores on

other ants passing by. Matt Simon, “Absurd Creature of the Week: The Zombie

Ant and the Fungus That Controls Its Mind,” wired.com, September 13, 2013,

http://www.wired.com/2013/09/absurd-creature-of-the-week-zombie-ant-

fungus/. For a delightful overview of some of the more colorful examples in

nature, see Ed Young, “Suicidal crickets, zombie roaches, and other parasite

tales,” TED, March 2014, http://www.ted.com/talks/ed_yong_suicidal_

wasps_zombie_roaches_and_other_tales_of_parasites#.

50. Hack a drone yourself using SkyJack freeware: Samy Kamkar, “Skyjack,”

samy.pl, December 2, 2013, http://samy.pl/skyjack/.

51. Sharing position data from inertial navigation systems among vehicles can

also modestly mitigate the problem of inertial navigation “drift” in accuracy

over time, although it does not solve the problem.

52. This need not be real-time supervision, of course.

53. For example, see “Self-Organizing Systems Research Group,” Harvard

University, http://www.eecs.harvard.edu/ssr/.

54. John Arquilla and David Ronfeldt, Swarming and the Future of Conict

(Santa Monica: The RAND Corporation, 2005), vii, http://www.rand.org/

content/dam/rand/pubs/documented_briengs/2005/RAND_DB311.pdf.

Some modern-day examples of swarming with human agents can be found

in the tactics of protestors and rioters, particularly when empowered with

information technology that allows them to rapidly communicate and

synchronize their actions. Bill Wasik, “#Riot: Self-Organized, Hyper-Networked

Revolts – Coming to a City Near You,” wired.com, December 16, 2011, http://

www.wired.com/2011/12/_riots/all/.

55. Arquilla and Ronfeldt, ibid, page vii.

56. Sean Edwards, Swarming and the Future of Warfare (Santa Monica: The

RAND Corporation, 2005), xvii, http://www.rand.org/content/dam/rand/pubs/

rgs_dissertations/2005/RAND_RGSD189.pdf.

57. “HART on-demand intelligence systems proves value to warghters ahead

of theater deployment,” Space War, May 20, 2011, http://www.spacewar.

com/reports/HART_On_Demand_Intelligence_System_Proves_Value_To_

Warghters_Ahead_of_Theater_Deployment_999.html.

58. Scott Littleeld, “Hydra,” darpa.mil, http://www.darpa.mil/Our_Work/

TTO/Programs/Hydra.aspx; and Andy Coon, “Upward Falling Payloads (UFP),”

darpa.mil, http://www.darpa.mil/Our_Work/STO/Programs/Upward_

Falling_Payloads_(UFP).aspx.

59. Patrick Tague, “Improving Anti-Jamming Capability and Increasing

Jamming Impact With Mobility Control” (paper presented at the Mobile

Adhoc and Sensor Systems (MASS), 2010 Institute of Electrical and Electronics

Engineers 7th International Conference, San Francisco, California, November

8-12, 2010), 501-506.

60. Oce of Naval Research.

61. This assumes that follow-on secondary and tertiary targets are within the

available range of the munition given its remaining fuel. Targets would have

to be reasonably clustered for this to work.

62. Mark V. Arena et al., Why Has the Cost of Navy Ships Risen?; and Mark V.

Arena et al., Why Has the Cost of Fixed-Wing Aircraft Risen?.

63. “Expendable Wave: Raytheon’s MALD and MALD-J Decoys,”

defenseindustrydaily.com, April 24, 2013, http://www.defenseindustrydaily.

com/raytheons-mald-decoys-gaining-versatility-04844/.

64. Brian Everstine, “Flight Restrictions Lifted for F-22s with Auto

Oxygen System,” defensenews.com, April 4, 2013, http://www.

defensenews.com/article/20130404/DEFREG02/304040017/

Flight-Restrictions-Lifted-F-22s-Auto-Oxygen-System.

65. The debate on monoculture vs. diversity has a long history in the

information technology sector. For a very brief and readable survey of the

arguments, see Bruce Schneier, “Software Monoculture,” Schneier on Security,

December 1, 2010, https://www.schneier.com/blog/archives/2010/12/

software_monocu.html.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

66. Sherrill Lingle et al., Methodologiees for Analyzing Remotely Piloted

Aircraft in Future Roles and Missions, (Santa Monica: The RAND Corporation,

2012), 18, http://www.rand.org/content/dam/rand/pubs/documented_

briengs/2012/RAND_DB637.pdf.

67. “DARPA-BAA-09-65 – Precision Electronic Warfare (PREW),” FedBizOpps.

gov, August 24, 2009, https://www.fbo.gov/index?s=opportunity&mode=for

m&id=85271e56b3b9aa657b2fd26cbeaa3944&tab=core&_cview=1.

68. Mark J. Mears, “Cooperative Electronic Attack Using Unmanned Air

Vehicles,” Air Force Research Lab, Wright-Patterson Air Force Base, http://

www.dtic.mil/dtic/tr/fulltext/u2/a444985.pdf.

69. For more, see David Scheidt, in Kimon P. Valavanis, George J. Vachtsevanos,

Handbook of Unmanned Aerial Vehicles, (Springer, 2014).

70. For a technical description of this eect, see David Scheidt, ibid.

71. Col Dr. Frans Osinga, “A Discourse on Winning and Losing,” July 13,

2007, http://www.au.af.mil/au/awc/awcgate/boyd/osinga_boydconf07_

72. United States Department of Defense, Autonomy in Weapon Systems,

3000.09 (November 21, 2012). In full disclosure, the author was a major

contributor to the development of DOD Directive 3000.09.

73. This oft-repeated aphorism of military operations is actually paraphrased

from Helmuth Von Moltke’s original statement: “No operation extends with

any certainty beyond the rst encounter with the main body of the enemy.”

74. Source: U.S. Air Force.

75. Lt Gen Dave Deptula, “Air Force Unmanned Aerial System (UAS) Flight

Plan 2009-2047,” slide 11, http://www.defense.gov/DODcmsshare/

briengslide/339/090723-D-6570C-001.pdf.

76. Brian Mekdici and M.L. Cummings, “Modeling Multiple Human Operators

in the Supervisory Control of Heterogenous Unmanned Vehicles,” September

2009, http://web.mit.edu/aeroastro/labs/halab/papers/PerMIS09.pdf.

77. For example, see “Aurora’s Autonomy and Flight Control,” aurora.aero,

http://www.aurora.aero/Research/Autonomy.aspx; and “Product and

Services: Cooperative Control/UDMS,” proxytechnologiesinc.com, http://www.

proxytechnologiesinc.com/systems.html.

78. Robert Brizzolara and Matthew Klunder, Oce of Naval Research media

roundtable, September 30, 2014.

79. Keith Button, “The MAC Attack,” DefenseNews.com, October, 1, 2009,

http://www.defensenews.com/article/20091001/C4ISR02/910010314/

The-MAC-attack.

80. This direction was issued as part of the Fiscal Year (FY) 2012 Defense

Budget, with $46 million allocated over the future years’ defense plan from

FY12-FY15.

81. “United States Air Force RPA Vector – Vision and Enable Concepts 2013-

2038,” (United States Air Force, February 17, 2014).

82. For example, current FAA guidelines that pilots may not perform crew

duties for more than one aircraft at a time are often cited as a rationale for

not elding multi-aircraft control technology. FAA guidelines do not apply to

military operations outside of U.S. national airspace, however. United States

Department of Transportation Federal Aviation Administration, Unmanned

Aircraft Systems (UAS) Operational Approval, N 8900.227 (July 30, 2013).

83. For example, see “Tactical Tomahawk Takes Flight,” navy.mil, August 27,

2002, http://www.navy.mil/submit/display.asp?story_id=3311.

84. “Dogghting Drones – Swarms of Unmanned Battle-Bots Take

to the Skies,” airforce-technology.com, July 23, 2012, http://www.

airforce-technology.com/features/featuredogght-drones-unmanned-battle-

bot-swarms/ and Radhika Nagpal, “The Kilobot Project,”eecs.harvard.edu,

http://www.eecs.harvard.edu/ssr/projects/progSA/kilobot.html.

85. For a more detailed examination of a wider array of possible command-

and-control models, see David Scheidt, Kevin Schultz, “On Optimizing

Command and Control Structures,” Proceedings of the 16th International

Command and Control Research and Technology Consortium (The Johns

Hopkins University Applied Physics Laboratory, 2011).

86. This process is called “particle swarm optimization.” Another downside to

this approach is that it only may result in localized optimization.

87. Coordination is still possible is very low bandwidth environments. For

example, DARPA’s Collaborative Operations in Denied Environment (CODE)

project restricts communications between the swarm and the command

element to 50 kilobits per second, or less than a 56k dial-up modem, circa

1997. “Collaborative Operations in Denied Environment (CODE),” fedbizopps.

gov, April 25, 2014, https://www.fbo.gov/index?s=opportunity&mode=form

&id=2f2733be59230cf2ddaa46498fe5765a&tab=core&_cview=1.

88. One could employ other analogous forms of “jamming” to disrupt

their method of implicit communication, however, such as blocking visual

co-observation of swarm elements with obscurants. Dave Scheidt, JHU/APL,

personal correspondence.

89. Firas Safadi, Raphael Fonteneau, Damien Ernst, “Articial Intelligence

Design for Real-time Strategy Games,” Proceedings of NIPS Workshop on

Decision Making with Multiple Imperfect Decision Makers, December 16, 2011.

90. John Hawley et al., The Human Side of Automation: Lessons for Air Defense

Command and Control, Army Research Laboratory, March 2005.

91. Timothy Chung, Naval Postgraduate School, personal correspondence.

92. John A. Sauter et al., “Distributed Pheromone-Based Swarming Control

of Unmanned Air and Ground Vehicles for RSTA,” (paper presented at SPIE

Defense and Security Conference, Orlando, Florida, March 2008).

93. “All Systems Go: Navy’s Laser Weapon Ready for Summer Deployment,”

Navy.mil, April 7, 2014, http://www.navy.mil/submit/display.asp?story_

id=80172; and Allen McDuee, “Navy’s New Railgun Can Hurl a Shell Over

5,000 MPH,” Wired.com, April 9, 2014, http://www.wired.com/2014/04/

electromagnetic-railgun-launcher/.

94. “Dogghting Drones – Swarms of Unmanned Battle-Bots Take to the Skies.”

95. Additionally, while having a person physically onboard a platform does not

render it immune to cyberattacks, having a person physically present could

at least prevent the adversary from taking control of the system via cyber

attacks, provided there are hardware-level physical overrides. A co-located

human operator could physically unplug, for example, a malfunctioning

missile battery. When systems are controlled remotely, this additional failsafe

does not exist.

96. Humans, of course, are not impervious to deception. Machines actually

might be better at detecting data that is manipulated slightly. A person

might not notice 1% dierence in position data, whereas a machine might

catch such a dierence. Humans, however, have a level of common sense that

can be applied to information that is grossly wrong. If sent the directive to

attack friendly forces, for example, a human would intuitively know that the

order was either garbled or the result of enemy deception. Some degree of

“skepticism” can be programmed into machines – if certain types of failures

or enemy attacks can be anticipated. Human intelligence is far more robust,

however, to unanticipated circumstances.

97. This, of course, depends on an eective communications link.

98. Andrew Herr, “Will Humans Matter in the Wars of 2030?,” forthcoming

publication.

99. Mark V. Arena et al., Why Has the Cost of Navy Ships Risen; and Mark V.

Arena et al., Why Has the Cost of Fixed-Wing Aircraft Risen?.

100. For example, the F-35 Joint Strike Fighter has over 24 million lines of code.

A Mercedes S-class sedan as approximately 100 million lines of code. “A Digital

Jet for the Modern Battlespace,” f35.com, https://www.f35.com/about/life-

cycle/software; and Robert N. Charette, “This Car Runs on Code,” spectrum.

ieee.org, February 1, 2009, http://spectrum.ieee.org/transportation/systems/

this-car-runs-on-code.

101. Admiral Jonathan W. Greenert, “Payloads over Platforms: Charting

a New Course,” usni.org, July 2012, http://www.usni.org/magazines/

proceedings/2012-07/payloads-over-platforms-charting-new-course.

102. Chris Urmson, “The Latest Chapter for the Self-Driving Car: Mastering City

Street Driving,” Google Ocial Blog on BlogSpot.com, April 28, 2014, http://

googleblog.blogspot.com/2014/04/the-latest-chapter-for-self-driving-car.

html.

103. For a good overview of the state of technology today, including current

limitations of autonomous vehicles, and likely future trends see “Self-

driving cars take a small step closer to reality,” TheStar.com, September

15, 2014, http://www.thestar.com.my/Tech/Tech-News/2014/09/15/

Self-driving-cars-take-a-small-step-closer-to-reality/.

104. Patrick Lin, “The Ethics of Autonomous Cars,” TheAtlantic.com,

October 8, 2013, http://www.theatlantic.com/technology/archive/2013/10/

the-ethics-of-autonomous-cars/280360/.

105. Takeo and landing are also the most prevalent conditions for accidents

for human-inhabited aircraft as well. The challenge of manually landing an

aircraft is compounded in an uninhabited aircraft because the pilot does not

get immediate kinesthetic feedback on the aircraft’s movement. Thus, there

is not only a slight time delay in communications between the aircraft and the

controller, but also a time delay in perception by the pilot, who must rely on

visual or other cues to perceive movement of the aircraft.

106. For example, the Air Force’s vision document for uninhabited aircraft out

to 2038 still refers to them as “remotely piloted,” “United States Air Force RPA

Vector – Vision and Enabling Concepts 2013-2038.” For a somewhat humorous

overview of the tortured language used to refer to uninhabited aircraft see

Joe Trevithick, “Learn to Speak Air Force – A Public Service Announcement

Regarding Drones,” Medium.com/War-Is-Boring, May 27, 2014, https://

medium.com/war-is-boring/learn-to-speak-air-force-e6ebc5614b25.

107. Department of the Army, Unmanned Aircraft System Flight Regulations,

Army Regulation 95-23 (July 2, 2010), http://www.apd.army.mil/pdles/

r95_23.pdf.

108. For more on what this shift to supervisory control will mean and the

challenges it will bring, see Hawley, ibid.

109. Michael C. Horowitz, “The Looming Robotics Gap.”

110. Ekso Bionics, “Yes, We Said Bionics,” http://www.eksobionics.com/ekso.

111. Ben Fitzgerald and Kelley Sayler, “Creative Disruption: Technology,

Strategy and the Future of the Global Defense Industry” (Center for a New

American Security, June 2013), http://www.cnas.org/sites/default/les/

publications-pdf/CNAS_FutureDefenseIndustry_FitzGeraldSayler.pdf.

112. General Mike Hostage, commander of the Air Force’s Air Combat

Command, has made this point about the software on the F-22, saying “The

F-22, when it was produced, was ying with computers that were already so

out of date you would not nd them in a kid’s game console in somebody’s

home gaming system. But I was forced to use that because that was the

[specication] that was written by the acquisition process when I was going to

buy the F-22.” “Interview: Gen. Michael Hostage, Commander, US Air Force’s

Air Combat Command,” 3 February 2014, http://mobile.defensenews.com/

article/302030017.

113. For an excellent critique of the belief that technology can make war quick

or easy, see H.R. McMaster, “The Pipe Dream of Easy War,” New York Times,

July 20, 2013, http://www.nytimes.com/2013/07/21/opinion/sunday/the-

pipe-dream-of-easy-war.html?pagewanted=all&_r=0.

114. This echoes a recommendation from Samuel J. Brannen in “Sustaining

the U.S. Lead in Unmanned Systems,” (CSIS International Security Program,

February 2014), 15, http://csis.org/les/publication/140227_Brannen_

UnmannedSystems_Web.pdf.

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

Appendix

TOTAL PROBABILITY OF KILL FOR SALVO 60

Robotics on the Battleeld Part II

The Coming Swarm

OCTOBER 2014

APPENDIX: TOTAL PROBABILITY OF KILL FOR SALVO*

By number of munitions red and probability of kill (Pk) for each munition.

Number of

Munitions per

Salvo

Single

Munition Pk =

0.9

Single

Munition Pk =

0.8

Single

Munition Pk =

0.7

Single

Munition Pk =

0.6

Single

Munition Pk =

0.5

1 0.9 0.8 0.7 0.6 0.5

2 0.99 0.96 0.91 0.84 0.75

3 0.999 0.992 0.973 0.936 0.875

4 0.9999 0.9984 0.9919 0.9744 0.9375

5 0.99999 0.99968 0.99757 0.98976 0.96875

6 1 0.99994 0.99927 0.9959 0.98438

7 1 0.99999 0.99978 0.99836 0.99219

8 1 1 0.99993 0.99934 0.99609

9 1 1 0.99998 0.99974 0.99805

10 1 1 0.99999 0.9999 0.99902

11 1 1 1 0.99996 0.99951

12 1 1 1 0.99998 0.99976

13 1 1 1 0.99999 0.99988

14 1 1 1 1 0.99994

15 1 1 1 1 0.99997

16 1 1 1 1 0.99998

17 1 1 1 1 0.99999

18 1 1 1 1 1

19 1 1 1 1 1

20 1 1 1 1 1

21 1 1 1 1 1

22 1 1 1 1 1

23 1 1 1 1 1

24 1 1 1 1 1

25 1 1 1 1 1

26 1 1 1 1 1

27 1 1 1 1 1

28 1 1 1 1 1

29 1 1 1 1 1

30 1 1 1 1 1

… … … … … …

39 1 1 1 1 1

40 1 1 1 1 1

41 1 1 1 1 1

* Total probability of a kill for a salvo (Pksalvo) is given by the following formula: Pksalvo = 1 – (1 – Pk)^N where Pk is the probability of kill for a single munition

and N is the number of munitions red in a salvo.

Single

Munition Pk =

0.45

Single

Munition Pk =

0.4

Single

Munition Pk =

0.3

Single

Munition Pk =

0.2

Single

Munition Pk =

0.11

Single

Munition Pk =

0.055

0.45 0.4 0.3 0.2 0.11 0.055

0.6975 0.64 0.51 0.36 0.2079 0.106975

0.83363 0.784 0.657 0.488 0.29503 0.156091

0.90849 0.8704 0.7599 0.5904 0.37258 0.202506

0.94967 0.92224 0.83193 0.67232 0.44159 0.246369

0.97232 0.95334 0.88235 0.73786 0.50302 0.287818

0.98478 0.97201 0.91765 0.79028 0.55769 0.326988

0.99163 0.9832 0.94235 0.83223 0.60634 0.364004

0.99539 0.98992 0.95965 0.86578 0.64964 0.398984

0.99747 0.99395 0.97175 0.89263 0.68818 0.43204

0.99861 0.99637 0.98023 0.9141 0.72248 0.463277

0.99923 0.99782 0.98616 0.93128 0.75301 0.492797

0.99958 0.99869 0.99031 0.94502 0.78018 0.520693

0.99977 0.99922 0.99322 0.95602 0.80436 0.547055

0.99987 0.99953 0.99525 0.96482 0.82588 0.571967

0.99993 0.99972 0.99668 0.97185 0.84503 0.595509

0.99996 0.99983 0.99767 0.97748 0.86208 0.617756

0.99998 0.9999 0.99837 0.98199 0.87725 0.638779

0.99999 0.99994 0.99886 0.98559 0.89075 0.658646

0.99999 0.99996 0.9992 0.98847 0.90277 0.677421

1 0.99998 0.99944 0.99078 0.91347 0.695163

1 0.99999 0.99961 0.99262 0.92298 0.711929

1 0.99999 0.99973 0.9941 0.93146 0.727773

1 1 0.99981 0.99528 0.939 0.742745

1 1 0.99987 0.99622 0.94571 0.756894

1 1 0.99991 0.99698 0.95168 0.770265

1 1 0.99993 0.99758 0.95699 0.7829

1 1 0.99995 0.99807 0.96172 0.794841

1 1 0.99997 0.99845 0.96593 0.806125

1 1 0.99998 0.99876 0.96968 0.816788

… … … … … …

1 1 1 0.99983 0.98938 0.889887

1 1 1 0.99987 0.99055 0.895943

1 1 1 0.99989 0.99159 0.901666

About the Center for a

New American Security

The mission of the Center for a New American

Security (CNAS) is to develop strong, pragmatic

and principled national security and defense

policies. Building on the expertise and

experience of its sta and advisors, CNAS

engages policymakers, experts and the public

with innovative, fact-based research, ideas

and analysis to shape and elevate the national

security debate. A key part of our mission is to

inform and prepare the national security leaders

of today and tomorrow.

CNAS is located in Washington, and was

established in February 2007 by co-founders

Kurt M. Campbell and Michèle A. Flournoy.

CNAS is a 501(c)3 tax-exempt nonprot

organization. Its research is independent and

non-partisan. CNAS does not take institutional

positions on policy issues. Accordingly, all views,

positions, and conclusions expressed in this

publication should be understood to be solely

those of the authors.

Center for a New American Security

1152 15th Street, NW

Suite 950

Washington, DC 20005

TEL 202.457.9400

FAX 202.457.9401

EMAIL info@cnas.org

www.cnas.org

Production Notes

Soy ink is a helpful component in paper recycling. It helps in this

process because the soy ink can be removed more easily than

regular ink and can be taken out of paper during the de-inking

process of recycling. This allows the recycled paper to have less

damage to its paper bers and have a brighter appearance.

The waste that is left from the soy ink during the de-inking

process is not hazardous and it can be treated easily through

thedevelopment of modern processes.

Paper recycling is reprocessing waste paper bers back into

ausable paperproduct.

1152 15th Street, NW

Suite 950

Washington, DC 20005

STRONG, PRAGMATIC AND PRINCIPLED

NATIONAL SECURITY AND DEFENSE POLICIES

TEL 202.457.9400

FAX 202.457.9401

EMAIL [email protected]rg

www.cnas.org

Printed on Post-Consumer Recycled paper with Soy Inks