SWARMING LOGISTICS FOR TACTICAL LAST-MILE DELIVERY

Sam Thornton, Guy Edward Gallasch

Land Division

Defence Science and Technology Group

Edinburgh, South Australia, Australia

[email protected]nce.gov.au

Abstract— Last-mile delivery in a military context can often

be dangerous, putting personnel and the supplies they carry at

risk. The emergence of aerial ‘delivery drones’ from the

commercial delivery sector highlights the possibilities of

uncrewed vehicles being used in last-mile delivery. However,

demonstrations of such technology have been limited to single

vehicle deliveries, where only small portions of supplies can be

delivered at once. This paper explores the concept of low-cost,

uncrewed vehicle swarming for tactical last-mile delivery in a

deployed setting. The benefits of uncrewed swarming systems

over conventional methods of resupply are discussed, as well as

the vulnerabilities and challenges faced by such systems.

Keywords—Last-mile Logistics, Swarming, Drone Delivery

I. INTRODUCTION

Last-mile delivery (LMD) in a military context is the

distribution of supplies from the last point of bulk

disaggregation to dispersed forces located in the area of

operations. It is a critical Combat Service Support (CSS)

process. The last-mile is not actually reference to a fixed

distance, but it is generally within a range of 30km [1]. While

there is scepticism of the potential for swarming to be applied

to LMD, it has many potential benefits over conventional

methods of resupply, most notably its scalability, flexibility,

and robustness against single point failure. The use of

uncrewed delivery vehicles also reduces risks to personnel

transporting supplies in contested environments, and can

provide extended delivery capabilities to locations inaccessible

by humans. This paper explores the concept, including

strengths and weaknesses, of using swarms of low-cost,

uncrewed platforms for last-mile military resupply.

II. SWARMING

Swarming can be defined as the collective physical

behaviour (often emergent and complex) of a group or ‘swarm’

of physical agents, where each agent’s own behaviour is based

on simple rules for interaction with other agents within the

swarm, and the environment. Biological examples of swarming

can be seen in social insects such as ants and honey bees, in

birds (‘flocking’), fish (‘schooling’), and many other organisms

[2]. Swarming can occur both with and without the use of

explicit communication between agents, making it an attractive

concept to explore for technological operations in

electromagnetically contested and degraded environments [2].

III. LAST-MILE DELIVERY

A. Military versus Civilian Contexts

LMD in a military context differs greatly from that of a

civilian context. Supply delivery to the tactical edge in

contested environments poses the threat of delivery systems

being disrupted by adversaries. Whether this is to hijack

provisions, or to simply stem the flow of supply, military

delivery systems can often be considered ‘soft’ targets and

require appropriate countermeasures for protection. The result

is the need for increased complexity of delivery platform

design and delivery planning, and/or the necessity of security

escort units. Additionally, military environments tend to be less

structured that civilian environments, and can change

throughout the course of an operation. High bandwidth

connectivity is also problematic close to the tactical edge.

B. Changing Nature of the Battlespace

The call for new LMD methods originates in part from the

nature of modern warfare. The ‘front lines’ of the battlespace

have evolved from describing the literal front lines of opposing

forces (such as trench warfare in the First World War), to

forces fighting in more dispersed and highly mobile tactical

groups. This in turn has forced (and will continue to force)

LMD methods to become more versatile and precise. As stated

in [3]: “Supplying widely dispersed units without traditional

CSS battalions present is a difficult problem that will probably

need to be addressed by a package of fixes” (p. 71).

C. Current Last-mile Delivery Methods

Current modes of direct delivery in military operations

primarily involve the use of crewed supply vehicles and air

drops. Due to the increasingly dispersed nature of modern

warfighting, delivery vehicles are required to undertake multi-

stop routes for distribution of supplies to ever more widely

dispersed forces. This is an inefficient way of operation, as

supplies after the first delivery are not transported directly to

their delivery points. The result of this indirect delivery is

increased delivery times and wasted resources. Air drop

methods such as the Joint Precision Air Drop System (JPADS)

can be expensive, requiring recovery on delivery by the

receiver so that they can be returned and reused [4]. Air drops

systems also need to be taken up to altitude before they can be

deployed, further increasing delivery times, and requiring

crewed cargo aircraft to fly over potentially contested areas.

Sam is an undergraduate student with the University of Adelaide. He

undertook this work while on a Summer Vacation Placement with Land

Logistics Group, Land Division, over the 2017/18 vacation period.

IV. SWARMING FOR LAST-MILE DELIVERY

A. Benefits of Autonomous Systems for Last-Mile Delivery

The benefits of using uncrewed swarming platforms for

tactical last-mile resupply form part of the bigger picture of

harnessing autonomous systems for CSS. The use of

autonomous systems for LMD could increase the safety of

personnel by reducing their exposure to risk. For example,

resupply convoys are often seen as ‘soft’ targets, and hence one

argument for autonomous convoys is to reduce the number of

personnel travelling in such convoys. Autonomous systems

physically integrated with human forces could also provide

faster decision making in time critical operations, such as

optimal route planning or rerouting in hostile environments [5].

The value of autonomous systems for tactical LMD comes

from not only the potential increase in the safety and wellbeing

of personnel, but also from the likely increase in LMD

efficiency. Autonomous systems do not become physically or

mentally tired from carrying out their tasks. This provides

potential for more continuous and sustained logistics

operations over longer ranges. Autonomous systems could also

be used to operate in environments considered too dangerous

or inaccessible for humans, extending current CSS capabilities.

B. Benefits of Swarming for Last-Mile Delivery

Although the use of swarming for both offensive and

intelligence, surveillance and reconnaissance (ISR) operations

is already being explored by militaries around the world, the

benefits of using swarms in logistics operations have not been

discussed outside of autonomous (and semi-autonomous)

ground convoys, which themselves do not exhibit swarm

intelligence or emergent behaviour [6].

Coupled with low-cost, potentially expendable individual

platforms (swarm agents do not need to be highly intelligent as

individuals), swarming offers a force multiplier that is more

scalable and flexible than complex, stand-alone systems.

Efficiencies can be gained through a better match of payload to

platform payload capacity. For instance, a small quantity of

supplies could be delivered by a few smaller platforms as

opposed to a single larger vehicle which may be transporting

empty space. It may additionally be the case that some

missions could benefit from having more delivery platforms

than needed, such as a way of breaking up valuable supplies, or

having some agents act as decoys.

Having scalable delivery swarms also means that multiple

locations can be serviced both directly (avoiding the

inefficiencies of indirect delivery via milk runs) and in parallel.

This also facilitates more ‘on demand’ delivery due to

increased availability of delivery platforms. Swarming also

allows for swarm formations and tactics to be altered to suit

dynamic and unknown environments. Furthermore, dispersed

multi-robot systems can be multi-purpose, e.g. used for both

situational awareness and delivery.

As autonomous and automated uncrewed systems become

more complex, the cost of staffing required for operation, data

management and analysis of the systems also increases [5]. In

this regard, another benefit swarming brings to LMD (and

automated and autonomous systems in general) is the ability

for relatively few humans to control large numbers of

autonomous platforms, through relatively simple rules for

swarm control and organisation. Multi-robot swarming could

act to reduce operating costs, as a single operator would be able

to control a large number of platforms – the operator would

likely see the swarm as a single entity, placing less investment

and attention into the individual platforms that actually make

up the swarm.

Finally, the use of decentralised and distributed multi-robot

systems increases LMD system robustness over using single, or

fewer platforms for delivery. If a single platform used to

resupply forces is destroyed or fails in its task, none of the

supplies are delivered. However if 100 platforms are used to

deliver the supplies in a distributed manner, then the failure of

a single platform, or even handful of platforms, has a much

smaller impact, as the majority of supplies will arrive. Lt. Gen.

Michael Dana, Deputy Commandant of Installations and

Logistics for the US Marine Corps, believes that drone swarms

for logistics would be especially handy in littoral or island

environments [7].

C. Air versus Ground Swarming

LMD for ground forces doesn’t need to be constrained to

ground platforms. Swarms of UAVs are likely to be the easiest

to realise, as they don’t have to adapt to changing ground

terrain, of particular significance for military environments

where there can be little or no pre-existing road infrastructure.

Additionally, the sophistication of mechanisms required for

ground movement are also greater than that for aerial

manoeuvres, further supporting the use of UAVs [6]. Swarm

configurations and obstacle avoidance methods are also more

flexible in the air domain, due to the freedom of elevation

control. Against countermeasures, UAV swarms have the

advantage that they are more manoeuvrable and are generally

less susceptible to IEDs, land mines, or being obstructed.

UAV swarms do have some downsides, however. One is a

need for sensing in 3-dimensional space (although there are

less obstacles to avoid in the sky). Another is that UAV

swarms, if flying in open skies, may reveal troop locations

and/or be actively targeted by countermeasures. In terms of

payload, UAVs are more constrained in payload capacity,

although the use of scalable swarms somewhat makes up for

this, as supplies can be distributed across multiple platforms.

Platform cost to payload capacity also tends to be greater for

UAVs when compared to UGVs.

Uncrewed maritime vehicles (UMVs), which can be

classed as either surface or underwater vessels (or both) do not

feature as prominently in LMD for ground forces, though they

can play roles in amphibious operations and when forces

operate close to bodies of water.

D. Use of UAV Swarming in Last-mile Delivery

Due to physical constraints, UAV swarms for LMD are

likely to be constrained in the near future to lighter, smaller

items such as medical supplies, small electrical components,

and bulk commodities that can be broken down into smaller

payloads (e.g. food, water, ammunition). Swarming would

allow the overall delivery of meaningful quantities of such

items. It is conceivable that UAVs could work in teams to

collectively transport larger, heavier items, but it may be the

case that conventional delivery methods for such items are

more efficient. Scalability and flexibility of tasking provides

the capability for parallel delivery to dispersed forces, e.g.

simultaneous emergency resupply of ammunition to dispersed

fighting elements engaged in a contested urban combat setting

– a time critical delivery of valuable supplies to a high risk or

denied environment. Another use case could be the delivery of

emergency medical supplies, similar to the RQ-7 Shadow UAV

demonstration of the QuickMEDS system [5]. However,

currently, we note that swarming for LMD is not

technologically feasible. We come back to this in Section VI.

E. Type of UAV

The type of UAV used for LMD swarming would depend

on the functional requirements of the swarming platforms.

There are two main variants: multi-copter (rotary wing) and

fixed-wing. We refer the reader to [8] for a more detailed

discussion of this, including Vertical Take-Off and Landing

(VTOL) platforms and gliders.

F. Size of UAV

Swarming platforms for LMD would need to be small

enough to be low-cost and expendable, but large enough to be

able to carry at least a few kilograms in payload weight

(depending on how divisible supplies are). Larger UAVs

generally have greater range capabilities, and for LMD a radius

of operation would be up to 30km. Smaller UAVs designed

with hybrid or even hydrogen fuel cell power systems (as

opposed to purely electric) could reach this target, at the cost of

greater running expenses. Smaller UAVs would be preferred

for swarming, as the decentralised functional benefits of

swarming increase with agent numbers, and smaller, cheaper

platforms would be easier to mass produce than larger ones.

Larger, less agile heavy-lift UAVs (e.g. [9]) would be more

suited to standalone deliveries. Smaller UAVs also have

increased resilience against collision due to having less

momentum [10]. The precise trade-off between effective

throughput, range, and operating costs of small vs. large UAVs

is beyond the scope of this paper.

G. Vulnerabilities

As with all uncrewed systems, data security is a

vulnerability of swarming systems. The potential for

communications jamming, spoofing and hijacking are all things

that must be considered when implementing swarming

systems, especially since the task of LMD places these systems

in contested zones. The physical protection of LMD swarms is

also an important consideration, as even though the individual

platforms may be considered expendable, the swarm as a whole

cannot be. The use of distributed control systems for swarm

interaction somewhat alleviates the risk of single point of

failure (both data-wise and physically) for entire swarms, but

the risk still exists nonetheless. Further adding to these

vulnerabilities is the fact that militaries have already begun

looking into and implementing anti-drone swarm technologies

(e.g. [6, 11]). There is also the threat that opposing weaponised

drone swarms could intercept and destroy delivery swarms.

V. STATE-OF-THE-ART

UAV LMD is a relatively new area that has only recently

gained prominence in the civilian and military logistics.

Because of this, there is little material in the open literature

around the idea of applying swarming to UAV LMD. The

UK’s Defence Science and Technology Laboratory (dstl) are

currently holding an Accelerator competition to challenge

private organisations to develop and demonstrate autonomous

last-mile resupply systems. Winners of the competition’s first

phase included Marble Aerospace, with their proposed project

titled “Swarm of high speed and small payload UAS with

robotic hangars, for high speed long range resupply of small

size item”, although details of the project have not been

publicly released [12]. Similarly, two first phase winners, 2iC

and Blue Bear Systems Research, have partnered for the

second phase of the competition to deliver “Autonomous UAV

swarm operation”, with Blue Bear Systems Research focusing

on developing modular UAVs for a “fractionated” last-mile

resupply system [13,14].

DARPA funded research firm Otherlab have taken to

exploring LMD through the use of small (about a metre

wingspan) GPS-driven gliders named APSARA – Aerial

Platform Supporting Autonomous Resupply/Actions – that can

travel up to 150 kilometres (when deployed from 35000 feet)

and have a 10 metre landing accuracy (presumably in ideal

conditions) [15]. The gliders are capable of delivering a one

kilogram payload. Their structures are made from cardboard,

allowing recipients to leave the airframes to degrade once the

gliders have landed (including their electronics). Although

these gliders do not display swarm intelligence, they are

intended to be deployed in large groups (up to hundreds). The

US Naval Research Lab has also developed small GPS-driven

gliders intended to be dropped in groups from aircraft [16]. The

Close-In Covert Autonomous Disposable Aircraft (CICADA)

gliders have 3D printed fuselages, and wings and tail fins

constituting of printed circuit boards for on-board avionics.

Currently, the gliders are designed to only carry sensory and

communications payloads, being able to transmit data back to

their launching planes – using the gliders for delivery of

supplies is yet to be explored.

VI. TECHNOLOGICAL CHALLENGES

There are numerous technological challenges to overcome

before swarming is viable for meaningful LMD in a military

context, beyond the usual Space, Weight and Power constraints

forcing a trade-off between range/endurance, payload capacity,

size, and cost (in this case largely centred around energy

density of batteries or other fuels). Many of these challenges

are also applicable to swarming for other purposes.

A. Whole-of-Swarm Positioning

Reliance on positioning systems such as GPS for swarm

coordination is a clear weakness when operating in a contested

environment where GPS can often be blocked or spoofed, or

where satellite coverage isn’t sufficient. This will need to be

accounted for through the use of inbuilt maps combined with

other localisation and positioning techniques including the use

of dead reckoning, inertial navigation, and optical-flow.

B. Internal Swarm Localisation

There are numerous methods for localisation of individual

agents within a swarm and for the detection of neighbours,

both active and passive, and each with their own pros and cons.

A more detailed discussion is presented in [8], but in brief,

these include wireless networking technologies, camera vision,

laser rangefinding (3D lidar), infrared sensors, ultrasonic

sensors, radar, and audio. From a signature management

perspective, passive methods are preferred, however such

methods (e.g. vision) are too financially and computationally

expensive to be viable for low-cost, small UAVs at the current

level of technology maturity (although this is expected to

improve rapidly). Furthermore, unless carefully designed, the

active sensors of many robots swarming in close formation

may interfere with each other in unwanted ways, e.g. the

sensors from one platform pick up the transmitted signals from

another platform instead of their own reflected signals [17].

C. Signature Management

Adding to the problem, any vehicles operating in warzones

should ideally have sufficiently low signatures (radio

frequency, acoustic, thermal, visual etc.) to avoid detection by

adversaries, limiting the number of sensory technologies and

acceptable communications bandwidth than can be used.

D. Human-Swarm Interface

Another big technological challenge for UAV swarming in

LMD, and for coordination of multi-robot systems in general is

the operational control of such systems. This not only relates to

the human-swarm interface, in which data needs to be

optimally presented to operators within human cognitive limits,

but also to the degree of autonomy each system exhibits.

Higher degrees of autonomy allow lesser needs of operator

control, but greater needs in operator analysis of autonomous

performance and decision making.

E. Communications and Networking

If wireless networks are used for both inter- and intra-

swarm communication, suitable networking protocols will need

to be developed. These networks must be scalable and adapt to

agents both entering and exiting a swarm’s network. Separating

different data into different channels of communication (e.g.

remote piloting of a swarm leader vs. intra-swarm

communications) into different channels of communication

may help to provide resilience against interference [18].

VII. CONCLUSION

The concept of swarming enables meaningful volumes of

supplies to be delivered by low-cost, uncrewed systems. The

use of uncrewed systems for tactical last-mile delivery reduces

the exposure of CSS personnel to potentially hostile

environments. Furthermore, swarming systems allow supplies

that can be divided into smaller parts to be transported in a

distributed manner that is scalable, flexible and robust.

Swarming delivery systems suit the increasingly dispersed and

mobile nature of modern warfighting, where conventional

methods of supply delivery can be time and cost inefficient.

Military organisations have acknowledged the potential for

UAV swarms to be used in future warfare in offensive and ISR

roles, and have already begun developing countermeasures.

However, numerous technological challenges remain.

REFERENCES

[1] Gov.uk, “Competition document: autonomous last mile resupply” last

accessed 22 May 2018, Available via

https://www.gov.uk/government/publications/accelerator-competition-

autonomous-last-mile-supply/accelerator-competition-autonomous-last-

mile-resupply

[2] Wikipedia, “Swarm behaviour”, last accessed 22 May 2018, available

via https://en.wikipedia.org/wiki/Swarm_behaviour

[3] Edwards, S., “Swarming on the Battlefield: Past, Present, and Future”,

Monograph Report, RAND Corporation, 2000, available via

https://www.rand.org/pubs/monograph_reports/MR1100.html

[4] Ackerman, E., “U.S. Marines Testing Disposable Delivery Drones”,

IEEE Spectrum, 17 April 2017, available via

https://spectrum.ieee.org/automaton/robotics/drones/marines-testing-

disposable-gliding-delivery-drones

[5] Ivanova, K., Gallasch, G. E., and Jordans, J., “Automated and

Autonomous Systems for Combat Service Support: Scoping Study and

Technology Prioritisation”, DST Group Technical Note DST-Group-

TN-1573, October 2016.

[6] Sanders, A., “Drone Swarms”, Master's Thesis, School of Advanced

Military Studies, United States Army Command and General Staff

College, Fort Leavenworth, Kansas, 2017.

[7] Seck, H., “Marines Want Swarming Delivery Drones for Resupply,

Disaster Relief”, Military.com, 25 July 2017, available via

https://www.military.com/defensetech/2017/07/25/marines-want-

swarming-delivery-drones-resupply-disaster-relief

[8] Thornton, S. and Gallasch, G. E., “Swarming Logistics for Tactical Last-

mile Delivery” draft DST Group General Document, 21 February 2018,

unpublished.

[9] Aitken, R., “What do China's Delivery Drones Look Like? - JD.Com

Spotlight”, UnmannedCargo.org, 26 October 2017, available via

http://unmannedcargo.org/chinese-delivery-drones/

[10] Kumar, V., “The Future of Flying Robots”, last accessed 23 May 2018,

available via https://www.youtube.com/watch?v=ge3--1hOm1s

[11] Lockheedmartin.com, “Technology that Counters Drone Swarms”, last

accessed 31 January 2018, available via

https://lockheedmartin.com/us/innovations/061416-webt-laser-swarms-

drones.html .

[12] Gov.uk, “Accelerator funded contracts 1 April 2017 to 31 March 2018”,

last accessed 8 December 2017, available via

https://www.gov.uk/government/publications/accelerator-funded-

contracts/accelerator-funded-contracts-1-april-2017-to-31-march-2018

[13] 2icworld.com, “News | 2iC”, last accessed 20 December 2017, available

via http://www.2icworld.com/news/11/45/UK-SMEs-partner-to-

develop-Autonomous-Last-Mile-Logistics-Solution-for-MoD

[14] Scott, R., “Innovative modular UAS concept [DSEI17D2]”, Janes 360,

13 September 2017, available via

http://www.janes.com/article/73890/innovative-modular-uas-concept-

dsei17d2

[15] Ackerman, E., “Swarms of Disposable Drones Will Make Critical

Deliveries and Then Vanish”, IEEE Spectrum, 1 Feb. 2017, available via

https://spectrum.ieee.org/automaton/robotics/drones/otherlab-apsara-

aerial-delivery-system

[16] Ackerman, E., “Naval Research Lab Tests Swarm of Stackable

CICADA Microdrones”, IEEE Spectrum, 25 July 2017, available via

https://spectrum.ieee.org/automaton/robotics/drones/naval-research-lab-

tests-swarm-of-stackable-cicada-microdrones

[17] Turgut, A., “Self-organised flocking with a mobile robot swarm”, Ph.D

thesis, Graduate School of Natural and Applied Sciences, Middle East

Technical University, 2008.

[18] Etter, W., Martin, P., Mangharam, R., “Cooperative Flight Guidance of

Autonomous Unmanned Aerial Vehicles”, CPS Week Workshop on

Networks of Cooperating Objects (CONET), April 11-14, 2011.