Available

Available online
online at
at www.sciencedirect.com
www.sciencedirect.com
Available online at www.sciencedirect.com
Available onlineonline
Available
Available at www.sciencedirect.com
online at
at www.sciencedirect.com
www.sciencedirect.com

ScienceDirect
Procedia Manufacturing
Procedia Manufacturing 00
00 (2019)
(2019) 000–000
000–000
Procedia Manufacturing 00 (2019) 000–000
Procedia Manufacturing
Procedia
Procedia 42 00
Manufacturing
Manufacturing (2020)
00 488–497
(2019)
(2019) 000–000
000–000
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

International Conference
International Conference on
on Industry
Industry 4.0
4.0 and
and Smart
Smart Manufacturing
Manufacturing (ISM
(ISM 2019)
2019)
International
International Conference on Industry 4.0 and Smart Manufacturing (ISM 2019)
International Conference
Conference on
on Industry
Industry 4.0
4.0 and
and Smart
Smart Manufacturing
Manufacturing (ISM
(ISM 2019)
2019)
Using
Using drones for parcels delivery process
Using drones
drones for
for
a,∗
parcels
parcels delivery
delivery process
a process aa , Giusy Macrinaaaa
Luigi
Luigi
Luigi Di
Di
Di Puglia
Puglia
Puglia Pugliese
Pugliese
Pugliese a,∗,, Francesca
a,∗
a,∗, Francesca
Francesca Guerriero
Guerriero
Guerriero , Giusy Macrina
Luigi Di Puglia Pugliese
Pugliese aa ,, Giusy Macrina aa
a Luigi Di
a Department
Department
a Department
Puglia
of Mechanical,
of Mechanical,
Mechanical, Energy
Energy and
a,∗
,, Francesca
Francesca
Energy and Management
and Management
Guerriero
Guerriero
Engineering,
Management Engineering,
Engineering, University
University of
GiusyRende
University of ,Calabria,
Giusy
of Calabria,
Macrina
Macrina
Calabria, Rende 87076, Italy
Rende 87076,
87076, Italy
Italy
a of
a Department
Department of
of Mechanical,
Mechanical, Energy
Energy and
and Management
Management Engineering,
Engineering, University
University of
of Calabria,
Calabria, Rende
Rende 87076,
87076, Italy
Italy

Abstract
Abstract
Abstract
Abstract
Abstract
Parcels delivery
Parcels delivery is is the
the most
most expensive
expensive phase
phase ofof the
the distribution logistics.
logistics. Everyday,
Everyday, several
several vehicles,
vehicles, usually
usually internal
internal combustion
combustion engine
engine vehicles,
vehicles,
Parcels
have delivery is thenumber
most expensive phase of the distribution
distribution logistics. Everyday, several vehicles, usuallyseveral
internal combustion engine vehicles,
have to
Parcels
Parcels
have to
serve
serve aaa high
to delivery
delivery
serve
is
is the
high
high
most
thenumber of
of customers
most expensive
number expensive
of
phase
customers
phase
customers
spatially
of
of the
spatially
distributed
the distribution
spatially distribution
distributed
in
in anan urban
distributed logistics.
logistics.
in an urban
urban
area.
area. Their
Everyday,
Everyday,
area.
several
Their
several
Their
presence
presence
presence
generates
vehicles, usually
generates
vehicles,
generates
internal
several
usuallyseveral
internalnegative
negative externalities,
combustion engine
externalities,
combustion
negative externalities,
such
such as,
engine vehicles, as,
vehicles,
such as,
noise,
have to
noise,
have tocongestion
serve aa high
congestion
serve high and
and pollutant
number
pollutant
number of emissions.
of customers
customers
emissions. Drones
spatially
Drones
spatially have
have become
distributed
become
distributed in aaan
in anvalid
urbanalternative
valid
urban area. Their
Their
alternative
area. to
to support
presence
support
presence the
the delivery
generates
delivery
generates process
several
process
several and
negative
and
negativeseveral
several big
big companies,
externalities, such as,
companies,
externalities, such as,
noise,
such congestion and pollutant emissions. Drones have become a valid alternative to support the delivery process and several big companies,
such as,
noise,
noise,
such
Amazon
Amazon and
congestion
as,
congestion
as, Amazon and DHL,
and
have
pollutant
DHL, have
pollutant
DHL, have
started
started to
emissions.
emissions.
started to
use
use them
toDrones
Drones
use
have
them
have
them
for
for
parcels
parcelsaadeliveries.
forbecome
become
parcels
valid
deliveries. On
On the
valid alternativetheto
alternative
deliveries. On the toone
one
one
hand,
support
support
hand,
drones
hand, the
drones
the
drones
drastically
delivery processreduce
process
drastically
delivery
drastically
and
reduce
reduce
negative
and several
negative
several
negativebigexternalities,
big companies,
externalities,
companies,
externalities,
allowing
such
allowingas, aa more
as, Amazon sustainable
and DHL,delivery
and DHL, have process.
have started to On
to use the
them other
for hand,
for parcelsseveral
parcels technical
deliveries. On aspects
On the one must
one hand,be
hand, carefully
drones taken
drastically into account.
reduce In
negativeparticular, they
externalities,
such
allowing
have a a more
Amazon
limited
sustainable
moreflight
sustainable
endurance
delivery
delivery
and
process.
started
process.
capacity.
On
use the
the other
them
Onaddition,
In other hand,
hand,
several
several
several technical
deliveries.
technical
restrictions
aspects
the
aspects
related to
must
must
safety
be
and
carefully
bedrones
carefully
flight area
taken
drastically
taken
must
into
into
be
account.
reduce
account.
considered.
In
In particular,
negativeparticular,
Indeed,
they
externalities,
they
not all
allowing
have a
allowing a more
limited sustainable
flight
a moreflight endurance
sustainable delivery
and
delivery process.
capacity. On
In
process. In the other
addition,
Onaddition, hand,
several
the otherseveral several technical
restrictions
hand, several aspects
related
technical to
aspects must
safety be
and
mustand carefully
flight
be carefullyareataken
must
taken into
be
into account.
considered.In particular,
Indeed,
account. In particular, they
not all
they
have a
countries limited
allow the endurance
use of dronesandin capacity.
the urban area. In this work, restrictions
we provide arelated to
qualitative safety
and flight
quantitative area must
analysis be
on considered.
benefits and Indeed, not
drawbacks all
in
have
have a
countries
a limited
allow
limited flight
the
flight endurance
use of drones
endurance and
andin capacity.
the urban
capacity. In
In addition,
area. In this
addition, several
work,
several restrictions
we provide
restrictions arelated to
qualitative
related to safety
and
safety and flight
quantitative
and flight area must
analysis
area must be
on
be considered.
benefits
considered.and Indeed, not
drawbacks
Indeed, not all
in
all
countries
using drones
drones allow the
in the use
the parcels of
parcels drones
deliveryin the urban
process. area.
We In
analyze this work,
three we
different provide a qualitative
transportation and
systems quantitative
with incrementalanalysis
incremental on
useon benefits
of benefits
drones forand
for thedrawbacks
delivery. Inin
In
countries
using
countries allow
allowin use of
use of drones
drones
deliveryin process.
in the urban
the urbanWe area. In this
analyze
area. In this work,
three we provide
different
work, we provide a qualitative
qualitative
transportation
a and quantitative
systems
and quantitative
with analysis
use
analysis on
of benefits
drones andthe
and drawbacks
delivery.
drawbacks in
in
using drones
particular, we in the
address parcels
the delivery
problem process.
of We
delivering analyze
parcels three different
without drone, transportation
known as systems
vehicle with
routing incremental
problem, the use of
problemdrones
in for
which the
thedelivery.
deliveriesIn
using
using drones
particular,
drones in the
weinaddress
address parcels
the parcels delivery
the problem
problem
delivery ofprocess.
of We
delivering
process. analyze three
parcels three
We analyze different
without transportation
drone,transportation
different known as systems
as vehicle
vehicle with
routing
systems incremental
problem,
withproblem,
incremental use
the use of
problemdrones for
in which
of droneswhich the delivery.
thedelivery.
for the deliveriesIn
In
particular,
are performed we by the
aa fleet delivering parcels without drone, known routing the problem in the deliveries
particular,
are performed
particular,
are performed
we
we address
by
address
by theof
the
fleet
a latter
fleet of
of
drones
problem
drones
problem
drones
starting
of from
of delivering
starting from
delivering
starting from
the
the
the
central
parcels
central
parcels
central
depot,
without
depot,
without
depot,
and
drone,
and aknown
hybrid
and aaknown
drone, hybrid
hybrid
transportation
as vehicle
vehicle routing
transportation
as
transportation
system
system
routing where
problem,
where
problem,
system
the
the
the
the
wheresome
classical
problem
classical
theproblem
classicalinvehicles
in which
which the
vehicles
vehicles
are
are
the equipped
deliveries
equipped
are deliveries
equipped
with
are drones.
performed
withperformed
are drones. In In the
by
Inby a fleet
thea latter
latter case,
of
case,
fleet of the
drones classical
starting
the classical
drones classical vehicles
from the
vehicles
starting from perform
central
perform
the perform the
depot, deliveries
and
the deliveries
central depot, a
deliveries and
hybrid the
and the
and a hybrid the drones
transportation
drones can
transportationcan get
can get
get in
system charge
where
in charge
system charge the
some
wheresome deliveries.
classical The
vehicles
deliveries.
the classical drone
are takes
takes off
are equipped
The drone
vehiclesdrone equippedoff
with
from drones.
the vehicle, the carries case,
out the
the delivery, vehicles
and lands to the the
same vehicle at and
a drones
randevouz-location. in
During the drone deliveries.
delivery, The
the classical takes off
vehicle
with
from
with drones.
the
drones. In
vehicle,
In the
the latter
carries
latter case,
out
case, the
the
the classical
delivery,
classical vehicles
and lands
vehicles perform
to the
perform the
same
the deliveries
vehicle
deliveriesat and
a the drones
randevouz-location.
and the drones can
can get
get in charge
During
in the
charge some
drone
some deliveries.
delivery,
deliveries. The
the
The drone
classical
drone takes off
vehicle
takes off
from the
continues vehicle,
its work. carries
work.carries out
The three
three the delivery,
transportation and lands
systems to
arethe same
formalized vehicle at a randevouz-location.
via mathematical
mathematical During
programmingDuring models. the drone
The delivery,
solutions the
obtained classical vehicle
by solving
solving the
from
from the vehicle,
continues
the vehicle,
its carries
The out the
out the delivery, and
transportation
delivery, and lands are
systems
lands to the
to the same vehicle
formalized
same vehicle
via at aa randevouz-location.
at randevouz-location.
programming During
models. theThe
the drone delivery,
solutions
drone the classical
obtained
delivery, the classical
by vehicle
the
vehicle
continues
models its
viaits work.
a work. The
general-purpose three transportation
solver systems
are compared are formalized via mathematical programming models. The solutions obtained by solving the
continues
models
models via
continues viaitsaa work.
The
The three
general-purpose
general-purpose three transportation
solver
solver are systemsand
systems
are compared
transportationcompared areinsights
are
and
and
formalized
insights
formalized
insights
on
on the
on the viause
via
the use of
of drones in
mathematical
of drones
drones in
mathematical
use
the
the urban area
programming
the urban
urban area
inprogramming area
are
models.
are provided.
The
The solutions
are provided.
models.provided. solutions obtained
obtained by by solving
solving the
the
models via
models via aa general-purpose
general-purpose solver solver are
are compared
compared and and insights
insights on on the
the use
use of
of drones
drones inin the
the urban
urban area
area are
are provided.
provided.

© 2020The
ccc 2020
2020 TheAuthors.
The Authors.
Authors. Published
Published
Published by Elsevier
by Elsevier
by Elsevier B.V. B.V.
B.V.


 c 2020
This
This
c
This isisan
2020
is
2020 anThe
an
The
Theopen
open
Authors.
open access
access
Authors.
access
Authors.
Published
article
Published
article
Published
by Elsevier
under
article by
under
by the CC
under CC
the
Elsevier
the
Elsevier
B.V.
BY-NC-ND
CC BY-NC-ND
B.V.
BY-NC-ND
B.V. licenselicense
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
This is an open
Peer-review access
under article under
responsibility of the scientific
the CC BY-NC-ND committeelicense
of the(http://creativecommons.org/licenses/by-nc-nd/4.0/)
the International Conference on Industry
Industry 4.0 and
and Smart
Smart Manufacturing.
This is an
Peer-review open access
under article under
responsibility CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
This is an open
Peer-review under
access
under responsibility
article under
responsibility of of
of thethe
the CCscientific
scientific
BY-NC-ND
scientific committee
committee
committee of
license
of the ofInternational
the International
International Conference
Conference
Conference on on Industry
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
on Industry 4.0 and4.0
4.0 andManufacturing.
Smart Smart Manufacturing.
Manufacturing.
Peer-review under
Peer-review under responsibility
responsibility of of the
the scientific
scientific committee
committee of of the
the International
International Conference
Conference on on Industry
Industry 4.0
4.0 and
and Smart
Smart Manufacturing.
Manufacturing.
Keywords:
Keywords: urban transportation; last-mile delivery; drones; vehicle routing problem
Keywords: urban urban transportation;
transportation; last-mile
last-mile delivery;
delivery; drones;
drones; vehicle
vehicle routing
routing problem
problem
Keywords:
Keywords: urban urban transportation;
transportation; last-mile
last-mile delivery;
delivery; drones;
drones; vehicle
vehicle routing
routing problem
problem

context,
context, the the expectation
expectation of of the
the customers
customers has has begun
begun high
high in in
context,
context, the
the expectation
expectation of
of the
the customers
customers has
has begun
begun high
high in
in
1.
1. Introduction
Introduction terms
context,
terms of
of quality
the
quality of
expectation
of service.
service.of In
the
In addition,
customers
addition, the
the number
has begun
number of
of online
high
onlinein
1.
1. Introduction
Introduction terms
terms of
of quality
quality of
of service.
service. In
In addition,
addition, the
the number
number of
of online
online
1. Introduction retailers
terms
retailers of is
is constantly
quality of
constantly growing
service.
growing In and customers
addition,
and customers the can
number
can easily
easilyof change
online
change
retailers
retailers is
is constantly
constantly growing
growing and customers
customers can
and completely can easily
easily change
change
The
The last-mile
last-mile delivery
delivery is
is the
the “end
“end node”
node” of of the
the logistic
logistic chain.
chain. their
their on-line
retailers is
on-line shop
shop if
constantly if they
growing
they are
are notand
not customers
completely satisfied.
can easily
satisfied. There-
change
There-
The
The last-mile
last-mile delivery
delivery is the “end node” of the logistic chain. their
their on-line
on-line shop
shop if
if they
they are
are not
not completely
completely satisfied.
satisfied. There-
There-
It refers
It refers
The to the
the delivery
last-mile
to delivery
deliveryofis
ofis the
parcels
the
parcels “end
“endto node”
to node”
the of
of the
the customers
customers
the logistic
and it
logistic
and itchain.
is the
the
chain.
is fore, fulfilling
their
fore, fulfilling
on-line customers’
shop if
customers’ they needs
are
needs notis completely
is one of
one of the
the biggest
biggest challenge
satisfied. There-
challenge
It
It refers
refers to
to the
the delivery
delivery of
of parcels
parcels to
to the
the customers
customers and
and it
it is
is the
the fore,
fore, fulfilling
fulfilling customers’
customers’ needs
needs is
is one
one of
of the
the biggest
biggest challenge
challenge
most
It
most expensive
refers to the
expensive process
delivery
process in
of
in distribution
parcels
distribution to logistics.
the customers
logistics. Indeed,
and
Indeed, the
it
the iscost
the
cost for
fore,
for on-line
fulfilling
on-line retailers.
customers’
retailers. Thus,
Thus, on
needs
on the
is
the one hand,
of the
hand, the
the shopping
biggest challenge
shopping on-
on-
most
most expensive
expensive process
process in
in distribution
distribution logistics.
logistics. Indeed,
Indeed, the
the cost
cost for
for on-line
on-line retailers.
retailers. Thus,
Thus, on
on the
the one
one hand,
hand, the
the shopping
shopping on-
on-
ranges
most from
expensive
ranges from
from 13%13% to 73%
process
13% toto 73%in with respect
distribution
73% with
with respect to the
logistics.
respect to to the total distribution
Indeed,
the total the
total distributioncost
distribution line
for increases
on-line
line increases
increases the the possibility
retailers. Thus,
the possibility
possibility of on of the
the
of the retailers
one hand,
the retailers
retailers to to
the increment
shopping
to increment
increment theirtheir
on-
their
ranges
ranges from 13% to line
line increases the possibility of the retailers to increment their
cost [12].
ranges
cost
cost
[12].
from
[12].
The
The
The13% to 73%
entity
entity
entity
of such
73%
of
of
with
such
such
respect
withaaa cost
cost
respect
cost
to
to the
depends
depends
depends theontotal
on
total
on
distribution
quality of ser-
distribution
quality
quality of
of
ser-
ser-
revenue,
line
revenue,
revenue,
on the
increases
on
on
the
the
the
other
other
other
hand, high
possibility
hand,
hand,
high
of
high theoperational
retailers
operational
operational to costs
costs
costs
have to
increment
have
have
to
to
be
their
be
be
cost
vice,
cost
vice, [12].
[12]. The
dangerous
The
dangerous entity
entity of
nature
of
nature such
of
such
of the
the a
a cost
cost depends
materials
depends
materials on
delivered,
on
delivered,quality
quality of
geographic
of
geographic ser-
ser- revenue,
paid in
revenue,
paid in on
order
on
order the
to
the
to other
guarantee
other
guarantee hand,
hand,the
the high
high operational
expected customers’
operational
expected customers’ costs
costs have
have to
satisfaction.
to
satisfaction.be
be
vice,
vice, dangerous
dangerous nature
nature of
of the
the materials
materials delivered,
delivered, geographic
geographic paid
paid in
in order
order to
to guarantee
guarantee the
the expected
expected customers’
customers’ satisfaction.
satisfaction.
area,
vice, market
dangerous
area, market
market share, and
nature
share, typologies
of the materials
and typologies
typologies of vehicles used.
delivered,
of vehicles
vehicles used. geographic This
paid
This intrend
order
trend force
force the logistic
to guarantee
the logistic
logistic operators
theoperators
expectedto
operators to efficiently
tocustomers’ manage
efficiently satisfaction.
manage the the
the
area, share, and of used. This
This trend
trend force
force the efficiently manage
area,
area, market which
Retailers
market
Retailers
Retailers
share, and
which
share,
which
and typologies
operate
typologies
operate
operate in
in
of vehicles
in e-commerce
of vehiclesface
e-commerce
e-commerce
used.
face
used.
face
several chal-
several
several
chal-
chal-
delivery
This trend
delivery
delivery force the
process;
process;
process;
logistic
therefore,
the logistic the
therefore,
therefore,
operators
the
operators
the
to
to efficiently
implementation
efficiently
implementation
implementation of
of
manage
of novel
novel
manage
novel
the
distri-
the
distri-
distri-
Retailers
lenges. Indeed,
Retailers
lenges. Indeed,which
which operate
on-line shopping
operate
on-line shoppingin
in e-commerce
is becoming
e-commerce
is becoming face
very
face
very several
common
several
common chal-
and
chal-
and delivery
bution
delivery
bution process;
paradigms
process;
paradigms therefore,
is a key
therefore,
is a key the
success
the
success implementation
factor. Since
implementation
factor. Since a
a of
high
of
highnovel
novel distri-
number of
distri-
number of
lenges.
lenges. Indeed,
Indeed, on-line
on-line shopping
shopping is
is becoming
becoming very
very common
common and
and bution
bution paradigms
paradigms is
is aa key
key success
success factor.
factor. Since
Since aa high
high number
number of
of
the demand
lenges.
the of
Indeed,
demand of same-day
on-line
same-day deliveries
shopping
deliveries is has exponentially
becoming
has very
exponentially grown
common
grown in
and
in delivery
bution
delivery requests
paradigms have
requests haveis a key
have to be
to be
be satisfied
success
satisfied in
factor. the
Since
in the
the same-day
same-day and
a high number
and inin the
of
in the
the
the demand
the demand
demand of
of same-day
same-day deliveries
deliveries has
has exponentially
exponentially grown
grown in
in delivery
delivery requests
requests have to
to be satisfied
be satisfied
satisfied in
innumber same-day
the same-day
same-day and
and inin the
the last years.
last years.of More and
same-day
years. More
More and more
deliveries
and more
more peoplepeople
has prefer to
exponentially
people prefer
prefer to use
to use shopping
grown
use shopping
shopping in same urban
delivery
same urban area, have
requests
area, it is
it is expected
expected
to a high
a highin number
the of delivery
of delivery
and vehi-
the
vehi-
the last same
same urban
urban area,
area, it
it is
is expected a high number of delivery vehi-
the last
on-line
the last
on-line
on-line
years.
instead
years.
instead More
to
More
to
instead to
and
buy
and
buy more
items
more
items
buy items
from
frompeople
people prefer
conventional
prefer
conventional
from conventional
to
to use
use shopping
shop. In
shopping
shop.
shop. InIn this
this
this
cles
same
cles
cles
on
on
on
the
urban
the
the
roads
area, at
roads
roads at
at is expected
it any
any
any
time
expected
time
time
of
of
of
aa high
the
the
the
day.
high
day.
day.
number
As
number
As
As
of
of delivery
aa result,
aa result,
result,
the
the
the
vehi-
increase
delivery vehi-
increase
increase
on-line
on-line instead
instead to
to buy
buy items
items fromfrom conventional
conventional shop.shop. InIn this
this cles
of
cles
of on
on the
same-day
the
same-day roads
roads at
deliveries
at
deliveriesany
any time
in
time
in the
theof
of the
urban
the
urban day.
area
day.
areaAs
As result,
causes
a result,
causes the
several
the
severalincrease
nega-
increase
nega-
of
of same-day
same-day deliveries
deliveries in the
in the urban
urban area
the presence area causes
causes several
several nega-
nega-
tive
of externalities.
tivesame-day
externalities. Indeed,
deliveries
Indeed, in the
the urban
presence area of this
this high
of causes number
several
high number nega-of
of
∗ Corresponding author. tive
tive externalities.
externalities. Indeed,
Indeed, the
the presence
presence of
of this
this high
high number
number of
of
∗
∗ Corresponding
Corresponding author.
author. vehicles
tive
vehicles on
on the
externalities.
the urban
Indeed,
urban road
road system
the presence
system increases
of
increases this congestion,
high
congestion,numbernoise,
noise,of
∗ E-mail address:author.
∗ Corresponding
E-mail address: [email protected]
address:author. (Luigi
[email protected] (Luigi Di
(Luigi Di Puglia
Di Puglia
Puglia vehicles
vehicles on
on the
the urban
urban road
road system
system increases
increases congestion,
congestion, noise,
noise,
Corresponding
E-mail [email protected] and
and has
vehicles
has a
on
a bad
the
bad impact
urban
impact on
road
on the
the air
system
air quality in
increases
quality in the urban
congestion,
the urban area, due
noise,
area, due
Pugliese).
E-mail
Pugliese). address: [email protected] (Luigi Di
E-mail address: [email protected] (Luigi Di PugliaPuglia and
and has
has a bad impact on the
the air quality in the
the urban area,
area, due
Pugliese).
Pugliese).
Pugliese).
to
to the
and
to thehas
the COaa222bad
CO
CO
bad impact In
emissions.
impact
emissions.
emissions.
onorder
on
In
In the
order
order
air
to
air quality
face
face the
to quality
to face thein
the
innegative
the
negative
negative
urban
urbanenvironmen-
area,
environmen-
environmen-
due
due
c
to
tal
to
talthe
the CO
impacts
CO
impacts emissions.
of the
2 emissions.
of the In
In order
traditional
order
traditional to
to face
deliveries,
face
deliveries,the
the negative
some
negative
some environmen-
companies
environmen-
companies have
have
2351-9789
2351-9789
2351-9789 cc ©2020
 2020 TheThe
2020
The Authors.
Authors. Published
Authors.
Published by Elsevier
by Elsevier
Published byB.V.
Elsevier B.V.
by Elsevier B.V.
tal impacts of 2 the traditional deliveries, some companies have
2351-9789
This isisan
Thisis anan
2351-9789

open

2020
cc open
2020
Thearticle
access Authors.
access
The
Published
under
article
Authors. the CC BY-NC-ND
under
Published BY-NC-ND
theElsevier
by
B.V. tal
tal impacts
impacts of of the
the traditional
traditional deliveries,
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
CC BY-NC-ND
B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/)
license (http://creativecommons.org/licenses/by-nc-nd/4.0/) deliveries, somesome companies
companies have have
This
2351-9789
This is an open

open access
2020 The
access article
article under
Authors.
under the
the CC
Published
CC by Elsevier
BY-NC-ND license
B.V.
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
Peer-review
This is an
Peer-review
This open
is an open
Peer-review under
under
access
under responsibility
responsibility
article
responsibility
access
under article
of
under the
of the
under
responsibility of of the
CC scientific
scientific
the scientific
scientific
CC
committee
BY-NC-ND
BY-NC-ND committee
of
license
committee
license
committee of
the of the International
International Conference on Conference
Industry 4.0 andonSmart
and
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
of the
the International Conference onon Industry
Industry 4.0 and
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
International Conference 4.0 Industry
Smart
Smart 4.0 and Smart Manufacturing.
Manufacturing.
Manufacturing.
Manufacturing.
Peer-review under
under responsibility
responsibility of
10.1016/j.promfg.2020.02.043
Peer-review of the
the scientific
scientific committee
committee of
of the
the International
International Conference
Conference on
on Industry
Industry 4.0
4.0 and
and Smart
Smart Manufacturing.
Manufacturing.
 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497 489
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 2

started to introduce the use of alternative vehicles. Electric ve- taken into account in [8]. In addition, [8] introduced the con-
hicles are common in several urban areas. They reduce noise cept of synchronization between truck and drone. In particular,
and CO2 emissions ([18, 19]) but the use of these vehicles con- the drone can wait the truck for a limited amount of time after it
tinue to generate congestion and do not solve the problem of performed the delivery. This assumption is made for safety and
efficiency in last-mile delivery process. security reasons. The introduction of a limitation on the waiting
With the aims of gaining from the possibility offered by time has an impact on the behaviour of the transportation sys-
the new behavior of the customers, managing in a profitable tem. The speed of the drones with respect to that of the trucks
way the own resources and finding new eco-friendly solutions, play a crucial role in the definition of deliveries when synchro-
several big on-line retailers, such as Amazon and DHL, have nization issues are taken into account. In particular, the higher
started to promote new delivery processes. the speed of the drone with respect the speed of the truck, the
One of the most promising distribution paradigm is based higher the possibility of exceed the maximum allowed waiting
on crowd-shipping. Ordinary people accept to deliver parcels time, making several drone-deliveries infeasible [8]. It follows
to other people for a small compensation ([6, 20, 21, 22]). Re- that the insight provided in [11] for the TSP-D is no longer
cently, Amazon and DHL have started to use unmanned aerial valid.
vehicles, common known as drones, for same-day last-mile de- In this paper we analyze the behavior of a transportation sys-
livery process [30]. On the one hand, the use of drones in the tem where trucks and drones are used to perform deliveries. We
delivery process reduces noise and avoids both the increase of take into account realistic scenario considering time window
congestion in the urban area and CO2 emissions. On the other and synchronization constraints. Benchmark instances, used by
hand, drones have a lower capacity (they can carry low weight the scientific literature to analyze RPs, are considered for the
parcels) and working time (low flight endurance) than classical empirical evaluation. The system is analyzed under both cost
vehicles. Overall, drones have a positive impact. Indeed, their and negative externalities efficiency. The resulting VRP-D is
use reduces the lead time and both the makespan [28] and the compared with the VRP and the RP where only drones are used
transportation costs [8] of the delivery process. for delivering parcels, i.e., RPD. The contribution of the paper
The scientific literature refers to the routing of vehicles for is to provide a comparative and systematic analysis of the con-
delivering parcels as routing problems (RPs). The RPs are clas- sidered transportation systems by highlighting the benefits and
sified into two main classes, i.e., travelling salesman problem the drawbacks of the delivery process in an urban area.
(TSP) when a vehicle without capacity constraint is used, and The reminder of the paper is organized as follows. Section 2
vehicle routing problem (VRP) when several capacitated ve- formally introduces the addressed problems. Section 3 presents
hicles are considered. Several variants of both TSP and VRP the results of the computational analysis. Section 4 concludes
have been studied by considering pickup and delivery opera- the paper.
tions, multi trip assumptions, the presence of one or more cen-
tral depots, intermediate depots, and time window constraints
that impose the interval of time in which each customer has to 2. Transportation systems definition
be served. For more details on both TSP and VRP and their
variants, the reader is referred to [4, 10, 14, 15, 17]. In this Section we provide a formal description of the con-
The RPs with the aid of drones was introduced in [28] where sidered transportation systems. Firstly, we give the common no-
the cooperation of trucks and drones was studied under a theo- tations, then we provide the mathematical formulations of the
retical point of view. In particular, the authors provided bounds three cases.
on the makespan for the TSP with drones (TSP-D) and the VRP Let V = N ∪{0, n+1} be the set of nodes composed of the set
with drones (VRP-D). A comparison with the TSP and VRP of n customers N, the depot 0 and its copy n+1. Let A be the set
was carried out concluding that the combination of trucks and that contains all possible connections (i, j) between each pair of
drones leads to a short makespan. This result suggests that the nodes i, j ∈ V. In particular, A = {(i, j) : i ∈ N ∪ {0}, j ∈ N ∪
use of drones in combination with trucks improves the perfor- {n + 1}}. We define the problems over a complete graph G(V, A).
mance of the delivery process in term of completion time. A delivery request qi is associated with each customer i ∈ N.
The TSP-D was studied in [1, 11, 23] providing several in- Let η(qi ) and w(qi ) be the number of parcels and the weight
sights on the behaviour of this transportation system. In par- of request qi , respectively. A time window [ai , bi ] is associated
ticular, in [11] was proved empirically that the best results with each customer i ∈ N, where ai is the earliest service time
are obtained when the speed of the drones is twice that of and bi is the latest service time to customer i ∈ N. This means
the trucks. The VRP-D was considered in [24, 28]. The au- that customer i have to be served within [ai , bi ]. Let sti and sdi be
thors addressed simplified version where trucks are not capaci- the time needed to serve customer i ∈ N when the delivery to i
tated, and no flight endurance limitation was considered for the is performed by the trucks and he drones, respectively.
drones. They minimized the completion time. From these sem- A distance is associated with each arc (i, j) ∈ A. We suppose
inal works, several contributions appeared by considering more a Manhattan metric for the trucks and the Euclidean metric for
realistic scenarios. In particular, in [7] the authors took into the drones. Let dit j and didj be the distance travelled by the truck
account the recharging of the battery of the drones after each and the drone, respectively, from node i ∈ N ∪ {0} to node j ∈
drone-delivery. In [8, 25], the minimization of the transporta- N ∪ {n + 1}. Let vt and vd be the speed of the trucks and the
tion cost was considered and time window constraints were drones, respectively. Thus, the time to traverse an arc (i, j) ∈ A
2
490 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 3

is given by tit j = dit j /vt and tidj = didj /vd for the trucks and the have to be served by exactly one truck. Constraints (6) define
drones, respectively. Costs per distance travelled/flight cti j and the weight of parcels that the trucks carry along the arcs (i, j).
cdij have to be paid along the arc (i, j) ∈ A by the trucks and the Constraints (7) impose that the weight carries along each arc
drones, respectively. (i, j) ∈ A must not exceed the capacity of the truck. Constraints
We assume a maximum allowed waiting time for the drone (8) impose the trucks to end their route empty. Constraints (9)
defined as T . Thus, a drone can wait at most T instant time for define the arrival time to each customer j ∈ N. Constraints (10)
the truck in order to land on. impose the service to each customer j ∈ N within its time win-
We assume that each truck and each drone has a capacity dow. Mi j is a sufficiently large number.
W t and W d , respectively. In addition, each drone can carry one
parcel for each drone-delivery. We assume that energy con-
sumption of the drones is proportional to the distance it flies. 2.2. Drone routing problem (RPD)
Given the battery fully charged at the beginning of each drone-
delivery, we retrieve the maximum distance E that each drone This problem addresses the case in which only drones per-
can fly for each drone-delivery. form deliveries. It is assumed that each drone starts/ends its de-
We consider a limited number of trucks and drones. Let K livery from/at the depot. Among all customers N, only a subset
and D be the sets of available trucks and drones, respectively. of them can be served by the drones due to both the character-
istics of the parcels and the limitation on the flight endurance.
2.1. Truck routing problem (VRP) In particular, we define the set N d = N d (q) ∪ N d (E), where
N d (q) = {i ∈ N : η(qi ) = 1, w(qi ) ≤ W d } is the set of cus-
In this problem we consider only trucks for the delivery pro- tomers whose request qi can be satisfied by the drones and
d d
cess. Let xikj be binary variables taking value equal to one if the N d (E) = {i ∈ N d (q) : d0i + din+1 ≤ E} is the set of customers
truck k ∈ K travel along the arc (i, j) ∈ A, zero otherwise. Let zi j whose distances from the depot is compatible with the flight
be the weight of the parcels carried by the trucks along the arc endurance of the drones.
(i, j) ∈ A. Let τki be continuous variables indicating the instant Let yd0in+1 be binary variables taking value equal to one if
time in which the customer i is served by the truck k ∈ K. The drone d ∈ D serves customer i ∈ N d . Let τdi be the instant time
variable τk0 is the starting time of a route performed by the truck in which the customer i ∈ N d is served by the drone d ∈ D. The
k ∈ K. The VRP can be mathematically formulated as follows. variable τd0 is the starting time of the delivery performed by the
drone d ∈ K. We assume that each drone can perform at most
 one delivery. The mathematical formulation for RPD is given
min C t (x) = cti j dit j xikj (1) below.
i∈V j∈V k∈K
s.t 
 
x0k j − k
xi,n+1 = 0, ∀k ∈ K; (2) min C d (y) = (cd0i d0id + cdin+1 din+1
d
)yd0in+1 (11)
j∈N i∈N i∈N d d∈D
  s.t
k
xih − xhk j = 0, ∀h ∈ N, k ∈ K; (3) 
i∈V j∈V
yd0in+1 = 1, ∀i ∈ N d ; (12)
 d∈D
x0k j ≤ 1, ∀k ∈ K; (4) 
j∈N
yd0in+1 ≤ 1, ∀d ∈ D; (13)
 i∈N d
xikj = 1, ∀ j ∈ N; (5) d
M(yd0in+1 − 1) + τd0 + t0i ≤ τdi , ∀d ∈ D, i ∈ N d ; (14)
i∈N k∈K
 M(yd0in+1 − 1) + τdi + sdi + tin+1
d
≤ τdn+1 , ∀d ∈ D, i ∈ N d ; (15)
  
ω(qi ),
 ∀i ∈ N,
z ji − zi j =  (6) M(yd0in+1 − 1) + τdi − t0i
d
− τd0 ≤ T, ∀d ∈ D, i ∈ N d ; (16)
i∈N −ω(qi ), i = 0,

j|( j,i)∈A j|(i, j)∈A
 a j ≤ τdj ≤ b j , ∀d ∈ D, j ∈ N d . (17)
zi j ≤ W t xikj , ∀(i, j) ∈ A; (7)
k∈K
zin+1 = 0, ∀i ∈ N; (8) Objective function (11) minimizes the flight cost. Con-
Mi j (xikj − 1) + τki + sti + tit j ≤ τkj , ∀k ∈ K, i ∈ V, j ∈ V; (9) straints (5) impose that each customer i ∈ N d is served by ex-
a j ≤ τkj ≤ b j , ∀ j ∈ N, k ∈ K. (10) actly one drone. Constraints (13) impose that each drone can
serve at most one customer i ∈ N d . Constraints (14) define
the arrival time to customer i when it is served by the drone d.
Objective function (1) minimizes the travelled cost. Con- Constraints (15) define the arrival time of drone d to the depot.
straints (2) impose that if truck k ∈ K starts a route, then it must Constraints (16) allow a maximum waiting time T of drone d at
end the route at node n + 1. Constraints (3) balance the flow at customer i before starting the service. Constraints (17) impose
each customer node h ∈ N. In particular, if the truck k visits the service of drone d to take place within the time window as-
customer i, then it must leave the node i after the service. Con- sociated with customer i. M is a sufficiently large number. We
straints (4) impose that each truck k ∈ K can perform at most highlight that the capacity constraint and the maximum allowed
one route. Constraints (5) impose that each customer j ∈ N flight distance E are included in the definition of set N d . This
3
 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497 491
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 4

means that all drone-deliveries are feasible with respect to the The emissions of CO2 in kg is defined by the following func-
capacity constraint and the flight endurance requirement. tion

2.3. Truck-Drone routing problem (VRP-D)

 
gt (z, x) = f (zi j )dit j xikj . (18)
We consider the transportation system described in [8]. We k∈K (i, j)∈A
briefly describe the characteristics of the delivery process and
the assumptions made with respect to the behaviour of the
drones, the trucks, and their cooperation. Each truck is equipped Function gt (z, x) is used to estimate the CO2 in [19] where
with a drone. A subset of customers N t is served by the trucks, the VRP with mixed fleet composed of conventional trucks
whereas the customers N d = N \ N t are served by the drones. (diesel) and electric vehicles is considered.
The trucks perform their route to serve the customers included The drone is characterized by zero emissions since it is an
in N t . The drones which take in charge the delivery of parcels electric vehicle. However, the energy consumed by the drone is
for the customers in N d perform the drone-delivery in parallel produced by power generation facilities. The process of power
with the associated truck. This means that the drone-deliveries generation produces CO2 emissions. Thus, we take into account
take place while the trucks perform their routes. We note that the CO2 emitted by power generation facilities. In [13] was esti-
the definition of subsets N t and N d is a decision. This means mated that 0.3773 kg of CO2 is emitted for each kWh produced.
that we do not assume a-priori information related to the cus- Let β be the Wh consumed by the drone per km, the CO2 emis-
tomers served by the drones and the trucks. A drone-delivery sion for the drone-deliveries can be estimated by considering
is characterized by the tuple i, w, j; in particular, i is the node the following function
associated with the customer where the drone takes off from
the truck, w is the customer served by the drone, and j is the
node associated with the customer where the drone lands on the    
truck. The customers i and j belong to the route of the truck gd (y) = β × 3.773(10−4 )× d
diw + dwd j ydiw j (19)
whose the drone, performing the drone-delivery i, w, j, is as- i∈V w∈N d j∈V d∈D
sociated. If the drone arrives to customer j with T instant time
earlier than the truck, then the drone-delivery i, w, j is de-
The value of β depends on the characteristics of the drone.
clared infeasible. In other words, the waiting time of the drone
In [13] the authors consider β ∈ [10, 100].
for the truck must not exceed T instant time. The mathematical
formulation for the VRP-D is a combination of models (1)–(10)
and (11)–(17) with further constraints related to the synchro-
nization between each truck and the associated drone. For the 3. Computational analysis
mathematical model of the VRP-D, the reader is referred to [8].
In this Section we analyze and compare the results obtained
by solving the three models associated with the three trans-
2.4. Discussion on CO2 emission models portation systems, i.e., VRP, RPD, and VRP-D. The mathemat-
ical formulations are implemented in Java language and solved
In [27] the CO2 emissions are estimated by considering two by using CPLEX 12.5.
elements, i.e., the distance travelled and the weight carried by The experiments are carried out on an Intel Core i7-4720HQ
the truck. In particular, it is computed an emission factor f (z) CPU 2.60 GHz with 8 GB RAM under Microsoft 10 operating
of CO2 in terms of kg/km as a function of the load of the truck. system. We consider instances inspired by the scientific litera-
By using the chemical reaction proposed by Lichty [16], the ture. The test problems are described in what follows.
kg of CO2 per litre of diesel consumed is 2.61. Thus, know-
ing the consumption of diesel, the emission factor f (z) is de-
termined. In [27] is reported the consumption of diesel for five 3.1. Test problems
different load configurations. The emission factors f (z) along
with the consumption of diesel for the considered load configu- The test problems are generated by starting from the well-
rations are reported in Table 1. known Solomon benchmarks for the VRP with time windows
[26]. These benchmarks are grouped in three classes: 12 ran-
Table 1. Estimation of emission factors for a truck with 10-tonne capacity [19]. dom instances (R2) where the customers are randomly dis-
Load of Weight laden Consumption f (z) tributed, 8 clustered instances (C2) where subsets of customers
the vehicle (%) (litre/100km) (kg CO2 /km) are randomly distributed, and 8 random clustered instances
Empty 0 29.6 0.77 (RC2) where some customers are randomly distributed and the
Low loaded 25 34.0 0.83 remaining ones are clustered. Each instance is characterized of
Half loaded 50 34.4 0.90 100 costumers and the depot 0. In order to generate the in-
High loaded 75 36.7 0.95 stances considered in this work, we have adapted the bench-
Full load 100 39.0 1.01 marks R2, C2, and RC2 as described in what follows.
4
492 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 5

The depot 0 and its copy n + 1 are located as in the original drone capacity and close enough to the depot such that drone
instances. We choose n ∈ {5, 10, 15} customers among the orig- can serve each customer and return back to the depot within the
inal ones randomly, maintaining the information related to the limited flight endurance.
position, the demand, and the time windows.
In order to take into account the limitations related to drone- Analysis on transportation cost. Looking at Table 5, we ob-
delivery, we impose that only the 80% of the customers can be serve that the most expensive transportation system is VRP,
served by the drones. In particular, we generate the set of cus- followed by VRP-D and RPD. Indeed, the latter shows a trans-
tomers N d (q) ⊂ N such that |N d (q)| = 0.80n. In particular, the portation cost equal to 334.33. However, due to the technical
0.80n customers with associated a low value of ω{qi } are in- limitations of the drones, only the 56% of the customers are
cluded into N d (q). The service time for the customers served by served, on average (see column #d). This means that, on aver-
the trucks, i.e. sti , ∀i ∈ N is set equal to the original service time, age, the 44% of the requests cannot be taken in charge. This
whereas, sdi = 1/2sti , ∀i ∈ N d (q). We assume η(qi ) = 1, ∀i ∈ N, drawback causes a severe reduction of the quality of service.
and we set E = max(i, j)∈A didj , cti j = 25 × cdij , vt = vd , β = 10, The transportation system VRP-D takes advantages in term of
and T equal to 10 for classes R2 and RC2 and 90 for class C2. cost related to the use of drones and overcome the drawback
We highlight that the values chosen for the speeds of the trucks related to the quality of service, thanks to the use of trucks
and the drones allow to obtain the highest probability of feasi- that can perform the deliveries that drones cannot fulfill. Thus,
ble drone-deliveries in presence of synchronization constraints VRP-D shows the best trade-off between quality of service and
between a truck and the associated drone (see [8]). In addition, effectiveness. Indeed, all customers are served and we observe
the higher transportation cost ct of the trucks than that of the an average reduction of the overall transportation cost C(x, y)
drone, i.e. cd , is justified by several reasons, like courier cost of 39% with respect to the cost obtained with VRP.
and fuel consumption cost among the others (see, e.g. [29]). In what follows, we analyze the average transportation cost
per customer served by the trucks and the drones, separately.
3.2. Numerical results We refer to Cct (α) and Ccd (α) as the cost per unit of customer
served by the trucks and the drones, respectively, considering
Tables 2–4 summarize the average results for each trans- the transportation system α ∈ {VRP, RPD, VRP − D}. On av-
portation system and each class of test problems, considering erage, Cct (VRP) is 797.59 whereas Cct (VRP − D) is 1037.59
the instances with 5, 10 and 15 customers, respectively. Table (see Table 5). The same observation can be done when con-
5 reports the numerical results for each transportation system sidering the use of drones. Indeed, Ccd (RPD) is 50.52 whereas
averaged over all instances and classes of test problems. Ccd (VRP − D) is 52.71. In addition, the number of trucks used
We impose a time limit for the resolution of each instance in VRP-D is 1.29 lower than the number of trucks used in VRP
of 30 minutes. Thus, if the considered instance is solved within and the number of drone-deliveries performed in VRP-D is 1.19
the time limit, the solution obtained is the optimal one. If the times less than that observed for RPD. These results suggest
resolution process takes more than 30 minutes, the solution re- that the use of trucks and drones in both VRP and RPD is more
turned by CPLEX is a feasible one but not optimal. In the latter efficient than that observed for VRP-D. However, the combina-
case, CPLEX provides the optimality gap that gives an idea on tion of trucks and drones leads to a better organization of the de-
how far is the returned feasible solution from the optimal one. livery process in terms of overall transportation cost (see Table
Tables 2– 5 show the execution time under column time, 5, column C(x, y)). The observed average trend is the same for
the optimality gap under column gap, the overall transportation each value of |N|, as shown in Fig. 1 and Fig. 2, where Cct (α) for
cost, i.e. C(x, y) = C t (x) + C d (y), under column C(x, y), and VRP and VRP-D and Ccd (α) for RPD and VRP-D are depicted,
the overall CO2 emissions, i.e. g(z, x, y) = gt (z, x) + gd (y), un- respectively.
der column g(z, x, y). The rest of each table is divided into two
parts, both of them have four columns. In the first part, named
“Truck”, we report information related to the trucks, whereas
the second one, named “Drones” we report information related
to the drones. The empty entries means that no information is
available from the resolution process. In particular, for the VRP
and the RPD the second and the first parts are empty since no
drones and no trucks are considered, respectively. The first and
the second parts of the Tables report the transportation costs un-
der columns C t (x) and C d (y), the CO2 emissions under columns
gt (z, x) and gd (y), the number of trucks and drones used for the
deliveries under columns #v and #d, respectively, the number of Fig. 1. Average values of Cct (VRP) and Cct (VRP − D) at varying the number of
arcs travelled by the trucks under column arcs, and the poten- customers |N|.
tial number of drone-deliveries, i.e., the number of customers
for which a feasible drone-delivery exists, under column |N d |. An interesting trend is observed for Cct (α), see Fig. 1. Indeed,
We recall that N d contains all customers that can be served by the higher the number of customers |N|, the lower the cost per
a drone. In other words, customers with requests compatible to customer served by the trucks. An inverted trend is observed
5
 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497 493
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 6

emissions. In addition, the higher the number of customers to

be served the higher the differences between the CO2 emitted
with VRP and VRP-D (see Tables 2–4). In particular, the CO2
emissions of VRP are 1.64, 2.03, and 2.04 times higher than
those emitted with VRP-D, considering |N| equal to 5, 10, and
15, respectively. This trend is depicted in Fig. 4, where the CO2
emitted for VRP and VRP-D is plotted at varying the number
of customers to be served.

Fig. 2. Average values of Ccd (RPD) and Ccd (VRP − D) at varying the number of
customers |N|.

when the transportation cost per customer served by the drones

Ccd (α) is considered, see Fig. 2.
Despite the lower values of Cct (VRP) than the values of
Cct (VRP−D), the overall transportation cost for VRP-D is lower
than that observed for VRP. The trend of C(x, y) for VRP and
VRP-D at varying the number of customers is depicted in Fig.
Fig. 4. Average values of g(z, x, y) for VRP and VRP-D at varying the number
3. of customers |N|.

It is interesting to note that the CO2 emitted per arcs trav-

elled by the trucks for VRP and VRP-D is the same and equal
to about 15. The low value of emissions for VRP-D is justified
by the lower arcs travelled than those travelled for VRP. In par-
ticular, the number of arcs used for VRP is about twice that for
VRP-D, on average, see Table 5. This trend is observed for each
value of |N| as shown in Tables 2–4.
The number of arcs travelled gives us a general idea on the
congestion generated by the deliveries performed by the trucks.
As expected, VRP-D uses a less number of arcs for the de-
Fig. 3. Average values of C(x, y) for VRP and VRP-D at varying the number of
livery process. In particular, #arcs for VRP is 1.65, 2.00, and
customers |N|.
2.22 times higher than that observed for VRP-D, considering
|N| equal to 5, 10, and 15, respectively (see Tables 2–4, col-
This trend can be justified by considering the number of cus- umn #arcs). Fig. 5 shows the average number of arcs travelled
tomers served by the trucks. Whilst for VRP all customers are for VRP and VRP-D at varying the number of customer to be
served by the trucks, for VRP-D only the 57%, 44%, and 40% served |N|.
are served by the trucks, considering |N| equal to 5, 10, and 15,
respectively. Thus, the incidence of the transportation cost re-
lated to the delivery performed by the trucks reduces for VRP-D
with increasing of the number of customers to be served. In-
deed, Fig. 3 highlights that the transportation cost for VRP-D
is 1.45, 1.59, and 1.60 times lower than the cost paid for VRP,
considering |N| equal to 5, 10, and 15, respectively.

Analysis on CO2 emissions and congestion. As expected, RPD

produces on average the lowest amount of emissions (see Ta-
bles 2–5). However, we highlight that these emissions are not
released in the urban area, but they refers to the emissions of
the power generation plants which produce the energy used by Fig. 5. Average number of arcs travelled considering VRP and VRP-D at vary-
the drone during the deliveries. We also highlight that not all ing the number of customers |N|.
customers can be served by the drones for RPD.
In what follows we focus our attention on VRP and VRP- The increasing difference of #arcs between VRP and VRP-D
D. Looking at Table 5, we observe that the CO2 emitted with for increasing values of |N| is justified by the number of cus-
VRP is 48% higher than the emissions with VRP-D. This is tomers served by the drones. Indeed, looking at Tables 2–4, the
an expected results, since the drones produce a very low CO2 percentage of requests taken in charge by the drones, calculated
6
494 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 7

as #d/|N| × 100, is 43%, 56%, and 60%, considering |N| equal data. However, some of them, like time to traverse the arcs, ser-
to 5, 10, and 15, respectively. vice time, fuel consumption of the truck, energy consumption
As expected, VRP-D uses a less number of trucks. In par- of the drones are uncertain. It should be interesting to analyze
ticular, #v for VRP is 1.29 times higher than that observed for the three transportation systems considering such uncertainties
VRP-D, on average. This means that VRP-D gives benefits in within a robust optimization framework [2, 3, 5, 9].
terms of congestion since less trucks enter in the urban area.
References
Final remarks. The computational results highlight the draw-
back of using only drones for the delivery process. Indeed, for [1] Agatz, N., Bouman, P., Schmidt, M., 2018. Optimization approaches for
the instances considered in this paper, for the 44% of the cus- the traveling salesman problem with drone. Transportation Science 52,
tomers, on average, the request is not satisfied. However, RPD 739–1034.
gives high benefits in terms of transportation cost, CO2 emis- [2] Agra, A., Christiansen, M., Figueiredo, R., Hvattum, L., Poss, M., Requejo,
sions and congestion. Indeed, the cost per customers served is C., 2013. The robust vehicle routing problem with time windows. Com-
puters and Operations Research 40, 856–866.
50.52 against the 797.59 observed for VRP, on average. The [3] Alves Pessoa, A., Di Puglia Pugliese, L., Guerriero, F., Poss, M., 2015. Ro-
CO2 emissions is 144.12 times lower than those observed for bust constrained shortest path problems under budgeted uncertainty. Net-
VRP and the delivery process performed by the drones does not works 66, 98–111.
cause congestion. VRP-D takes the advantages of using drones [4] Bektas, T., 2006. The multiple traveling salesman problem: An overview
in the delivery process overcoming the drawbacks. Indeed, the of formulations and solution procedures. Omega 34, 209–219.
[5] Bertsimas, D., Sim, M., 2003. Robust discrete optimization and network
request that cannot be satisfied by the drones, are fulfilled by the flows. Mathematical Programming 98, 49–71.
trucks. VRP-D presents a lower transportation cost than VRP, [6] Dahle, L., Andersson, H., Christiansen, M., Speranza, M.G., 2019. Crowd-
i.e., a reduction of about 39% is observed. The emissions are shipping with time windows and transshipment nodes. Computers and Op-
drastically reduced thanks to the possibility of using drones. erations Research 109, 122–133.
A reduction of about 48% is obtained for VRP-D in compari- [7] Daknama, R., Kraus, E., 2018. Vehicle Routing with Drones. Technical
Report. arXiv preprint arXiv: 1705.06431v1.
son with VRP. In addition, a less number of trucks are used for [8] Di Puglia Pugliese, L., Guerriero, F., 2017. Last-mile deliveries by using
VRP-D and they travel along a less number of arcs with respect drones and classical vehicles, in: Sforza, A., Sterle, C. (Eds.), International
to the number of trucks and number of arcs travelled consider- Conference on Optimization and Decision Science, ODS 2017, Springer
ing VRP. These results highlight the potential of VRP-D of re- Proceedings in Mathematics and Statistics. Springer New York LLC. pp.
ducing the congestion in the urban area. Overall, VRP-D is the 557–565.
[9] Di Puglia Pugliese, L., Guerriero, F., Poss, M., 2019. The resource con-
transportation system that provides the better trade-off among strained shortest path problem with uncertain data: A robust formulation
transportation cost, CO2 emissions and congestion. and optimal solution approach. Computers and Operations Research 107,
140–155.
[10] Dixit, A., M.A.S.A., 2019. Vehicle routing problem with time windows
4. Conclusions using meta-heuristic algorithms: A survey. dvances in Intelligent Systems
and Computing 741, 539–546.
[11] Ferrandez, M., Harbison, T., Weber, T., Sturges, R., Rich, R., 2016. Opti-
In this paper we analyzed the drawbacks and the benefits mization of a truck-drone in tandem delivery network using k-means and
of using drones in the delivery process in an urban area. We genetic algorithm. Journal of Industrial Engineering and Management 9.
considered three types of transportation systems, i.e., a deliv- [12] Gevaers, R., V.D., Voorde, E., Vanelslander, T., 2009. Characteristics of
ery process where only trucks are used, referred to as VRP, one innovations in last-mile logistics - using best practices, case studies and
where only drones are taken into account, referred to as RPD making the link with green and sustainable logistics, in: European Trans-
port Conference, Leiden, Netherlands. p. 21.
and a hybrid transportation system where trucks are equipped [13] Goodchild, A., Jordan, T., 2018. Delivery by drone: An evaluation of un-
with drones, in this configuration both typologies of vehicles manned aerial vehicle technology in reducing CO2 emissions in the deliv-
are used to deliver parcels. The three delivery processes are ery service industry. Transportation Research Part D: Transport and Envi-
analyzed in terms of transportation cost, CO2 emissions and ronment 61, 58–67.
congestion. We provided mathematical formulations used to [14] Laporte, G., 1992a. The traveling salesman problem: an overview of exact
and approximate algorithms. European Journal of Operational Research
solve instances inspired by the scientific literature. The com- 59, 231–247.
putational results highlighted the drawback of RPD that is not [15] Laporte, G., 1992b. The vehicle routing problem: An overview of exact
able to fulfill all the requests and the benefits in terms of trans- and approximate algorithms. European Journal of Operational Research
portation cost, emissions and congestion. VRP is the less effi- 59, 345–358.
cient transportation system. Indeed, it presents the highest cost, [16] Lichty, L.C., 1967. Combustion Engine Processes. McGraw-Hill Educa-
tion, New York.
emissions, and congestion among the considered transportation [17] Lust, T., T.J., 2010. The multiobjective traveling salesman problem: A
systems. VRP-D is able to serve all customers, overcoming the survey and a new approach. Studies in Computational Intelligence 272,
drawbacks related to RPD taking all the advantages from the 119–141.
drones. Indeed, VRP-D presents a lower cost, emission and [18] Macrina, G., Di Puglia Pugliese, L., Guerriero, F., 2019a. An energy-
congestion than those observed for VRP. Thus, the numerical efficient green-vehicle routing problem with mixed vehicle fleet, partial
battery recharging and time windows. European Journal of Operational
results suggested that VRP-D has the best trade-off between ef- Research 276, 971–982.
ficiency and reduction of negative externalities, i.e., CO2 emis- [19] Macrina, G., Di Puglia Pugliese, L., Guerriero, F., 2019b. The green mixed
sions and congestion. In this work we considered deterministic fleet vehicle routing problem with partial battery recharging and time win-
7
 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497 495
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 8

dows. Computers and Operations Research 101, 183–199.

[20] Macrina, G., Di Puglia Pugliese, L., Guerriero, F., Laganà, D., 2017. The
vehicle routing problem with occasional drivers and time windows, in:
Sforza, A., Sterle, C. (Eds.), Optimization and Decision Science: Method-
ologies and Appliations, ODS, Sorrento, Italy. Springer, Cham, Switzer-
land. pp. 577–587.
[21] Macrina, G., Di Puglia Pugliese, L., Guerriero, F., Laporte, G., 2020.
Crowd-shipping with time windows and transshipment nodes. Computers
and Operations Research 113.
[22] Macrina, G., Guerriero, F., 2018. The green vehicle routing problem
with occasional drivers, in: Daniele, P., Scrimali, L. (Eds.), New Trends
in Emerging Complex Real Life Problems, Springer International Publish-
ing. Springer New York LLC.
[23] Murray, C.C., Chu, A.G., 2015. The flying sidekick traveling salesman
problem: Optimization of drone-assisted parcel delivery. Transportation
Research Part C: Emerging Technologies 54, 86–109.
[24] Poikonen, S., Wang, X., Golden, B., 2017. The vehicle routing problem
with drones: Extended models and connections. Networks 70, 34–43.
[25] Sacramento, D., Pisinger, D., Ropke, S., 2019. An adaptive large neigh-
borhood search metaheuristic for the vehicle routing problem with drones.
Transportation Research Part C: Emerging Technologies 102, 289–315.
[26] Solomon, M.M., 1987. Algorithms for the vehicle routing and scheduling
problems with time window constraints. Operations Research 35, 254–265.
[27] Ubeda, S., Faulin, J., Serrano, A., Arcelus, F.J., 2014. Solving the green
capacitated vehicle routing problem using a tabu search algorithm. Lecture
Notes in Management Science 6, 141–149.
[28] Wang, X., Poikonen, S., Golden, B., 2017. The vehicle routing problem
with drones: several worst-case results. Optimization Letters 11, 679.
[29] Wang, Z., Sheu, J.B., 2019. Vehicle routing problem with drones. Trans-
portation Research Part B: Methodological 122, 350–364.
[30] Yoo, W., Yu, E., Jung, J., 2018. Drone delivery: Factors affecting the pub-
lic’s attitude and intention to adopt. Telematics and Informatics 35, 1687–
1700.

8
496 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497
L. Di Puglia Pugliese et al. / Procedia Manufacturing 00 (2019) 000–000 9

Table 2. Average numerical results for the instances with 5 customers considering the three transportation systems and the three classes of test problems when
minimizing the transportation cost.
Trucks Drones
|N| Problem Class time gap C(x, y) g(z, x, y) C t (x) gt (z, x) #v #arcs C d (y) gd (y) #d |N d |
R2 1.12 0.00% 4636.36 100.04 4636.36 100.04 1.45 6.45
5 VRP C2 0.63 0.00% 5818.75 118.39 5818.75 118.39 1.63 6.63
RC2 0.96 0.00% 6092.86 130.73 6092.86 130.73 1.38 6.43
R2 0.11 0.00% 87.53 0.33 87.53 0.33 2.27 2.27
5 RPD C2 0.10 0.00% 90.37 0.34 90.37 0.34 2.00 2.00
RC2 0.10 0.00% 185.35 0.70 185.35 0.70 3.25 3.25
R2 0.82 0.00% 3358.93 64.71 3277.27 64.40 1.09 4.09 81.66 0.31 2.00 2.27
5 VRP-D C2 0.63 0.00% 4350.78 90.15 4275.00 89.86 1.25 4.75 75.78 0.29 1.50 2.00
RC2 0.39 0.00% 3715.80 57.70 3543.75 57.05 1.00 3.00 172.05 0.65 3.00 3.25
VRP 0.90 0.00% 5515.99 116.38 5515.99 116.38 1.48 6.50
AVG5 RPD 0.10 0.00% 121.08 0.46 121.08 0.46 2.51 2.51
VRP-D 0.62 0.00% 3808.50 70.85 3698.67 70.44 1.11 3.95 109.83 0.41 2.17 2.51

Table 3. Average numerical results for the instances with 10 customers considering the three transportation systems and the three classes of test problems when
minimizing the transportation cost.
Trucks Drones
|N| Problem Class time gap C(x, y) g(z, x, y) C t (x) gt (z, x) #v #arcs C d (y) gd (y) #d |N d |
R2 218.34 0.01% 6945.45 166.03 6945.45 166.03 1.73 11.73
10 VRP C2 51.49 0.00% 8300.00 174.56 8300.00 174.56 2.50 12.50
RC2 57.89 0.01% 9600.00 226.80 9600.00 226.80 1.63 11.63
R2 0.95 0.00% 271.86 1.03 271.86 1.03 6.64 6.64
10 RPD C2 0.94 0.00% 354.45 1.34 354.45 1.34 6.50 6.50
RC2 0.88 0.00% 410.47 1.55 410.47 1.55 7.25 7.25
R2 74.96 0.01% 4369.22 89.00 4122.73 88.07 1.55 5.82 246.49 0.93 5.73 6.64
10 VRP-D C2 20.44 0.01% 4763.66 86.38 4456.25 85.22 1.75 6.25 307.41 1.16 5.50 6.50
RC2 43.74 0.01% 5575.58 104.49 5225.00 103.16 1.50 5.88 350.58 1.32 5.63 7.25
VRP 109.24 0.01% 8281.82 189.13 8281.82 189.13 1.95 11.95
AVG10 RPD 0.92 0.00% 345.59 1.30 345.59 1.30 6.80 6.80
VRP-D 46.38 0.01% 4902.82 93.29 4601.33 92.15 1.60 5.98 301.49 1.14 5.62 6.80

Table 4. Average numerical results for the instances with 15 customers considering the three transportation systems and the three classes of test problems when
minimizing the transportation cost.
Trucks Drones
|N| Problem Class time gap C(x, y) g(z, x, y) C t (x) gt (z, x) #v #arcs C d (y) gd (y) #d |N d |
R2 1164.93 1.22% 8527.27 209.05 8527.27 209.05 2.00 17.00
15 VRP C2 334.99 0.65% 9818.75 215.77 9818.75 215.77 2.50 17.50
RC2 764.19 2.32% 12043.75 294.79 12043.75 294.79 2.25 17.25
R2 6.03 0.00% 428.75 1.62 428.75 1.62 10.27 10.27
15 RPD C2 5.74 0.00% 568.17 2.14 568.17 2.14 10.63 10.63
RC2 5.71 0.00% 612.05 2.31 612.05 2.31 10.75 10.75
R2 1181.05 10.20% 4796.08 97.67 4395.45 96.15 1.64 7.82 400.63 1.51 8.82 10.27
15 VRP-D C2 466.51 0.86% 5794.78 107.61 5312.50 105.79 1.75 7.50 482.28 1.82 9.25 10.63
RC2 779.39 2.63% 7246.48 148.03 6718.75 146.04 1.75 8.00 527.73 1.99 8.75 10.75
VRP 754.70 1.40% 10129.92 239.87 10129.92 239.87 2.25 17.25
AVG15 RPD 5.83 0.00% 536.32 2.02 536.32 2.02 10.55 10.55
VRP-D 808.98 4.56% 5945.78 117.77 5475.57 115.99 1.71 7.77 470.21 1.77 8.94 10.55

9
 Luigi Di Puglia Pugliese et al. / Procedia Manufacturing 42 (2020) 488–497 497

Table 5. Average numerical results over all instances considering the three transportation systems when minimizing the transportation cost.
Trucks Drones
Problem time gap C(x, y) g(z, x, y) C t (x) gt (z, x) #v #arcs C d (y) gd (y) #d |N d |
VRP 288.28 0.47% 7975.91 181.79 7975.91 181.79 1.90 11.90
AVG RPD 2.29 0.00% 334.33 1.26 334.33 1.26 6.62 6.62
VRP-D 285.33 1.52% 4885.70 93.97 4591.86 92.86 1.47 5.90 293.84 1.11 5.57 6.62

10