10th IFAC Symposium on Intelligent Autonomous Vehicles

Gdansk,
10th Poland, July
July 3-5, 2019
10th IFAC
Gdansk, Symposium
IFACPoland,
Symposium on
on Intelligent
3-5, 2019
Intelligent Autonomous
Autonomous Vehicles
Vehicles
Gdansk,
Gdansk, Poland,
Poland, July
July 3-5,
3-5, 2019
2019 Available online at www.sciencedirect.com
10th IFAC Symposium on Intelligent Autonomous Vehicles
Gdansk, Poland, July 3-5, 2019
ScienceDirect
IFAC PapersOnLine 52-8 (2019) 283–288
A
A spatially
spatially wind
wind aware
aware quadcopter
quadcopter (UAV) (UAV) path
path planning
planning approach
approach
A spatially wind aware quadcopter (UAV) path planning approach
A spatially wind aware quadcopter (UAV) path planning approach
Georgios
Georgios A. A. Thanellas*,Vassilis
Thanellas*,Vassilis C. C. Moulianitis**
Moulianitis**
Georgios A.
Georgios A.Nikos Nikos A. Aspragathos***
Thanellas*,Vassilis
Thanellas*,Vassilis
A. Aspragathos*** C. Moulianitis**
C. Moulianitis**
Nikos A. 
Aspragathos***
Nikos A. Aspragathos***
Georgios A. Thanellas*,Vassilis C. Moulianitis**

Nikos A. Aspragathos***
** Department
Department of of Mechanical
Mechanical Engineering
Engineering 
and Aeronautics,
and Aeronautics, University
University ofof Patras,
Patras,
** Department
Department of Greece, (e-mail:
of Mechanical
Mechanical
Greece, (e-mail: geothanellas@gmail.com).
Engineering
Engineering and
and Aeronautics,
Aeronautics, University
geothanellas@gmail.com). University ofof Patras,
Patras,
** Department
** Department of
of Product
Product and Systems
Greece,
Greece,
and Design
(e-mail:
(e-mail:
Systems Design Engineering, University of
geothanellas@gmail.com).
geothanellas@gmail.com).
Engineering, University of Aegean,
Aegean, Greece,
Greece, (e-mail:
(e-mail:
* Department of Mechanical Engineering and Aeronautics, University of Patras,
** Department of Product
** Department of Product and and moulianitis@syros.aegean.gr)
Systems
Systems Design
Design Engineering,
Engineering,
moulianitis@syros.aegean.gr) University
University of
of Aegean,
Aegean, Greece,
Greece, (e-mail:
(e-mail:
Greece, (e-mail: geothanellas@gmail.com).
*** Department
*** Department of of Mechanical
Mechanical Engineering and and Aeronautics,
moulianitis@syros.aegean.gr)
moulianitis@syros.aegean.gr)
Engineering Aeronautics, University
University of Patras,
Patras,
** Department of Product and Systems Design Engineering, University of Aegean,ofGreece, (e-mail:
*** Department of
*** Department of MechanicalGreece,
Mechanical(e-mail: asprag@mech.upatras.gr)
Engineering
Engineering and
and Aeronautics, University of Patras,
Greece,moulianitis@syros.aegean.gr)
(e-mail: Aeronautics,
asprag@mech.upatras.gr) University of Patras,
*** Department of MechanicalGreece,
Greece, (e-mail:
(e-mail: asprag@mech.upatras.gr)
asprag@mech.upatras.gr)
Engineering and Aeronautics, University of Patras,
Abstract: Wind aware aware path
path planningGreece,
planning is (e-mail:
is essential asprag@mech.upatras.gr)
essential towards
Abstract: Wind towards the
the development
development of of autonomous
autonomous quadcopters
quadcopters
(UAVs)
Abstract: which
Wind
Abstract:which
(UAVs) Wind will will be
aware
aware able
path
path to
be able to fly efficiently
planning
planning is in
essential
is essential
fly efficiently outdoor
towards
towards
in outdoor environments.
the In
development
the development
environments. this
of paper, the
the wind
autonomous
In thisofpaper,
autonomous wind formation
quadcopters
quadcopters
formation
and quadcopter
(UAVs)
(UAVs) which
which kinematics
will be ableare to presented
fly so
efficientlyas to
in provide
outdoor cost functions
environments. for
In A*
this based
paper,on distance
the wind and wind
wind
formation
and
Abstract: Wind will
quadcopter be able
kinematics
aware pathare to fly efficiently
presented
planning is so asin
to outdoor
essentialprovide
towardsenvironments.
cost
thefunctions In this
for
development paper,
A*ofbased onthe wind
distance
autonomous formation
and
quadcopters
information.
and
and quadcopter
quadcopter
information. Furthermore,
kinematics
kinematics
Furthermore, aa collision
are
are presented
presented
collision checking
so
so as
as
checking to
to method
provide
provide
method is
cost
cost
is proposed,
functions
functions
proposed, which
for
for A*
A*
which is incorporated
based
based
is on
on distance
distance
incorporated to
to aa flight
and
and wind
wind
flight
(UAVs) which will be able to fly efficiently in outdoor environments. In this paper, the wind formation
scenario in
in an
information.
information.
scenario outdoor
anFurthermore, terrain
Furthermore,
outdoor terrain under
aa collision
collision
under various
checking
checking
various windmethod
formations.
method is A
A computational
is proposed,
proposed, which
which is comparison
is incorporated
incorporatedof aa uniform
of to
to aa flight
flight
and quadcopter kinematics are presented so aswind formations.
to provide cost computational
functions for A* basedcomparison
on distance uniform
and wind
graph and
scenario
scenario
graph and a
in
ina PRM
an
an
PRM based
outdoor
outdoor
based graph
terrain
terrain
graph is
is finally
under
under evaluated.
various
various
finally wind
wind
evaluated. Experiments
formations.
formations.
Experiments on
A
A
on artificial wind
computational
computational
artificial wind fields, through
comparison
comparison
fields, of
of
through illustrative
a
a uniform
uniform
illustrative
information. Furthermore, a collision checking method is proposed, which is incorporated to a flight
simulations,
graph and
andinaa an
PRMhighlight
basedthe
based benefit
graph is of
of wind
finally aware
evaluated. A* searching
searching algorithm.
Experiments on artificial
artificial wind
wind fields, through
graph
simulations,
scenario PRMhighlight
outdoor graph
the benefit
terrain isunder
finally evaluated.
wind aware
various wind Experiments
A* formations. on
algorithm.
A computational fields,
comparison ofillustrative
through illustrative
a uniform
simulations, highlight the benefit of wind aware A* searching algorithm.
© 2019,
simulations,
graph and
Keywords:IFAC highlight
a quadcopter,
PRM the benefit
(International
based graph
wind, is of
Federation wind
finally
path of aware
Automatic
evaluated.
planning, A* searching
Control)
Experiments
collision avoidance,algorithm.
Hosting by Elsevier
on minimum
artificial wind
time,Ltd. Allthrough
fields,
minimumrights reserved.
illustrative
distance.
Keywords: quadcopter, wind, path planning, collision avoidance, minimum time, minimum distance.
simulations,
Keywords: highlight the benefit of wind aware A* searching algorithm.
Keywords: quadcopter,
quadcopter, wind,
wind, path
path planning,
planning, collision
collision avoidance, minimum time,
avoidance, minimum

time, minimum
minimum distance.
distance.
Keywords: quadcopter, wind, path planning, collision  avoidance,
Evolution
Evolution path minimum
path planner
planner time,forminimum
for navigationdistance.
navigation in
in ocean
ocean environments
environments
1.
1. INTRODUCTION
INTRODUCTION considering
Evolution path
Evolution
considering obstacles.
path planner
obstacles. Kularatne
planner Kularatnefor navigation
for navigationet
et al.
al. (2016)
ocean extracted
in(2016)
in ocean environments
environments
extracted cost
cost
1.
1. INTRODUCTION
INTRODUCTION 
Unmanned Aerial Vehicles (UAVs) were developed initially functions
considering
considering
functions based on
obstacles.
obstacles.
based on time
time and energy
Kularatne
Kularatne
and energy et
et to be
al.
al.
to be minimized
(2016)
(2016)
minimized along
extracted
extracted
along the
cost
cost
the
Unmanned Aerial Vehicles (UAVs) were developed initially Evolution path planner for navigation in ocean environments
for military
Unmanned Aerial 1.
applications,
VehiclesINTRODUCTION
however
(UAVs) they
were currently
developed became
initially determined
functions
functions
determined path.
based
based
path. on
on time
time and
and energy
energy to
to be
be minimized
minimized along
along the
the
Unmanned
for militaryAerial Vehicles however
applications, (UAVs) were they developed
currently initially
became considering obstacles. Kularatne et al. (2016) extracted cost
extremely
for
for
extremely popular
militarypopular
military in civil
applications,
applications,
in civil however
applications
however
applications they such
theysuch as agriculture,
currently
currently
as agriculture,
became determined
became determined
functions
Accordingbased path.
path. on time and(2018),
to Langelaan
Langelaan energy tothe
(2018), be minimized
the vertical air airalong
motionthe
Unmanned Aerial Vehicles (UAVs) were developed initially According to vertical motion
environmental
extremely
extremely
environmental popular
popularprotection,
in
in
protection, civil
civil aerial
aerial photography,
applications
applications such
such
photography, as
aspublic
public safety
agriculture,
agriculture,
safety (thermal
determined
According instabilities,
path.
to Langelaan orographic
(2018), lift or
thewave), the
vertical spatial
air wind
motion
for military applications, however they currently became (thermal According to Langelaan
instabilities, (2018),
orographic lift the
or vertical
wave), the air
spatial motion
wind
and traffic
traffic flow
environmental
environmental control
protection,
protection, (Mohammed
aerial
aerial et
et al,
photography,
photography, 2014). Moulianitis
public safety gradients
and
extremely flow
popular control
in civil(Mohammed
applications al,such
2014).aspublic safety
Moulianitis
agriculture, (thermal
(thermal
gradients
According
such as
as horizontal
instabilities,
instabilities,
such orographic
orographic
horizontal
to Langelaan
shear
lift
shear
(2018),
layers
lift or
or wave),
wave),
layers
the
and
and thethe
the
vertical
temporal
spatial
spatial
the wind
wind
airtemporal
motion
et
and
and
et al.
al. (2018)
traffic
traffic
(2018) concluded
flow
flow control
control
concluded that
that UAVs
(Mohammed
(Mohammed
UAVs constitute
et
et al,
al,
constitute a
2014).
2014).
a cost-effective
Moulianitis
Moulianitis
cost-effective gradients like
such gusts
as are the
horizontal main sources
shear of
layers energy
and theavailable
temporal in
environmental protection, aerial photography, public safety gradients such
like as
gusts horizontal
are the main shear
sourceslayers
of
(thermal instabilities, orographic lift or wave), the spatial wind and
energy the temporal
available in
solution
et
et al.
al.
solution early
(2018)
(2018)
early forest
concluded
concluded
forest fire
fire detection
that
that UAVs
UAVs
detection compared
constitute
constitute
compared with
a
a
with other
other state
cost-effective
cost-effective
state the atmosphere.
gradients like Gusts
gusts are have
the short
main duration
sources of and
energy are stochastic
available in
and traffic flow control (Mohammed et al, 2014). Moulianitis the gradients like
atmosphere. gusts
Gustsare the
have main
shortsources
duration
gradients such as horizontal shear layers and the temporal of energy
and are available
stochastic in
of
of the
solution
solution
the art
art relevant
early
early
relevant mechatronic
forest
forest fire
fire
mechatronic systems.
detection
detection
systems. Among
compared
compared
Among with
withcommercially
other
other
commercially state
state in
the nature, while
atmosphere. the
Gusts vertical
have air
short motion
duration have
and long
are duration,
stochastic
et al. (2018) concluded that UAVs constitute a cost-effective gradients the
in atmosphere.
nature, while
like Gusts
guststhe are have
vertical
the short
main air duration
motion
sources of and
have
energy are
long stochastic
duration,
available in
available
of
of the art
the
available UAVs
artearly
relevant
relevant
UAVs (fixed
mechatronic
mechatronic
(fixed wing, systems.
single
systems. rotor
Among
Among and multirotor),
commercially
commercially
solution forest fire wing,
detection single rotor
compared and
with multirotor),
other state in while
in nature,
while
the
the
the main
nature, whilesource
while
main
atmosphere. the
source
Gusts
of
the vertical the
the wind
vertical
ofhave air
wind
short
field
air motion
motion
field
duration
is
is the
have
have
theand
wind
long
long
wind forecasting
duration,
duration,
areforecasting
stochastic
multi-rotors
available
available
multi-rotors UAVs
UAVs offer
(fixed
(fixed
offer greater
wing,
greater flexibility
single
wing, systems.
flexibilityrotor
single rotor inand control
control and
multirotor),
inandcommercially
multirotor), services
and while the based
main on Numerical
source Numerical
of the wind Weather
field is Predictions
the wind (NWP),
forecasting
of the art relevant mechatronic Among while
services
in the
nature, main
based
while source
onthe of the
vertical wind
air field
Weather
motion is the wind
Predictions
have long forecasting
(NWP),
duration,
maneuverability
multi-rotors offer with greatervertical take-offin
flexibility and landing
multi-rotors
maneuverability
available UAVs offer with
(fixed greatervertical
wing, flexibilityrotorinandand
singletake-off control
control and (Windfinder.com,
and
landing
multirotor), services
services based
based
(Windfinder.com, on
on 2018).
Numerical
Numerical
2018). Weather
Weather
while the main source of the wind field is the wind forecasting Predictions
Predictions (NWP),
(NWP),
capabilities.
maneuverability
maneuverability Even though,
with
with multi-rotors
vertical
vertical have
take-off
take-off a and
limited flight
landing
capabilities.
multi-rotors Even offerthough,
greater multi-rotors
flexibility have in a and
limited
control landing
flight
and (Windfinder.com,
Usually, UAVs
(Windfinder.com,
Usually, UAVs
services based are are2018).
largely
2018).
on largely
Numerical limited
limited to primitively
to primitively
Weather Predictions follow
follow user-
user-
(NWP),
time, whose
capabilities.
capabilities.
time, energy
Even
whose Even
energythough,is
is spent
though, spent mainly
multi-rotors
multi-rotors on fighting
have
on have
mainly take-off fightingaa and gravity
limited
limited
gravity and
flight
flight
and Usually,
maneuverability with vertical landing defined waypoints
UAVs are in outdoor
largely limitedenvironments,
to primitively which
follow are not
user-
stabilizing
time,
time, in
whoseinEven
whose
stabilizing
capabilities.
the
energy
energy air.
the air. Lawrance
isLawrance
is
though, spent
spent mainly
mainlyet al. (2014)
on(2014)
et al.on
multi-rotors fighting
fighting
have
suggested and always energy efficient and time optimal under wind. Inuser-
two
gravityflight
gravity
asuggested
limited and
two
Usually,
defined UAVs
waypoints
(Windfinder.com, are largely
in
2018). outdoorlimited to primitively
environments, follow
which are not
this
approaches for the energy autonomy problem, either aerial defined
defined
always
Usually, waypoints
waypoints
energy
UAVs efficient
are in
in outdoor
outdoor
largely and time
limitedenvironments,
environments,
optimal
to under
primitively which
which
wind.
follow are
are
In not
not
this
user-
stabilizing
stabilizing
approaches in
time, whoseinfor the
energy air. Lawrance
thetheair.energy
isLawrance
spentautonomy
mainlyet al. (2014)
et al.onproblem,
(2014)
fighting suggested
suggested
either two
two
gravityaerial
and paper, we propose aa path planning procedure, considering a
recharging always
always
paper, energy
energy efficient
efficient and
and time
time
defined waypoints in outdoor environments, which are nota
we propose path planning optimal
optimal under
under
procedure, wind.
wind.
considering In
In this
this
recharging or
approaches
approaches
stabilizing or ongoing
for
inforthethe
ongoing harvesting
theair.energy
energy autonomy
autonomy
harvesting
Lawrance
of energy
of al.
et energy
(2014)
from
problem,
problem,
from the
the vehicle’s
either
either
suggested aerial
aerial
vehicle’s
two collision
paper, avoidance technique using A* with new alternative
surrounding
recharging environment.
orfor
ongoing The
harvesting latter approach
of energy
energy fromcan the be proved
vehicle’s paper, we
collision
always we propose
propose
avoidance
energy aatechnique
efficient path
path andplanning
planning
using
time procedure,
procedure,
A* with
optimal under considering
considering
newwind. In thisaa
alternative
recharging
surroundingor
approaches ongoing
environment.
the energyharvesting
The
autonomyof
latter approach from
problem, the
can vehicle’s
be
either proved
aerial cost
collision functions,
avoidance which techniqueare compared
using A* with
with distance
new cost
alternative
collision
cost avoidance
functions, which techniqueare using
compared A* with
with new alternative
distance costa
very
very efficient
surrounding
surrounding
efficient
recharging
in
in wind
wind path
environment.
orenvironment.
ongoing path planning
The latter
The
planning
harvesting latterwhich
approach
approach
ofwhich
energy
could
could provide
can
fromcanprovide paths paper,
paths
bevehicle’s
thebe proved
proved functions
cost
we propose a path planning procedure, considering
of
functions,standard
which A*. To
are reduce
compared computational
with searching
distance cost
with
very reduced
efficient inenergy
wind path consumption
path planning as
which well as
could shorter
provide flight cost
functions
paths collision functions,
of which
standard A*. are
To compared
reduce
avoidance technique using A* with new alternative with
computational distance cost
searching
very
with efficient
surrounding in wind
reducedenvironment.
energy planning
The latterwhich
consumption asapproach
well could provide
as can
shorter paths
flight
be proved cost,
functions we introduce
of standard a graph,
A*. based
To reduceon surface veil,
computational using either
searching
time. reduced
with reduced inenergy
energy consumption as well
well as provide
shorter flight
flight functions
cost, we of standard
introduce a A*.
graph, To reduce
based
cost functions, which are compared with distance cost on computational
surface veil, searching
using either
with
time. consumption as as shorter aacost,
uniform or aa PRM
very efficient
time.
wind path planning which could paths cost, we
we introduce
uniform
functions introducePRMaa node
or standard
of graph,
graph,
node distribution
based
based
A*. distribution
To reduceon in
in aa simulated
on surface
surface veil,
veil, using
simulated
computational usingterrain.
either
either
terrain.
searching
time.
with concept
The reducedofenergy
exploiting consumption
the energy asfrom
wellthe as environmental
shorter flight aa uniform or a PRM node distribution in a simulated terrain.
The concept of exploiting the energy from the environmental cost, Theuniform
we or a
introducePRM a node
graph, distribution
based on in a
surfacesimulated
veil, usingterrain.
either
time.
forces has The remainder
remainder of of this
this paperpaper is is organized
organized as as follows:
follows: Section
Section 2 2
The
The
forces has been
concept
concept of
been applied
applied for
of exploiting
exploiting themarine
forthe energyas
energy
marine well
from
asfrom
wellthe as
as aerial
the aerial vehicles
environmental
environmental
vehicles aThe uniform
presents or aformation
the
remainder PRM
of this node of
paper distribution
wind
is as
organized in aassimulated
well as the
follows: terrain.
quadcopter
Section
(Liu
forces and
has Sukhatme,
been applied 2018;
for Mahmoud.Zadeh
marine as well as et al.
aerial (2016);
vehicles The remainder of this paper is organized
presents the formation of wind as well as the quadcopter as follows: Section 22
forces
(Liu has Sukhatme,
and been applied forthemarine
2018; energyasfrom
Mahmoud.Zadeh wellthe as aerial
et al. vehicles
(2016); kinematics.
The concept
Kularatne
(Liu and et
of exploiting
al.,
Sukhatme, 2016; Techy
2018; and Woolsey,
Mahmoud.Zadeh
environmental
2009;et Al-Sabban
al. (2016);
presents
kinematics.
presents
The theSection
the
remainder formation
Section 3 describes
of this3 paper
formation of
of wind
describeswind theasform
theas
is organized well
form of
ofas
wellas asaa veil-graph
the
the quadcopter
veil-graph
follows:
in
in an
quadcopter
Section an
2
Kularatne
(Liu and et al.,
Sukhatme, 2016; Techy
2018; and Woolsey,
Mahmoud.Zadeh 2009;et
forces has been applied for marine as well as aerial vehicles kinematics. Al-Sabban
al. (2016); unstructured terrain
Section as
3 well as
describes a collision
the form avoidance
of a technique
veil-graph in an
et al.,
Kularatne 2012).
et al.,Path
2016; planning
Techy andbased
Woolsey,on Markov
2009; Decision
Al-Sabban kinematics.
unstructured
presents the Section
terrain
formation 3
as describes
well
of as
winda theas form
collisionwell ofas a veil-graph
avoidance
the in
technique
quadcopter an
Kularatne
(Liu et al.,Path
et al.,and2012).
Sukhatme, 2016;planning
TechyMahmoud.Zadeh
2018; and Woolsey,
based 2009;
on Markov Al-Sabban
et al. Decision
(2016); in aa triangular
unstructured mesh.
terrain In well
addition, two path planning approaches
processes under ocean currents for underwater robots or unstructured
in triangular mesh. as
terrain as well
3Indescribesas
as aa the
addition, collision
collision
twoformpath ofavoidance
avoidance
planning technique
technique
approaches
et
et al., 2012).
al.,
processes
Kularatne 2012). Path
et al.,Path
under planning
2016;planning
ocean currents
Techy and based
based
for on Markov
on
underwater
Woolsey, Markov
robots
2009; or under
under kinematics.
Decision
Decision
Al-Sabban are
in a introduced
triangular
Section
and
mesh. evaluated
In addition, withtwo a comparison
path
a veil-graph
planning metric
in an
based
approaches
winds
processesfor UAV
under have
ocean been
currents introduced
for (Liu
underwater and
robots Sukhatme,
or under in
are a triangular
introduced
unstructured mesh.
and
terrain In
evaluated
as addition,
well as with
a two a path
collision planning
comparison
avoidance approaches
metric based
technique
processes
winds
al., for under
UAVPath ocean currents
haveplanning for
been introduced underwater
based on(Liu robots or
and Sukhatme, under on
et 2012).
2018; Techy
winds Techy
for under
UAV andhaveWoolsey,
have 2009; Al-Sabban
been introduced
introduced Al-Sabban
Markov
(Liu and andet al.,
Decision are
al., 2012). on
Sukhatme, areatime.
in time. Section
introduced
introduced
Section
triangular
4
4 presents
and
and
mesh. evaluated
evaluated
presents the
In addition,
experimental
thewith
with aa path
comparison
comparison
experimental
two
evaluation
metric
metric
evaluation
planning
of
approaches
path
ofbased
based
path
winds
2018;
processesfor UAV and Woolsey,
ocean been
currents 2009;
for underwater (Liu et
robots Sukhatme,
or2012).
under on planning
time. approaches.
Section 4 presents Finally,
the section
experimental 5 presents
evaluation theof main
path
Mahmoud.Zadeh
2018; Techy et
andhave al.
Woolsey, (2016)2009; implemented
Al-Sabban a Differential
et Differential
al., on
are time.
planning Section
introduced
2012). conclusions and future work. 4
approaches.
and presents
evaluated the
Finally, experimental
withsection
a 5
comparisonevaluation
presents metrictheof path
main
based
2018;
winds Techy
Mahmoud.Zadeh
for UAV and etWoolsey,
al.been
(2016)2009; Al-Sabban
implemented
introduced (Liu and aet al., 2012).
Sukhatme, planning approaches. Finally, section 5 presents the main
Mahmoud.Zadeh etWoolsey,
al. (2016)
(2016) implemented planning
conclusions
on time. approaches.
and
Section future
4 presents Finally,
work. the section
experimental 5 presents
evaluation theof main
path
Mahmoud.Zadeh
2018; Techy and et al. 2009; Al-Sabban aaet Differential
implemented Differential
al., 2012). conclusions
conclusions and
and future work.
future Finally,
work. section 5 presents the main
planning approaches.
Mahmoud.Zadeh
Copyright © 2019
et
IFAC
al. (2016) implemented a Differential
Copyright 2019 IFAC
2405-8963 © 2019, conclusions and future
IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. work.
Copyright
Peer review©
Copyright 2019
©under IFAC
2019 responsibility
IFAC of International Federation of Automatic Control.
10.1016/j.ifacol.2019.08.084
Copyright © 2019 IFAC
2019 IFAC IAV
284
Gdansk, Poland, July 3-5, 2019 Georgios A. Thanellas et al. / IFAC PapersOnLine 52-8 (2019) 283–288

⃗ 𝑑𝑑 = 𝑢𝑢𝑑𝑑𝑑𝑑 𝑖𝑖̂ + 𝑢𝑢𝑑𝑑𝑑𝑑 𝑗𝑗̂ + 𝑢𝑢𝑑𝑑𝑑𝑑 𝑘𝑘̂ , (2)

𝑢𝑢
and given the wind velocity 𝑣𝑣𝑤𝑤 with respect to the ground (fig
2. PROBLEM FORMULATION
Path planning efficiency in UAVs depends from distance

𝑢𝑢𝑑𝑑𝑑𝑑 𝑢𝑢𝑤𝑤𝑤𝑤
criteria, especially under obstacle existence, while in outdoor 2) the total ground velocity of the UAV is the following:

⃗ 𝑡𝑡 = 𝑢𝑢
𝑢𝑢 𝑢𝑢 𝑢𝑢
⃗ 𝑤𝑤 ⇒ [ 𝑑𝑑𝑑𝑑 ] + [ 𝑤𝑤𝑤𝑤 ]. (3)
⃗ 𝑑𝑑 + 𝑢𝑢
environments, time index can improve flight endurance under

𝑢𝑢𝑑𝑑𝑑𝑑 0
windy conditions regard to the limitation of the possible on
board (fuel/battery) which can be carried by the UAV. In the
following a brief description of an outdoor windy environment
as well as a quadcopter kinematics model are presented.

2.1 Windy environment

The success of the efficient wind navigation model depends on

the density of measurements and the wind prediction map
accuracy. The wind forecast maps provided present high
uncertainty and the wind distribution is really stochastic,
however a known wind distribution is considered in this paper
to show the effectiveness of the proposed method. In real
conditions navigation the stochastic nature should be
considered or on-line wind measurement will be provided. In
this paper, the wind field is considered as a two-dimensional
time-invariant vector field, representing velocity at each point

⃗ 𝑤𝑤 (𝑥𝑥, 𝑦𝑦, 𝑧𝑧) = 〈𝑣𝑣𝑤𝑤𝑤𝑤 , 𝑣𝑣𝑤𝑤𝑤𝑤 , 0〉,

of a map:
𝑢𝑢 (1)
where 𝑢𝑢𝑤𝑤𝑤𝑤 = 𝑢𝑢𝑤𝑤 𝑐𝑐𝑐𝑐𝑐𝑐𝜃𝜃𝑤𝑤 and 𝑢𝑢𝑤𝑤𝑤𝑤 = 𝑢𝑢𝑤𝑤 𝑠𝑠𝑠𝑠𝑠𝑠𝜃𝜃𝑤𝑤 define the wind
Fig 2. Analysis of UAV related velocities
⃗ 𝑡𝑡 is used to determine the flight time from a start position to
𝑢𝑢
components of 𝑣𝑣𝑤𝑤 in x-axis and y-axis respectively referenced

magnitude (𝑢𝑢𝑤𝑤 = 5𝑚𝑚/𝑠𝑠) and direction (𝜃𝜃𝑤𝑤 ) in each point

a goal position taking into account the wind field as well as
to the inertial frame. Fig 1 illustrates a wind map with a mean

(𝑥𝑥, 𝑦𝑦, 𝑧𝑧) of the map. Various formations of artificial wind

obstacle avoidance. The UAV speed limit is considered since
the path duration is calculated with the maximum velocity. The
acceleration limit is not considered, however this could be
fields based on wind data formation given by weather
taken into account in a smoothing stage of the determined
agencies, are presented in section 4.
polyline type path.

3. PATH PLANNING
In this work, an outdoor unstructured terrain is modelled with
triangular meshes whose edges (defined by vertices) are stored
in a circular list that is traversed in counterclockwise order
with respect to the outward normal vector of the face. A grid
is adapted to the terrain surface like a veil in order to minimize
the computational cost of path planning among nodes. In the
aftermath, a collision avoidance technique is employed based
on computational geometry between path and triangles. For the
path planning, a graph searching algorithm A* is used, with
cost functions based either on wind deviation from UAV speed
direction or distance criteria. The evaluation of their cost
functions relies on the determined total flight time, taken into
account the wind velocity.

Fig. 1. Snapshot of a wind map (Windfinder.com, 2018).

Graph Search-based Trajectory Planning

in a flight area (𝑊𝑊). The number of vertices (nodes) are either

2.2 Quadcopter kinematics Α discrete graph G=(V,E) is used to represent the workspace

The quadcopter is an underactuated system since four centroids of cells in a uniform square discretization or
actuators are used to control 6 DoF of its body. In this paper, randomly generated nodes in workspace, based on PRM
the rotational DoF are excluded and the state of the quadcopter (Kavraki et. al, 1994). The latter one offers an optimization in
is expressed by its position coordinates. The path planning computational cost due to reduced nodes in 2D. To keep the

⃗ 𝑑𝑑 :
flying object with velocity 𝑢𝑢
problem is simplified by considering the quadcopter as a free computational cost low in 3D, the concept of veil above the
surface area is adopted, discarding multiple vertices in z-axis.
Vertices which lie inside obstacles are elevated above them in

284
 2019 IFAC IAV
Gdansk, Poland, July 3-5, 2019 Georgios A. Thanellas et al. / IFAC PapersOnLine 52-8 (2019) 283–288 285

order to keep as much as possible edges between neighboring 𝑒𝑒 = ⃗⃗𝑠𝑠𝑖𝑖 + 𝜆𝜆𝑏𝑏⃗, (6)

where 𝜆𝜆 is a scalar quantity and 𝑏𝑏⃗ is the unit vector of the line
nodes. Fig.3 shows a 3D view map of a triangular surface plot,

defined by the nodes [𝑠𝑠𝑖𝑖 , 𝑠𝑠𝑗𝑗 ] given by:

which contains a veil with uniform and random generated
nodes.

𝑠𝑠𝑗𝑗 − 𝑠𝑠𝑖𝑖
𝑏𝑏⃗ = , (7)
‖𝑠𝑠𝑗𝑗 − 𝑠𝑠𝑖𝑖 ‖

The collision detection method rejects edges which have a

common point with a triangle of the mesh surface. This

point 𝑝𝑝𝑘𝑘 of the edge 𝑒𝑒 and the vertices of triangles 𝑣𝑣𝑙𝑙,(1,2,3) with
achieved by comparing the sum of areas of an intermediate

the area 𝐴𝐴𝑣𝑣⃗𝑙𝑙 formed by the vertices of triangle 𝑣𝑣𝑙𝑙 . Therefore,

if:

𝐴𝐴𝑣𝑣⃗𝑙𝑙 = 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 + 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 + 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 , 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶

𝑝𝑝 𝑝𝑝 𝑝𝑝

{ (8)
𝑙𝑙,12 𝑙𝑙,13 𝑙𝑙,23

𝐴𝐴𝑣𝑣⃗𝑙𝑙 ≠ 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 + 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 + 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 , 𝑁𝑁𝑁𝑁 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶

Fig. 3 3D view of simulated outdoor area includes nodes above
the surface. Among them there are a start node and a goal node. 𝑝𝑝 𝑝𝑝 𝑝𝑝
𝑙𝑙,12 𝑙𝑙,13 𝑙𝑙,23

where 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 , 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 , 𝐴𝐴𝑣𝑣⃗ 𝑘𝑘 are the areas, formed by (𝑝𝑝𝑘𝑘 , 𝑣𝑣1 , 𝑣𝑣2 ),
𝑝𝑝 𝑝𝑝 𝑝𝑝
A wind aware A* searching algorithm is proposed to
𝑙𝑙,12 𝑙𝑙,13 𝑙𝑙,23
(𝑝𝑝𝑘𝑘 , 𝑣𝑣1 , 𝑣𝑣3 ) and (𝑝𝑝𝑘𝑘 , 𝑣𝑣2 , 𝑣𝑣3 ) respectively. Fig 4 illustrates a
determine the path between the start and the goal node. At each

paths to be extended. Based on the cost 𝑔𝑔(𝑛𝑛) between start and point 𝑝𝑝𝑘𝑘 of edge vector 𝑒𝑒, which forms 3 triangles with vertices
iteration of its main loop, A* needs to determine which of its

current node and an estimation of the cost ℎ(𝑛𝑛) between
𝑐𝑐(𝑛𝑛) = 𝑔𝑔(𝑛𝑛) + ℎ(𝑛𝑛). Each step of A* includes the following:
current node and goal node, A* selects the path that minimizes
{𝑣𝑣1 , 𝑣𝑣2 , 𝑣𝑣3 } of an 𝑙𝑙 triangle between nodes 𝑠𝑠𝑖𝑖 and 𝑠𝑠𝑗𝑗 . According
to equation (8), the vector segment between 𝑠𝑠𝑖𝑖 and 𝑝𝑝𝑘𝑘 is
collision free in contrast to 𝑝𝑝𝑘𝑘+1 , whose sum of areas is the

Possible States (𝒔𝒔): the finite set of possible states which are area of triangle.

coordinates and is written as si = [xi yi zi ]𝑇𝑇 , where

the graph nodes. Each state is described by the Cartesian

xi , yi and zi denote the cartesian coordinates of the UAV

position in the graph.
Possible Actions (𝒂𝒂𝒔𝒔 ): It is the finite set of actions, in terms of
motion, available from state si . Depending on graph node
density, a spatial radius 𝑟𝑟𝑠𝑠 is defined, which ensures
connectivity of si with its neighbors sj , where 𝑖𝑖 ≠ 𝑗𝑗. A node sj
is considered neighbor of si , if and only if:

(𝑥𝑥𝑗𝑗 − 𝑥𝑥𝑖𝑖 ) + (𝑦𝑦𝑗𝑗 − 𝑦𝑦𝑖𝑖 ) + (𝑧𝑧𝑗𝑗 − 𝑧𝑧𝑖𝑖 ) ≤ 𝑟𝑟𝑠𝑠

2 2 2
(4) Fig 4. The collision checking method based on computational
geometry.
So, actions represent all possible directions along each edge of

sphere with 𝑟𝑟𝑠𝑠 radius and are dependent from cost functions
UAV to neighboring states, which are interior of a predefined
Cost functions
and obstacle presence. The formulation of A* cost functions is based on edge

distance values, while the velocity 𝑢𝑢⃗ 𝑑𝑑 of the UAV coincides

Collision Avoidance
Valid edges of the graph are considered those which don’t
collide with obstacles, ensuring a collision free route between
two neighboring nodes. Given that, the simulated terrain
consists of triangle meshes and nodes of the graph, we are
looking for edges which connect nodes and are always above
attributes among nodes. An edge has orientation and Euclidean
with the orientation of path. In this work it is assumed that the
winds are instant time-invariant and the cost functions are the
following:
Distance minimizing cost: the Euclidean distance is
determined between states and rewards the nodes that

triangles 𝑣𝑣𝑙𝑙 which lie between the nodes 𝑠𝑠𝑖𝑖 and 𝑠𝑠𝑗𝑗 , as follows:
the surface. To achieve that, we keep a subset of terrain minimize the path length. The UAV moves to the nearest
nodes, until it reaches its goal (𝑠𝑠goal ). In this way, the
𝑠𝑠𝑖𝑖,𝑥𝑥 ≤ 𝑣𝑣𝑙𝑙,𝑥𝑥 ≤ 𝑠𝑠𝑗𝑗,𝑥𝑥
{ 𝑖𝑖,𝑦𝑦 ≤ 𝑣𝑣𝑙𝑙,𝑦𝑦 ≤ 𝑠𝑠𝑗𝑗,𝑦𝑦
𝑠𝑠 𝑠𝑠𝑗𝑗 with distance (d) from state 𝑠𝑠𝑖𝑖 , resulting a cumulative cost
calculation of path cost in each A* iteration, defines the state

𝑠𝑠𝑖𝑖,𝑧𝑧 ≤ 𝑣𝑣𝑙𝑙,𝑧𝑧 ≤ 𝑠𝑠𝑗𝑗,𝑧𝑧 𝑔𝑔𝑑𝑑 (𝑛𝑛) to current node (n) and computes its remaining distance
(5)

to target by the heuristic cost ℎ𝑑𝑑 (n). The cost functions in each
An edge is given by: iteration are the following:

285
2019 IFAC IAV
286
Gdansk, Poland, July 3-5, 2019 Georgios A. Thanellas et al. / IFAC PapersOnLine 52-8 (2019) 283–288

𝑛𝑛

𝑔𝑔𝑑𝑑 (𝑛𝑛) = ∑ 𝑑𝑑(𝑠𝑠𝑖𝑖 , 𝑠𝑠𝑖𝑖−1 ) where wi , i = 1,2 are the weights which determine the priority
{ , (9)
𝑖𝑖=2
ℎ𝑑𝑑 (𝑛𝑛) = 𝑑𝑑(𝑠𝑠𝑛𝑛 , 𝑠𝑠goal )
of A* in convergence; accordingly, to angle and distance
respectively, which synthesize the minimum cost of path
The above cost functions of A* result a total cost c𝑑𝑑 (𝑛𝑛), based
between the start and goal position.
on Euclidean distance without taking into account the wind Metric of costs
field.
In order to compare A* cost function variants, we examine the
Minimum angle cost: This cost formula computes the execution time of path. For the computation of local flight
minimum angle between wind vector and UAV speed vector speed, it is assumed that the drone follows an edge without

𝑔𝑔𝜃𝜃 (𝑛𝑛). For example, in Fig 5, from the current position of

and keeps the node with minimum angle, resulting a cost deviations:

UAV in state 𝑠𝑠𝑖𝑖 the state 𝑠𝑠𝑗𝑗+1 is chosen which has the smallest 𝑢𝑢𝑡𝑡𝑖𝑖 = 𝑢𝑢𝑑𝑑𝑖𝑖 + 𝑢𝑢𝑤𝑤
𝑖𝑖
∙ 𝑐𝑐𝑐𝑐𝑐𝑐𝜑𝜑 (12)

Where 𝜑𝜑 is the angle between the drone velocity and the wind
angle between the UAV direction of motion and the wind
velocity, 𝑢𝑢𝑑𝑑𝑖𝑖 is the local drone speed in each segment and 𝑢𝑢𝑤𝑤
𝑖𝑖
∙
direction.
𝑐𝑐𝑐𝑐𝑐𝑐𝜑𝜑 is the wind component along path’s segment. 𝑢𝑢𝑤𝑤 ∙ 𝑠𝑠𝑠𝑠𝑠𝑠𝜑𝜑
𝑖𝑖

component of wind is assumed that it is counterbalanced by

the controller. So, the quadcopter will navigate along each
segment, resulting a flight time under wind impact, regardless
the cost function, as follows:
𝑛𝑛
𝑑𝑑(𝑠𝑠𝑖𝑖 , 𝑠𝑠𝑖𝑖−1 )
time = ∑ , (13)
𝑢𝑢𝑡𝑡𝑖𝑖−1
𝑖𝑖=2

Fig 5. Procedure for determination of optimal path 𝒙𝒙𝒐𝒐𝒐𝒐𝒐𝒐 among 4. APPLICATION AND DISCUSSION OF
RESULTS
states.
Regarding the heuristic cost, we compute the mean angle of In this section artificial vector fields are used to simulate wind

a cost ℎ𝜃𝜃 (𝑛𝑛). This assumption for the computation of heuristic

wind vectors between the current and the target node, resulting
function deals with wind direction fluctuation. These cost
fields, which are applied in a discrete simulated flight area x∈[-
2,2] and y∈[-2,2], with measurement units expressed in mesh
resolution (mr), in which are juxtaposed the two A* s with the
proposed cost functions compared with distance based A*. In

considering that the drone speed ‖𝑢𝑢𝑑𝑑 ‖ = 2𝑚𝑚𝑚𝑚/𝑠𝑠 and the

functions are given by: each simulated case, the execution time is determined,

average speed of wind ‖𝑢𝑢𝑤𝑤 ‖ = 1𝑚𝑚𝑚𝑚/𝑠𝑠 are predefined. Finally,

𝑛𝑛
‖u
⃗ d,𝑖𝑖−1 × 𝑢𝑢
⃗ w,i ‖
𝑔𝑔𝜃𝜃 (𝑛𝑛) = ∑ arctan ( )
⃗ d,i−1 ∙ 𝑢𝑢
𝑢𝑢 ⃗ w,i
𝑖𝑖=2
, (10)
uniform and random graph based on PRM time duration of

∑𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔
𝑖𝑖=𝑛𝑛 𝜃𝜃𝑖𝑖
flight are compared.
ℎ𝜃𝜃 (𝑛𝑛) =
{ ∑𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔
𝑖𝑖=𝑛𝑛 i
Three cases of simulated wind fields are considered:

where 𝑢𝑢𝑑𝑑 expresses UAV velocity that coincides with path

Case 1, Tailwinds. In this field, the wind has constant direction

orientation and 𝑢𝑢
⃗ 𝑤𝑤 is the wind vector in state 𝑠𝑠𝑗𝑗 . The above
with varied magnitude:
(1 − y 2 )î
cost functions of A* result a minimum angle total cost c𝜃𝜃 (𝑛𝑛), 𝑢𝑢𝑤𝑤 (x, y) =
⃗⃗⃗⃗⃗ ∀ y ∈ |y| ≤ 1, (14)
which ensures an optimal edge 𝑥𝑥𝑜𝑜𝑜𝑜𝑜𝑜 , which will be traversed √(1 − y 2 )2
by the UAV with maximum possible total velocity (𝑢𝑢 ⃗ 𝑡𝑡 ), The start and end position of the UAV are located so that the
resulting to minimum time. field will always provide a tailwind component during the
motion.

between wind and UAV with 𝑔𝑔𝜃𝜃 (𝑛𝑛),as minimum angle cost
Hybrid angle cost: This cost formula computes the angle
Case 2, Headwinds and Tailwinds. This field provide both
headwinds and tailwinds with different magnitudes during the
and target with ℎ𝑑𝑑 (n), as distance minimizing cost does. In
function does and Euclidean distance between current node
motion of the UAV from the start to end location.

order to homogenize the quantities of 𝑔𝑔𝜃𝜃 (𝑛𝑛) and ℎ𝑑𝑑 (n) both (𝑥𝑥 − 2)î + (x + 1)ĵ
𝑢𝑢𝑤𝑤 (x, y) =
⃗⃗⃗⃗⃗ , (15)
√(𝑥𝑥 − 2)2 + (x + 1)2
values are normalized in the range [0,1]:

c(n) = 𝑤𝑤1 ∙ + 𝑤𝑤2 ∙

𝑔𝑔𝜃𝜃 (𝑛𝑛) ℎ𝑑𝑑 (n) Case 3, Rotated wind field. It is used to validate the algorithm:
𝜋𝜋 min(ℎ𝑑𝑑 (n),𝑑𝑑(𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 ,𝑠𝑠goal ))
(11)

286
 2019 IFAC IAV
Gdansk, Poland, July 3-5, 2019 Georgios A. Thanellas et al. / IFAC PapersOnLine 52-8 (2019) 283–288 287

−yî + xĵ
𝑢𝑢𝑤𝑤 (x, y) =
⃗⃗⃗⃗⃗ , (16)
√𝑦𝑦 2 + 𝑥𝑥 2
Table 1. Simulation Results
Minimum Distance Minimum angle Wind

For each case, the simulation results are shown in Fig 6 (a-c). Case 1 3.11s 2.30s
Each simulation case includes two paths. The first one (red) is
Case 2 3.06s 3.63s
extracted with minimum distance cost and the second (green)
is extracted with minimum angle cost, proving a wind optimal Case 3 4.16s 2.51s
path.

is evaluated with under various 𝑤𝑤1 and 𝑤𝑤2 values. Experiments

In Table 2, the flight time with the hybrid angle cost function

reveal that the increase of 𝑤𝑤2 leads to considerably shorter

flight times for 1st and the 2nd case. If 𝑤𝑤2 = 1 the minimum
angle wind function can be trapped in local optimum since it
does not take into account the distance to the goal position, as
happened in case 2, while the hybrid angle function can find

shows a top view of map with high and low 𝑤𝑤1 and 𝑤𝑤2
the best path, combining both distance and wind field. Fig. 7

respectively and vice versa. Also, the execution times of table

2 for 1st and 3rd case are examined with respect to optimal
weights of 2nd case and show that the hybrid angle A* performs
always better than the minimum distance A*. Also, hybrid
(a) angle A* presents better results compared to the minimum
angle A*, which is unable to converge to a good position with
opposed wind fields, while in the case of goal targeted wind
the minimum angle A* performs slightly better than the hybrid
one, as shown in table 2.

Table 2. Hybrid angle A* weights

Case Distance Angle Exec. Time
weight Weight
2 0.9 0.1 3.06s

2 0.6 0.4 3.02s

(b) 2 0.4 0.6 2.95 s

2 0.1 0.9 2.95s

1 0.1 0.9 2.15s

3 0.1 0.9 2.56

(c)
Fig. 6 Simulation of path planning with distance and minimum
angle criteria in: (a) Case 1. (b) Case 3. (c) Case 3.
The durations of distance and angle paths, considering the

𝑤𝑤1 > 𝑤𝑤2 and 𝑤𝑤2 > 𝑤𝑤1 respectively.

average wind speed are shown in Table 1. Using time equation,
it is shown in table 1 that in the first and third case the
minimum angle flight is executed in shorter time than this of
minimum distance, unless the wind is completely reversed to
drone’s flight path.
Fig. 7 Simulation of path planning hybrid angle with weights:
Finally, comparative results of the processing time of path
generation, using uniform graph as well as PRM graph for the

287
2019 IFAC IAV
288
Gdansk, Poland, July 3-5, 2019 Georgios A. Thanellas et al. / IFAC PapersOnLine 52-8 (2019) 283–288
case of minimum distance and angle criteria, are shown in
Tables 3 and 4 respectively. The maximum size of PRM graph
random nodes was set to three quarters of the total amount of
uniform graph nodes, due to the fact that path formation
remains quite similar. The above processing times have been
measured for our MATLAB implementation running in single
core Intel Core i7 workstation, operating at 3.5GHz with 16
Gb of RAM.
Table 3. Processing Time with minimum distance criteria
Uniform Graph PRM Graph
All cases 0.51s 0.15s
Table 4. Processing Time with minimum angle criteria
Uniform Graph PRM Graph
All cases 1.07s 0.3s
Tables 3,4 show that in terms of processing time, PRM is much
more efficient than this of Uniform graph. This speed up of
PRM can offset the processing bottleneck between minimum
angle and minimum distance cost function calculation in
uniform graph, as processing time of the minimum angle cost
function is slightly longer than this of minimum distance.
Hybrid angle A* has the same processing cost compared to the
minimum distance A*.
5. CONCLUSIONS
This paper presents a smart creation of graph, introducing a
veil pattern of states (nodes) above the simulated terrain
environment. The proposed variants of A* cost functions,
based on either the use of wind orientation only or the use of
distance and wind orientation, compared to a minimum
distance A*, outperform in terms of path’s execution time.
However, it was shown that minimum angle A* traps into local
minimum with completely opposed wind fields, in contrast to
the hybrid angle A*, which ensures a path in shorter time than
the distance-based A* under various wind criteria and obstacle
presence, using the proposed collision avoidance technique.
The use of PRM offers a speedup in the path planning process,
highlighting the potentials for UAV onboard processing.
Finally, the hybrid angle A* sets the foundations for a wind
optimal global planner, which can be extended with a local
planner, which will be able to handle local wind variances in
orientation and speed during navigation, which are common in
real-time waypoint missions. Finally, the wind aware local
planner, running onboard, can be utilized in applications with
limited communication resources of a quadcopter with ground
station.
ACKNOWLEDGEMENT
This research is a part of work of the project funded by the
Interreg Balkan-Mediterranean program: “SFEDA-Forest
288
Monitoring System for Early Fire Detection and Assessment
in the Balkan-Med Area (MIS: 5013503)”.
REFERENCES
Al-Sabban, W. H., Gonzalez, L. F., Smith, R. N., & Wyeth, G.
F. (2012, February). Wind-energy based path planning for
electric unmanned aerial vehicles using markov decision
processes. In Proceedings of the IEEE/RSJ International
Conference on Intelligent Robots and Systems. IEEE.
Kavraki, L., Svestka, P., & Overmars, M. H. (1994).
Probabilistic roadmaps for path planning in high-
dimensional configuration spaces (Vol. 1994). Unknown
Publisher.
Langelaan, J. (2008). Biologically inspired flight techniques
for small and micro unmanned aerial vehicles. In AIAA
guidance, navigation and control conference and exhibit
(p. 6511).
Lawrance, N. R. J., Acevedo, J. J., Chung, J. J., Nguyen, J. L.,
Wilson, D., & Sukkarieh, S. (2014). Long endurance
autonomous flight for unmanned aerial vehicles.
AerospaceLab, (8), p-1.
Liu, L., & Sukhatme, G. S. (2018). A Solution to Time-
Varying Markov Decision Processes. IEEE Robotics and
Automation Letters, 3(3), 1631-1638.
MahmoudZadeh, S., Powers, D. M. W., Sammut, K., &
Yazdani, A. (2016). Differential evolution for efficient
AUV path planning in time variant uncertain underwater
environment. Robotics (cs. RO).
Mohammed, F., Idries, A., Mohamed, N., Al-Jaroodi, J., &
Jawhar, I. (2014, May). UAVs for smart cities:
Opportunities and challenges. In Unmanned Aircraft
Systems (ICUAS), 2014 International Conference on (pp.
267-273). IEEE.
Moulianitis, V. C., Thanellas, G., Xanthopoulos, N., &
Aspragathos, N. A. (2018, June). Evaluation of UAV
Based Schemes for Forest Fire Monitoring.
In International Conference on Robotics in Alpe-Adria
Danube Region (pp. 143-150). Springer, Cham.
Techy, L., & Woolsey, C. A. (2009). Minimum-time path
planning for unmanned aerial vehicles in steady uniform
winds. Journal of guidance, control, and dynamics, 32(6),
1736-1746.
Windfinder.com (2018), Wind, wave, & weather reports,
forecasts and statistics worldwide, Last accessed: 25 Dec
2018.