-
Wind Efficient Path Planning and Reconfigurationof UAS in Future
ATM
Leopoldo Rodriguez, Fotios Balampanis, Jose A. Cobano, Ivan
Maza, Anibal OlleroRobotics, Vision and Control Group
University of SevilleSeville, Spain
Email: lrodriguez15, fbalampanis, jcobano, imaza,
[email protected]
Twelfth USA/Europe Air Traffic Management Research and
Development Seminar (ATM2017)
Abstract—Unmanned Aerial Systems (UAS) integration tofuture
airspace is one of the greatest challenges in Air
TrafficManagement. The use of UAS for covering wide areas
impliesthe consideration of airspace restrictions and static and
dy-namic obstacle avoidance. This results in complex shapes
thatneed to be partitioned adequately to ensure coverage.
Anotherimportant element for consideration in the generation of
safeand efficient trajectories of UAS is the wind field.
Typically,in severe wind scenarios, wind is considered often a
hazardouscondition. However, recent studies show that proper
identificationof the wind field could be used to increase the
energy efficiencyof the mission. This paper presents a novel method
of areadecomposition and partition that ensures coverage by
generatinga triangular mesh to optimize the coverage in the
presence ofurban areas, airspace restrictions or even the presence
of anobstacle. The waypoint sequencing considers the wind field
inorder to perform on-line adjustments to ensure energy gainsor to
minimize energy losses with the identified wind field. Forthis
purpose, an innovative method for wind identification isproposed
which analyses the statistical behavior of wind vectorestimates in
order to identify specific features and characterizegiven models.
Given the design philosophy and architecture, thissystem can be
integrated into next generation autonomous UASflight management
systems as part of the waypoint sequencingand trajectory
optimization functions. A test case in the north-Seattle area is
presented, which is simulated using a 6DOF modelwith different wind
scenarios which resulted into considerableenergy gains either by
heeding the wind field during the waypointsequencing and during the
mission execution. Results show thatthere is a significant
improvement on the energy efficiency withan energy consumption
reduced by 10% in the presence of wind.
I. INTRODUCTION
Numerous ongoing research intend to provide the
necessaryrequirements and procedures for the safe integration of
UASinto non-segregated airspace in the context of the future
AirTraffic Management (ATM) system, proposed in the NextGeneration
Air Traffic Management System (NextGen) and theSingle Sky European
Research (SESAR). Potential researchand commercial UAS applications
including goods delivery,search and rescue, among others, require a
precise set ofrules to ensure safety and reliability of the
involved actors.In the context of applications which require area
coveragein a non-segregated airspace, there are many aspects
whichneed to be considered. Smart path planning is a key area
thatneeds to be studied in order to ensure that any given task
does not compromise the safety of the airspace in which itis
being performed. Current and future ATM imply complexand dynamic
areas which represent numerous challenges inmission planning.
Regarding light and small UAS, the NationalAeronautics and Space
Administration (NASA) has proposedthe development of the Unmanned
Air Traffic Management(UTM) [1] system for Low-Altitude UAS as a
response tosafely manage UAS in airspace that are not regulated by
theFederal Aviation Administration (FAA). This effort involves athe
participation of very important partners such as Amazon,Google,
Lockheed Martin Corporation, Honeywell, etc. Inaddition of this set
of rules, one important concern of theairspace that is managed by
the civil authority is the need toprovide with a reliable
Detect-and-Avoid capability as exposedby Haessig et al. [2], in
which technologies such as Au-tonomous Dependent System-Broadcast
(ADS-B) is exploredas a means to resolve potential conflicts with
cooperativeobstacles. There are vast studies about such as the
onespresented by Cordon et al. [3], [4] and Paczan et al [3],
[4].that provide an insight, in a systems perspective, of
differentaspects of the integration of UAS in both NextGen and
SESARcontexts. In [3] the authors present the requirements of
theinterfaces that are needed for the different phases of flightof
a UAS including the mission/flight preparation, as definedby the
SESAR Concept of Operations (CONOPS) [5]. TheBusiness or Mission
Development Trajectory (BDT/MDT) isa key element on the future ATM
and UTM operations, sincemission planning, specially in areas in
which the air traffic isdense, represents a considerable
challenge.
One important element that can be considered in the earlystages
of mission planning, and has proven to affect the oper-ation of all
types of UAS, is the weather data [6], particularlythe wind field.
Nowadays, the use of Commercial-On-The-Shelf (COTS) components
permits that the UAS navigationsystems, even those classified as
small or very light, providethe users with very accurate
information on their state vector.However, the use of sensors to
determine accurately the windvector during the flight are expensive
and does not providea significant impact neither during the mission
planning norduring the early stages of flight, in which accurate
knowledgeof the wind field may result into a more comprehensive
pathplanning. Different research efforts, such as the one
presentedby Langelaan et al. in [7] and Wenz et. al. [8] present
methods
-
for wind estimation with low-cost sensors even
consideringhazardous wind conditions. The instantaneous online
estima-tion of the wind vector at any stage of flight, or during
thepre-fight operations may not be sufficient to take advantage
ofthe wind field to increase flight efficiency. The knowledge
ofhazardous wind conditions, which can be even inferred withweather
reports or by observation, may impose restrictions inthe use of
UAS, even in segregated airspace. Nevertheless,the use of wind as a
means to harvest energy in order toincrease the efficiency has been
a subject of research such as in[9]. By characterizing the wind
field and identifying punctualphenomena such as wind-shear or
thermals, the UAS may gainenergy that permit to fulfill its mission
with less fuel or toincrease the duration of flight. In the cases
presented in theliterature [7], [9], and also in those that has
been studied by theauthors [10], the identification of wind is
performed withoutany restriction in airspace. Therefore, it has
been noticed thatthe integration of the wind identification with
the path planningproblem has to be taken into account for future
ATM and UTMUAS operations.
The path planning problem and wide area coverage repre-sents by
itself a challenge if the complex shapes of airspace aretaken into
account. The plentiful restrictions, the considerationof static and
moving obstacles, specially in operations close tourban areas, and
the potential appearance of sudden air trafficrestrictions require
complex algorithms to quickly respond tothe eventuality and
prioritizing the mission accomplishment.As it is mentioned before,
the wind field identification mayresult into the imposition of
greater restrictions which will addcomplexity to the area
decomposition, path planning (waypointsequencing) or the path
re-planning in an ongoing mission.However, if the wind field is
identified it also may representan advantage to increase the energy
efficiency throughout themission. Area decomposition and path
planning problems havebeen widely studied before [11] and the use
of algorithmssuch as boustrophedon movements typically represent
solu-tions which do not ensure 100% coverage or complex areas.The
authors have previously proposed an area decompositionmethod [12]
[13] which seeks complete coverage waypointplans and considers
potential airspace restrictions regardlessits shape, allowing the
reconfiguration of the decomposed areaand the waypoint
re-sequencing considering different aspectsor weights.
The purpose of this paper is to merge the area decompo-sition
and path planning problem with a wind identificationmethod in order
to provide a mission planning and reconfig-uration solution that
considers the complex shapes of currentand future airspace to
ensure the highest possible coverage,also taking into account the
efficiency and duration of themission by acknowledging the
identification of the wind fieldas a potential solution.
II. UAS-BASED WIND FIELD IDENTIFICATION
The wind field identification problem, as proposed in
[10],involves two stages. One that identifies the instantaneous
windfield at a given rate and other that performs a statistical
ybxbzb
yI
zI
xI^
^
^^
^^
r
va
w
Fig. 1. UAS located in an inertial frame.
analysis to these estimates in order to identify wind
features,which can be as simple as constant wind or as complex
asdiscrete and continuous gusts or wind shear.
A. Wind Vector Estimation
The wind vector estimation is performed with a DirectComputation
(DC) method. It consists in the use of the GlobalNavigation
Satellite System (GNSS) navigation solution to-gether with the air
mass relative speed in order to determinethe wind vector estimate
without the use of a Bayesian filter.
Let a UAS to be located in a vector r in an inertial frameI with
unit vectors defined as (x̂I , ŷI , ẑI). Heeding a bodyframe
located in the UAS center of mass with unit vectors(x̂b, ŷb, ẑb),
the wind vector w with components (wx, wy, wz)and the
air-mass-relative velocity (absolute airspeed) va areshown in Fig.
1.
From this point, one can infer the total velocity, based onthe
computation of the so-called speed triangle ṙ in the inertialframe
as shown in (1):
ṙ = va + w (1)
Therefore, the wind vector w can be calculated in the
inertialframe as in (2):wxwy
wz
I
=
ẋẏż
I
− (CbI )−1uvw
b
(2)
The ground position with coordinates (ẋ, ẏ, ż)I can
beobtained with good precision in the navigation solution of
theGNSS system, which typically has a Kalman filter
functionembedded to increase the precision. The airspeed
components(u, v, w) which are typically given in the body frame can
bedetermined with the knowledge of angle of attack and sideslip,or
with the use of an multi-axis airspeed sensor as proposedby Wenz et
al in [8]. COTS autopilots have the capability tocompute the
airspeed, AOA and sideslip which is sufficient todetermine the wind
components with relative good precision.
The main source of error in (2) cannot be inferred easilysince
the sources of error are independent. However, in [7],the
computation of the wind speed acceleration, which canbe also
observed in [10], permits the determination of theairspeed
measurements as the most important source of error.It has been
determined that there is no need of a Bayesian
-
filter to keep the wind vector estimation error bounded
withinacceptable limits (between 0.7 m/s and 1 m/s for low
flightpath angles and without GNSS augmentation) [7].
B. Wind Field Prediction
In order to characterize the surrounding wind field, wecan
consider a wind field as the “sum” of four features:constant wind
in a given direction, wind shear, discrete gustsand continuous
gusts. The Wind Identification System (WIS)proposed by the authors
in [10] performs a statistical anal-ysis of accumulated wind
estimates together with off-boardinformation such as weather
reports or a wind database.
Other important considerations that need to be taken intoaccount
are that wind measurements are typically distributedaltitude-wise
following a Weibull distribution. Given a data setof wind vector
estimations W = (W1, ...,Wn), the Weibulldistribution can be
expressed as:
f(W) =κ
ν
(W
ν
)(κ−1)eWν κ (3)
where κ and ν are respectively the shaping and scalingparameters
of the Weibull distribution.
From here, one can infer the most probable wind speed‖w‖r at a
particular location as a function of κ and ν:
‖w‖r = ν(
1− 1κ
) 1κ
(4)
The identification of the Weibull parameters to calculatethe
wind speed magnitude is not trivial, therefore, a solutionwhich has
been implemented as in [10] is to use a GeneticAlgorithm (GA) in
order to estimate the Weibull parametersand to determine the most
probable wind speed at a givenlocation. An implementation of the GA
to find the Weibullparameters can be found in [10].
Once the Weibull parameters are identified in altitudegroups, a
statistical analysis of the running mean and standarddeviation of
the wind estimates is performed in order to fitthe wind estimates
into feature models either by analyzing thewind magnitude change
over altitude with an Empirical PowerLaw (EPL) or by performing a
short term Gaussian regressionto characterize complex features. The
selected models arebased on the U.S. Military Specification
MIL-F-8785C [14],[15].
1) Wind Shear Identification: In the presence of a WindShear,
wind estimates show a growth in the airspeed overaltitude as shown
in Fig 2. It is represented by the followingexpression:
‖w‖shear = W20ln hz0
ln 6.096z0, 1m ≤ h ≤ 300m (5)
Shear is a typically undesired phenomena in terminal
areaoperations for manned aircraft. However, in the case of UASand
while operating in the UTM, it can be useful if the surfacelayer is
identified properly, i.e. the altitude in which the wind
0 2 4 6 8 10 12
Wind speed (m/s)
0
50
100
150
200
250
300
Alt
itude (
h)
Fig. 2. Wind shear model.
-20 0 20 40 60 80 100 120 140 160 180 200
Distance (m)
0
1
2
3
4
5
6
Win
d s
peed (
m/s
)
Gust length
Gust
am
plitu
de
Fig. 3. Discrete gust model.
speed is decreasing faster allowing a energy gain in the formof
speed or altitude.
If a group of most probable wind speeds Ω =Wmp1,Wmp2, ...,Wmpn
can be fitted with a polynomialapproximation into the empirical
power law, then the WISassumes that there is a presence of shear on
the system.By analyzing the components of the wind vector
estimates,the wind direction can be inferred by averaging the
directioncosines between them.
2) Discrete Gust Identification: The discrete gust can beseen as
an increase of the wind velocity magnitude within adistance. The
considered model is the 1− cos as in [14], [15]:
‖w‖gust =
0 x < 0Wm
2
(1− cos πxdm
)0 ≤ x ≤ dm
Wm x > dm
(6)
And it is represented in Fig. (3)In order to characterize the
discrete gust one can refer to the
work presented by Zbrozek [17] in which the distribution
ofdiscrete gusts, while knowing the power spectrum of
normalacceleration, can be utilized in order to determine the
intensityof the discrete gust by a continuous distribution of Root
MeanSquared (RMS) turbulence. Hence, the analytic expression
forprobability density distribution of a gust velocity is:
f̂(σw) =
√2
π
1
bexp
(−1
2
(σwb
)2)(7)
-
Therefore, by observing the RMS distribution of the
windestimates one can determine the intensity if the gust is
con-sidered a stationary process.
3) Continuous Gust Identification: For the continuous gust,the
Dryden spectral representation is used. This representation,also
approved in [14], [15] treats the linear and angularwind velocities
as spatially varying stochastic processes inwhich each component is
defined by a power spectral density.Therefore, the power spectral
densities Φ for linear velocities(ug, vg, wg) and angular
velocities are shown in (8) and in (9).
Φug (Ω) = σ2u
2Luπ
1(1+LuΩ)2
Φvg (Ω) = σ2v
2Lvπ
1+12(LvΩ)2
(1+4(LuΩ)2)2
Φwg (Ω) = σ2w
2Lwπ
1+12(LwΩ)2
(1+4(LwΩ)2)2
(8)
Φpg (ω) =σ2w
2V Lw0.8(
2πLω4b
) 13
Φqg (ω) =±( ωV )
2
1+( 4bωπV )2 Φwg (ω)
Φqg (ω) =∓( ωV )
2
1+( 3bωπV )2 Φvg (ω)
(9)where σi is the root-mean-square vertical or lateral gust
velocity and Li is an integral length scale of the
turbulenceeddies in the ith. velocity or angular component. Ω is
thespatial frequency and b represents the aircraft wingspan.
Trying to characterize the spectral density definitions inorder
to predict a turbulence is a complex problem thatrequires high
computational power and the results may not beuseful for online
use. However, a standard Gaussian Process(GP) regression can be
incorporated in order to perform ashort-term prediction to
determine a covariance vector q(X, x)and a linear prediction p̄(X)
of a linear combination of thewind estimates in the ith direction.
The expression of theprediction is:
p̄(X) = q(x,X)[Q(X,X) + σ2nI
]−1 Ŵz (10)where q(x,X) is the covariance vector between two
obser-
vations at location X, Q(X,X) is the covariance matrix and σ2nis
the measurement noise covariance. Additional details on
theimplementation of the regression can be found in [10].
III. COMPLEX AREA PARTITIONING CONSIDERING
AERIALRESTRICTIONS.
The increased interest of UAS usage for commercial pur-poses
reflects on several mission scenarios which surpass theuse of
specific waypoint airways. These missions are oftenhandled by the
means of a grid decomposition of areas inorder to accomplish
complete coverage [11], for instancein crop spraying or aerial
photography. As we’ve shown inprevious studies [12] [18], complex
scenarios and geographicattributes are not treated properly by the
use of a simple griddecomposition of an area. More specifically,
coastal area taskswith their numerous no fly zones or complex
shores, impose adynamical approach which has been developed in the
contextof the MarineUAS project.
Fig. 4. A test case area, north of Seattle. The red polygon
defines the missionarea whereas several other restrictions apply,
as can be seen in the next figures.
In a test case scenario area as seen in Fig.4,
severalheterogeneous UAS have the task of covering the area inorder
to obtain sea life information by using their sensors.In order to
perform a fair and successful partitioning of theconfiguration
spaces for each UAS, respecting their relativesensing capabilities,
like their Field of View (FoV), as also asthe strict borders of the
area or future airspace restrictions, aninitial Constrained
Delaunay Triangulation (CDT) is proposed.
A CDT introduces forced edge constrains as part of the inputand
in such a way, complex areas can be triangulated, creatinga
triangular mesh. Then, each centroid of every triangle can
beconsidered as a waypoint in the flight plan. Moreover,
everytriangle can be given a cost, based on several task,area or
agentrelated criteria. As we will describe, this cost can be used
forwind information and will further facilitate the extraction
ofwaypoint flight plans.
Consider the region presented in Fig.4 as C, includingobstacles,
as shown in Fig.5. For waypoint list planning, thisarea is treated
as a two dimensional grid such that C = R2,where Cobs are the
restricted areas and Cfree = C \Cobs isthe area to be partitioned
for the UASs. By triangulating andpartitioning Cfree for M vehicles
in a sum of triangles (ψ)such that
Cfree =M∑i=1
N∑j=1
ψij , (11)
the waypoint planning is actually a graph search problemof N
nodes organized in the CDT. The algorithmic strategiespresented in
[13] are not computationally expensive and permitthe online
reconfiguration and planning for either detect andavoidance
purposes, task updates or emergency situations.
IV. WIND EFFICIENT WAYPOINT SEQUENCING
In order to perform the wind predictions, as describedin Section
II-B, wind estimations have to be stored until asufficient number
of them allows such process. This processtakes in average 60 s from
the moment the UAS goes airborne[10].
During pre-flight phase, the meteorological reports and theuse
of external sensors, such as an anemometer, allow anestimation of
the predominant wind direction and to do a
-
Fig. 5. Several low altitude restrictions apply in the area,
like those that canbe seen on the bottom right part. Moreover, the
system must be capable torespond in any online restriction.
Screenshot is a courtesy of SkyVector.com
Start
Preflight Wind
Yes
DetermineStarting Position
Perform Pre-Flight
WEWS
No
PerformInitial
Sequencing
No
Yes
Obtain Wind
Estimates
UASAirborne
Wind Est.Sufficient
No
Yes
Perform FeatIdentification
DetermineWind Direction
DetermineNeighborghs
Weights
AdjustNew Sequence
DetermineTrajectory
Curve
TrackTrajectory
Yes
Fig. 6. High level view of the WEWS algorithm.
initial wind-based sequencing in order to improve the
flightefficiency.
The Wind Efficient Waypoint Sequencing (WEWS) processweights the
cells of the decomposed area in such a way thatprioritizes the
information on the wind field in two stages: thefirst one aims to
determine the predominant wind directionin the presence of constant
wind, shear and/or discrete gustsand determines a sequence in which
the sequence tries tokeep a positive component aligned to the wind
velocity. Theother proposed process intents to provide information
to theautopilot in order to adjust the trajectory on a
waypoint-to-waypoint basis to maximize the use of the wind as a
meansof gaining energy. The WEWS high level process is depictedin
Fig. 6.
The initial sequencing is considered as an iteration into
theShared Business or Mission Trajectory (SBT/SMT) process.However,
during the execution of the mission, the trajectorymay suffer
adjustments which are to be used in the update
of the Reference Business or Mission Trajectory (RBT/RMT).During
anytime during the mission execution, the path mightsuffer from
temporary restrictions or even the appearance of anobstacle. The
area decomposition and partitioning algorithmsare able to
reconfigure the waypoint sequence accordinglyensuring coverage of
the area still with the wind informationas part of the weight.
The energy as a function of the trajectory q(t):
Ē(q(t)) =∫ T
0
[c1‖v(t)‖3 +
c2‖v(t)‖
(1 +‖a(t)‖2 −m
g2)
]dt
+1
2m(‖v(T )‖2 − ‖v(0)‖2)
(12)where
m =(aT (t)v(t))2
‖v(T )‖2(13)
v(t) corresponds to the first derivative of the
trajectoryfunction:
v(t) ∆= ˙(q)(t) (14)
a(t) is the second derivative of the trajectory function:
a(t) ∆= ˙(q)(t) (15)
and g is the gravitational constant.In general, avionics power
consumption is significantly
smaller in proportion to the propulsion energy. Therefore
onlykinematic related components are included into (12).
V. TEST CASES, SIMULATIONS AND RESULTS
The area of Fig.4 has been chosen as a test case, by takinginto
consideration a simple aerial restriction as can be seenin Fig.5.
The area has been partitioned for two UASs, havingthe same FoV but
different coverage capabilities. A coveragewaypoint flight plan has
been extracted for one of them, havingas a sequence criterion the
outer to inner complete coverage(Fig.7). The simulated experiments
show that the resultingtrajectories manage to respect the
aforementioned restrictionswhile successfully performing the
coverage task (Fig.8).
For simulation purposes the UAS with 20% relative areacoverage
capability was selected. The simulated platform cor-responds to the
Aerosonde UAV (see Fig. 9. Its characteristicsare enumerated in
Table I.
TABLE IAEROSONDE UAV CHARACTERISTICS
Length 1.7 mWidth 2.9 mHeight 1.97 mWeight (MTOW) 25 kgMaximum
Speed 140 knotMaximum Range 3000 kmMaximum Ceiling 4500 m (14760
ft)
-
Fig. 7. North Seattle area with a flight restriction on the
bottom right.Left: initial partition for 2 UASs, having an 80% and
20% relative areacoverage capabilities. Black triangles represent
the initial positions of theUASs, while the red dots represent the
waypoints. The two configurationspaces are distinguished by the
different shades of grey. Right: Borders tocenter cost applied to
each cell and a coverage flight plan has been extractedfor one of
the UASs.
Fig. 8. An APM screenshot during flight.
A 6DOF model was selected in order to perform
Software-In-The-Loop (SITL) experiments with the WEWS algo-rithms
in place. Simulations were performed using theMATLAB/SIMULINK R©.
environment together with the Pix-hawk SITL environment.
Three scenarios are considered, the first one decomposesthe area
as in [12], sequencing the waypoints from the edgesto the center in
order to ensure full coverage. The secondscenario considers the
presence of sustained wind. Finally the
Fig. 9. Aerosonde UAV.
X position (m) #104-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
Y Po
sitio
n (m
)
#104
-5
-4
-3
-2
-1
0
1
2
Starting Point
DIrection of sequence
Fig. 10. Initial area decomposition showing the starting point
and the winddirection.
TABLE IIFIRST SEQUENCING SIMULATION RESULTS
Flight Duration 3.2 hMedium Altitude 100 m ASLTotal Energy
Consumption ≈22385.66 W/hAverage Wind Speed 0 m/sAverage Airspeed
20.93 m/s
third scenario considers shear and gust phenomena.
A. First scenario: zero wind consideration
Taking the UAS located at the bottom left of Fig. 7.The proposed
area is decomposed and sequenced initially asindicated in Fig. 10,
in which the coordinates are expressedas relative position to the
starting point. Note that this basesequencing does not consider
prior information of the windfield.
While executing this trajectory the approximated resultedenergy
as per 12 in a Software-In-The-Loop (SITL). Theresults of the
simulation are shown in Table II
For the energy calculation a standard gasoline engine with
adisplacement of 55 CC and a power of 5.6 HP was consideredas a
test case. Hence, the avionics consumption is not includedin the
calculation.
B. Second scenario: constant wind consideration
For the second scenario, an average sustained wind of 5 m/sis
considered with a wind direction of (−90o). This allowsto
appreciate the impact of the wind in the initial
waypointsequencing. If this wind is considered then the system
triesto minimize the amount of times that the trajectory of
theaircraft has a negative component relative to the direction
ofthe wind. In addition, the sustained wind was identified
andcharacterized as per the Weibull distribution indicated in
(3).
The resulting sequencing with the online adjustment is de-picted
in Fig. 11. In addition, the identified wind velocity withthe
corresponding Weibull characterization can be observed inFig.
12.
The simulation results and energy consumption are shownin Table
III. It can be observed and improvement of approxi-mately 11% in
the energy consumption. Note that during thesimulation the flight
mode was on speed control, therefore,the UAS travels a similar
distance with a different sequence.Hence, the flight duration is
not significantly impacted. Notethat the average airspeed is
decreased due to the wind effect.
-
X position (m) #104-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
Y Po
sitio
n (m
)
#104
-5
-4
-3
-2
-1
0
1
2
Starting Point
Fig. 11. Area decomposition considering sustained east wind.
Time (s)0 2000 4000 6000 8000 10000 12000
Win
d Ve
locit
y (m
/s)
2
2.5
3
3.5
4
4.5
5
5.5
6
Fig. 12. Estimated wind speed over time with a mean value of 5
m/s.
Wind Speed (m/s)2.5 3 3.5 4 4.5 5 5.5 6 6.5
Freq
uenc
y
0
50
100
150
200
250
300
350
400
Fig. 13. Distribution of wind estimates following a Weibull
shape.
TABLE IIISECOND SEQUENCING SIMULATION RESULTS
Flight Duration 3.16 hMedium Altitude 100 m ASLTotal Energy
Consumption ≈19913.408 W/hAverage Wind Speed 5 m/sAverage Airspeed
16.74 m/s
During the simulation, online adjustments are done in
thetrajectory allowing reshaping the remaining waypoint
sequenc-ing. The main inconvenience of this is that there are sharp
turnsthat may result into higher energy consumption considering
thefull dynamics and the actual trajectory tracking.
C. Second scenario: gust wind consideration
The last scenario considers the presence of gust and
shear.however, since there are no changes in altitude the shear
doesnot affect directly the energy gain. The simulation starts
withno knowledge of the wind field and after a minute of flight,the
first wind predictions start to occur. Once the systemdetermines
the wind velocity and direction (5 m/s, −90o)the algorithm starts
to reconfigure the path during the firstpart of the flight. At 3000
s, a discrete gust of 2 m/s occurs
X position (m) #104-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
Y Po
sitio
n (m
)
#104
-5
-4
-3
-2
-1
0
1
2
Starting Point
Fig. 14. Area decomposition considering discrete gust.
Time (s)0 2000 4000 6000 8000 10000 12000
Win
d ve
locit
y (m
/s)
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
Fig. 15. Estimated wind speed over time with a mean value of 5
m/s.
Wind velocity (m/s)3 4 5 6 7 8 9 10
Freq
uenc
y
0
50
100
150
200
250
300
350
400
450
Fig. 16. Distribution of wind estimates following a Weibull
distribution.
TABLE IVSECOND SEQUENCING SIMULATION RESULTS
Flight Duration 3.19 hMedium Altitude 100 m ASLTotal Energy
Consumption ≈20421.783 W/hAverage Wind Speed 6.24 m/sAverage
Airspeed 15.23 m/s
which results into the increase of the wind velocity
magnitude.The direction of the wind is also affected hence the
systemperforms a more severe reconfiguration of the flight path.
Theidentification of the gust can be determined by detecting
theincrease of the wind speed (see Fig. 15) and also with
thegeneration of two Weibull distributions as seen in Fig. 16.
The simulation results can be observed in Table IV. In herea
energy improvement of 8% was observed. Note that thewaypoint
sequence starts as in Fig. 10, however as the wind isidentified and
the gust is detected there are significant changeson the
sequence.
In the two cases, the energy improvement occurred onlyby
rearranging the original waypoint sequence. Even thoughsome
overlapping of way points may be observed, there is stilla
significant improvement by considering the wind force.
-
VI. CONCLUSION AND FUTURE WORK
The presented algorithms for area decomposition,
waypointsequencing and wind identification are a promising
combi-nation for generating wind efficient trajectories for UAS
infuture airspace. Normally, a decomposition method takes
littleconsideration of the wind phenomena as an aid to improvethe
energy efficiency. The cell weighting method allows thegeneration
online reconfiguration of the intended waypointsequence considering
potential airspace restrictions in thecontext of current and next
generation airspace. The designedarchitecture allows an easy
incorporation into next generationBusiness or Mission Trajectory
(B/MT) procedures. The areadecomposition ensures the maximum
coverage of a givenarea due to the triangular cells that permit the
inclusionof complex shapes, which are given by the surface of
thesurveyed area and/or airspace restrictions. The results
showsimprovements up to 11% of efficiency with low winds andup to
9% in the presence of gust and shear with onlinesequence
reconfiguration. Efficiency is one of the key aspectsin future
airspace with more stringent carbon-dioxide emissionregulations,
and the wind energy harvesting is a promisingarea for UAS operating
in low altitudes in both the UTMand ATM systems. The incorporation
of wind identificationand smart area decomposition into UAS flight
managementfunctions shall permit a more efficient use of airspace
evenin hard meteorological conditions. Future work includes
theincorporation of the described system to a trajectory
generationsystem in order to determine, on a waypoint-to-waypoint
basis,the 4D optimal trajectories for maximum energy gain given
awind field. This will allow a more realistic quantification of
theenergy harvested to the wind by generating soaring
trajectorysolutions even for the sharp turns that the sequencing
systemgenerates. The safety and reliability of UAS systems
includingCOTS components allow identification of the wind vector
andwind field with the necessary precision to perform the
initialsequencing, online reconfiguration and trajectory
optimization.In addition, a full test campaign is being prepared in
orderto validate and verify the different functions for
safe-longduration missions.
ACKNOWLEDGMENT
This work has been supported by the MarineUAS
project(MSCA-ITN-2014-642153), funded by the European Com-mission
under the H2020 Programme as part of the MarieSklodowska Curie
Actions and the AEROMAIN project(DPI2014-5983-C2-1-R), funded by
the Science and Innova-tion Ministry of the Spanish Government.
REFERENCES
[1] UTM: Air Traffic Management for Low-Altitude Drones,
National Aero-nautics and Space Administration. Washington DC,
United States ofAmerica, 2015.
[2] D.A. Haessig, R.T. Organ, M. Olive. “ “Sense and Avoid” -
What’srequired for aircraft safety”, SoutheastCon 2016 . ,
Northfolk, VA.United States of America, March 30-April 3. 2016.
[3] R.R. Cordón, F. J. Saz, C. Cuerno. “RPAS Integration in
Non-segregatedAirspace: the SESAR Approach. Systems Needed for
Integration”, FourthSESAR Innovation Days . Madrid, Spain, 25th
27th November 2014.
[4] N.M. Paczan, J. Cooper, E. Zakrewski. “Integrating Unmanned
AircraftInto NextGen AUtomation Systems”, 31st Digital Avionics
SystemsConference. Williamsburg, VA, United States of America,
October 14-18, 2012.
[5] SESAR Concept of Operations Step 1, EUROCONTROL, SESAR
JointUndertaking. Brussels, Belgium, 2013.
[6] L. Nelson. “More than Just a Weather Forecast. The Critical
Role ofAccurate Weather Data in UAV Missions”, Defense Update.
UnitedStates of America, 2009.
[7] J.W. Langelaan, N. Alley, J. Neidhoefer. “WInd Field
Estimaton for SmallUnmanned Aerial Vehicles”, AIAA Guidance,
Navigation and ControlConference. Toronto, Canada, 2010.
[8] A. Wenz, T.A. Johansen, A. Cristofaro. “Combining model-free
andmodel-based Angle of Attack estimation for small fixed-wing UAVs
usinga standard sensor suite” International Conference on Unmanned
AircraftSystems (ICUAS), 2016.
[9] A. Chakrabarty. J.W. Langelaan. “Flight Path Planning for
UAV Atmo-spheric Energy Harvesting Using Heuristic Search”, AIAA
Guidance,Navigation and Control Conference. Toronto, Canada,
2010.
[10] L. Rodriguez. J.A. Cobano. A. Ollero “Small UAS-Based Wind
FeatureIdentification System Part 1: Integration and Validation”,
Sensors 2017,17(1).
[11] E. Galceran, M. Carreras. “A survey on coverage path
planning forrobotics. Robotics and Autonomous Systems”, 61(12),
pp.1258-1276,2013.
[12] F. Balampanis, I. Maza, A. Ollero. “Area decomposition,
partition andcovarage with multiple remotely piloted aircraft
systems operating incoastal regions”. International Conference on
Unmanned Aircraft Systems(ICUAS). Arlington, VA, United States of
America, 2016, pp. 275-283,2016.
[13] F. Balampanis, I. Maza, A. Ollero. “Coastal Areas Division
and Cov-erage with Multiple UAVs for Remote Sensing”, Sensors,
17(4), 808,2017.
[14] U.S. Department of Defense “Military Specification, Flying
Qualities ofPiloted Airplanes MIL-F-8785C”, Department of Defense.
WashingtonDC, United States of America, 1980
[15] U.S. Department of Defense “Handbook, Flying Qualities of
PilotedAirplanes MIL-F-8785C”, Department of Defense. Washington
DC,United States of America, 1980
[16] C. Gao. “Autonomous Soaring and Surveillance in Wind Fields
with andUnmanned Aerial Vehicle”, PhD Thesis, University of
Toronto. Toronto,Canada, 2015.
[17] J.K. Zbrozek. “The Relationship between the Discrete Gust
and PowerSpectra Representations of Atmospheric Turbulence With a
SuggestedModel of Low-Altitude Turbulence”, Aeronautical Research
CouncilReports and Memoranda No. 3216 (1). Cranfield, United
Kingdom,March, 1960.
[18] F. Balampanis, I. Maza, A. Ollero. “Area Partition for
Coastal Regionswith Multiple UAS”, Journal of Intelligent &
Robotic Systems (2017),
doi:10.1007/s10846-017-0559-9.