You are on page 1of 6

Proceedings of the 3rd IFAC Workshop on Multivehicle Systems WePL.

11
Proceedings
May 18, 2015.of the 3rd
Porto IFACGenova,
Antico, Workshop on Multivehicle Systems
Italy WePL.11
Proceedings of the 3rd IFAC Workshop on Multivehicle Systems WePL.11
Proceedings
May 18, 2015.of the 3rd
Porto IFACGenova,
Antico, Workshop on Multivehicle Systems
Italy WePL.11
Available online at www.sciencedirect.com
May 18, 2015. Porto Antico, Genova, Italy
May 18, 2015. Porto Antico, Genova, Italy

ScienceDirect
IFAC-PapersOnLine 48-5 (2015) 059–064
Model
Model Predictive
Predictive Control
Control for
for UAV
UAV Automatic
Automatic
Model Predictive
Model Predictive Control
Control for UAV
forDeck Automatic
UAVwith
Automatic
Landing
Landing on
on Moving
Moving Carrier
Carrier Deck with Heave
Heave
Landing on Moving Carrier
Carrier Deck with Heave
Landing on MovingMotion Deck with Heave
Motion
Motion
Motion
Soyeon Koo, Seungkeun Kim, Jinyoung Suk ∗∗
Soyeon
Soyeon Koo, Seungkeun Kim, Jinyoung Suk ∗∗
Soyeon Koo, Koo, Seungkeun
Seungkeun Kim, Kim, Jinyoung
Jinyoung Suk Suk
∗ Department of Aerospace Engineering, Chungnam National University,
∗ Department of Aerospace Engineering, Chungnam National University,
∗ Department of Aerospace Engineering, Chungnam National University,
∗ Department Daejeon of Aerospace 305-764, Republic Chungnam
Engineering, of Korea, (e-mail:
National University,
Daejeon
Daejeon 305-764,
305-764, Republic
Republic of
of Korea, (e-mail:
esther9112@hanmail.net,skim78,jsuk@cnu.ac.kr)
Daejeon 305-764, Republic
esther9112@hanmail.net,skim78,jsuk@cnu.ac.kr) of Korea,
Korea, (e-mail:
(e-mail:
esther9112@hanmail.net,skim78,jsuk@cnu.ac.kr)
esther9112@hanmail.net,skim78,jsuk@cnu.ac.kr)
Abstract: This paper investigates a model predictive automatic carrier landing system considering the
Abstract:
Abstract: This
This paper investigates aa model predictive automatic carrier landing system considering the
motion
Abstract:
motion of This paper
of carrier.
carrier.
Withinvestigates
paper
With
the medium-altitude
investigates
the medium-altitudea model
model predictive
predictive automatic
long-endurance
automatic
long-endurance
UAVcarrier
UAV carrier
and
landing
and carrier
landing
carrier
system
models,
systemthe
models,
considering
guidance and
considering
the guidance
the
the
and
motion
control
motion of of
for carrier.
safe
carrier. With
shipboard the medium-altitude
landing
With the medium-altitude use the long-endurance
relative geometry
long-endurance UAV and
between
UAV carrier
the
and carrier UAVmodels,
and
models, the guidance
carrier
the deck
guidance and
into
and
control
control for
for safe shipboard landing use the relative geometry between the UAV and carrier deck into
account.
control Thesafe
for safe shipboard
automatic
shipboard landing
carrier
landing landing use the
the relative
use system geometry
is operated
relative geometry between
by two
betweentypesthe UAV
UAV and
of control
the carrier
systems.
and carrierOne deck
deckis into
the
into
account.
account. The
The automatic
automatic carrier
carrier landing
landing system
system is
is operated
operated by
by two
two types
types of
of control
control systems.
systems. One
One is
is the
the
linear
account. quadratic
The tracker
automatic with
carrier integral
landing designed
system to
is track the
operated desired
by two glide
types slope.
of The
control other
systems. is the
One model
is the
linear
linear quadratic
quadratic tracker
tracker with
with integral
integral designed
designed to
to track
track the
the desired
desired glide
glide slope.
slope. The
The other
other is
is the
the model
model
predictive
linear quadratic control which
tracker finds
with the
integral optimizing control
designed control input
to trackinput sequences
the desired over
glideover a
slope. certain
The othertime horizon
ishorizon
the model by
predictive
predictive control
control which
which finds
finds the
the optimizing
optimizing control input sequences
sequences over a
a certain
certain time
time horizon by
by
predicting
predictive the the future
control whichmotions of
findsofthe the UAV and
optimizing the carrier
control deck.
input Numerical
sequences simulations
over are
a certainare timeperformed
horizon forfor
by
predicting
predicting the future
future motions
motions of the
the UAV
UAV and
and the
the carrier
carrier deck.
deck. Numerical
Numerical simulations
simulations are performed
performed for
an UAV
predicting and a carrier
thea future with
motions heaveof the motion to
UAVtoand verify the feasibility
the carrier of the
deck. Numerical proposed approach.
simulations are performed for
an
an UAV
UAV and
and carrier with heave motion verify the feasibility of the proposed approach.
an UAV IFAC
© 2015, and aa (International
carrier
carrier with with heave
heave motion
motionofto
Federation
verify
verify the
toAutomatic feasibility
theControl)
feasibility of
of the
Hosting thebyproposed
proposed approach.
approach.
Elsevier Ltd. All rights reserved.
Keywords: Unmanned Aerial Vehicles, Automatic carrier landing, Model predictive control, Linear
Keywords:
Keywords: Unmanned
Unmanned Aerial Vehicles, Automatic carrier landing, Model predictive control, Linear
Quadratic
Keywords: Tracker
Unmanned withAerial
Integral.
Aerial Vehicles,
Vehicles, Automatic
Automatic carriercarrier landing,
landing, Model
Model predictive
predictive control,
control, Linear
Linear
Quadratic
Quadratic Tracker with Integral.
Quadratic TrackerTracker with with Integral.
Integral.
1. INTRODUCTION future command to determine current control action. Thus, the
1. INTRODUCTION
1. INTRODUCTION future command
future command to to determine
determine current
current control
control action.
action. Thus,
Thus, the the
1. INTRODUCTION model
future predictive to
command control is suitable
determine currentfor automatic
control action. UAV Thus,carrier
the
model
model predictive
predictive control is suitable for automatic UAV carrier
landing
model because itcontrol
predictive control is
is suitable
can consider
suitablethefor
for automatic
future geometry
automatic UAV
UAV carrier
between
carrier
Unmanned Aerial Vehicles (UAVs) have operated various mis- the landing because
landing because it it can
can consider
consider the the future
future geometry
geometry between between
Unmanned Aerial Aerial Vehicles
Vehicles (UAVs)
(UAVs) have have operated
operated various various mis-mis- the UAVbecause
landing and theitcarrier with heave
can consider motion.geometry
the future The previous betweenre-
Unmanned
sions in military and civil domains. Recently, operating UAVs the UAV and
UAV and thethe carrier
carrier with heave
with heave motion. The
motion. The previous
previous re-
re-
Unmanned
sions in Aerial
military Vehicles
and civil (UAVs)
domains. have operated
Recently, various
operating mis-
UAVs searches
the UAV Kang Kang
and theandand Hedrick
carrier (2009),
with (2009), Chao
heave motion. et
Theal. (2012)
previous and
re-
sions
thein military and searches Hedrick Chao et al. (2012) and
on
sions
on the inaircraft
aircraft and civil
militarycarrier
carrier civil
is
domains.
is high
domains.
high
interest
interest
Recently,
as the operating
Recently,
as the
X-47B case.
operating
X-47B
UAVs
UAVs
case.
In Iskandarani
In
searches
searches
Iskandarani
Kang
et al.and
Kang
et and
al.
Hedrick
(2014) used (2009),
Hedrick
(2014) used
the kinematic
(2009),
the
Chao
kinematic
et
et al.
Chao model
modelal. of(2012)
UAV and
(2012)
of UAV
dy-
and
dy-
on
on the
doing
theso, aircraft
the UAVs carrierare is high
high interest
required as
as the
to automatic X-47B
thetakeoff
X-47Band case.land-In Iskandaraniand et
theal. (2014) used
doing aircraft
so, the UAVs
UAVscarrierare isrequired
required interest
to automatic
automatic takeoff case.
and land-In namics,
Iskandarani
namics, andand et
theal.
MPC
MPC (2014) used the
determined
determined
kinematic
the the control model
kinematic
the
input toof
control model
input to
UAV
UAVaa dy-
track
tooftrack
track
de-
dy-
de-
doing
ing on so,
thethemoving are
aircraft carrierto deck.(Whittenbury takeoff and
(2011)) land-In sired
namics, the MPC determined the control input a
doing
ing on so,
on the the
the movingUAVs
moving aircraftare required
aircraft carrier to automatic
carrier deck.(Whittenbury takeoff
deck.(Whittenbury (2011)) and land-
(2011)) In namics,
In sired line or
lineand
to
or the
keep formation
MPCformation
to keep
keep determined flight. After
the control
flight.
the
After theinput
the
error
to track
error a de-
dynamics
dynamics de-
ing
comparison with automatic take off and landing on the ground, In was sired line or to formation flight. After error dynamics
ing on the moving
comparison with aircraft carrier
automatic take deck.(Whittenbury
off and landing on (2011))
the ground, siredderived,
was line or to
derived,
tracking
tracking
problem was
keep formation
problem was
transformed
flight. After the error
transformed
into regulating
into dynamics
regulating
comparison
the runway on with automatic
a carrier decktake off
off and
is narrow andlanding
short, and on
on the
theground,
motion problem was derived,fromtracking problem
line. was
Also,transformed into
comparison
the runway
runway on with
on automatic
a carrier
carrier decktake is narrow
narrow andandlanding
short, and and the
theground,
motion was derived,
problem
the desired
fromtracking
the desiredproblem
desired line. was
Also,transformed
to track into regulating
to track a moving
track aa moving
moving
ground
regulating
ground
the
of a carrier
thearunway decka is not deck
fixed is and and
affected short,
by externalthe motion
factors problem
vehicle, from
Kim the
applied line.
nonlinear Also,
model topredictive control ground
frame-
of
of a carrieron
carrier deck
deck
a carrier
is notdeck
is not fixed
fixed
is narrow
and affected
and
and short,
affected and the motion
by external
by external factors problem
factors vehicle, fromapplied
Kim the desired line. model
nonlinear Also, topredictive
track a moving
control ground
frame-
such
of a as ship movement
carrier deck is not and seaand
fixed state. Thus, robust
affected by guidance
external and work
factors vehicle,
vehicle,forKim
a applied
pair of nonlinear
unmanned model
aerial predictive
vehicles. control
This frame-
approach
such as
such
control
as ship
ship movement
law
movement and
are required
and sea
for
sea state.
state. Thus,
safe
Thus, robust
shipboard
robust guidance
landing.
guidance and
The
and work
Marc work forKim
for a
applied
a pair
pair of
nonlinear model
of unmanned
unmanned aerial predictive
aerial vehicles. This
vehicles.
control
This frame-
approach
approach
such
controlas ship
law movement
are required andforsea state.
safe Thus, robust
shipboard landing. guidance
The and is
Marc is
to get
work
to foroptimal
get a pair of
optimal
performance
unmanned compared
performance compared
to using
aerial vehicles.
to using Thisa decoupled
a approach
decoupled
control
L. law
law are
compares sixrequired
control for safe
safe shipboard
approaches for anlanding.
automated The Marc
Thecarrier is to
control
L. compares
compares aresixrequired
control for approaches shipboard anlanding.
for an automated Marc controller
carrier is to get
get optimal
controller
structure:
optimal
structure:
performance
performance compared
heading control
compared
heading control
control
to
to using
using aa decoupled
for standoff-distance
for standoff-distance
standoff-distance
keep-
decoupled
keep-
L.
landing
L. compares design six control
problem. approaches
This paper for
considers automated
limit of carrier
control controller
ing and structure:
speed controlheading
for for
phase-keeping.(Kim et al. keep-
(2013))
landing
landing designsixproblem.
design
control This
problem. approaches
This paper
for an automated
paper considers
considers limit of
limit
carrier controller
of control
control ing and structure:
speed controlheading
for control for standoff-distance
phase-keeping.(Kim et al. keep-
(2013))
authority
landing for carrying
design problem. condition
This paper for ship motion,limit
considers air turbulence
of control At ing and
ing the speed
speed control
andautomatic landing
control for phase-keeping.(Kim
for system, Juliano designed
phase-keeping.(Kim et
et al.
al.the(2013))
MPC
(2013))
authority
authority for
for carrying
carrying condition
condition for
for ship
ship motion, air turbulence At the automatic landing system, Juliano designed the MPC
and ship air
authority for wake.
carrying And, the control
condition for law motion,
ship approaches
motion, air
air turbulence
examined for
turbulence At
At the
the automatic
automatic
automatic landinglanding
which
landing system,
tracks
system, Juliano
flight-path
Juliano designed
angle
designed the
and
the MPC
total
MPC
and
and ship
ship air wake. And, the control law approaches examined for automatic landing which tracks flight-path angle and total
are:
are: ship air
and fuzzy
fuzzy air wake.
logic,
wake.
logic,
twoAnd,
two
neural
And,
neural
the control
thenetwork
network
law
law approaches
control approaches, approaches
approaches,
examined
indirect inver- forward
examined
indirect inver-
for
for automatic
velocity.
automatic
forward velocity.
landing
landing
(de
which
(de Bonfim
which
Bonfim
tracks
Gripp
tracks
Gripp
flight-path
and Sampaioangle
flight-path
and Sampaio
(2014))
angle
(2014))
and
and total
total
are:
are: fuzzy
sion, and alogic,
hybrid two neural
neural network
twoapproaches were approaches,
demonstrated indirect inver-
to compare
inver- forward
sion, fuzzy
and alogic,
hybrid approaches network
were approaches,
demonstrated indirect
to compare forward velocity.
velocity. (de (de Bonfim
Bonfim Gripp
Gripp and and Sampaio
Sampaio (2014))
(2014))
sion, and
and aa hybrid
performance
sion, on a approaches
hybrid 6 Degree ofwere demonstrated
Freedom simulation. to
to compare
(Stein-
performance
performance on aa approaches
on 6 Degree
6 Degree of ofwere demonstrated
Freedom
Freedom simulation.
simulation.
compare
(Stein-
(Stein-
berg and
performance Page on (2001))
a 6 Denison
Degree of researched
Freedom utilized
simulation. concepts
(Stein-
berg
berg and
and Page (2001)) Denison researched utilized concepts
of
of and Page
Dynamic
berg
Dynamic
(2001))
Inversion
Page (2001))in
Inversion in
Denison
order toresearched
Denison
order to researched
develop a
utilized
utilized concepts
develop a controller
controller
to land
concepts
to land
of
of Dynamic
an Unmanned
Dynamic Inversion
Combatin
Inversion order
inAerial to
to develop
order Vehicle develop(UCAV) aa controller to
to land
on an aircraft
controller land
an
an Unmanned
Unmanned Combat
Combat Aerial
Aerial Vehicle (UCAV) on an aircraft
carrier
an with varying
Unmanned Combat wind, windVehicle
Aerial Vehicle (UCAV)
turbulence,
(UCAV) and on seaan
on an aircraft
state con-
aircraft
carrier
carrier with
with varying wind, wind turbulence, and sea state con-
carrier with varying
ditions.(Denison
ditions.(Denison varying wind,
(2007))
(2007))
wind
wind turbulence,
wind,Boskovic
Boskovic
proposedand
turbulence,
proposedandan
ansea
ansea
state
state con-
autonomous
autonomous con-
ditions.(Denison
carrier landing
ditions.(Denison (2007))
system
(2007)) for Boskovic
UAVs,
Boskovic proposed
referred
proposedto as the
an autonomous
Carrier
autonomous Mo-
carrier
carrier landing
landing system
system for
for UAVs,
UAVs, referred
referred to
to as the Carrier Mo-
tion Prediction
carrier
tion
&
landing system
Prediction &
Autonomous Landing
for UAVs,Landing
Autonomous to as the Carrier Mo-
as the
system.(Boškovic
referredsystem.(Boškovic Carrier and
Mo-
and
tion
tion Prediction
Redding (2009))&
Prediction & Y.Autonomous
Zhang studied
Autonomous Landing system.(Boškovic
the a design
Landing effort to improve
system.(Boškovic and
and
Redding
Redding (2009))
(2009)) Y.
Y. Zhang
Zhang studied
studied the a design effort to improve
the integrated
Redding (2009)) flight/thrust
Y. Zhang controlthe
studied the aa design
system design effort
in automatic
effort to
to improve
carrier
improve
the integrated
the integrated flight/thrust
flight/thrust control
control system
system in in automatic
automatic carriercarrier
landing systemflight/thrust
the integrated using LMI-based control H infinity
system synthesis.(Zhang
in automatic carrier
landing
landing system using
system using LMI-based
LMI-based H
H infinity
infinity synthesis.(Zhang
synthesis.(Zhang
et al. (2005)
landing system) Inusing
this paper, to trackHthe
LMI-based desiredsynthesis.(Zhang
infinity glide slope and
et
et al.
al. (2005) ) In this paper, to track the desired glide slope and
et al. (2005)
perform
perform
safe))landing
(2005)
safe
In
In this
this paper,
landing
on theto
paper,
on the
track
track the
tomoving
moving the desired
carrier,
desired
carrier,
glide
Model slope
slope and
glide Predictive
Model Predictive and
perform
Control
perform safe
(MPC)
safe landing
landing on
is proposed
on the
the moving
and switched
moving carrier,
carrier, Model
according
Model Predictive
to flight
Predictive
Control (MPC)
Control (MPC) is is proposed
proposed and and switched
switched according
according to to flight
flight
time
Controlbefore
(MPC)touchis down.
proposed The andModel Predictive
switched Controlto(MPC)
according flight
time
time before
before touch down.
touch down. The Model
The Model Predictive
Predictive Control (MPC)
Control (MPC)
uses
time dynamic
before touch modelsdown.to propagate
The Model thePredictive
state intoControl the future,(MPC) and
uses
uses dynamic
dynamic models
models to propagate the state into the future, and Fig. 1. Overview of automatic carrier landing system
subsequently
uses dynamic uses
models theto
to propagate
estimated
propagate the
future
the state
state
state into
into andthe
the future,
the desired
future, and
and Fig. Fig. 1.
1. Overview
Overview of of automatic
automatic carrier
carrier landing
landing system
system
subsequently
subsequently uses the estimated future state and the desired
subsequently uses uses thethe estimated
estimated future future state
state and and thethe desired
desired Fig. 1. Overview of automatic carrier landing system
Copyright 2015 IFAC
2405-8963 © 2015, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved. 59
Copyright © 2015 IFAC 59
Copyright
Peer review© 2015 responsibility
IFAC 59
Copyright ©under
2015 IFAC of International Federation of Automatic Control. 59
10.1016/j.ifacol.2015.06.464
60 Soyeon Koo et al. / IFAC-PapersOnLine 48-5 (2015) 059–064

Fig. 1 presents an overview of the proposed automatic land- The trim conditions of control inputs are
ing system. Firstly, the UAV is controlled by Linear Quadratic
• δttrim = 0.4deg
Tracker with Integral (LQTI) until it can track the desired glide
slope command. The LQTI controller is designed to reduce the
• δetrim = 9.4deg
distance between the desired glide slope fixed to the moving the
carrier and the position of the UAV regardless of carrier motion.
Then, at 6 seconds before touch down the MPC is activated To predict the UAV position as a function of the longitudinal
in order to consider the heave motion of the carrier. The MPC states, consider two dimensional UAV kinematic model without
propagates the states of UAV and carrier over a finite horizon disturbances as :
of N steps into the future. By using these states, the control dx
action is obtained by solving an optimal control problem. Then, = V cos(θ − α) (4)
dt
first control input is used at the current time step. To make dz
the UAV land on the moving carrier deck, the cost function is = −V sin(θ − α) (5)
dt
designed to minimize the sum of tracking between the UAV The throttle and elevator deflection angle are constrained by the
and the carrier deck positions error history during the horizon. following operational limits :
The error at the final touch down instant is weighted in the cost 
0 ≤ δt ≤ 1 (6)
function. The performance of the proposed MPC with the LQTI
are verified by numerical simulations with longitudinal UAV −20◦ ≤ δe ≤ 20◦ (7)
and Carrier dynamic models. Note that this work main focuses The continuous model can be discretized by Euler integration
on longitudinal motion in a vertical plane, but can be easily into
extended to three dimensional space. XU k+1 = XU k + f (XU k , uUk )Ts (8)
The overall structure of this paper is given as: Section 2 con- where Ts is a sampling time. Thus, the discrete UAV dynamic
tains the UAV and Carrier dynamic models. Section 3 describes model from (1)-(3) given as:
design of the LQTI and MPC to track the desired glide slope.
Section 4 shows the comparative simulation results using two 
Vk+1
 
Vk
  
Vk
 
 
types of controllers for automatic carrier landing of the UAV. αk+1 αk   AU  αk  + BU δt k 
       
  
Lastly, conclusions and future works are given in Section 5. 
 θk+1  
= θk  
+
 θk  δek  Ts
      (9)
 qk+1   qk   qk 
 xk+1   xk   V cos(θk − αk ) 
2. DYNAMIC MODEL zk+1 zk −V sin(θk − αk )

2.1 UAV dynamic model 2.2 Carrier dynamic model

To verify the feasibility of the automatic carrier landing by the This study considers the carrier dynamic model which was
MPC, this paper considers only longitudinal motion. This study derived by SidarSidar and Doolin (1983). They analyzed fre-
considers the linearized UAV as Lee et al. (2014) quency response and power spectral density functions as a
means of representing heave motion characteristics of a carrier.
ẊU = fk (XU , uU ) = AU XU + BU uU (1)
The process transfer function of the carrier heave motion is
Where XU = [V α θ q]T are perturbed states from the trimmed represented as:
state, u is the velocity component in x-axis of UAV, α is angle of
Gs (s) = gs/(s2 + as + d) (10)
attack, θ is pitch angle, and q is pitch rate. And uU = [δth δe ]T
where g = 0.6, a = 0.06, and d = 0.36. Thus, the natural fre-
are throttle and elevator deflection as control inputs, and has
quency of the carrier motion is ω0 = 0.6 rad/s, and damping
operational constraints with respect to trim control input. The
ratio is ξ = 0.05. Then, the transfer function in (10) is equiva-
longitudinal system matrices of the UAV is given by
  lently rearranged as the state-space form with kinematic model
0.04 0.77 −8.992 −13.12 presented by
 −0.007 −1.046 0.016 0.956  ẋ = V (11)
AU =  (2)   C  ship
0 0 0 1 
0.008 −8.838 −0.005 −2.542 żC zC
= Ah + Bh us (12)
  z̈C żC
1.1079 5.366
where xC is position of the carrier in the x-axis, zC is the heave
 0.0396 0.0671 
BU =  (3) motion, żC is the heave motion rate, and us is the scalar random
0 0 
input. However, this study assumes that Vship and us are constant
−0.13 3.096
value. The system matrices of heave motion is given by
When the UAV approaches to the carrier, the flight velocity    
maintains about 51m/s (100kts). Thus, the UAV linear dynamic 0 0 g
Ah = , Bh = (13)
model is derived for the following trim condition −d −a −ga
The discrete-time representation for the ship motion is given by
• Vtrim = 51.44m/s(100knots)
XCk+1 = XCk + fk (XCk , uCk )Ts (14)
• αtrim = 6.8754deg
     
• θtrim = 3.7999deg
xCk+1 xCk  Vship

 zCk+1  = zCk +  z  Ts (15)
Ah Ck + Bh usk
żCk+1 żCk żCk
• qtrim = 0deg/s

Copyright © 2015 IFAC 60


Soyeon Koo et al. / IFAC-PapersOnLine 48-5 (2015) 059–064 61

3. AUTOMATIC CARRIER LANDING SYSTEM where linearized flight path angle γ can be expressed in terms
of longitudinal states α and θ as γ = θ − α. This paper assumes
3.1 Geometry for automatic Carrier Landing Algorithm the flight path angle at the trim condition γtrim follows γR . Thus,
the derivative of d is represented by
As the motion of a carrier deck is affected by external factors d˙ = V sin(θ − α). (18)
such as sea state, an automatic carrier landing algorithm gen-
erates the guidance command using relative position/attitude Now, define the states and control input with (1) and (18) as
between a carrier and an UAV. The relative glide slope geom-
X p = [ue α θ q d]T u = [δt δe ]T (19)
where, X p is the state vector to track the velocity command
and the distance error. Ve is the error velocity between velocity
command and UAV velocity. The plant is described by linear
state-space equations:

X˙p = A p X p + B p u (20)
y p = Cp X p (21)
Through the optimal control theory, the control input minimiz-
ing the performance index is given by
Fig. 2. Relative glide-slope geometry for automatic carrier u(t) = −R−1 (t)BT K(t)x(t) (22)
landing Then, the Riccati equation is given as :
etry between a carrier deck and an UAV is shown in Fig. 2. K̇(t) = −K(t)A p − A p T K(t) − Q(t) + K(t)B p R−1 (t)B p T K(t)
To track the desired glide slope line, firstly the UAV defines (23)
the reference glide slope angle (γR ) from the carrier. Then, where Q(t) and R(t) are a real symmetric positive semi-definite
the UAV calculates the error distance between the desired line state-weighting matrix and a real symmetric positive definite
and the UAV position using the difference angle (γerror ). The control input weighting matrix, respectively. K(t) is control
error vertical distance of the UAV on the desired line can be gain.
represented as The errors of velocity and distance are integrated to improve the
d = Rsin(γR − γ) (16) tracking performance. Considering the integral term, dynamic
where d is the glide path deviation toward desired line, R is the model matrices can be augmented as
  
relative distance between the carrier and UAV. The aim of this X = [X p y p T
y p ] = [α θ q Ve d Ve d]T (24)
study is guide the UAV for safe ship board landing so that d is
driven back to zero.(Stevens and Lewis (2003))    
Ap 0 Bp
AΣ = , BΣ = (25)
Cp 0 0
3.2 Controller 1 : LQ Tracker with integral The final control input can be obtained as

The UAV dynamic model, as mentioned in section 2, is u(t) = −G p x p (t) − Gy e(t) − Gz e(t) (26)
a coupled Multi Input Multi Output(MIMO) system. Linear
Quadratic Regulator(LQR) has been shown to be efficient and where G p , Gy and Gz are state feedback gain, proportional gain
relatively simple than classical control system design to apply and integral gain, respectively.
to the MIMO system.(Jeong et al. (2013)) This study applies
this optimal LQR controller with the integral term, as shown in
Fig. 3, to reduce the steady state error. 3.3 Controller 2 : Model predictive control

The geometry in Fig. 2 shows that the derivative of d is given To tracking the runway on the moving carrier, the MPC al-
by Stevens and Lewis (2003) gorithm is applied to generate control inputs of the UAV. The
d˙ = V sin(γ − γR ) (17) structure of the MPC for automatic carrier landing is illustrated
in Fig. 4. Firstly, the motions of the UAV and carrier are pre-
dicted using the dynamic models in section 2, for finite horizon
of N steps into the future. Then, it is required to solve an optimal
control problem for the horizon at each time step. The optimal
control inputs are computed to minimize the distance error
between the UAV and desired glide slope (r), that is particularly
weighted to final time by prediction of the UAV and carrier
motion.
The prediction horizon N is denoted by
N = Te /Ts p + 1 (27)
where Te is a time size of the horizon, Ts p is the sampling time
Fig. 3. LQTI block diagram during the prediction horizon. By using the discretized the UAV

Copyright © 2015 IFAC 61


62 Soyeon Koo et al. / IFAC-PapersOnLine 48-5 (2015) 059–064

Table 1. Simulation parameters


Parameter Value Unit
Sampling time Ts 0.02 sec
Horizon size Te 2 sec
Sampling time of prediction Ts p 0.2 sec
Desired glide slope angle γR 3 deg
Carrier velocity Vship 15.43 (30) m/s (knots)
Weighting factor pd 10 -

Then, the MPC finds the control input for landing by taking
into account the heave motion. The final time is presented as:
t f = Rcosγ/(VU −Vship ) (32)
where t f is final time to the carrier landing. R is distance
Fig. 4. Structure of MPC algorithm for automatic carrier land- between the UAV and carrier, γ is glide path angle of UAV.
ing system Also, trade-off can be done between tracking performance,
computation load by adjusting sampling frequency and the size
and carrier dynamic models, in (9), (15), the predicted state X of recoding horizon. Throttle command is required to stay over
at the instant k can be written as: 70 % while the UAV approaches to the carrier for possible wave
 +s+1|k) = X(k
X(k  +s|k)+ f (X(k  +s|k), u(k +s|k))Ts p (28) off or bolter. It is assumed that the carrier and UAV move to the
negative x direction. The initial conditions of the simulation are
 + s|k) denotes the prediction state of the instant s (s
where X(k setup as:
= 0,1,2,3,...,N) from the current instant k.
UAV : [ xU0 zU0 ] = [ 1000m 800m ] (33)
Then, in order to find the optimal control inputs during the hori-
zon, consider the cost function that minimizes the difference Carrier : [ xC0 zC0 żC0 ] = [ 0m 5m 2m/s ] (34)
between the predicted glide slope and UAV trajectories as For comparison view, simulations are carried out for both the
N only LQTI controller and the MPC with the LQTI.
Jd = pd dN + ∑ ds (29)
s=0
Fig. 5 depicts absolute trajectories while an UAV approaches
to the carrier. The UAV flies close enough to the reference
where d is the distance error between desired trajectory and the glide slope in using both controllers before the final phase. At
UAV position. pd is constant weighting factors. pd is used for the about -100m (x-axis), the controller calculates the control
more precise tracking of the carrier motion at final touchdown. inputs considering the heave motion of the carrier. Fig. 6 and 7
To prevent excessive control energy and oscillations, the addi- display the detailed error analysis of the carrier landing.
tional cost function is setup given by
N In Fig. 6, the LQTI controller fails to land on the carrier with
Ju = ∑ (δts + δes ) (30) approximately two-meter altitude error from the deck. Whereas
s=0
Where δt and δe are throttle and elevator deflection, respec-
tively. Now, the final cost function is written as the sum of (29)
and (30) 6 Carrier
UAV−LQTI
J = Jd + Ju (31)
4
To solve this optimal problem, this study uses trust region re-
z(m)

flective algorithm(Conn et al. (2000)) and considers the control 2


inputs in as (7) inequality constraints. Note the this optimiza-
tion is repeated at each sampling time, and then determines the 0
control input history during the horizon , u = [ u(k + 1|k), u(k +
2|k), u(k + 3|k), ..., u(k + N|k) ] at instant k, and the first control −400 −395 −390 −385 −380
x(m)
inputs are used to control the UAV at the current time.

6 Carrier
4. NUMERICAL SIMULATIONS UAV−MPC
4
z(m)

To verify the feasibility of the designed automatic carrier land-


ing system, this section presents the results of numerical sim- 2
ulation. The parameters used for the simulation are shown in
Table 1, and the optimization solution is computed numerically 0
in the Matlab environment. −400 −395 −390 −385 −380
To prevent unnecessary control input consumption and com- x(m)
putation burden at the far distance from the carrier, the landing
scenario is divided into two type of controller parts. Before
the final landing phase, the LQTI controller is used to track Fig. 6. Absolute trajectories of the UAV and carrier at the final
the desired glide slope without considering the heave motion. phase with only LQTI(top) and MPC with LQTI(bottom)

Copyright © 2015 IFAC 62


Soyeon Koo et al. / IFAC-PapersOnLine 48-5 (2015) 059–064 63

200
Carrier
UAV−LQTI
100 UAV−MPC
z(m)

−100
−400 −200 0 200 400 600 800 1000
x(m)
Fig. 5. Absolute trajectories of UAV and carrier during automatic carrier landing by LQTI and MPC

x−axis distance error


1000 1
UAV−MPC
0.8
UAV−LQTI

throttle(%)
error(m)

0.6
500
0.4
UAV−LQTI
0.2 UAV−MPC
0 0
0 5 10 15 20 25 0 5 10 15 20 25
time(sec) time(sec)
z−axis distance error
20
40
UAV−MPC
elevator(deg) 10
UAV−LQTI
20
error(m)

0 −10 UAV−LQTI
UAV−MPC
−20
−20 0 5 10 15 20 25
0 5 10 15 20 25
time(sec)
time(sec)
Fig. 7. Horizontal position error between the UAV and car- Fig. 8. Control inputs with LQTI and MPC
rier(top) and vertical distance error from the glide slope This result demonstrates that the designed MPC controller with
attached to the carrier UAV and carrier heave dynamics can contribute on increasing
Table 2. the error of Simulation result
Parameter Only LQTI MPC with LQTI Unit 65
Horizontal error 0 0 m
60
V(m/s)

Vertical error 2.073 0.03 m


55

the MPC accomplishes successful landing on the carrier. Fig. 7 50


0 5 10 15 20 25
and Table. 2 shows the distance error between the UAV and the time(sec)
desired glide slope attached to the carrier, and horizontal and 50
vertical distance errors at the final phase. The x-axis distance
α(deg)

error linearly decreases to zero, and the z-axis distance error 0


converges to 0.03m after about 20 seconds using the MPC. On
the other hand, the result of the LQTI controller is oscillated by −50
0 5 10 15 20 25
following the heave motion in real time and has 2.073m error time(sec)
at the final phase. 20

Fig. 8 shows the control inputs with using only LQTI and 0
θ(deg)

LQTI/MPC. From start to before 6 seconds to landing, throttle −20


has a minimum value since the velocity is increased by gliding,
−40
and the elevator is moving for following the desired glide slope. 0 5 10 15 20 25
time(sec)
At the 6 seconds before landing, the control inputs start to
100
change for considering the heave motion of the carrier. The
LQTI controller finds the control inputs by considering the real
q(deg/s)

0
time heave motion. On the other hand, the MPC determines the
optimal control inputs via the propagated motions of the carrier
−100
and the UAV. And at the final phase the control inputs have 0 5 10 15 20 25
time(sec)
rapid response because in the MPC, the final motion is weighted
to touchdown on the carrier deck. Fig. 9 shows the states of
the UAV during the carrier landing using the MPC with LQTI. Fig. 9. States of UAV during carrier landing

Copyright © 2015 IFAC 63


64 Soyeon Koo et al. / IFAC-PapersOnLine 48-5 (2015) 059–064

the precision of the carrier deck landing. Stevens, B.L. and Lewis, F.L. (2003). Aircraft control and
simulation. John Wiley & Sons.
Whittenbury, J.R. (2011). Configuration design development of
the navy ucas-d x-47b. In AIAA Centennial of Naval Aviation
5. CONCLUSIONS AND FUTURE WORK
Forum” 100 Years of Achievement and Progress”.
Zhang, Y., Yang, Y., and Yu, Y. (2005). Integrated flight
This paper presented the automatic carrier landing system using thrust control via lmi-based hinfty synthesis in automatic
the LQTI and MPC. The benefit of this approach is that it can carrier landing system. In Computational Intelligence for
find the optimal control inputs which reflect the motion of the Modelling, Control and Automation, 2005 and International
UAV and the carrier. The automatic carrier landing algorithm Conference on Intelligent Agents, Web Technologies and
was designed by relative geometry between the UAV and Car- Internet Commerce, International Conference on, volume 1,
rier. In the frame work of MPC, the guidance and control was 1147–1152. IEEE.
designed to track the final the heave motion of the carrier. The
MPC predicted the motion of the UAV and carrier using the
longitudinal dynamic models, and found optimal control input
to track the glide slope and safely land on the carrier deck
at the final time. Simulation results showed that the proposed
carrier landing system improves landing accuracy on the carrier.
In the future, we will investigate the effect of highly coupled
carrier motion with the various sea states and disturbances with
considering the emergency situations.

REFERENCES
Boškovic, J.D. and Redding, J. (2009). An autonomous carrier
landing system for unmannned aerial vehicles.
Chao, Z., Zhou, S.L., Ming, L., and Zhang, W.G. (2012). Uav
formation flight based on nonlinear model predictive control.
Mathematical Problems in Engineering, 2012.
Conn, A.R., Gould, N.I., and Toint, P.L. (2000). Trust region
methods, volume 1. Siam.
de Bonfim Gripp, J.A. and Sampaio, U.P. (2014). Automatic
landing of a uav using model predictive control for the
surveillance of internal autopilot’s control. In Interational
Conference on Unmanned Aircraft Systems.
Denison, N.A. (2007). Automated carrier landing of an
unmanned combat aerial vehicle using dynamic inversion.
Technical report, DTIC Document.
Iskandarani, M., Givigi, S., Fusina, G., and Beaulieu, A. (2014).
Unmanned aerial vehicle formation flying using linear model
predictive control. In Systems Conference (SysCon), 2014 8th
Annual IEEE, 18–23. IEEE.
Jeong, J., Kim, S., and Suk, J. (2013). Optimal tracking control
system design for a ring-wing type uav. In Unmanned
Aircraft Systems (ICUAS), 2013 International Conference
on, 413–422. IEEE.
Kang, Y. and Hedrick, J.K. (2009). Linear tracking for a fixed-
wing uav using nonlinear model predictive control. Control
Systems Technology, IEEE Transactions on, 17(5), 1202–
1210.
Kim, S., Oh, H., and Tsourdos, A. (2013). Nonlinear model
predictive coordinated standoff tracking of a moving ground
vehicle. Journal of Guidance, Control, and Dynamics, 36(2),
557–566.
Lee, D., Koo, S., Ra, C., Kim, K., Kim, S., and Suk, J. (2014). A
study of system identification of an uav for carrier operation.
In KIMST Annual Conference Proceedings.
Sidar, M. and Doolin, B. (1983). On the feasibility of real-
time prediction of aircraft carrier motion at sea. Automatic
Control, IEEE Transactions on, 28(3), 350–356.
Steinberg, M. and Page, A. (2001). A comparison of neural,
fuzzy, evolutionary, and adaptive approaches for carrier
landing. Defense Technical Information Center.

Copyright © 2015 IFAC 64

You might also like