Professional Documents
Culture Documents
SantiaguilloSalinas-2017 Observ Based Time Varying Parrot-Drone
SantiaguilloSalinas-2017 Observ Based Time Varying Parrot-Drone
The International
Proceedings Federation
of the 20th Worldof Congress
Automatic Control
The International
Proceedings Federation
of the 20th9-14,
Worldof Congress
Automatic Control
Toulouse,
The France,
International July
Federation 2017
of Automatic Control
Available online at www.sciencedirect.com
Toulouse, France,
The International July 9-14, 2017
Toulouse, France,Federation of Automatic Control
July 9-14, 2017
Toulouse, France, July 9-14, 2017
ScienceDirect
IFAC PapersOnLine 50-1 (2017) 10305–10310
Observer-based
Observer-based Time-varying Backstepping
Observer-based Time-varying
Time-varying Backstepping
Backstepping
Control
Observer-based
Control for Parrot’s
Time-varying AR.Drone 2.0
Backstepping
Control for Parrot’s AR.Drone 2.0
for Parrot’s AR.Drone 2.0
Control for Parrot’s AR.Drone 2.0∗
J.
J. Santiaguillo-Salinas
∗
Santiaguillo-Salinas ∗ M.A. Rosaldo-Serrano ∗
∗ M.A. Rosaldo-Serrano
J. Santiaguillo-Salinas
E. M.A. Rosaldo-Serrano
Aranda-Bricaire ∗ ∗
J. E.
Santiaguillo-Salinas Aranda-Bricaire
∗
E. Aranda-Bricaire
∗
M.A. Rosaldo-Serrano
∗ ∗
E. Aranda-Bricaire ∗
∗ Electrical Engineering Department, Mechatronics Section,
∗
∗ Electrical
Electrical Engineering
EngineeringMéxico Department,
Department, Mechatronics
Mechatronics Section,
Section,
CINVESTAV,
CINVESTAV, México City,
City, México
México (e-mail:
(e-mail:
∗
Electrical Engineering Department,
City, México (e-mail:Section,
CINVESTAV, Méxicomarcosarosaldos@gmail.com,
jsantiaguillo@cinvestav.mx, Mechatronics
jsantiaguillo@cinvestav.mx,
CINVESTAV,
jsantiaguillo@cinvestav.mx, Méxicomarcosarosaldos@gmail.com,
City, México (e-mail:
marcosarosaldos@gmail.com,
earanda@cinvestav.mx)
earanda@cinvestav.mx)
jsantiaguillo@cinvestav.mx, marcosarosaldos@gmail.com,
earanda@cinvestav.mx)
earanda@cinvestav.mx)
Abstract:
Abstract: This
This paper studies
studies the 3D
the 3D trajectory
trajectory tracking
tracking for for the the quadrotor
quadrotor commercial
Abstract: AR.Drone
prototype This paper paper 2.0 studies
of the the
French 3D company
trajectoryParrot.
tracking The forproposed
the quadrotor control
commercial
commercial
strategy is
prototype
Abstract: AR.Drone
This paper 2.0 studies
of the theFrench 3D company
trajectory Parrot.
tracking Theforproposedthe control commercial
quadrotor strategy
strategy is is
prototypeusing
designed AR.Dronea 2.0 of the
time-varying French
version of company
the Parrot. technique.
backstepping The proposed For control
the implementation
designed
prototype
designed using
AR.Drone
using aa time-varying
2.0assumed
time-varying of theversion
French
version of
of the
company
the backstepping
Parrot.
backstepping technique.
The
technique. proposed For
Forofthe implementation
control
the strategyare
implementation is
of
of the
the control
control law,
law, it
it is
is assumed that
that only
only the
the position
position and
and orientation
orientation of the
the AR.Drone
AR.Drone are
designed
of the
known; using
control
linear and a
law, time-varying
it is
angular assumed version
that
velocities of of
only
thethe thebackstepping
position
AR.Drone are and technique.
orientation
estimated For
using ofthe
the implementation
AR.Drone
suitable Luenberger are
known;
of
known; linear
the control
linear and
law,
and angular
angular velocities
it is assumed that
velocities of the
of only AR.Drone
the position
theallows
AR.Drone are
areand estimated
orientation
estimated using
using ofsuitable Luenberger
the asymptotically
suitable AR.Drone
Luenberger are
observers.
observers. The
The proposed
proposed control
control strategy
strategy allows an
an AR.Drone
AR.Drone 2.0
2.0 to
to converge
converge asymptotically
known;
observers.
to linear
a predetermined and
The proposed angular
flight velocities
control of
strategy
trajectory. the
The AR.Drone
allows are
an AR.Drone
theoretical resultsestimated using
2.0validated
are to converge suitable Luenberger
asymptotically
through real-time
to
to aa predetermined
observers. The proposed
predetermined flight trajectory.
flightcontrol strategy
trajectory. The theoretical
The allows results
results are
an AR.Drone
theoretical 2.0validated
are to converge
validated through real-time
asymptotically
through real-time
experiments.
experiments.
to a predetermined flight trajectory. The theoretical results are validated through real-time
experiments.
© 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
experiments.
Keywords: Time-varying
Time-varying Backstepping, Luenberger Observers, Trajectory Tracking, Parrot’s
Keywords:
Keywords:
AR.Drone Time-varying Backstepping,
2.0. Backstepping, Luenberger
Luenberger Observers,
Observers, Trajectory
Trajectory Tracking, Tracking, Parrot’sParrot’s
AR.Drone
Keywords:
AR.Drone 2.0. 2.0.
Time-varying Backstepping, Luenberger Observers, Trajectory Tracking, Parrot’s
AR.Drone 2.0.
1. INTRODUCTION (2013)
1. INTRODUCTION
1. INTRODUCTION (2013) aa process
process to to identify
identify the the model
model of of the
the AR.Drone
AR.Drone is is
is
(2013) a process
presented; also, atopath
identify the model
following strategyof the is AR.Drone
proposed
1. INTRODUCTION presented;
(2013) a also, atopath
process identifyfollowing
the strategy
model of the proposed to
is AR.Drone to
is
In recent years, unmanned aerial vehicles (UAVs) have presented;
control its also, a path
position. Paper following
Bristeau strategy
et al. is proposed
(2011) shows to
the
In recent years, unmanned aerial vehicles (UAVs) have control
presented; its position.
also, a Paper
path Bristeau
following et al.
strategy (2011)
is shows
proposed the
to
In recent
become years,
very unmanned
popular, since aerial
they can vehicles
perform (UAVs)
autonomous have controlinternal its position. Paper
architecture of the Bristeau
AR.Drone. et al. (2011) the
However, shows the
values
become
In recent very popular,
years, since they
unmanned they
aerial canvehicles
perform(UAVs)
autonomous have internal
control
internal architecture
its position.are
architecture of
Paper the AR.Drone.
the Bristeau
ofundisclosed.
AR.Drone. However,
et al. (2011) the
However, shows
the values
the
values
become
tracking very
of popular,
pre-programmed since paths. can perform
This autonomous
enables them to of the parameters
tracking
become
tracking of pre-programmed
very
of pre-programmed
popular, since paths.
they
paths. can This enables
perform
This them to
autonomous
enables them to of of the
internal
the parameters
architecture
parameters are
are ofundisclosed.
the AR.Drone.
undisclosed. However, the values
carry
carry out exploration,
outof exploration,
exploration, reconnaissance
reconnaissance and
and transport
transport of
of In
tracking
carry out pre-programmed paths.
reconnaissance This enables
and them
transport to
of of
In the
the quadrotor literature
parameters it
are undisclosed. is common to use filters and
objects
objects
carry
that
that
out
represent
represent
exploration,
a
a high
high
reconnaissance
risk
risk to
to crew.
crew.
and
The
The
transport
quadro-
quadro- of In the quadrotor
quadrotor
observers, due to
literature
literature
noise present
it
it is common
is in
common
the
to
to use
use filters
measurements
and
filtersfrom
and
objects
tor that represent
helicopter is a type a ofhigh UAV riskwhich
to crew.has The
a quadro- observers,
particular In the due
quadrotor to noise
literaturepresentit is in the
common measurements
to use filters from
and
tor helicopter
objects is
that represent a type of UAV which has a particular observers,
sensors and due to
because noise in present
some in
cases the itmeasurements
is not possible fromto
tor helicopter
configuration that typea of
is aconsist high
of UAV
four risk to crew.
which
rotors has The
placed a on quadro-
particular
the tips sensors
observers, and due because
to noise in some
present cases
in the it is not
measurements possible fromto
configuration
tor helicopter that
is a consist
type of
of four
UAV rotors
which placed
has a on the
particulartips sensors and
measure all because
the states inofsomethe cases it In
system. is Mokhtari
not possible et to
al.
configuration
of a cross-shaped thatstructure.
consist of four rotors placed on the tips measure sensors all
and the
because states in of the
some system.
cases it In
is Mokhtari
not possibleet al.
to
of a cross-shaped
configuration that structure.
consist of four rotors placed on the tips measure
(2012) all the linearization
feedback states of theand system.
a In observer
linear Mokhtarifor et the
al.
of a cross-shaped structure. (2012)
measure feedback
all the linearization
states of the and aa linear
system. In observer
Mokhtari for
et the
al.
The quadrotor
of a cross-shaped helicopter
structure.has received special attention (2012)
control offeedback linearization
a quadrotor helicopter and linear observer
is presented. In for
Wangthe
The
The quadrotor helicopter
quadrotor helicopter has received
has received special attention control
special attention (2012)
control of
of aa quadrotor
feedback linearization
quadrotor helicopter
helicopterand ais presented.
linear
isobservers observer
presented. In Wang
for
Inused
Wangthe
from research community. The great interest in the use of and
and Shirinzadeh
Shirinzadeh (2015)
(2015) nonlinear
nonlinear observers are
are used to
from
The
from
this
research
quadrotor
research
type of UAV
community.
helicopter
community.
comes
The
has
Thethe
from
great
received
great fact
interest
interest
that
in
special the
in the
they are
use
attention
useeasy
of
of andcontrol
estimate of
Shirinzadeha
the quadrotor(2015)and
velocities helicopter
nonlinear is
uncertainties presented.
observers of a Inused to
Wang
arequadrotor to
this
from
this
to
type
type
control
of UAV
research
of UAV
compared
comes from
community.
comes
to
from
other
the
Theaircrafts
great
the fact
fact that they
interest
that
and
they
forinitstheareuse
are easy
easy
capacity of estimate
and
estimate
helicopter
the
Shirinzadeh
the
when
velocities
(2015)
velocities
its
and
and
position
uncertainties
nonlinear
uncertainties
is known. observers
For
of
oftheaaare
quadrotor
used
quadrotor
case of to
an
to control
thiscontrol
type compared
of UAV to other
comes other
from aircrafts
the fact and
that for its capacity
they capacity
are easy helicopter
estimate
helicopter when
the
when its
velocities
its position
and
position is
is known.
uncertainties
known. For
For ofthe
thea case of
quadrotor
case of an
an
to perform compared
aggressive to maneuvers aircrafts and
Mellinger for
et its
al. (2012), AR.Drone, in Vago-Santana et al. (2014) a Kalman filter
to control
performcompared
aggressivetomaneuvers
maneuvers
other Mellinger
aircrafts and etits
for al.capacity AR.Drone,
(2012), helicopter
AR.Drone, in
in Vago-Santana
when its the
position
Vago-Santana et
is al.
etofknown.
al. (2014)
(2014) Foraa the
Kalman
Kalmancaseon filter
of an
filter
to perform
navigate aggressive
autonomously in structured Mellinger
and et
not al. (2012),
structured is used to estimate states the system, based the
navigate
to perform autonomously
aggressive in structured
maneuvers and
Mellinger not
et structured
al. (2012), is
is used
AR.Drone, to estimate
used to estimatein the
Vago-Santana states
the states etof the
al.
of the system,
(2014) a based
Kalman
system, based on the on the
filter
navigate
environments autonomously
Achtelik etinal.structured
(2009); and notet structured
Blondsch al. (2010), combination of inertial and visual data.
environments Achtelik et
navigate cooperatively
environments autonomously
Achtelik etinal.
al.structured
(2009); Blondsch
(2009); Blondsch
and notet al. (2010), combination
et structured
al. (2010), is used to estimate
combination of
of inertial and
and visual
the states
inertial of the
visual data.
system, based on the
data.
perform
perform cooperatively tasks
tasks for proper
for proper
proper manipulation
manipulation and
and
environments
perform Achtelik
cooperatively et al.
tasks (2009);
for Blondsch et al.
manipulation (2010),
and This
This paper addresses the design and experimental
paper
combination addresses
of inertial theand design
visual and data.experimental valida- valida-
transport
transport of
of objects
objects Michael
Michael et
et al.
al. (2010).
(2010). The
The main
main advan-
advan-
perform
transport cooperatively tasks
of objects helicopters
Michael et al. for proper
(2010). manipulation
The maintake-off and
advan- tion This
tion paper
of a addresses
control the
strategy design
along andwith experimental
state observersvalida-to
tages of quadrotor are hover, vertical This
tion of
paper
of aa control
addresses
control strategy
the
strategy along
design
along andwith
with state
experimental
state observers
observersvalida-to
to
tages
transport
tages of quadrotor
of objects
of quadrotor helicopters
Michael
helicopters et are
al.
are hover,
(2010). vertical
The main
hover, vertical take-off
advan-
take-off achieveachieve 3D trajectory tracking of a quadrotor helicopter,
and landing, so it can be used in crowded environments. tion
achieveof 3D
a
3D trajectory
control
trajectory tracking
strategy
trackingalong of
of a
with
a quadrotor
state
quadrotor helicopter,
observers
helicopter, to
and landing,
tageslanding,
and so
of quadrotor it
so it cancan be
helicoptersused
beofused in
are
in crowded
hover,
crowded environments.
vertical take-off
environments. specifically, for the Parrot’s AR.Drone 2.0. The backstep-
The great disadvantage this type of UAV is its flight specifically,
achieve 3D for the
trajectory Parrot’s
tracking AR.Drone
of a 2.0.
quadrotor The backstep-
helicopter,
The great disadvantage
and landing,
The great disadvantage
so it can ofusedthis in
beperform
of this type
type of UAV
crowded
of UAV is its
environments.
is flight specifically,
its flight ping technique for istheused Parrot’s
for theAR.Drone
design of 2.0.control
the The backstep-
strategy.
time, since they can not long duration flights. ping technique
specifically,
ping technique is used
forAR.Drone
isthe
used for
for the
Parrot’s the design
AR.Drone
design of
of the control
2.0.
the The
control strategy.
backstep-
strategy.
time,
The since
great they can
disadvantage not perform
of this long
type
time, since they can not perform long duration flights. duration
of UAV is flights.
its flight Although the has a variety of sensors, these
Although
ping technique
Although the
thenoisy AR.Drone
is used
AR.Drone has
forerroneous
the a
hasdesign variety
a variety of sensors,
of theofcontrol
sensors, these
strategy.
these
The
time, AR.Drone
since they 2.0
can is
not a
perform commercial
long quadrotor
duration flights. heli- may
may present
present noisy and
and erroneous measurements.
measurements. Because
Because
The
The AR.Drone
AR.Drone 2.0
2.0 is
is a
a commercial
commercial quadrotor
quadrotor heli-
heli- Although
may present the AR.Drone
noisy and has
erroneous a variety of
measurements. sensors, these
Because
copter manufactured by the French company Parrot. The of
of this,
this, Luenberger
Luenberger observers
observers are
are used
used to
to estimate
estimate the
copter
The
copter
AR.Drone
manufactured
AR.Drone
manufactured
has 2.0
been
by
is the
a
by the as
chosen
French
commercial
French company Parrot.
quadrotor
company platform
experimental Parrot. The The
heli-
for may present
of this, Luenberger
longitudinal noisy
and and erroneous
observers
angular velocities measurements.
are of usedthe to estimate the
AR.Drone. Becausethe
AR.Drone has been
copterresearchers
AR.Drone manufactured
has been chosen
chosen
by the as
French
as experimental
company
experimental platform
Parrot.
platform Thefor longitudinal
for of this, Luenberger
longitudinal and
and angular
angular velocities
observers
velocitiesare of used
of the
the toAR.Drone.
estimate the
AR.Drone.
many
many researchers because
because of
of as its low cost
its experimental
low cost and
cost and large
and large
large number
number
AR.Drone
many
of hasboard.
researchers
sensors on been chosen
because
However,of its low
since it possess platform
an number
internalfor The The main
main contribution
longitudinal and angular
contribution of this
this paper
paperofis
of velocities the
is the implementation
the AR.Drone.
implementation
of
many
of sensors on
researchers
sensors on board.
board. However,
because
However, since
ofititsissince
low it
costpossess
and large
itpossible
possess an internal
number
an internal The
of a main contribution
time-varying of
version this
of paper
the is the
backstepping implementation
technique
controller to stabilize it, not to use the of
The
of a time-varying
main contribution
a time-varying version
of of
this
versionLuenberger the
paper backstepping
is
of the backstepping the technique
implementation
technique
controller
of sensors
controller to
on
to stabilize
board.
stabilize it,
However, it is not
since
it, it isquadrotor itpossible
possess
not possible to
an use the
internal
to useit.the combined with suitable observers to achive
dynamical model of a generic to model In combined
of atrajectory with
time-varying
combined suitable
withtracking version
suitable Luenberger
of the
Luenberger observers
backstepping
observers to achive
technique
to achive
dynamical
controller
dynamical model
to
model of
stabilize a generic
it,
of a generic it isquadrotor
not
quadrotor to
possible model
to
to related
model to useit. In
the
it.theIn 3D 3D control for the AR.Drone 2.0. As a
the literature it is possible to find some work trajectory
combined
3D trajectory with tracking
suitable
tracking control for
Luenberger
control for the
the AR.Drone
observers
AR.Drone 2.0.
to
2.0. As
As isa
achive a
the literature
dynamical
the literature it
model
it is possible
of a
is possible to
generic find some
quadrotor
toAR.Drone.
find some In work
workto related
model to
it.
related toetthe the
In first step, a complete dynamical model of the AR.Drone
modeling and control of the Hernandez al. first
3D
first step,
step, aawhich
trajectory complete
tracking
complete dynamical
controlfrom
dynamical model
for the
model of
of the
the AR.Drone
AR.Drone 2.0.ofAs
AR.Drone is
isa
modeling
the and
literature control
it is of
possiblethe to AR.Drone.
find some In Hernandez
work
modeling and control of the AR.Drone. In Hernandez et al. proposed, related toet al.
the proposed, is obtained the combination the
first step, awhich
proposed, complete
which
is obtained
is aa dynamical
obtained
from
from modelthe combination
ofhelicopter
the AR.Drone
the combination
of the
of the is
modeling
This paper and control supported
is financial of the AR.Drone. by CONACyT, In Hernandez
México, throughet al. dynamical
dynamical model
model of
of generic
generic quadrotor
quadrotor helicopter and
and
This
theThis
paper
scholarships
is financial
No. 243226
paper is financial
supported by
and No.by280703
supported
CONACyT, México,
CONACyT, México, through
through proposed, which
dynamical model is of obtained
a generic from quadrotorthe combination
helicopter and of the
the
the scholarships
This No. 243226
paper is financial and No.by280703
supported CONACyT, México, through dynamical model of a generic quadrotor helicopter and the
the scholarships No. 243226 and No. 280703
the scholarships No. 243226 and No. 280703
Copyright © 2017 IFAC 10792
2405-8963 © 2017,
Copyright 2017 IFAC 10792
IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2017 responsibility
IFAC of International Federation of 10792
Automatic Control.
Copyright © 2017 IFAC 10792
10.1016/j.ifacol.2017.08.1497
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 J. Santiaguillo-Salinas et al. / IFAC PapersOnLine 50-1 (2017) 10305–10310
10306
where Ω = W η̇ and W is
− sin θ 0 1
W = cos θ sin φ cos φ 0
cos θ cos φ − sin φ 0
The dynamical model of a generic quadrotor is given by
u
ẍ = (cos ψ sin θ cos φ + sin ψ sin φ) (3a)
m
u
ÿ = (sin ψ sin θ cos φ − cos ψ sin φ) (3b)
m
u
z̈ = (cos θ cos φ) − g (3c)
m
ψ̈ =τ̃ψ (3d)
θ̈ =τ̃θ (3e)
φ̈ =τ̃φ (3f)
where the main thrust u and the angular momenta τ̃ψ ,τ̃θ ,
Fig. 1. quadrotor helicopter scheme. and τ̃φ are the control inputs. To model the internal
controller of the AR.Drone, the following differential equa-
modeling of the internal controller of the AR.Drone. Real- tions are proposed
time experiments are presented to validate the theoretical
results. z̈ = − a1 ż + a3 uz (4a)
10793
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 J. Santiaguillo-Salinas et al. / IFAC PapersOnLine 50-1 (2017) 10305–10310 10307
(5f)) it follows straightforwardly that z and ψ converge variable that depends of time, so does Bξ (t). To begin the
exponentially to zd and ψd , respectively. design process by backstepping, we consider the subsystem
(14a), where eη1 is taken as a virtual input. We propose a
For the design of the motion control strategy for the
virtual control law eη1 = α1 (eξ , t) given by
AR.Drone in the horizontal plane we use the backstepping
technique Kokotovic (1992); Krstic et al. (1995); Khalil 1
eη1 = α1 (eξ , t) = A−1 (ψ) [−kp eξ1 − kd eξ2 ] (15)
(2002). In this case, we use an extension of the classi- gz (t)
cal backstepping technique to deal with a time-varying where kp , kd > 0 are the control gains. The closed-loop
system, which we refer to as time-varying backstepping system (14a)-(15) produces
Malisoff and Mazenc (2009). Define the state for X, Y , θ
and φ dynamics as ėξ = Heξ (16)
T with
ξ1 = [x, y] (9a)
0 1
T H= ⊗ I2
ξ2 = [ẋ, ẏ] (9b) −kp −kd
η1 = [θ, φ]
T
(9c) where ⊗ denotes the Kronecker product. We propose the
T following Lyapunov function
η2 = θ̇, φ̇ (9d) 1
V1 = eTξ P eξ (17)
Taking the first order approximation of (5a)-(5b) the 2
dynamics of these states is where the matrix P has the form
ξ˙1 =ξ2 (10a) a b
P = ⊗ I2 (18)
˙ξ2 =gz (t)A(ψ1 )η1 b c
(10b)
η̇1 =η2 (10c) and satisfies
1 1
η̇2 = − β1 η2 − β2 η1 + β3 uη (10d) V̇1 = eTξ P H + H T P eξ ≤ − eTξ Qeξ (19)
2 2
T
where uη = [uθ , uφ ] , gz (t) = g − a1 z2 + a3 uz ,
where a = 1 + kp−1 kd−1 kp + kd kp−1 , b = kp−1 and
cos ψ1 sin ψ1 ci 0 c = 1 + kp−1 kd−1 . Therefore the closed-loop system (14a)
A(ψ1 ) = and βi = , i = 1, 2, 3
sin ψ1 − cos ψ1 0 bi with the virtual control input (15) is exponentially stable.
are constant matrices corresponding to the parameters For the first step of backstepping, we make a change of
obtained in the modeling of the AR.Drone 2.0 internal variables
controller. Taking uη = β3−1 [β1 η2 + β2 η1 + vη ] system y2 = eη1 − α1 (eξ , t) (20)
(10a)-(10d) is reduced to resulting in
ξ˙1 =ξ2 (11a) ėξ = [Aξ eξ + Bξ (t)α1 (eξ , t)] + Bξ (t)y2 (21a)
ξ˙2 =gz (t)A(ψ1 )η1 (11b) ẏ2 =u1 (21b)
η̇1 =η2 (11c) where eη2 = α̇1 (eξ , t) + u1 . Using
η̇2 =vη (11d) 1
V2 = V1 + y2T y2 (22)
Define now the system errors as 2
eξ1 =ξ1 − ξ1d (12a) as a Lyapunov function candidate, we obtain
eξ2 =ξ2 − ξ˙d (12b) 1
1 V̇2 ≤ − eTξ Qeξ + y2T [BξT (t)P eξ ] + y2T u1 (23)
2
eη1 =η1 − η1d (12c)
Choosing
eη2 =η2 − η̇1d (12d) u1 = −[BξT (t)P eξ ] − k1 y2 (24)
where is the desired position of the AR.Drone in the
ξ1d with k1 > 0 as the control gain, produces
XY plane, η1d is the desired orientation of the AR.Drone 1
2.0, that is, the desired roll and pitch angles. Subsequently V̇2 ≤ − eTξ Qeξ − y2T k1 y2 . (25)
2
it will be shown that η1d is a function of ξ1d . The error
dynamics in X, Y , θ and ψ results Therefore the subsystem (21a)-(21b)-(24) is asymptoti-
cally stable. Substituting u1 , α̇1 (eξ , t) and y2 in the control
ėξ1 =eξ2 (13a) eη2 = α2 (eξ , eη1 , t) we have
ėξ2 =gz (t)A(ψ1 )eη1 (13b)
eη2 =α̇1 − BξT (t)P eξ − k1 (eη1 − α1 (eξ )) (26)
ėη1 =eη2 (13c)
After one step of backstepping, we rewrite the system
ėη2 =vη − η̈1d (13d) (14a)-(14b) as
Rewriting the error dynamics with vη = + wη we haveη̈1d
ẽ˙ ξ = Ãξ (t)ẽξ + B̃ξ eη2 (27)
ėξ =Aξ eξ + Bξ (t)eη1 (14a)
(14b) eξ Aξ Bξ (t) 0
ėη1 =eη2 where ẽξ = , Ãξ (t) = and B̃ξ = .
eη1 0 0 I2
ėη2 =wη (14c)
which can be stabilized by the virtual control law (26). For
e
where eξ = ξ1 , Aξ =
0 I2
and Bξ (t) = gz (t)
0
, the second step of backstepping, we make another change
eξ2 0 0 A(ψ1 ) of variables
with I2 ∈ R2×2 the identity matrix. Since ψ1 is a state y3 = eη2 − α2 (eξ , eη1 , t) (28)
10794
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 J. Santiaguillo-Salinas et al. / IFAC PapersOnLine 50-1 (2017) 10305–10310
10308
resulting in T T
where xξη = [ξ1 , ξ2 , η1 , η2 ] , Bξη = [0, 0, 0, β3 ] ,
ẽ˙ ξ = Ãξ (t)ẽξ + B̃ξ α2 (eξ , eη1 , t) + B̃ξ y3 (29a) 0 I2 0 0
0 0 gz (t)A(ψ1 ) 0 I 0 0 0
ẏ3 =u2 (29b) Aξη (t) = and Cξη = 2 ,
0 0 0 I2 0 0 I2 0
where wη = α̇2 (eξ , eη1 , t) + u2 . Using 0 0 β2 β1
1 It can be verified that the subsystems for X, Y , θ and φ
V3 = V2 + y3T y3 (30)
2 dynamics is observable. We propose the following Luen-
as a Lyapunov function candidate, we obtain berger observer for X, Y , θ and φ dynamics
1
V̇3 ≤ − eTξ Qeξ − y2T k1 y2 + y3T y2 + y3T u2 (31) x̂˙ ξη =Aξη (t)x̂ξη + Bξη uη + Lξη (yξη − ŷξη ) (37a)
2 ŷξη =Cξη x̂ξη (37b)
Choosing T
u2 = −y2 − k2 y3 (32) where x̂ξη = ξˆ1 , ξˆ2 , η̂1 , η̂2 are the observed states and
with k2 > 0 as the control gain, produces Lξη is the matrix gain of the observer.
1 Proposition 2. Consider the subsystem (36a)-(36b) and
V̇3 ≤ − eTξ Qeξ − y2T k1 y2 − y3T k2 y3 (33)
2 the observer (37a)-(37b). Then, there exists Lξη such that
After some algebraic manipulations, we can write wη as the system (37a)-(37b) is an observer for the subsystem
wn = −K1 eξ1 − K2 eξ2 − K3 eη1 − K4 eη2 (34) (36a)-(36b), i.e. limt→∞ (xξη (t) − x̂ξη (t)) = 0.
with Proof. We define the observation error as
K1 =gz (t)−1 A−1 (ψ1 )(kp + k2 bgz2 (t) + k2 k1 kp ) x̃ξη = xξη − x̂ξη (38)
+ gz−1 (t)Ȧ−1 (ψ1 )(k2 kp + bgz2 (t) + k1 kp ) The observation error dynamics are
+ ġz−1 (t)A−1 (ψ1 )(k2 kp + bġz2 (t) + k1 kp ) x̃˙ ξη = (Aξη (t) − Lξη Cξη ) x̃ξη (39)
+ (gz−1 (t)Ä−1 (ψ1 ) + g̈z−1 (t)A−1 (ψ1 ))(kp ) and since the pair (Cξη , Aξη (t)) is observable, we can
choose Lξη such that Aξη (t) − Lξη Cξη is Hurwitz and the
+ ġz−1 (t)Ȧ−1 (ψ1 )(2kp ) observation errors converge asymptotically to zero.
K2 =gz−1 (t)A−1 (ψ1 )(bgz2 (t) + k1 kp + kd + k2 kp Remark 1. It is worth mentioning that we have a local
+ k2 cgz2 (t) + k2 k1 kd ) observer for X, Y , θ and φ dynamics, since it is based on
the first order approximation.
+ gz−1 (t)Ȧ−1 (ψ1 )(k2 kd + cgz2 (t) + k1 kd + 2kp )
Remark 2. A similar procedure can be reproduced for the
+ ġz−1 (t)A−1 (ψ1 )(k2 kd + cġz2 (t) + k1 kd + 2kp ) Z and ψ dynamics. For the sake of conciseness, it is
+ (gz−1 (t)Ä−1 (ψ1 ) + g̈z−1 (t)A−1 (ψ1 ))(kd ) omitted here.
+ ġz−1 (t)Ȧ−1 (ψ1 )(2kd )
5. CONTROL STRATEGY WITH OBSERVER
K3 =(kp + k1 kd + cgz2 (t) + I + k2 kd + k2 k1
+ 2kd Ȧ−1 A−1 + ġz−1 (t)gz (t)) We modify the control law (35) under the separation
K4 =(k2 + k1 + kd ) principle using the estimated longitudinal and rotational
After two steps of backstepping we get the control law uη velocities ξˆ2 and η̂2 , respectively, obtained from the ob-
for motion in the XY plane for the AR.Drone given by server (37a). This produces
uη =β3−1 β1 η2 + β2 η1 + η̈1d + wn (35) uη =β3−1 β1 η̂2 + β2 η1 + η̈1d + ŵη (40)
Proposition 1. Consider the subsystem (5a)-(5b)-(5d)-(5e) with
and the control law (35). Then in the closed-loop system ŵη = − K1 eξ1 − K2 (ξˆ2 − ξ˙1d ) − K3 eη1 − K4 (η̂2 − η̇1d )
(5a)-(5b)-(5d)-(5e)-(35) the AR.Drone converges locally We identify the matrix Hξη (t), from the tracking errors
asymptotically
to the desired trajectory in the plane XY , (12a)-(12d). To do this, we have that the error dynamics
i.e. limt→∞ ξ1 (t) − ξ1d (t) = 0, and to the desired orien- for X, Y , θ and φ in closed-loop with the control law
tation angles, i.e. limt→∞ η1 (t) − η1d (t) = 0. designed by backstepping (35) results
ėξη = Hξη (t)eξη (41)
Proof. The backstepping design procedure ensures expo-
T
nential convergence of the closed-loop system (10a)-(10d)- whit eξη = [eξ1 eξ2 eη1 eη2 ] and
(35), where (10a)-(10d) is a first order approximation
of (5a)-(5e). Therefore, local exponential convergence is 0 I2 0 0
guaranteed for the closed-loop system (5a)-(5e)-(35). 0 0 gz (t)A(ψ1 ) 0
Hξη (t) = .
0 0 0 I
−K1 −K2 −K3 −K4
4. STATE OBSERVERS DESIGN
Proposition 3. Consider the subsystem (36a), the observer
For X, Y , θ and φ dynamics, we rewrite the subsystem (37a) and the control law (40). The matrix Hξη (t) and
(10a)-(10d) in the form Aξη (t) − Lξη Cξη are Hurwitz. Then in the closed-loop
system (36a)-(37a)-(40), the tracking and observation er-
ẋξη =Aξη (t)xξη + Bξη uη (36a) rors converge asymptotically to zero, i.e. limt→∞ (xξη (t) −
yξη =Cξη xξη (36b) xdξη (t)) = 0 and limt→∞ (xξη (t) − x̂ξη (t)) = 0.
10795
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 J. Santiaguillo-Salinas et al. / IFAC PapersOnLine 50-1 (2017) 10305–10310 10309
10796
Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 J. Santiaguillo-Salinas et al. / IFAC PapersOnLine 50-1 (2017) 10305–10310
10310
1.4
1.2
REFERENCES
0.2 1
0 0.5
−0.2 0
−0.4 −0.5
−0.6
−0.8 −1.5
−1
Achtelik, M., Bachrach, A., and He, R. (2009). Stereo
vision and laser odometry for autonomous helicopters
Y (m) X (m)
−0.1
5 10 15 20 25 30 35 40 45 50 55
28.
Time (s)
Position error in Y axys
Bristeau, P.J., Callou, F., Vissiére, D., and Petit, N.
0.08
(2011). The navigation and control technology inside
Error (m)
0.04
the ar.drone micro uav. 18th IFAC World Congress.
García-Carrillo, L.R., Dzul-López, A.E., Lozano, R., and
0
−0.04
5 10 15 20 25
Time (s)
30 35 40 45 50 55
Pégard, C. (2013). Modeling the quad-rotor mini-
0
position error in Z axys
rotorcraft. In Quad Rotorcraft Control, Advances in
−0.05
Industrial Control, chapter 2, 23–34. Springer London.
Error (m)
−0.1
−0.15
Hernandez, A., Copot, C., De keyser, R., Vlas, T., and
Nascu, I. (2013). Identification and path following con-
−0.2
5 10 15 20 25 30 35 40 45 50 55
Time (s)
trol of an ar.drone quadrotor. 2013 17th International
Conference on System Theory, Control and Computing
Fig. 3. AR.Drone position errors (ICSTCC), 583–588.
Khalil, H.K. (2002). Nonlinear Systems. Prentice Hall.
Orientation error in the angle φ
Kokotovic, P.V. (1992). The joy of feedback: nonlinear and
adaptive. IEEE Control Systems Magazine, 12(3), 7–17.
Error (rad)
0
−0.02 trol Engineering. Springer London.
−0.06
5 10 15 20 25 30 35 40 45 50 55
Mellinger, D., Michael, N., and Kumar, V. (2012). Tra-
Time (s)
Orientation error in the angle ψ
jectory generation and control for precise aggressive
0.05
manuevers with quadrotors. International Journal of
Error (rad)
0
Robotics Research.
−0.05
Michael, N., Fink, J., and Kumar, V. (2010). Cooperative
5 10 15 20 25 30
Time (s)
35 40 45 50 55
manipulation and transportation with aerial robots.
Proceedings of Robotics: Science and Systems.
Mokhtari, A., M’Sirdi, N.K., Meghriche, K., and Belaidi,
Fig. 4. AR.Drone orientation errors K. (2012). Feedback linearization and linear observer
for a quadrotor unmanned aerial vehicle. Advanced
use of state observers to estimate the longitudinal and
Robotics, 20(1), 71–91.
rotational velocities of the AR.Drone. As it can be seen
Vago-Santana, L., Santos-Brandão, A., Sarcinelli-Filho,
in the experimental results, the AR.Drone achieves 3D
M., and Carelli, R. (2014). A trajectory tracking and
trajectory tracking with aceptable performance.
3d positioning concontrol for the ar.drone quadrotor.
For this type of vehicle, the error in the trajectory tracking 2014 International Conference on Unmanned Aircraft
can be considered acceptable because the nonlinear effects Systems (ICUAS), 756–767.
due to imbalance and deformations in the propellers, shafts Wang, X. and Shirinzadeh, B. (2015). Nonlinear aug-
and gears of the AR.Drone are neglected. Also, turbulence mented observer design and application to quadrotor
generated by the propellers and other aerodynamics effects aircraft. Nonlinear Dynamics, 80(3), 1463–1481.
were neglected, i.e. ground effect.
It is recalled that the results of this work are local, because
for the design of the control strategy and state observers a
10797