See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/222818469

An integral predictive/nonlinear HâĹđ control

structure for a quadrotor helicopter

Article in Automatica · January 2010

DOI: 10.1016/j.automatica.2009.10.018 · Source: DBLP

CITATIONS READS

188 345

3 authors, including:

G.V. Raffo Francisco R. Rubio

Federal University of Minas Gerais Universidad de Sevilla
36 PUBLICATIONS 385 CITATIONS 123 PUBLICATIONS 1,601 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by G.V. Raffo on 02 March 2015.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
 Automatica 46 (2010) 29–39

Contents lists available at ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

An integral predictive/nonlinear H∞ control structure for a

quadrotor helicopter✩
Guilherme V. Raffo ∗ , Manuel G. Ortega, Francisco R. Rubio
Departamento de Ingeniería de Sistemas y Automática, Universidad de Sevilla, Camino de los Descubrimientos s/n, 41092, Sevilla, Spain

article info abstract

Article history: This paper presents an integral predictive and nonlinear robust control strategy to solve the path following
Received 2 March 2009 problem for a quadrotor helicopter. The dynamic motion equations are obtained by the Lagrange–Euler
Received in revised form formalism. The proposed control structure is a hierarchical scheme consisting of a model predictive
21 September 2009
controller (mpc) to track the reference trajectory together with a nonlinear H∞ controller to stabilize
Accepted 29 September 2009
Available online 13 November 2009
the rotational movements. In both controllers the integral of the position error is considered, allowing
the achievement of a null steady-state error when sustained disturbances are acting on the system.
Keywords:
Simulation results in the presence of aerodynamic disturbances, parametric and structural uncertainties
Nonlinear H∞ control are presented to corroborate the effectiveness and the robustness of the proposed strategy.
Predictive control © 2009 Elsevier Ltd. All rights reserved.
Robust control
Unmanned aerial vehicle

1. Introduction an advanced control strategy is required to achieve good perfor-

mance in autonomous flight or at least to help the piloting of the
The development of unmanned aerial vehicles (uav’s) have gen- vehicle, with high maneuverability and robustness with respect to
erated great interest in the automatic control area in the last few external disturbances.
decades. These kinds of vehicles have been used in tasks such as Concerning this matter, it must be taken into account that these
search and rescue, building exploration, security and inspection. vehicles are underactuated mechanical systems, which complicate
The uav’s are most useful, mainly, when these desired tasks are the control design stage even more. Techniques developed for fully
executed in dangerous and inaccessible environments. actuated robots cannot be directly applied to this class of systems
In the last few years the uav in the quadrotor configuration has (Fantoni & Lozano, 2002). Therefore, nonlinear modelling tech-
been highlighted in a lot of papers. This vehicle is based on vtol niques and modern nonlinear control theory are usually employed
(Vertical Take-Off and Landing) concepts and it is usually used to to achieve autonomous flight with high performance (Castillo et al.,
develop control laws. This kind of helicopter tries to reach a stable 2005b).
hovering and flight using the equilibrium forces produced by four Many efforts have been made to control quadrotor-based he-
rotors (Castillo, Lozano, & Dzul, 2005b). One of the advantages of licopters and some strategies have been developed to solve the
the quadrotor configuration is its load capacity. Moreover, this he- path following problems for this type of system. In Mistler, Benal-
licopter is highly maneuverable, which allows take-off and landing legue, and M’Sirdi (2001), the aerodynamic forces and moments
as well as flight in tough environment. One drawback, is that this acting on this model were considered. The path following prob-
type of uav presents a weight and energy consumption augmenta- lem was solved using exact linearization techniques and noninter-
tion due to the extra motors. acting control via dynamic feedback. In Bouabdallah, Murrieri, and
Nevertheless, these kinds of uav’s have a high nonlinear and Siegwart (2004) the rotor dynamics were considered in the model.
time-varying behavior and they are constantly affected by aerody- The model was split up into two subsystems: the angular rota-
namic disturbances. In addition, uav’s are usually models subject tions and the linear translations. Backstepping and sliding mode
to unmodelled dynamics and parametric uncertainties. Therefore, techniques were used to control the helicopter. In Raffo, Ortega,
and Rubio (2008) a control law based on a standard backstepping
approach for translational movements and a nonlinear H∞ con-
✩ This paper was not presented at any IFAC meeting. This paper was
troller to stabilize the helicopter are combined to perform path
following in the presence of external disturbances and paramet-
recommended for publication in revised form by Associate Editor Warren E. Dixon
under the direction of Editor Andrew R. Teel. ric uncertainties. However, this strategy is only able to reject
∗ Corresponding author. Tel.: +34 95 448 7487; fax: +34 95 448 7340. sustained disturbances applied to the rotational motion. In sev-
E-mail address: [email protected] (G.V. Raffo). eral publications the backstepping technique was used to perform
0005-1098/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.automatica.2009.10.018
30 G.V. Raffo et al. / Automatica 46 (2010) 29–39

both path following and stabilization problems. Backstepping ap- In this paper, an integral predictive and nonlinear robust control
proaches applied to the quadrotor helicopter are found in Madani strategy to solve the path following problem of the quadrotor
and Benallegue (2007), Zemalache, Beji, and Maaref (2007) and helicopter is proposed. The main idea is to combine the advantages
Guenard, Hamel, and Mahony (2008). of the predictive control methodology to follow a predefined
However, most of the control strategies tested on the quadro- trajectory in a smooth way, with the capacity of the nonlinear
tor helicopter do not consider external disturbances on the six de- H∞ theory to cope with unknown disturbances. To carry out these
grees of freedom and parametric uncertainties of the model. In objectives, a state-space predictive controller with integral action
the last few years researchers have begun to consider these ef- based on the time variant error model is used to track the reference
fects at the control law design stage. In Mokhtari, Benallegue, and trajectory, which is an improvement of the controller presented in
Orlov (2006) a feedback linearization-based controller with a slid- Raffo, Gomes, Normey-Rico, Kelber, and Becker (2009). To stabilize
ing mode observer was designed for the quadrotor helicopter. An the helicopter rotational movements a nonlinear H∞ controller is
adaptive observer was added to the control system to estimate the synthesized. Both controllers consider the position error integral
effect of external disturbances. In Bouabdallah and Siegwart (2007) that allows the sustained disturbances rejection.
a backstepping approach using integral action was used to improve The nonlinear H∞ approach used in this paper consists in an
the quadrotor helicopter path following performance when main- adaptation of a previous work, presented in Ortega, Vargas, Vivas,
tained winds disturb whole the system.
and Rubio (2005), formulated via game theory, to control mechan-
In some papers the quadrotor helicopter has also been con-
ical systems considering the tracking error dynamic equation. This
trolled using a linear H∞ controller based on linearized models.
strategy provides, through an analytical solution, a time variant
In Chen and Huzmezan (2003), a simplified nonlinear model of the
control law which is strongly model-dependent and it is similar
uav movements was presented. The path following problem was
to the results obtained with the feedback linearization procedures.
divided into two parts, the first one to achieve the angular rates
The remainder of the paper is organized as follows: in Section 2,
and vertical velocity stabilization by a 2DOF linear H∞ controller
a description of the quadrotor helicopter model is given. The
using the loop shaping technique. The same technique was used to
control the longitudinal and lateral velocities, the yaw angle and control strategy is exposed in Section 3. In Section 4 the nonlinear
the height in the outer loop. In the second part a predictive control H∞ controller for the rotational subsystem is developed. The
was designed to solve the path following problem. In Mokhtari, Be- predictive controller for the translational movements is presented
nallegue, and Daachi (2006) a robust feedback linearization with a in Section 5. Simulation results are presented in Section 6. Finally,
linear H∞ controller was applied to deal with the path following the major conclusions of the paper are drawn in Section 7.
problem with parameter uncertainties and external disturbances.
Furthermore, there are two issues that are worth pointing out. 2. System modelling
On one hand, most of the above control applications assume that
the computed control actions will never reach the saturation limits
of the actuators, although in practice it is possible. For instance, 2.1. Description
when the uav is far away from its destination, the generated
control signals are normally higher than the admissible values. The autonomous aerial vehicle used in this paper is a miniature
Moreover, the vehicles are composed of mechanical and electrical four rotor helicopter. The movement of the uav results from
parts, which are also subject to physical constraints. changes on the lift force caused by adjusting the velocities of the
When on-line constraints must be considered, model predictive rotors. Longitudinal motions are achieved by means of front and
controller (mpc) algorithms appear as an interesting choice. The rear rotors velocity, changing the forces f1 and f3 (see Fig. 1), while
objective of mpc is to compute a future control sequence in a de- lateral displacements are performed through the speed of the right
fined horizon in such a way that the prediction of the plant output and left propellers, which causes a variation on forces f2 and f4 . Yaw
is driven close to the reference. This is accomplished by minimiz- movements are obtained from the difference in the counter-torque
ing a multistage cost function with respect to the future control between each pair of propellers, (f1 , f3 ) and (f2 , f4 ), i.e., accelerating
actions. An analytical solution can be obtained for a quadratic cost the two clockwise turning rotors while decelerating the counter-
function if the model is linear and there are no constraints; other- clockwise turning rotors, and vice versa. Finally, the total thrust,
wise an iterative method of optimization should be used (Cama- which displaces the helicopter in the perpendicular plane with
cho & Bordons, 1998). Moreover, the mpc formulation generates respect to the propellers, is obtained by the sum of the four forces
(implicitly) a nonsmooth control law. Given that trajectories are generated by propellers.
normally known and using an appropriate vehicle instrumentation
This kind of system is a flight vehicle of lightweight structure
to inform about position, orientation and movements (e.g. using
and, therefore, gyroscopic effects resulting from the rotation of
a gps, digital maps, etc.), the predictive controller becomes even
the rigid body and the four propellers should be included in the
more suitable for this task. Apart from the fact that mpc guides the
dynamic model (Bouabdallah, Noth, & Siegwart, 2004). However,
system smoothly, it presents an enhanced autonomy and can easily
in this paper the dynamic model of the system is obtained under
be extended to multivariable systems.
the assumption that the vehicle is a rigid body in the space subject
On the other hand, uav’s are constantly affected by model
to one main force (thrust) and three torques. This simplification
uncertainties and wind gusts, which can easily destabilize the
implies that gyroscopic effects caused by the propellers will be
vehicle. A good choice to reject these disturbances is the nonlinear
considered as disturbances for the rotational control law.
H∞ control theory. The goal of this control theory, presented by
Van der Schaft in Van der Schaft (1992), is to achieve a bounded ra- Besides, this helicopter is an underactuated mechanical system
tio between the energy of the so-called error signals and the energy with six degrees of freedom and only four control inputs. Due to
of the disturbance signals. In general, the nonlinear approach of the complexities presented, some assumptions are made to com-
this theory considers two Hamilton–Jacobi–Bellman–Isaacs partial pute the model for control purposes. The ground effect is neglected
differential equations (HJBI PDEs), which replace the Riccati equa- and the center of mass and the body fixed frame origin are assumed
tions in the case of the linear H∞ control formulation. The main congruent. Moreover, for controller synthesis purposes, the heli-
problem in the nonlinear case is the absence of a general method copter structure is assumed to be symmetric, which results in a
to solve these HJBI PDEs. moment of inertia tensor with just diagonal inertia terms.
 G.V. Raffo et al. / Automatica 46 (2010) 29–39 31

The rotational kinematics can be obtained from the relationship

between the rotation matrix and its derivative with a skew-
φ symmetric matrix (Craig, 1989; Olfati-Saber, 2001) as follows:
η̇ = Wη−1 ω
  " #" #
φ̇ 1 sin φ tan θ cos φ tan θ p
 θ̇  = 0 cos φ − sin φ q (3)
ψ̇ 0 sin φ sec θ cos φ sec θ r

where η = [φ θ ψ]′ , and ω = [p q r ]′ are the angular

velocities in the body fixed frame.
ψ
θ 2.3. Lagrange–Euler equations
ξ
The helicopter motion equations can be expressed by the
Lagrange–Euler formalism based on the kinetic and potential
energy concept:
   
fξ d ∂L ∂L
= − (4)
τη dt ∂ q̇i ∂ qi
L(q, q̇) = EcTrans + EcRot − Ep (5)
where L is the Lagrangian of the helicopter model, EcTrans is the
Fig. 1. Quadrotor helicopter scheme.
translational kinetic energy, EcRot is the rotational kinetic energy,
′
Ep is the total potential energy, q = [ξ η′ ]′ ∈ ℜ6 is the
3
2.2. Helicopter kinematics generalized coordinates vector, τ η ∈ ℜ represents the roll, pitch
and yaw moments, and fξ = RI b f + αT is the translational force
applied to the helicopter due to the main control input U1 in z axis
The helicopter as a rigid body is characterized by a frame linked
direction, with RI bf = RIe U1 ,2 and αT = [Ax Ay Az ]′ are the
to it. Let B = {B1 , B2 , B3 } be the body fixed frame, where the B1 3

axis is in the helicopter normal flight direction, B2 is orthogonal to aerodynamic forces vector, whose components are in the Ex , Ey and
Ez axes, respectively. The aerodynamic forces are considered like
B1 and positive to starboard in the horizontal plane, whereas B3 is
external disturbances for the control design purposes.
oriented in the ascendant sense and orthogonal to the plane B1 OB2 .
Since the Lagrangian does not contain kinetic energy terms
The inertial frame I = {Ex , Ey , Ez } is considered fixed with respect
combining ξ̇ and η̇, the Lagrange–Euler equations can be divided
to the earth (see Fig. 1).
into translational and rotational dynamics (see Fig. 2 — Quadrotor
The vector ξ = [x y z ]′ represents the position of the he-
Helicopter block). The translational movement can be expressed by
licopter mass center expressed in the inertial frame I .1 The ve-
the following equation (Castillo, Lozano, & Dzul, 2005a; Raffo et al.,
hicle orientation is given by a rotation matrix RI : B → I ,
2008):
where RI ∈ SO(3) is an orthonormal rotation matrix (Fantoni &
Lozano, 2002). The rotation matrix can be obtained through three mξ̈ + mge3 = fξ . (6)
successive rotations around the axes of the body fixed frame. In
this paper, the XYZ fixed Euler angles have been used to describe Eq. (6) can be expressed by means of state vector ξ , yielding:
the helicopter rotation with respect to the ground. These angles  1 Ax
are bounded as follows: roll angle, φ , by (−π /2 < φ < π /2);

 ẍ = (cos ψ sin θ cos φ + sin ψ sin φ) U1 +

 m m
pitch angle, θ , by (−π /2 < θ < π /2); and yaw angle, ψ , by 
1 Ay
(−π < ψ < π ). ÿ = (sin ψ sin θ cos φ − cos ψ sin φ) U1 + (7)

 m m
From these three rotations, the following rotation matrix from 

z̈ = −g + 1 (cos θ cos φ) U + Az
B to I is obtained: 1
m m
" # where m is the helicopter mass and g is the gravitational
CψCθ C ψ Sθ Sφ − Sψ C φ C ψ Sθ C φ + Sψ Sφ
RI = Sψ C θ Sψ Sθ Sφ + C ψ C φ Sψ Sθ C φ − C ψ Sφ (1) acceleration.
−S θ C θ Sφ CθCφ The rotational kinetic energy equation must be rewritten to
represent the equations of the rotational motion as a function of
where C · = cos(·) and S · = sin(·). the generalized coordinate η. Let Wη name the Jacobian from ω to
The kinematic equations of the rotational and translational η̇ in (3), the following matrix is defined:
movements are obtained by means of the rotation matrix. The
translational kinematic can be written as: J = J (η) = Wη′ JWη (8)
where J is the diagonal moment of inertia tensor. Then the
vI = RI · vB (2)
rotational kinetic energy equation can be expressed as follows:
where vI = [u0 v0 w0 ]′ and vB = [uL vL wL ]′ are linear ve- 1 ′
locities of the mass center expressed in the inertial frame and body EcRot = η̇ J η̇. (9)
2
fixed frame, respectively.

2 The notation e represents the vector e = [0 0 1]′ . Thus, the therm R

3 3 Ie3
1 The notation prime ′ denotes transpose. denotes the third column of the rotation matrix.
32 G.V. Raffo et al. / Automatica 46 (2010) 29–39

Fig. 2. Quadrotor helicopter control structure.

From the above expression, the Lagrange–Euler rotational 3. Control strategy

equations in terms of η can be written, in general form, as follows
(Castillo et al., 2005a): To achieve a robust path following for the quadrotor helicopter,
two techniques, capable of controlling the helicopter in presence
M (η)η̈ + C (η, η̇)η̇ = τ η (10)
of sustained external disturbances, parametric uncertainties and
where M (η) = J (η), that is: unmodelled dynamics, are combined. The proposed control
strategy is based on the decentralized structure of the quadrotor
M (η)
  helicopter system (4), which is composed of the dynamic Eqs. (6)
Ixx 0 −Ixx
Sθ and (10). The overall scheme of the control strategy is depicted in
= 0 2
Iyy C φ + Izz S φ


2
Iyy − Izz C φ S φ C θ 
Fig. 2.
−Ixx S θ Iyy − Izz C φ S φ C θ Ixx S 2 θ + Iyy S 2 φ C 2 θ + Izz C 2 φ C 2 θ Firstly, the reference trajectory for the translational movements
(11) is provided off-line by the Trajectory Generator block. The com-
putation of this trajectory is based on a virtual reference vehicle
and
" # whose model is the same as the one of the quadrotor helicopter
c11 c12 c13 for the translational motion. Thereby, starting from a desired route
C (η, η̇) = c21 c22 c23 for the translational movements, xr , yr , and zr , and their deriva-
c31 c32 c33 tives, the reference control inputs U1r , uxr and uyr , are computed.
where The yaw reference angle is defined separately. This trajectory is
generated under the following assumptions: there are no external
c11 = 0 disturbances acting on the virtual vehicle; and the attitude of the




c12 = Iyy − Izz θ̇ C φ S φ + ψ̇ S 2 φ C θ + Izz − Iyy ψ̇ C 2 φ C θ virtual vehicle is supposed to be stabilized.
A predictive controller is proposed to control the quadrotor
− Ixx ψ̇ C θ

translational movements in an outer loop, using the references
c13 = Izz − Iyy ψ̇ C φ S φ C 2 θ provided by the trajectory generator. The state-space predictive




c21 = Izz − Iyy θ̇ C φ S φ + ψ̇ S 2 φ C θ + Iyy − Izz ψ̇ C 2 φ C θ controller based on the error model (e-sspc) also includes the
integral of the position error in the state vector in order to
+ Ixx ψ̇ C θ achieve null steady-state error when sustained disturbances are


c22 = Izz − Iyy φ̇ C φ S φ considered.
The translational motion control is performed in two stages.
c23 = −Ixx ψ̇ S θ C θ + Iyy ψ̇ S 2 φ C θ S θ + Izz ψ̇ C 2 φ S θ C θ

In the first one, the helicopter height, z, is controlled and the
c31 = Iyy − Izz ψ̇ C 2 θ S φ C φ − Ixx θ̇ C θ total thrust, U1 , is the manipulated signal. In the second stage,



the reference of pitch and roll angles (θr and φr , respectively) are
c32 = Izz − Iyy θ̇ C φ S φ S θ + φ̇ S 2 φ C θ + Iyy − Izz φ̇ C 2 φ C θ generated through the two virtual inputs, computed to follow the
+ Ixx ψ̇ S θ C θ − Iyy ψ̇ S 2 φ S θ C θ − Izz ψ̇ C 2 φ S θ C θ desired xy movement. In this second step, the control variable U1

is used as a time-varying parameter.
c33 = Iyy − Izz φ̇ C φ S φ C 2 θ − Iyy θ̇ S 2 φ C θ S θ − Izz θ̇ C 2 φ C θ S θ
Finally, a nonlinear H∞ controller for the rotational subsys-
+ Ixx θ̇ C θ S θ. tem is used in an inner loop to perform the quadrotor helicopter
stabilization. The angular position and velocity are controlled in
Therefore, the mathematical model (used for the controller
this loop, being the torques applied on the three axis, τ ηa =
synthesis) that describes the helicopter rotational movement
[τφa τθa τψa ]′ (see Eq. (19)), the manipulated variables. To obtain
obtained from the Lagrange–Euler formalism is given by:
null steady-state error in presence of sustained external distur-


η̈ = M (η)−1 τ η − C (η, η̇)η̇ . (12) bances, the integral of the angular position error is also consid-
ered. Due to the cascade structure of this strategy and taking into
 G.V. Raffo et al. / Automatica 46 (2010) 29–39 33

account the closed-loop performance achieved by the inner nonlin- 4.2. Rotational subsystem nonlinear H∞ control
ear H∞ controller loop, the Euler angles can be considered as time-
varying parameters on the design of the translational controller. The rotational movements dynamic model (10), obtained from
The development of these controllers is analyzed in the the Lagrange–Euler formalism, is used in order to develop the
following sections. nonlinear H∞ controller. τ η joins the control torques and external
disturbances, and is redefined as:
4. Nonlinear H∞ controller for stabilization
τ η = τ ηa + τ ηd (19)
In this section, a nonlinear H∞ controller for the rotational sub- where τ ηa is the applied torques vector and τ ηd represents the total
system is developed to achieve robustness in the presence of sus- effect of system modelling errors and external disturbances.
tained disturbances, and parametric and structural uncertainties. As a first step to synthesize the control law, the tracking error
vector is defined as follows:
4.1. Nonlinear H∞ control approach    
η̃˙ η̇ − η̇r
The dynamic equation of a nth order smooth nonlinear system  η̃  Z η − ηr 
xη = Z =
 (20)
which is affected by an unknown disturbance can be expressed as
follows: η̃dt η − ηr dt
ẋ = f (x, t ) + g (x, t )u + k(x, t )d , (13)
where ηr and η̇r ∈ ℜn are the desired trajectory and the corre-
p q
where u ∈ ℜ is the vector of control inputs, d ∈ ℜ is the sponding velocity, respectively. Note that an integral term has been
vector of external disturbances and x ∈ ℜn is the vector of states. included in the error vector. This term will allow the achievement
Performance can be defined using the cost variable ζ ∈ ℜ(m+p) of a null steady-state error when persistent disturbances are acting
given by the expression: on the system (Ortega et al., 2005).
 
h(x) The following control law is proposed for the rotational
ζ=W , (14) subsystem (Ortega et al., 2005):
u
where h(x) ∈ ℜm represents a function of the vector of states τ ηa = M (η)η̈ + C (η, η̇)η̇
to be controlled and W ∈ ℜ(m+p)×(m+p) is a weighting matrix. If

− T1−1 M (η)T ẋη + C (η, η̇)Txη + T1−1 u. (21)
the states x are assumed to be available for measurement, then
the optimal H∞ problem can be posed as follows (Van der Schaft, The proposed control law can be split up into three different
1992): parts: the first one consists of the first two terms of that equation,
Find the smallest value γ ∗ ≥ 0 such that for any γ ≥ γ ∗ there which are designed in order to compensate the system dynamics
exists a state feedback u = u(x, t ), such that the L2 gain from d to ζ (10). The second part consists of two terms including the error
is less than or equal to γ , that is: vector xη and its derivative, ẋη . Assuming τ ηd ≡ 0, these two terms
Z T Z T of the control law enable perfect tracking, which means that they
kζk22 dt ≤ γ 2 kd k22 dt . (15) represent the essential control effort needed to perform the task.
0 0 Finally, the third part includes a vector u, which represents the
The internal term of the integral expression on the left-hand additional control effort needed for disturbance rejection.
side of inequality (15) can be written as: It can also be pointed out that, despite it seeming that the
 
  h(x) preceding control law might not seem a well posed system, it will
kζk22 = ζ ′ ζ = h′ (x) u′ W ′ W be shown afterwards that the computed torque does not rely on
u
joint accelerations, but on their references.
and the symmetric positive definite matrix W ′ W can be parti- Matrix T in (21) can be partitioned as follows:
tioned as follows:  
  T = T1 T2 T3
Q S
W ′W = . (16)
S′ R with T1 = ρ I , where ρ is a positive scalar and I is the nth order
identity matrix.
Matrices Q and R are symmetric positive definite and the fact
Substituting the expression of the control law from (21) into
that W ′ W > O guarantees that Q − SR −1 S ′ > O, where O is the
the Lagrange–Euler equation of the system (10) and defining d =
nth order zero matrix.
Under these assumptions, an optimal control signal u∗ (x, t ) M (η)T1 M −1 (η)τ ηd , the following expression is obtained:
may be computed for system (13) if there is a smooth solution
M (η)T ẋη + C (η, η̇)Txη = u + d . (22)
V (x, t ), with V (x0 , t ) ≡ 0 for t ≥ 0, to the following HJBI equation
(Van der Schaft, 2000): The above expression represents the dynamic equation of the
∂V ∂ ′V system error. Taking into account this nonlinear equation, the
+ f (x, t ) nonlinear H∞ control problem can be posed as follows:
∂t ∂x   ‘‘Find a control law u(t ) such that the ratio between the energy of
1 ∂ ′V 1 ∂V  ′
+ k ( x, t )k′
(x , t ) − g (x, t )R −1 ′
g (x, t ) the cost variable ζ = W h′ (xη )u′ and the energy of the disturbance
2 ∂x γ 2 ∂x signals d is less than a given attenuation level γ ’’.
∂V′
1
Taking into account the definition of the vector error, xη , and
− g (x, t )R −1 S ′ h(x) + h′ (x) Q − SR −1 S ′ h(x) = 0 (17) the definition of the cost variable, ζ , the following structures are
∂x 2
√ considered for matrices Q and S in (16):
for each γ > σmax (R ) ≥ 0, where σmax stands for the maximum " # " #
singular value. In such a case, the optimal state feedback control Q1 Q12 Q13 S1
law is derived as follows (Feng & Postlethwaite, 1994): Q = Q12 Q2 Q23 , S = S2 .
  Q13 Q23 Q3 S3
∗ −1 ′ ∂ V (x, t )
′
u = −R S h(x) + g (x, t ) . (18) To apply the theoretical results presented in Section 4.1, it is
∂x
necessary to rewrite the nonlinear dynamic equation of the error
34 G.V. Raffo et al. / Automatica 46 (2010) 29–39

(22) into the standard form of the nonlinear H∞ problem (see A particular case can be obtained when the components of
(13)). This can be done by defining the following expressions: weighting compound W ′ W verify:
ẋη = f (xη , t ) + g (xη , t )u + k(xη , t )d , (23) Q1 = ω1 2 I , Q2 = ω2 2 I , Q3 = ω3 2 I , R = ωu 2 I , (29)


f xη , t
 
−M (η)−1 C (η, η̇) O O Q12 = Q13 = Q23 = O, S1 = S2 = S3 = O.
−1
= T0  T −1 I − T1−1 T2 −I + T1 (T2 − T3 ) T0 xη ,
−1
1
O I −I In this case, the following analytical expressions for the gain
  matrices have been obtained:
M (η)−1


−1  p  
g xη , t = k xη , t = T0 O  ω2 2 + 2ω1 ω3 1
O KD = I + M (η)−1 C (η, η̇) + I ,
ω1 ωu 2
where p  
ω3 ω2 2 + 2ω1 ω3 −1 1
" # KP = I+ M (η) C (η, η̇) + I ,
T1 T2 T3 ω1 ω1 ωu 2
T0 = O I I . (24)  
ω3 1
O O I KI = M (η)−1 C (η, η̇) + I
ω1 ωu 2
As stated in Section 4.1, the solution of the HJBI equation de-
pends on the choice of the cost variable, ζ , and particularly on the where the parameters ω1 , ω2 , ω3 and ωu can be tuned by a
selection of function h(xη ) (see (14)). In this paper, this function is systematic procedure keeping in mind a linear PID control action
taken to be equal to the error vector. That is, h(xη ) = xη . Once this interpretation.
function has been selected, computing the control law u will re- These expressions have an important property: they do not
quire finding the Lyapunov function, V (xη , t ), to the HJBI equation depend on the parameter γ . So, we obtain an algebraic expression
posed in the previous section (see (17)). for computing the general optimal solution for this particular case.
The following theorem will help do this.
5. E-SSPC for path following
Theorem. Let V (xη , t ) be the scalar function:
" # In this section a control law to solve the path following problem
1
M (η) O O
V xη , t = x′η T0′ O Y X −Y T0 xη , (25) by translational movements is designed. A linear state-space mpc
2 O X −Y Z +Y strategy based on the error model (e-sspc) is performed. From the
error model, two predictive controllers are synthesized. The first
where X , Y and Z ∈ ℜn×n are constant, symmetric, and positive one controls the height through the input U1 , whereas the second
definite matrices such that Z − XY −1 X +2X > O, and T0 is as defined one makes use of this signal as a time variant parameter in the
in (24). Let T be the matrix appearing in (22). If these matrices verify linear x and y motions to compute two virtual inputs, ūx and ūy .
the following equation: ˙
" # The system
(7) can be rewritten in a state-space form as ξ̄(t ) =
O Y X 1 f ξ̄(t ), uξ (t ) for the controller design, where ξ̄(t ) = [x(t ) u0 (t )
Y 2X Z + 2X +Q + T ′T y(t ) v0 (t ) z (t ) w0 (t )]′ stands for the state-space vector of the
X Z + 2X O γ2
system, where u0 (t ), v0 (t ), and w0 (t ) are the components of the

′
linear velocity of the vehicle mass center expressed in the inertial
− S ′ + T R −1 S ′ + T = O (26)
frame (see Eq. (2)).
then, function V (xη , t ) constitutes a solution to the HJBI, for a From (7) and the new state-space vector, the system dynamic
sufficiently high value of γ . equation to control design can be written in the following form:

The proof of this theorem is obtained following the steps ˙ t ) = f ξ̄(t ), u (t )

ξ̄( ξ
presented in Ortega etal. (2005).  ⋄
Once matrix T = T1 T2 T3 is computed by solving some
 u0 (t ) 

U1 (t ) Ax (t )
Riccati algebraic equations, substituting V xη , t in (18), control  ux ( t ) + 
 
law u∗ corresponding to the H∞ optimal index γ is given by  m m 
 v0 (t ) 

 
u∗ = −R −1 S ′ + T xη . (27) = U1 (t ) Ay (t )  (30)
 uy ( t ) + 
 m m 
Finally, if the control law (27) is replaced into (21), and after  w0 (t ) 
 
some manipulations, the optimal control law can be written as: U1 (t ) Az (t )
−g + (cos θ(t ) cos φ(t )) +
τ ∗ηa = M (η)η̈r + C (η, η̇)η̇ m m
 Z  with:
˙
− M (η) KD η̃ + KP η̃ − KI η̃dt (28)
ux (t ) , cos ψ(t ) sin θ(t ) cos φ(t ) + sin ψ(t ) sin φ(t )
(31)
where uy (t ) , sin ψ(t ) sin θ(t ) cos φ(t ) − cos ψ(t ) sin φ(t ).



KD = T1−1 T2 + M (η)−1 C (η, η̇)T1 + M (η)−1 R −1 S1′ + T1 Eqs. (7) show that the movement through the x and y axes
−1


depends on the control input U1 . In fact, U1 is the designed total
KP = T1 T3 + M (η)−1 C (η, η̇)T2 + M (η)−1 R −1 S2′ + T2 thrust magnitude to obtain the desired linear movement, while ux
−1



KI = −T1 M (η)−1 C (η, η̇)T3 + M (η)−1 R −1 S3′ + T3 . and uy can be considered as the directions of U1 that cause the
movement through the x and y axes, respectively.
 G.V. Raffo et al. / Automatica 46 (2010) 29–39 35

1 ∆t 0 0 0 0

The objective of this approach is to guarantee that the uav
follows a previously defined reference trajectory minimizing the 0 1 0 0 0 0
displacement error. However, due to the fact that the target
∆ t 0 1 0 0 0
Axy = 
0
,
coordinates vary in time, a virtual reference vehicle with the same 0 0 1 ∆t 0
 
quadrotor helicopter mathematical model is defined: 0 0 0 0 1 0
0 0 0 ∆t 0 1

 
ξ̄˙ r (t ) = f ξ̄ r (t ), uξr (t ) (32) 0 0 (37)
 ∆t U1 (k) 0
m 
where ξ̄ r (t ) = [xr (t ) u0r (t ) yr (t ) v0r (t ) zr (t ) w0r (t )]′ and  
 0 0 
uξr (t ) = [uxr uyr U1r ]′ are the reference states and the control Bxy =  
 0 0 
inputs, respectively. Null external disturbances are assumed in  ∆t 
the virtual reference vehicle. This virtual reference vehicle is
 0 U1 (k)
m
used to obtain the reference control inputs for translational 0 0
movements under the assumption that the helicopter height has
where ∆t is the sampling time, which has been chosen sufficiently
been stabilized. Therefore, for the case of this vehicle, the reference
small to capture all translational motion error dynamic and high
values are given by:
enough to consider the rotational closed-loop dynamics in steady
x¨r · m y¨r · m state.
U1r = m · (z̈r + g ), uxr = , uyr = . Based on this analysis, the path following problem for a uav can
U1r U1r
be understood as: find the control inputs in a bounded group of
possible values that drive the state variables in (35) from an initial
By subtracting the virtual reference vehicle (32) from the sys-
position xξ0 to the origin (Sun, 2005), i.e. limt →∞ xξ = 0.
tem (30), the proposed translational error model is given by:
Therefore, from the height and longitudinal–lateral error mod-
els the control laws can be designed in such a way that the system is
ξ̄(t ) = A(t ) · ē
ė ξ(t ) + B(t ) · e
uξ (t ) (33) forced to track the reference trajectory. The first law computes the
control input U1 in such a way that the following cost is minimized:
ē
where ξ(t ) = ξ̄(t ) − ξ̄ r (t ) represents the error vector, and e
uξ (t ) =  ′   h i′ h i
Jz = b
xξz −b
xξrz Qz b xξrz + b
xξz −b uξ z − b
e e
uξrz Rz b
uξ z − b
e e
uξrz
uξ (t ) − uξr (t ) is the control input error. Matrices A(t ) and B(t ) are
the Jacobians of the system (30) in relation to ξ̄(t ) and uξ (t ), re-


+Ω b
xξz (k + N2z |k) −b
xξrz (k + N2z |k) (38)
spectively. Besides, the integral of the position error term has been
included in the error vector to perform an appropriate path follow- where Qz and Rz are diagonal definite positive weighting matrices,
ing in presence of sustained disturbances. Therefore, the following N2z is the prediction horizon (Rossiter, 2003), and Ω is the terminal
augmented error vector is considered: state cost defined by:


Ω b
xξz (k + N2z |k) −b xξrz (k + N2z |k)
 x̃(t )   x(t ) − xr (t )   ′
= b xξz (k + N2z |k) −b xξrz (k + N2z |k)
Z 0 (t ) 
ũ Z u0 (t ) − u0r (t )
  
    × Pz b xξz (k + N2z |k) −b xξrz (k + N2z |k)
 x̃(t )dt   (x(t ) − xr (t ))dt 
   
    with Pz ≥ 0 (Kühne, Lages, & Gomes Da Silva, 2005).
 ỹ(t )   y(t ) − yr (t ) 
    The predictions of the model output b xξz (k + j|k) are computed
 ṽ0 (t )  Z v0 (t ) − v0r (t ) 
xξ (t ) = Z = . (34) using a linearized time-varying state-space model of the vehicle by
 ỹ(t )dt   (y(t ) − y (t ))dt 
   r  the Eqs. (35) and (36), obtaining:
   
 z̃ (t )   z (t ) − zr (t ) 
    xξz = Pz (k|k) · xξz (k|k) + Hz (k|k) · b
b e
uξ z , (39)
Z w̃0 (t )   Z w0 (t ) − w0r (t ) 
   
where euξz (k|k) = U1 (k) − U1r (k), and xξz (k) is the height state
z̃ (t )dt (z (t ) − zr (t ))dt
error vector. The height reference vectors are:
 
Using Euler’s method, a time-varying discrete linear model is xξrz (k + 1|k) − xξrz (k|k)
 
..
obtained, i.e.: b
xξrz ,  ,
.
xξrz (k + N2z − 1|k) − xξrz (k|k)
xξ (k + 1) = A · xξ (k) + B(k) · e
uξ (k). (35)  
U1r (k|k) − U1r (k − 1|k)
b  .. 
The input control U1 (t ) is considered as a time-varying e
uξrz , . 
parameter for the reference x and y motions. Moreover, because of U1r (k + Nuz − 1|k) − U1r (k − 1|k)
the decentralized control structure, the roll, pitch and yaw angles
are also considered as time-varying parameters. where Nuz is the control horizon.
The error model (35) can be split up into two subsystems: the Minimizing the Eq. (38) when the constraints are not consid-
ered, the control law can be obtained as:
height error and the x and y motions error. Matrices A and B for
each subsystem are the following:  −1 h ′
i
b
uξz = Hz′ Qz Hz + Rz
e xξrz − Pz xξz (k) + Rzb
· Hz Qz b e
uξrz , (40)
 
" # 0
1 ∆t 0
 ∆t  although only beuξz (k|k) is needed at each instant k (Camacho &
Az = 0 1 0 , Bz =  cos θ(k) cos φ(k) (36) Bordons, 1998). Therefore, the following control signal is applied
m
∆t 0 1
0 to the helicopter: U1 (k) = be
uξz (k|k) + U1r (k).
36 G.V. Raffo et al. / Automatica 46 (2010) 29–39

Fig. 3. Path following.

Fig. 4. Position (x, y, z ).
The second control law computes the x and y motion control
inputs. If the same previous procedure is carried out using the error with a more accurate model, which emulates a real quadrotor he-
model (35) and (37), the following control signal is obtained: licopter, using the crossed inertia terms in the moment of iner-
 ′ −1 tia tensor and saturated control inputs. Moreover, the simulations
b
e
uξxy = Hxy Qxy Hxy + Rxy
have been executed considering external disturbances on the six
h
i
degrees of freedom, and structural and parametric uncertainties.
× Hxy ′
Qxy bxξrxy − Pxy xξxy (k) + Rxyb
e
uξrxy , (41)
Taking into account that the simplified model derived in Section 2
is used just for control synthesis purposes, structural uncertainties
where b
uξxy (k|k) = [ũx (k|k) ũy (k|k)]′ , and
e are present because that model considers a moment of inertia ten-
sor with only diagonal inertia terms.
uξxy (k) = uξrxy (k) + b
e
uξxy (k|k). (42)
The initial conditions of the helicopter are ξ 0 = [0 0.5 0.5]′ m
The reference vectors of the error states, b
xξrxy , and the error and η0 = [0 0 0.5]′ rad. The values of the model parameters
control inputs, b e
uξrxy , are obtained by the same way as the one of used for simulations are the following: m = 0.74 kg, l = 0.21 m,
the height controller case. g = 9.81 m/s2 and Ixx = Iyy = 0.004 kgm2 , Izz = 0.0084 kgm2 . An
Taking into account that the required virtual inputs, uξxy (k), to uncertainty of ±30% in the elements of the mass and inertia matrix
follow the path reference in the xy plane have been obtained, the has been considered in the simulations.
necessary values of φ and θ could be computed by (31). At this Assuming that the quadrotor helicopter needs, under ideal
point, it should be noted that Eq. (31) constitutes a definition of conditions, a thrust value of about U1 ≈ 7.23 N to perform
the system to be controlled. Therefore, these values cannot be set hovering flight, the following persistent light gusts of wind are
directly since these angles are two of the outputs of the rotational considered as external disturbances on the aerodynamic forces and
subsystems; being the nonlinear H∞ inner loop in charge of moments: Ax = 1 N at t = 5 s; Ap = 1 Nm at t = 10 s; and Ay = 1 N
carrying out this task. at t = 15 s; Aq = 1 Nm at t = 20 s; Az = 1 Nm at t = 25 s; and
In consequence, a desired virtual directions vector, ūξxy (k) = Ar = 1 Nm at t = 30 s.
[ūx ūy ]′ , must be defined in the same sense of Eq. (31) as follows: The e-sspc parameters were adjusted as follows:
N2z = Nuz = 3Inz , Qz = diag(1, 0.8, 10), Rz = 0.05
ūx (t ) = cos ψ(t ) sin θr (t ) cos φr (t ) + sin ψ(t ) sin φr (t )
(43) N2xy = 10Inxy , Nuxy = 10Inxy ,
ūy (t ) = sin ψ(t ) sin θr (t ) cos φr (t ) − cos ψ(t ) sin φr (t ).
Qxy = diag(26, 35, 20, 26, 35, 20), Rxy = diag(75, 75).
Thereby, if uξxy (k) is substituted by its desired value in Eq. (42),
the following expression is obtained: The nonlinear H∞ controller gains were tuned with the
      following values: ω1 = 0.1, ω2 = 3, ω3 = 9 and ωu = 1.5.
ūx (k) ũ (k|k) u (k) The first reference path used is a circle evolving in the ℜ3
= x + xr . (44)
ūy (k) ũy (k|k) uyr (k) Cartesian space defined by:
   
Once the desired virtual inputs have been computed, the 1 πt 1 πt
xr = cos m, yr = sin m,
reference of the roll and pitch angles, φr and θr respectively, are 2 20 2 20
derived using Eq. (43). These references are necessary for the  
πt
helicopter rotational loop. zr = 3 − 2 cos m, ψr = 0 rad.
20
6. Simulation results Figs. 3–7 show the simulation results of the path following of
this reference trajectory. The way in which the helicopter follows
The proposed control strategy has been tested by simulation the reference for different vehicle parameters is presented in Figs. 3
in order to check the performance attained for the path follow- (in the 3D space) and 4. Besides, some marks have been included
ing problem. Besides, simulations comparing the control structure in this last figure, indicating the moments the disturbances have
developed in this paper with the one of Bouabdallah and Siegwart been applied.
(2007) have been performed in order to show the improvement ob- It can be seen how, starting from an initial position far from the
tained with the proposed strategy. This backstepping control strat- reference, the proposed control strategy is able to make the vehicle
egy has been chosen for the comparison analysis because it is able follow the reference trajectory. In addition, the vehicle trajectory in
to present similar performance results, as well as being also able the case of no integral action is considered in the control strategy,
to reject sustained disturbances. Simulations have been carried out is also presented in this figure. It can be clearly observed that, in
 G.V. Raffo et al. / Automatica 46 (2010) 29–39 37

Fig. 5. Position error (x, y, z ). Fig. 7. Orientation error (φ, θ, ψ).

Fig. 8. Path following.

Fig. 6. Orientation (φ, θ, ψ).

this last case, the vehicle leaves the trajectory of reference when a
disturbance is introduced, and it never reaches the reference again.
Fig. 5 shows the translational coordinates errors. It can be seen
that null steady-state error is achieved for all coordinates, even
if structural uncertainty is considered in the vehicle. Besides, this
figure also shows that a null steady-state error is not obtained in
the case of no integral action is included in the controller synthesis.
The way the inner nonlinear H∞ controller makes the vehicle
track its rotational references is presented in Figs. 6 and 7. It can
be observed how highly-coupled the system is, since each degree
of freedom is affected by the disturbances applied to the whole
helicopter. The first figure shows how the references generated
by the (e-sspc) translational controller, i.e. φr and θr , varies in
its attainment of an appropriate performance in the translational
loop. On the other hand, Fig. 7 corroborates the fact that null
steady-state error is also achieved for the inner loop variables,
unless in the case of no integral action is considered by the inner Fig. 9. Position (x, y, z ).
loop controller.
A second simulation collection has been carried out with a In order to make a quantitative comparison of the results at-
reference trajectory made up of a set of several kinds of stretches, tained by these two control strategies, some performance indexes
starting from xr0 = 0.5 m, yr0 = 0.0 m, zr0 = 1.0 m and have been computed.
ψr0 = 0 rad. In these simulations, results attained by the integral On one hand, the Integral Square Error (ISE) performance
mpc with the nonlinear H∞ control strategy are compared with the indexes obtained from the simulation results are presented in
ones achieved by the integral backstepping controller proposed by Table 1. It can be observed that the performance is improved by
Bouabdallah and Siegwart (2007). The parameters for both control the Integral MPC/Nonlinear H∞ control strategy for all states (x ↓
structures have been synthesized to obtain a smooth reference 30.05%, y ↓ 25.47%, z ↓ 40.62%, φ ↓ 76.13%, θ ↓ 40.66% and
tracking, with a quick disturbance rejection and a minimum ψ ↓ 11.48%).
transient error. The simulation results are depicted in Figs. 8–11. Although higher overshoots in the x and y error responses are
These figures show that both control strategies present a robust presented at the beginning of the trajectory (see Fig. 10), the accu-
path following when abrupt changes of references and sustained mulated error along the path is lesser than the error achieved by
disturbances are applied to the system. the backstepping controller.
38 G.V. Raffo et al. / Automatica 46 (2010) 29–39

Fig. 10. Position error (x, y, z ).

Fig. 12. Control inputs (U1 , τφa , τθa , τψa ).

Table 2
IADU index performance analysis.
Control signals MPC + NL H∞ BackStepping

U1 17.6692 20.8502
τφa 63.9550 199.9717
τθa 65.4172 220.3820
τψa 18.7141 43.1515

on all degrees of freedom. The control strategy has been split into
two stages.
On one hand, a state-space predictive controller for the trans-
lational movements has been proposed for the outer loop, which
achieves a good and smooth performance in the reference tracking.
Besides, to reject sustained disturbances affecting the translational
motion, the integral of the position error has been considered in the
Fig. 11. Orientation (φ, θ, ψ). error model used by the predictive controller.
On the other hand, a robust control for the helicopter stabiliza-
Table 1 tion based on nonlinear H∞ theory has been designed for the inner
ISE index performance analysis. loop. This controller is also able to reject sustained disturbances
States MPC + NL H∞ BackStepping due to the use of the integral action in the state vector.
The robustness, the smoothness and the predictive feature of
x 18.2883 26.1467
y 16.4420 22.0603
the proposed control strategy have been corroborated by simula-
z 11.2947 19.0209 tions, where parametric and structural uncertainties, and unmod-
φ 4.6388 19.4346 elled dynamics, besides sustained disturbances, have been taken
θ 4.7846 8.0633 into account.
ψ 4.6225 5.2219 The results have presented an excellent tracking of the several
classes of trajectories, and have illustrated the robust performance
On the other hand, the Integral Absolute Derivative control sig- provided by the nonlinear H∞ inner controller in the case of
nal (IADU) index has been computed for all control signals in both parametric uncertainties in the mass and inertia terms. Moreover,
control strategy (depicted in Fig. 12). This performance index is the use of integral action in the inner and outer loop controllers has
very appropriate to check the control signals’ smoothness. As pre- provided the capability to deal with sustained disturbances when
sented in Fig. 12, the integral MPC/Nonlinear H∞ control strategy all degrees of freedom are affected by this kind of perturbation in
different moments of time.
generates smoother input control signals than the other strategy,
To show the improvements achieved by the proposed control
underlining the quality and the feasibility of the proposed control
strategy, a comparative analysis between the proposed control
structure. The results obtained from the simulation are presented
strategy and other recent controller has been carried out by means
in Table 2. It can be seen that the smoothness is also improved
of the ISE and IADU performance indexes.
by the Integral MPC/Nonlinear H∞ control strategy for all con-
Finally, future work will involve the implementation of this
trol signals (U1 ↓ 15.26%, τφa ↓ 68.02%, τθa ↓ 70.32%, and control strategy in a real quadrotor helicopter. A new vehicle is
τψa ↓ 56.63%). being built, which will include an appropriate control hardware
to compute control signal. Since explicit control laws have been
7. Conclusions synthesized, no problem is expected regarding sampling times.

An integral predictive and robust nonlinear H∞ control strategy Acknowledgements

to solve the path following problem for a quadrotor helicopter has
been presented in this paper. The proposed control strategy has The authors would like to thank CICYT for funding this work
been designed under consideration of external disturbances acting under grants DPI2006-07338 and DPI2007-64697.
 G.V. Raffo et al. / Automatica 46 (2010) 29–39 39

References Van der Schaft, A. (1992). L2 -gain analysis of nonlinear systems and nonlinear state
feedback control. Institute of Electrical and Electronics Engineers. Transactions on
Bouabdallah, S., Murrieri, P., & Siegwart, R. (2004). Design and control of an indoor Automatic Control, 37(6), 770–784.
micro quadrotor. In Proc. IEEE Int. conf. on rob. and automat., Vol. 5, New Orleans, Van der Schaft, A. (2000). L2 -gain and passivity techniques in nonlinear control. New
USA (pp. 4393–4398). York: Springer-Verlag, p. 249.
Bouabdallah, S., Noth, A., & Siegwart, R. (2004). PID vs LQ control techniques applied Zemalache, K. M., Beji, L., & Maaref, H. (2007). Two inertial models of x4-flyers
to an indoor micro quadrotor. Proc. IEEE int. conf. on intelligent robots and dynamics, motion planning and control. Integrated Computer-Aided Engineering,
systems, Vol. 3, Sendai, Japan (pp. 2451–2456). 14(2), 107–119.
Bouabdallah, S., & Siegwart, R. (2007). Full control of a quadrotor. In Proc. of the
intelligent robots and systems. IROS 2007, San Diego, USA (pp. 153–158).
Camacho, E., & Bordons, C. (1998). Model predictive control. New York: Springer- Guilherme V. Raffo was born in Porto Alegre, Brazil, in
Verlag. 1979. He received the B.Sc. degree in automation and
Castillo, P., Lozano, R., & Dzul, A. (2005a). Stabilization of a mini rotorcraft with four control engineering from Pontifical Catholic University of
rotors. IEEE Control Systems Magazine, 45–55. Rio Grande do Sul – PUC-RS, Brazil, in 2002, his specialist
Castillo, P., Lozano, R., & Dzul, A. E. (2005b). Modelling and control of mini-flying degree in industrial automation from Federal University of
machines. London, UK: Springer-Verlag. Rio Grande do Sul (UFRGS), Brazil, in 2003 and his M.Sc.
Chen, M., & Huzmezan, M. (2003). A combined MBPC/2DOF H∞ controller for a quad degree in electrical engineering from Federal University
rotor UAV. In Proc. AIAA guidance, navigation, and control conference and exhibit, of Santa Catarina (UFSC), Brazil, in 2005. He is currently
TX, USA. working toward the Ph.D. degree at University of Seville,
Craig, J. J. (1989). Introduction to robotics — Mechanics and control (2nd edn). USA: Spain, in robust control and H∞ theory applied to UAV.
Addison-Wesley Publishing Company, Inc. From August to December 2008 he had a research
Fantoni, I., & Lozano, R. (2002). Non-linear control for underactuated mechanical stay at the University of Leicester, England, with the Control and Instrumentation
systems. London: Springer-Verlag. Research Group working in air traffic control, as part of his Ph.D. His current
Feng, W., & Postlethwaite, I. (1994). Robust nonlinear H∞ /adaptive control of robot research interests include robust control, H∞ theory, predictive control and robotic
manipulator motion. Proceedings of the Institution Mechanical Engineers, 208, systems.
221–230.
Guenard, N., Hamel, T., & Mahony, R. (2008). A practical visual servo control for an
unmanned aerial vehicle. Robotics, IEEE Transactions on [see also Robotics and
Manuel G. Ortega was born in Jaén, Spain, in 1969.
Automation, IEEE Transactions on], 24(2), 331–340.
He received the M.Sc. degree in Industrial Electrical
Kühne, F., Lages, W. F., & Gomes Da Silva, J. M. (2005). Point stabilization of mobile
Engineering, and the Ph.D. degree in System Engineering
robots with nonlinear model predictive control. In Proc. of the IEEE mechatronics
and Automation (with doctoral award), both from the
and robotics, Vol. 3, Niagara Falls, Canada (pp. 1163–1168).
University of Seville, Spain, in 1995 and 2001, respectively.
Madani, T., & Benallegue, A. (2007). Sliding mode observer and backstepping control
From 1996 to 2004, he was a Research Assistant under
for a quadrotor unmanned aerial vehicles. In Proc. of American control conference,
several grants and was an Assistant Professor at the
2007. ACC ’07, NY, USA (pp. 5887–5892).
Department of Systems Engineering and Automation of
Mistler, V., Benallegue, A., & M’Sirdi, N.K. (2001). Exact linearization and
the University of Seville. Since 2004, he has been an
noninteracting control of a 4 rotors helicopter via dynamic feedback. In Proc.
Associate Professor at the same Department. Dr. Ortega
IEEE int. workshop on robot and human inter. communic.
has worked in various research and development projects
Mokhtari, A., Benallegue, A., & Daachi, B. (2006). Robust feedback linearization and
in cooperation with industry. He is the author or co-author of more than 35
GH ∞ controller for a quadrotor unmanned aerial vehicle. Journal of Electrical
publications including journal papers, book chapters, and conference proceedings.
Engineering, 57(1), 20–27.
His current research interests include robust control, H∞ control theory, nonlinear
Mokhtari, A., Benallegue, A., & Orlov, Y. (2006). Exact linearization and sliding
control systems, robotics, and process control.
mode observer for a quadrotor unmanned aerial vehicle. International Journal
of Robotics and Automation, 21(1), 39–49.
Olfati-Saber, R. (2001). Nonlinear control of underactuated mechanical systems
with application to robotics and aerospace vehicles. Ph.D. thesis. Massachusetts Francisco R. Rubio received the Industrial Electrical En-
Institute of Technology. gineering degree and Doctorate from the Escuela Técnica
Ortega, M. G., Vargas, M., Vivas, C., & Rubio, F. R. (2005). Robustness improvement Superior de Ingenieros Industriales de Sevilla in 1981 and
of a nonlinear H∞ controller for robot manipulators via saturation functions. 1985, respectively. He received the CITEMA award for the
Journal of Robotic Systems, 22(8), 421–437. best work on automation by a young engineer in 1980. He
Raffo, G. V., Gomes, G. K., Normey-Rico, J. E., Kelber, C. R., & Becker, L. B. (2009). A is a Professor in the Department of Systems Engineering
predictive controller for autonomous vehicle path tracking. IEEE Transactions on and Automatic Control of the University of Seville. He has
Intelligent Transportation Systems, 10(1), 92–102. worked on various research and development projects in
Raffo, G. V., Ortega, M. G., & Rubio, F. R. (2008). Backstepping/nonlinear H∞ control cooperation with industry. His current interests are in the
for path tracking of a quadrotor unmanned aerial vehicle. In Proc. of the 2008 areas of adaptive control, robust process control, nonlinear
American control conference — ACC2008, Seattle, USA (pp. 3356–3361). control systems and robotics. He has written two books:
Rossiter, J. A. (2003). Model-based predictive control: A practical approach. New York: Advanced Control of Solar Plants published by Springer-Verlag and Control Adap-
CRC Press. tativo y Robusto published by Seville University. He has authored and co-authored
Sun, S. (2005). Designing Approach on trajectory-tracking control of mobile robot. more than 150 technical papers published in international journals and conference
Robotics and Computer-Integrated Manufacturing, 21(1), 81–85. proceedings.

View publication stats