1 s2.0 S026322411730324X Main

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Measurement 109 (2017) 51–64

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Nonlinear dynamic modeling and hybrid control design with dynamic


compensator for a small-scale UAV quadrotor
Yi-Rui Tang a, Xiao Xiao a, Yangmin Li b,c,⇑
a
Department of Electromechanical Engineering, University of Macau, Macau, China
b
Department of Industrial and Systems Engineering, Hong Kong Polytechnic University, Hong Kong, China
c
Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control, Tianjin University of Technology, Tianjin, China

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents a detailed nonlinear dynamic model and the development of a flight control scheme
Received 14 March 2017 for an unmanned aerial vehicle (UAV) quadrotor. In the dynamic modeling, significant aerodynamic
Received in revised form 29 April 2017 effects arising from the blade deformation of the rotor and the induced momentum are taken into
Accepted 12 May 2017
account. The model is capable of representing the quadrotor behavior for more flight regimes, including
Available online 17 May 2017
not only hover but also high-speed and translational flights. Such a model can benefit the controller
design from the improvement of the stability as well as the reduction of development period.
Keywords:
Furthermore, a hybrid control scheme incorporating optimal linear quadratic regulator (LQR) and robust
UAV quadrotor
Dynamic modeling
sliding mode control (SMC) techniques is designed in a hierarchical multiple-layer structure. In actual
Aerodynamic effect practice, only partial states of the quadrotor are readily measurable. A reduced-order observer is hence
Flight control developed and integrated into the compensator to estimate the unmeasured states. The simulation
Observer design results show that the designed controller can achieve outstanding performance which is greatly superior
than conventional control techniques can do.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction challenging. Academic institutes have reported many progress


and demonstrated some kinds of flight capabilities with the UAV
Small-scale unmanned aerial vehicle (UAV) quadrotors have quadrotors. Nevertheless, the reported results are still far from
experienced a rapid development over recent years. Many poten- the theoretical potential of the UAV quadrotors. This is mainly
tial applications have been successfully exploited with a broad because of the absence of an accurate model of the quadrotor
range of examples in the civilian and military field, e.g., air photog- dynamics by taking the advantage of which the flight control law
raphy, tactical surveillance, environmental sensing, agricultural can be analyzed and designed. The majority of previous studies
monitoring, infrastructure inspection, traffic surveillance, inspec- have used the common dynamic model with moderate level of
tion for dam or high-voltage electrical lines, on-site fire control, complexity to represent the quadrotor behavior [1,2]. However,
air combat and battlefield reconnaissance. Although impressive this model holds only for near hovering flights and it cannot repre-
achievements were made, the development of the UAV system is sent the flight behavior that an actual quadrotor exhibits. The sig-
still at the very beginning stage. A great deal of effort needs to be nificant aerodynamic effects resulting from the deformation of
made before the UAVs become reliable and trustworthy for appli- rotor blades and the effects from induced momentum are usually
cations. Unreliable control law design resulting from inaccurate neglected [3]. In fact, the quadrotor undergoes many aerodynamic
dynamic modeling is the critical factor that prevents UAVs’ effects during the flight in various operating regimes.
advance in technology. If the problem is addressed, it would enable The modeling errors in the common dynamic model can be sig-
researchers to broaden their horizon and explore more potential nificantly magnified for translational flights of the quadrotor, par-
applications of UAV quadrotors. ticularly for aggressive maneuvers. The modeling errors are
Quadrotor is a complex multi-variable and inherently unstable generally compensated by the design of controllers with control
system, which makes controller design particularly difficult and techniques, such as fuzzy-logic [4,5], iterative learning [6–9],
model predictive control [10,11] and neural network [12,13], etc.
⇑ Corresponding author at: Department of Industrial and Systems Engineering,
Nevertheless, the controllers developed with these control
Hong Kong Polytechnic University, Hong Kong, China.
approaches need to be tuned and implemented with abundant
E-mail address: yangmin.li@polyu.edu.hk (Y. Li). real-time computational efforts. It needs to carry out flight

http://dx.doi.org/10.1016/j.measurement.2017.05.036
0263-2241/Ó 2017 Elsevier Ltd. All rights reserved.
52 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

experiments by employing the real quadrotor, which is normally a rotor blade is not rigid and particularly has a tendency to bend
risky and time-consuming process full of repetitive tests. Another under inertial and aerodynamic forces [20]. In order to obtain an
reason is due to the very confined flight space in some laboratories, explicit and accurate dynamic model of the quadrotor, the com-
it is almost impossible to develop the control algorithms straight mon model is refined to adapt to the aerodynamic effects and
by using physical hardware. For example, the quadrotor has the changes of surroundings [21]. The model is designed with free
already flown out of the tracking area before the transient response inputs that can be readily employed for the control law design.
of the system can be clearly presented. Thus, a method needs to be For a quadrotor involving various physical effects, the derived
devised to greatly narrow down the searching range for the values equations of motion are typically high-order differential equations
of ideal control gains perfectly accommodating the control of with nonlinear coupled vectors. A Proportional-integral-derivative
actual quadrotors. Only minor modifications and refinements are (PID) controller is incompetent and unable to ensure the stability
required to achieve satisfactory control performance of the of the vehicle. More sophisticated and advanced control techniques
quadrotor after employing physical controller in flight tests. These are therefore desired. In general, modern control techniques are
problems can be efficiently addressed by conducting the controller competent to address control problems for complex multi-
development in simulation before performing real flight tests. variable system. Linear quadratic regulator (LQR) control approach
Whereas, the achievement of good performance of the developed was successfully applied to the controller design for solving such
controller in simulation needs the derivation of mathematical problems [22–24]. Full-state feedback is always desired in the con-
model of the quadrotor with high fidelity. For this reason, an accu- trol design with LQR. However, some states of the quadrotor can-
rate dynamic model of the quadrotor is desired. Such a model can not be directly measured from actual practice. For instance, rotor
benefit the controller design from both the improvement of stabil- flapping as part of aerodynamic effects is usually unmeasurable
ity and the reduction of development period. Likewise, the control during quadrotor flights. This kind of states is generally ignored
scheme can be especially designed according to the characteristics in the LQR control design in the former study [22]. As such, control
of the model. errors can be significant in the flight regimes where aerodynamic
To design an accurate and reliable control system for the UAV and momentum effects commonly occur, particularly in transla-
quadrotor, the dynamics of the vehicle platform should be first tional and high-velocity flights. Even though the derived dynamic
understood. Generally speaking, first-principles modeling and sys- model can be used to simulate the quadrotor’s behavior with an
tem identification modeling are two main dynamic modeling acceptable level of accuracy, a small class of unmodeled dynamics
approaches for the derivation of nonlinear dynamic models of and system uncertainty still exists. The LQR control is not robust
the UAV quadrotors. The system identification modeling approach enough to handle this situation. The problems were addressed by
is to use collected flight data to verify linearized models. The accu- employing the sliding mode control (SMC) to deal with modeling
racy of the identified models highly depends on flight data quality imprecision, system uncertainty and disturbances [25–28]. To
and the applied identification algorithms. The quality of collected remove steady-state errors of the controller, the previous
flight data cannot be fully guaranteed, although micro electro researches [22,29] have designed and incorporated an integral
mechanical system (MEMS) technology greatly advanced and model in the control loops.
many previous works have been conducted to improve the data This research proposes a high-fidelity nonlinear dynamic model
quality. Moreover, the iterative parameter update process com- and a hybrid control scheme for a small-scale UAV quadrotor.
bined with the large number of historical points can extremely Mathematical equations of the nonlinear model are presented with
increase the computational load and difficulties for parameter the consideration of several major aerodynamic and induced
identification, especially for inherently unstable systems such as momentum effects for the vehicle flight in different operating
UAV quadrotors. The previous study in [14] has shown that the regimes, such as rotor flapping, gyroscopic action and rolling
identification method is not efficient and satisfactory enough to moment of propellers. The developed model covers the aerody-
match the predictive histories against measured ones even for a namics for more flight regimes, not only in hover, but also in
hovering condition. First-principles modeling approach has been high-speed and translational flights. Furthermore, based on the
used to obtain nonlinear dynamic models that are capable of com- newly developed dynamic model, a control scheme is proposed
prehensively representing the complex quadrotor behavior over to efficiently regulate the whole quadrotor system to achieve sta-
the full flight envelope, especially for aggressive and acrobatic bility and optimize the vehicle’s capabilities. A reduced-order
flight conditions. It is generally impractical and too dangerous to observer as a part of compensator is designed to estimate the
perform flight tests for data collection in those flight conditions. unmeasurable states of the vehicle. The developed controller can
In physical approach, the quadrotor dynamics are derived by estab- be applied to physical quadrotor system with minor modifications.
lishing the equations of motion, using the fundamental laws of
mechanics and aerodynamics. This approach is referred to as
first-principles modeling [15,16]. The quadrotor model is derived 2. Nonlinear dynamic modeling of the quadrotor
based on the equations of unsteady motion of the rigid body.
Because of the nonlinear mutual dependence of mechanical and The quadrotor has four fixed-pitch rotors installed on a rigid
aerodynamic components of the vehicle, it’s necessary to divide cross platform, as shown in Fig. 1. The vehicle is actuated to move
the model into several parts and build each part of the model sep- by changing the rotational speed of the rotors. Each rotor produces
arately [17]. First-principle models are especially useful for simula- a torque about its center of rotation. The net aerodynamic torque of
tion applications since they are capable of representing the the quadrotor is zero when all of the rotors spin at the equal angu-
complex quadrotor behavior for a large flight envelope [18]. As lar velocity since the torque produced by the counter-rotating pairs
opposed to conventional helicopters which require complex of rotors is in opposite direction. The yaw is induced by increasing
mechanical linkages to tilt the swashplate as well as vary the angle the spinning speed of one pair of rotors while decreasing for
of rotor blades to perform the attitude control, the quadrotor has another pair so as to unbalance the aerodynamic torque about
relatively simpler mechanical configuration. The rotors mounted yaw axis. In addition to the torque, each rotor produces a thrust
on the airframe are the unique actuation in the system and major perpendicular to the plane of rotation of blades upwards. Either
source contributed to aerodynamic effects. The quadrotor has four roll or pitch is controlled by differential thrust generated by the
identical plastic rotors with fixed pitch, although some publica- pair of rotor blades rotating in the same direction. Increasing the
tions have demonstrated the variable-pitch platforms [19]. The rotational speed of one of the pair rotors while decreasing the
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 53

along the respective body axes. The angular motion is described by


the angular velocity x b with fp; q; rg around the respective body
axes and Euler angles denoted by f/; h; wg with respect to NED
inertial axes in fWg, referred to as the roll, pitch, and yaw,
respectively.
The equations of motion of the airframe with respect to fBg are
expressed in Newton-Euler formalism as

F tot ¼ mv  b  ðmv
_ b þ x  bÞ ð1Þ
and

_ b þ x
M tot ¼ J x  b  ðJx
 bÞ ð2Þ
where F tot and M tot stand for the vector of total force and moment
acting on the vehicle CoG, respectively, m denotes the mass of the
quadrotor, and J is the matrix of rotational inertial of the vehicle
Fig. 1. A custom-made instrumented quadrotor.
with J ¼ diagfJxx ; J yy ; Jzz g where off-diagonal terms are negligible
due to the symmetric mechanical structure of the quadrotor. In
speed of the other one will cause a rotational motion in pitch or roll
accordance to the common definitions in aviation, the transforma-
axis. However, this motion is decoupled from affecting the one in
tion from the body-fixed frame fBg to inertial coordinate fWg is
yaw axis since the torque balance required for yaw stability is con-
described by using the rotational transformation matrix denoted
stantly maintained. Likewise, the yaw motion can be produced
by RWB . The transformation follows the rotation sequence of the
without affecting either pitch or roll axis because of the same
Z-Y-X aeronautical convention and yields
thrust continually kept by the rotors in the pair. The main symbols 2 3 2 3
used in the derivation of quadrotor dynamic models are listed in Ex Xb
6 7 6 7
Table 1. 4 Ey 5 ¼ RWB 4 Y b 5 ð3Þ
Ez Zb
2.1. Quadrotor dynamics of the rigid body
It should be noted that the transpose of RWB equals its inverse since
In the derivation of the nonlinear dynamic equations, the it is an orthogonal matrix. The rotation matrix RWB is determined by
quadrotor can be considered as a rigid body combined with a instantaneous Euler angles f/; h; wg that are not directly measurable
mechanism to produce forces and moments. The dynamic model with onboard sensors. The angular velocity fp; q; rg provided by the
is derived in the body-fixed frame fBg and the North-East-Down IMU sensor can be used for the calculation of Euler angles, as
(NED) inertial coordinates fWg. As shown in Fig. 2, the body- expressed in the following equation [30],
fixed frame is attached to the quadrotor fuselage at its center of 2 3 2 32 3
/_ 1 tan h sin / tan h cos / p
gravity (CoG). Let fEx ; Ey ; Ez g denote unit vectors along the respec- 6_7 6 76 7
tive NED inertial axes in fWg and fX b ; Y b ; Z b g denote unit vectors 4 h 5 ¼ 40 cos /  sin / 54 q 5 ð4Þ
along forward, rightward and downward axes in fBg, respectively. w_ 0 sin /= cos h cos /= cos h r
The linear velocity of the quadrotor is denoted by v  b with fu; v ; wg
2.2. Equations of force and moment
Table 1
Definition of symbols for the quadrotor dynamic model. As illustrated in Fig. 2, the total force acting on the quadrotor is
Symbols Definitions
the gravitational force combined with the sum of the thrust gener-
ated by each rotor T i in body-fixed frame. The total moment acting
v b Linear velocity in body-fixed frame with fu; v ; wg
on the vehicle is principally from the applied forces produced by
x
b Angular velocity in body-fixed frame with fp; q; rg
rotor thrust to rotate the airframe about body-fixed axes. Besides,
F tot Total force acting on the vehicle
M tot Total moment acting on the vehicle each rotor produces a reaction moment as the result of rotor drag.
m mass of the quadrotor In high-speed and translational flights, complicated motions of
J Rotational inertial of the vehicle rotor blades occur since the blade sections undergo some periodic
RWB Rotational transformation matrix
variation in local velocity, which produces lots of aerodynamic
F t;i Thrust produced by the ith rotor
M t;i Thrust moment of the ith rotor
effects. These effects can exert a powerful influence over the
Fg Gravitational force vector acting on the vehicle quadrotor dynamics, leading to a poor control design without con-
M gyr Gyroscopic moment acting on the vehicle sideration of the effects. Thus, the moments caused by aerody-
Mr Rotor moment namic effects are summed up in the calculation of the total
M rol;i Rolling moment of the ith propeller moment. The rotor blades are made of plastic and have some level
Q r;i Drag moment of the ith propeller
of flexibility. The rotor tends to bend up and down while the lift
a Flapping angle
a0 Coning angle
generated by the airfoil on one side of rotor blades is different from
a 1 Longitudinal flapping angle that on the other side. During forward flight, more lift is produced
a 2 Lateral flapping angle by the advancing blade of the rotor than the retreating one since
kt;i Unit vector of rotor thrust the advancing blade has a higher airspeed [31]. Hence, the rolling
Ti Magnitude of rotor thrust
moment of a rotor occurs due to the unbalanced lift acting on
di Displacement of the ith rotor from the CoG
dl The arm length
the blades throughout the rotor disk. The rotor naturally tends to
hl The rotor height above CoG counter the rolling moment in the form of rotor flapping. More lift
Mh The hub moment exerted on the advancing blade causes it to bend upwards. This
Kh Constant stiffness coefficient motion reduces its angle of attack so that the lift is accordingly
kh Unit vector of hub moment
diminished, and vice versa [32]. Moreover, the gyroscopic torque
54 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

Fig. 2. Free body diagram of the quadrotor.

is also a source of aerodynamic and momentum effects, which may 2.2.2. Forces and moments of rotors
not be neglected. The torque is generated as the rotor speed differ- In hover flight, the thrust forces generated by four rotors should
ence of the pair of motors creates roll and pitch rates of the fuse- counteract the gravitational force F g arising from the mass of the
lage. Consequently, the total force and moment acting on the quadrotor. Whereas, in translational flight, the propelling force
quadrotor are written as should additionally resist fuselage and rotor drag. The rotor flap-
ping as a major aerodynamic effect is considered in the derivation
X
4
F tot ¼ F t;i þ F g of rotor force and moment equations. Fig. 3 demonstrates the blade
i¼1 flapping of a single rotor. The blades represented as a rigid beam
ð5Þ
X
4 X
4 are fixed to motor shaft at rotor hub with some degrees of flexibil-
M tot ¼ M r þ Q r;i þ M rol;i þ M gyr ity. The induced forces tilt the blade beam about the rotor hub dur-
i¼1 i¼1 ing rotor flapping, resulting in a flapping angle denoted by a with
where F t;i stands for the vector of the thrust produced by the ith respect to the hub plane which is orthogonal to the motor shaft.
rotor, F g and M gyr separately denote the gravitational force vector The equation for the calculation of the flapping angle is commonly
and gyroscopic moment vector acting on the vehicle, M r is the vec- derived by using infinite Fourier series. Generally speaking, the
tor of rotor moment, M rol;i and Q r;i stand for the rolling moment and terms related to the second and higher harmonics are negligible
drag moment of the ith propeller, respectively. In what follows, the as very limited contribution can be made to the thrust and moment
derivation of equations for each force and moment is introduced in of the rotor. The equation is written by considering only constant
detail. terms and the first harmonic,

2.2.1. Gravitational force


a ¼ a0  a 1 cos wr  a 2 sin wr ð7Þ
The gravitational force is externally exerted on the vehicle coin-
where the coning angle is denoted by a0 ; wr stands for blade azi-
ciding with the Ez axis in fWg, as shown in Fig. 2. The gravitational
muth angle, a  1 is the longitudinal flapping angle with positive val-
force as a function of acceleration with respect to the gravity is
ues when the blade flaps up about the rotor hub from the axis of
transformed from fWg to fBg,
3 2
2 3 positive X b counterclockwise, and a  2 denotes the lateral flapping
0 mgcos/sinh angle that is positive when the blade flaps up from positive Y b axis
6 7 6 7
F g ¼ R1
WB 4 0 5 ¼ 4 mgsin/ 5 ð6Þ counterclockwise. The resultant thrust associated with X b and Y b
axes can be considered as the projection of the rotor thrust vector
mg mgcos/cosh
on the rotor hub plane. It is assumed that the rotor thrust vector
where g is acceleration of gravity. is always perpendicular to the rotor tip-path-plane (TPP). The unit
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 55

Fig. 3. Blade flapping of the rotor.

vector of the rotor thrust kt;i in fBg can be described in terms of As shown in Fig. 4, the quadrotor experiences rotor flapping
rotor flapping, during translational flight. The thrust generated by each rotor is
2 3 tilted to produce a moment tending to rotate the fuselage. This
 sin a 1
6 rotor thrust moment can be written as
kt;i ¼ 4 cos a 1 sin a2 75 ð8Þ
 cos a
 1 cos a2 M t;i ¼ di  F t;i ð10Þ

and then the thrust produced by the ith rotor is given by where M t;i is the thrust moment of the ith rotor and di stands for
F t;i ¼ T i kt;i ð9Þ displacement vector of the ith rotor from the CoG of the quadrotor.
Regarding the mechanical structure of the vehicle, the displacement
where T i is the magnitude of rotor thrust. vector for the rotors is summarized as

Fig. 4. Quadrotor diagram with blade flapping effects in translational flight.


56 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

2 3 2 3
di;1 dl 0 hl Mrol;i ¼ C rol qAð-i r i Þ2 r i ð17Þ
6d 7 6 0 dl hl 7
6 i;2 7 6 7
6 7¼6 7 ð11Þ where C rol denotes the coefficient of rolling moment.
4 di;3 5 4 dl 0 hl 5 The rotor spins about the motor shaft that is regarded as an axis
di;1 0 dl hl of rotation in the local reference frame of the motor-rotor. This axis
of rotation is driven to change orientations with the movement of
where dl denotes the arm length and hl is the rotor height above
the fuselage of the quadrotor, which leads to a gyroscopic effect. An
CoG.
extra moment can be created to act on the fuselage under the gyro-
Moreover, the rotor flapping caused the stiffness effect where
scopic effect. The generated gyroscopic moment is given by the
the rotor hub moment is produced. The hub moment denoted by
equation
M h is computed by considering it as a torque created by a simple
torsional spring working about the rotor hub (see Fig. 3). The X
4
 b 
hub moment is then expressed as M gyr ¼  J r -i x
  kt;i ð18Þ
i¼1
M h ¼ kh akh ð12Þ
For the computation of flapping angle, only the coning angle
where K h is constant stiffness coefficient and kh represents the unit and the first harmonic are taken into account, which is applicable
vector of hub moment. The vector is normal to the plane constituted for most cases. Due to the approximate linear motion of the fix-
by the motor shaft and thrust vector. The unit vector of hub pitch rotor, the coning angle can be solved by using the blade ele-
moment is therefore written as ment theory [32],
2 3  
 cos a
 1 sin a
2 qacr4i   4k
6 7 a0 ¼ hb 1 þ l2  ð19Þ
kh ¼ Rz ð90 Þkt;i ¼ 4  sin a1 5 ð13Þ 8J r 3
0
where a is lift curve slope, c is rotor blade chord, hb stands for the
As a result, the total rotor moment is the sum of the rotor thrust pitch angle of the blade, and l is rotor advance ratio. The coeffi-
moments and hub moments, cients of the first harmonic in Eq. (7) is expressed as
X
4 lð16hb  12kÞ
Mr ¼ ðM t;i þ M h Þ ð14Þ a 1 ¼ ð20Þ
6  3l2
i¼1
and
2.3. Equations of aerodynamics and flapping 3lhb
a 2 ¼ ð21Þ
4 þ 2l2
The symbols describing aerodynamics and flapping on the rotor
are summarized in Table 2. The forces and moments are obtained Since parameter identification is not the major concern of the
by applying the blade element theory to integrate separate forces paper, the determination of parameters is briefly introduced. The
acting on every small blade section over the whole rotor. By means physical parameters of the mathematical model of the quadrotor
of dimensionless quantities [32], the thrust of the ith rotor is deter- are obtained from direct measurements (such as m; dl and hl ),
mined by extracted from the faithful Solidworks model (such as the matrix
of rotational inertial J) and estimated from test-bench experiments
T i ¼ C t qAð-i r i Þ2 ð15Þ (such as C t and C q ).
where C t stands for thrust coefficient, q denotes air density, A is the
area of rotor disk, -i and ri separately denotes rotational velocity 3. Flight controller design
and radius of the ith rotor. The drag torque of the ith rotor is given
by The flight control scheme is designed in a hierarchical multiple-
layer structure. Fig. 5 shows the control architecture of the pro-
Q r;i ¼ C q qAð-i r i Þ r i2
ð16Þ posed dynamic compensator of the quadrotor. In actual practice,
where C q is the torque coefficient. The problem of the determina- only partial states of the quadrotor are readily measurable. A state
observer is designed to estimate the unmeasurable states. It pro-
tion of rolling moment is addressed by employing the similar
vides convergence of the generated estimates to the real states
method used for rotor drag torque. The rolling moment of the ith
by means of the measurements of specified outputs of the quadro-
rotor is written as
tor system. The full states of the system are hence reconstructed in
order to make the state variable feedback control applicable to the
closed-loop system design. In this flight system, certain states do
Table 2 not need to be estimated. They can be either measured with sen-
Description of symbols for aerodynamics and flapping on the rotor. sors of the UAV quadrotor or calculated easily from system out-
Symbols Definitions puts. Thus, an reduced-order observer with a lower order than
Ct Thrust coefficient
the quadrotor plant is designed for the state variable estimation.
q Air density In addition, the reduced-order observer has a superior performance
A The area of rotor disk on improving the robustness of the control system in contrast with
-i Rotational velocity of the ith rotor its full-order counterpart.
ri Radius of the ith rotor
Full state feedback control law coupled with a reduced-order
Cq Torque coefficient
C rol Rolling moment Coefficient observer is employed for the compensator design. An integral aug-
Jr Rotational inertial of the rotor mentation model is incorporated into the compensator, which
a Lift curve slope equips the system with the capability of asymptotic tracking of a
c Rotor blade chord reference input with zero steady-state error. The challenge is to
hb Pitch angle of the blade
l Rotor advance ratio
determine the gains which regulate the closed-loop system to
achieve a guaranteed stability. Solving the gain problems directly
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 57

Fig. 5. Architecture of dynamic compensator.

by the closed-loop system formulations can increase the order of where xa denotes the components of the state vector that are
feedback loop resulting in the consumption of additional comput- directly measurable and available for feedback, and the vector
ing power. It obviously brings extra complexity to the construction involving the components to be estimated is denoted by xb ðtÞ. The
of the compensator. To facilitate the design, the separation princi- dynamic equations are written in terms of xa ðtÞ and xb ðtÞ,
ple is applied since the state variable design process allows the
x_ a ðtÞ ¼ A11 xa ðtÞ þ A12 xb ðtÞ þ B1 uðtÞ ð24Þ
problem to separate into an estimation component and a control
component. The separation principle provides an approach to inde-
x_ b ðtÞ ¼ A21 xa ðtÞ þ A22 xb ðtÞ þ B2 uðtÞ ð25Þ
pendently design a full state feedback control law and an observer
because of their independent stability. Since it is directly measurable and there is no observation required
The overall closed-loop system is guaranteed to be stable if the for the substate xa ðtÞ, the equation can be written as
feedback control assuming access to the complete state and the e
x a ðtÞ ¼ xa ðtÞ ¼ yðtÞ ð26Þ
observer are individually stable. The determination of the feedback
control and the observer gains is the objective of the compensator where e x a ðtÞ denotes the estimate of the substate xa ðtÞ. Hence, xb ðtÞ
design. The gains need to be regulated to not only stabilize the sys- is the only subset of the states that need to be estimated. Substitut-
tem but also make the transient response meet the desired perfor- ing the state xa ðtÞ in Eq. (26) and rearranging terms in Eqs. (24) and
mance. As for the quadrotor flight system, the desired performance (25) yield the plant dynamics
is generally associated with certain practical components. The
x_ b ðtÞ ¼ A22 xb ðtÞ þ A21 yðtÞ þ B2 uðtÞ ð27Þ
properties of the performance can also be specified in terms of per-
formance indices. The system design problems come to the search _
yðtÞ ¼ A12 xb ðtÞ þ A11 yðtÞ þ B1 uðtÞ ð28Þ
for minimum performance indices. The optimal control method
LQR is therefore employed to obtain the optimized gains of the where the plant dynamics have two sets of input uðtÞ and yðtÞ, and
closed-loop system. The residual system uncertainty is further _
yðtÞ is considered as a measured set of outputs.
tackled by a robust control approach SMC. Hence, a hybrid control The observer for the system can be described by the equation,
scheme incorporating the LQR and SMC is devised to ensure the
stability of the whole system.
x_ b ðtÞ ¼ A22 e
e _ e
x b ðtÞ þ A21 yðtÞ þ B2 uðtÞ þ Gk ðyðtÞ y_ ðtÞÞ ð29Þ
where the estimate of the substate xb ðtÞ is denoted by e x b ðtÞ, the
3.1. Dynamic reduced-order observer design
matrix Gk denotes the observer gain, and e y_ ðtÞ is the time-

The proposed compensator requires only the system input uðtÞ derivative of the estimate of yðtÞ. Referring to the Eq. (28), e y_ ðtÞ is
and the measured outputs yðtÞ without the knowledge of the full expressed as
state xðtÞ. A dynamic observer is built to estimate the internal y_ ðtÞ ¼ A12 e
e x b ðtÞ þ A11 yðtÞ þ B1 uðtÞ ð30Þ
states that are either unmeasurable or indirectly measured. Since
some states are already available for feedback, the number of state _
As opposed to yðtÞ, the output yðtÞ is not really measured since it
variables for estimation in the observer is reduced. Taking the lin- can bring extra cost and complexity of the control system. It is
ear approximation to the quadrotor dynamic equations, not taken as the input of the observer but an intermediate variable.
Instead, uðtÞ and yðtÞ are employed as input for the design of the
_
xðtÞ ¼ AxðtÞ þ BuðtÞ þ fðt; x; uÞ
ð22Þ reduced-order observer. Two variables A e and B
e are defined as
yðtÞ ¼ CxðtÞ
e ¼ A22  Gk A12
A and e ¼ B2  Gk B1
B
where xðtÞ 2 Rn denotes the state vector, uðtÞ 2 Rm is the control
input, yðtÞ 2 Rq is the system output, and fðt; x; uÞ represents the Let
system uncertainties due to parameter variations. The term ee e
z_ r ¼ A x b ðtÞ þ ðA21  Gk A11 ÞyðtÞ þ BuðtÞ ð31Þ
fðt; x; uÞ as an separate function to be considered only in the sliding
mode control. The control input uðtÞ is the sum of LQR control input Substituting Eq. (30) into Eq. (29) and taking the integral, the esti-
ul and SMC control input us . mated states are written as,
The full state vector xðtÞ is partitioned into two subsets,
e
x b ðtÞ ¼ Gk yðtÞ þ zr ðtÞ ð32Þ
 
xa ðtÞ
xðtÞ ¼ ð23Þ By introducing Eq. (32) into Eq. (31), it yields
xb ðtÞ
58 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

e r ðtÞ þ Lr yðtÞ þ BuðtÞ


z_ r ðtÞ ¼ Az e ð33Þ condition ee b ð0Þ. This means the observer has estimated the system
states to some acceptable level of accuracy. In this regard, the
where Lr ¼ A21  Gk A11 þ AG e k. design of the matrix Gk can be implemented to place the system
The goal of the observer design is to provide an estimate e x b with poles in predesigned locations so as to achieve satisfactory observer
properly designed gain matrix Gk . As time goes on, the state esti- performance.
mate e
x b ðtÞ converges to the actual xb ðtÞ, even though the initial
estimate e x b ð0Þ differs from the actual initial xb ð0Þ. Define the 3.2. Closed-loop state feedback control design
reduced-order observer estimation error as
Consider state feedback control:
e b ðtÞ ¼ xb ðtÞ  e
e x b ðtÞ ð34Þ
uðtÞ ¼ K e ^ ðtÞ
x ðtÞ þ u ð39Þ
Note that ee b ðtÞ is the only estimation error existing as the substates
in regard to xb ðtÞ. Taking the derivative of the estimation error in Eq. K ¼ ½ KaKb 
(34) yields
where K denotes the total feedback control gain, u ^ is the reference
e x_ b ðtÞ
e_ b ðtÞ ¼ x_ b ðtÞ  e ð35Þ signal introduced into the state feedback compensator by the aug-
mented integral model, and e x ðtÞ denotes the estimated state vector
Using the plant dynamics in Eq. (27) and the system observer in Eq. T
(29) gives for xðtÞ with e x a ðtÞ e
x ðtÞ ¼ ½ e x b ðtÞ  . By introducing Eq. (39) to Eq. (22),
the closed-loop system becomes
e_ b ðtÞ ¼ A22 ðxb ðtÞ  e
e _ e
x b ðtÞÞ  Gk ðyðtÞ y_ ðtÞÞ ð36Þ
_
xðtÞ ¼ AxðtÞ  BK e ^ ðtÞ
x ðtÞ þ Bu ð40Þ
To ensure that the estimation error of the substate e x b ðtÞ vanishes
The error in estimate of the state xðtÞ is defined as
asymptotically, the equation of the estimation error should be inde-
pendent of the observer’s input uðtÞ and yðtÞ. Combining Eq. (36) e
e ðtÞ ¼ xðtÞ  e
x ðtÞ ð41Þ
with Eqs. (28) and (30), the input terms cancel and the error in esti-
Since xa ðtÞ is directly measurable, there is no error in estimation.
mation of xb ðtÞ is given by
Referring to Eq. (34), the estimation error can be expressed as
e_ b ðtÞ ¼ A
e eee b ðtÞ ð37Þ  
0
e
e ðtÞ ¼ ð42Þ
e denotes observer system matrix with A e ¼ A22  Gk A12 . The e
e b ðtÞ
where A
system is in state-space representation as in Eq. (37) so its stability From Eqs. (40)–(42), the dynamic equation of closed-loop state
can be assessed with the characteristic equation associated with the feedback system is given by
system matrix A.e The characteristic equation of the reduced-order
_
xðtÞ ¼ ðA  BKÞxðtÞ þ BK b e ^ ðtÞ
e b ðtÞ þ Bu ð43Þ
observer system is
Combining Eqs. (37) and (43), the state feedback system model cou-
DðsÞ ¼ detðsI  ðA22  Gk A12 ÞÞ ¼ 0 ð38Þ
pled with observer dynamics is written in matrix form as
If all of the eigenvalues of A22  Gk A12 have strictly negative real       
d xðtÞ A  BK BK b xðtÞ B
parts or all roots of the characteristic equation lie in the left half ¼ þ ^ ðtÞ
u ð44Þ
dt ee b ðtÞ 0 A22  GK A12 e
e b ðtÞ 0
of the s-plane, then it is a stable system such that the estimation
error ee b ðtÞ tends asymptotically to zero over time for any initial The characteristic equation associated with Eq. (44) is

Fig. 6. Comparison of step responses of two controllers with respect to the translational velocity in u.
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 59

Fig. 7. Comparison of step responses of two controllers with respect to the translational velocity in v.

Fig. 8. Comparison of step responses of two controllers with respect to the translational velocity in w.

DðsÞ ¼ detðsI  ðA  BKÞÞ detðsI  ðA22  GK A12 ÞÞ ¼ 0 ð45Þ algorithm and the observer system can in principle be designed
independently. The desired performance of the closed-loop system
It can be readily found in Eq. (45) that the set of eigenvalues of the can be met by means of the design of the feedback control gain
closed-loop system is the union of the eigenvalues of the state feed- matrix K and the observer gain matrix GK .
back system matrix A  BK and the eigenvalues of the observer sys-
tem matrix A22  GK A12 . The closed-loop system is stable if both the 3.3. Augmented integral model design
system poles of the state feedback and those of the observer lie in
the left half of the s-plane. The closed-loop control design separates The quadrotor’s dynamic model contains nonlinear compo-
into a state feedback component and an observer design compo- nents. Even though the plant model is linearized at certain operat-
nent. Hence, the separation principle also holds for the compensator ing points and the feedback control strategies are properly applied,
design with a reduced-order observer. The state feedback control there still exist steady-state errors after the transient response has
60 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

Fig. 9. Comparison of square wave responses and tracking errors of two controllers with respect to the translational velocity in u.

decayed. The earlier designed control law can not ensure the refer- Taking the integral of Eq. (46) yields
ence tracking since it is vulnerable to model uncertainties and Z t
exogenous disturbances. The system output needs to be directly z ¼ r e ðsÞds ð48Þ
fed back to produce the tracking error of the reference input. If 0

the system output is regulated to make the tracking error converge Introducing Eq. (46) to Eq. (47), the state space representation of the
to zero, the steady-state error of the system can be eliminated. This system augmented with the integral of the tracking error is given by
can be achieved with a type-one system which has a pole at the         
origin [33]. The transfer function of the closed-loop system should
_
xðtÞ A 0 xðtÞ B 0
¼ þ uðtÞ  r f ðtÞ ð49Þ
contain an integration element. An augmented integral model is z_ ðtÞ Cr 0 zðtÞ 0 1
introduced in the closed-loop controller design to solve the prob-
where the plant model of Eq. (22) is used and C r denotes the output
lems resulting from steady-state errors. It provides the developed
matrix for the tracking error. A stabilizing feedback control law uðtÞ
compensator with the capability of asymptotic tracking of a refer-
can be designed for the augmented system,
ence input with zero steady state error. As mentioned in Sec-
tion 3.2, the separation principle is still applicable for the control h i e
x ðtÞ
 Z t
uðtÞ ¼  K b
K ¼ K e ^
x ðtÞ  K r e ðsÞds ð50Þ
system design and therefore the state feedback control laws can zðtÞ 0
be devised independently. To formulate the augmented system
model, the integral of the tracking error should be included in where K stands for state feedback gain, K ^ denotes the gain of the
the design. Define the intermediate variable z as integral model and it is assumed that the full state can be estimated
and fed back for control in accord with the separation principle. If
z_ ðtÞ ¼ re ðtÞ ð46Þ
control gains K and K ^ are properly selected, the stability of Eq.
where r e denotes the tracking error. It can be expressed as (49) is achieved. This signifies that the integral of the tracking error
is stable and the controller is capable of implementing asymptotic
r e ðtÞ ¼ yr ðtÞ  r f ðtÞ ð47Þ tracking with zero steady-state error. The feedback control gain K
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 61

Fig. 10. Comparison of square wave responses and tracking errors of two controllers with respect to the translational velocity in v.

consists of a LQR gain K l and a SMC gain K s . The hybrid control is parts. That is to say, the stabilizing gain Hl can be found by placing
addressed by applying each of the control gains to its own con- the poles in the left-half of s-plane from a control point of view. To
troller design. achieve optimal asymptotical stability of the feedback control sys-
tem, the LQR technique is applied to the design. One way to address
3.4. LQR optimal control design the LQR problem is to minimize the cost function given by [34]
h i Z 1
Let integral optimal control gain ^
Hl ¼ K l K and Jðxh ; ul Þ ¼ ðxTh Qxh þ uTl Rul Þdt ð53Þ
0
T
x ðtÞ zðtÞ . Eq. (50) is then simplified as
xh ðtÞ ¼ ½e
where Q and R stand for user-selected weight matrices. The cost
ul ðtÞ ¼ Hl xh ðtÞ ð51Þ function can be minimized with

Substituting for ul ðtÞ, as in Eq. (49), the state space representation of ^T P


Hl ¼ R1 B ð54Þ
the closed-loop system with respect to LQR is written as
b  BH
b l Þxh ðtÞ where P is determined by solving the Riccati equation. The similar
x_ h ðtÞ ¼ ð A ð52Þ
approach can be applied to determine the observer gains.
where
    3.5. SMC control design
^¼ A 0 ^¼ B :
A and B
Cr 0 0
Similarly, the augmented integral model is also incorporated in
The closed-loop system described in Eq. (52) is asymptotically stable the SMC control design. The state equation of the augmented sys-
only when the system has eigenvalues with strictly negative real tem in Eq. (49) is rewritten as
62 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

Fig. 11. Comparison of square wave responses and tracking errors of two controllers with respect to the translational velocity in w.
 
^ h ðtÞ þ Bu
x_ h ðtÞ ¼ Ax ^ s ðtÞ ð55Þ 0
eB
Q ^¼ ð60Þ
b2
B
The sliding function is defined as
Applying the transformation to the augmented system in Eq.
rðtÞ ¼ Pxh ðtÞ ð56Þ
(55), the governing equation of the normal system is expressed as
with
^x_ 1 ðtÞ ¼ A
^ 11 ^x1 ðtÞ þ A
^ 12 ^x2 ðtÞ
P ¼ fxh : Pxh ¼ 0g ð57Þ ð61Þ
_^x ðtÞ ¼ A ^ ^ 22 ^x2 ðtÞ þ B
^ 2 us ðtÞ
21 ^x1 ðtÞ þ A
h i 2
^ . The feedback
Let integral sliding mode control gain Hs ¼ K s K The corresponding partitions of P is established as P ¼ ½P1 ; P2 .
control law for SMC is given by The sliding motion is achieved by

Hs xh ðtÞ if Pxh ðtÞ – 0 P1 ^x1 ðtÞ þ P2 ^x2 ðtÞ ¼ 0 ð62Þ
us ðtÞ ¼ ð58Þ
0 otherwise
To cancel out the state of second subsystem ^x2 in Eq. (61), the
The sliding surface is designed by transforming the augmented expression in terms of ^
x1 is given by
system into a proper canonical form. The system is transformed
^x2 ðtÞ ¼ P1 ^ e ^
into two subsystems. Define the state variables in the new coordi- 2 P1 x1 ¼  K s x1 ðtÞ ð63Þ
nates as e s ¼ P1 P1 . The switching function can be expressed with
where K
  2
^x1 ðtÞ e s governing the sliding mode dynamics as
e xh ðtÞ
Q ð59Þ respect to the matrix K
^x2 ðtÞ h i
P¼ Ke s P2 Qe ð64Þ
where Q e is an orthogonal nonsingular transformation that satisfies
the following equation, Substituting ^x2 ðtÞ in Eq. (63) with that in Eq. (61) yields
Y.-R. Tang et al. / Measurement 109 (2017) 51–64 63

 
^ 11  A
x_ 1 ðtÞ ¼ A ^ 12 K
e s x1 ðtÞ ð65Þ controllers have extraordinary capabilities to rapidly track the
reference input signals. Meanwhile, it indicates that the unmeasur-
The design process becomes finding K e s in the state feedback control able states have been successfully estimated for the construction of
^ 11  A
so that all the eigenvalues of A ^ 12 K
e s lie in the left-half plane. full-state feedback. In addition, it can be observed clearly from the
This is done by using pole placement method. The feedback control zoomed in view that the high-performance controller has a
with integral SMC is then written as response with shorter settling and rise time, and smaller overshoot
8 compared with the optimal controller. The similar results can also
b 1 Mp rðtÞ if rðtÞ – 0
< bðt; x ÞðP BÞ
been seen in Figs. 7 and 8. It proves that the high-performance
us ðtÞ ¼
h
kMp rðtÞk ð66Þ
: controller has a better control ability even though both can deliver
0 otherwise
satisfactory control performance.
where Mp is a symmetrical positive-defined matrix and the solution Moreover, another set of tests are performed by feeding square
of the well-known Lyapunov equation. wave inputs of the velocities as the references to the control sys-
tem as shown in Figs. 9, 10 and Fig. 11. The velocity tracking
4. Simulation results responses of the optimal controller and the high-performance con-
troller are compared and the corresponding tracking errors are
The proposed control scheme is evaluated in a series of simula- demonstrated. Both of the controllers exhibit remarkable tracking
tions. The nonlinear dynamic model developed in Section 2 is ability in response to square wave reference signals and the track-
employed as the simulated UAV quadrotor to be controlled. As ing errors are kept very small within an acceptable level. Whereas,
aforementioned, the dynamic model incorporating aerodynamic the high-performance controller still performs better.
and momentum effects is capable of representing the flight behav- Finally, perturbations are injected to the system as the external
ior not only in hover but also for translational motions. Flight tests disturbances to validate the robustness of the proposed control
are thus carried out to validate the control design especially for scheme. As shown in Fig. 12, two controllers run under step refer-
translational flights. The velocities of the quadrotor are given as ence inputs. A perturbation is injected at the 10th second. It can be
control reference inputs to evaluate the tracking ability of the con- seen that both controllers can get recovery to keep on tracking the
trol system. The tracking performance of the optimal controller by reference signal without steady-state errors after disturbances.
applying LQR approach alone and the high-performance controller Nevertheless, the high-performance controller has a much faster
designed by the combination of LQR and SMC are compared. The recovery speed and smaller tracking errors. It indicates that the
system responses with respect to step input signals of the transla- incorporation of SMC algorithm can effectively reduce the influ-
tional velocities are illustrated in Figs. 6–8, respectively. The seg- ence of external disturbances and increase the robustness of the
ments of system responses right before approaching the steady- overall compensator. Thus, the proposed control scheme can effi-
state are zoomed in. The quadrotor is commanded to reach a veloc- ciently stabilize the UAV quadrotor with outstanding control per-
ity of 0.8 m/s for the translational flight from a hover as shown in formance. The high-performance controller can give superior
Fig. 6. Both of the controllers can have quick responses to the ref- control performance and it can be used to effectively compensate
erence inputs without steady-state errors. They exhibit very short the modeling imprecision, system uncertainty and external distur-
rise time with less than 1.8 s and settling time within 3.8 s, and bances which are hardly addressed by using the LQR optimal con-
small overshoots of 0.02 m/s. It proves that the developed trol law alone.

Fig. 12. Comparison of step responses and tracking errors of two controllers with respect to the translational velocity in u under perturbation injection.
64 Y.-R. Tang et al. / Measurement 109 (2017) 51–64

5. Conclusion [8] S. Lupashin, A. Schollig, M. Sherback, R.D. Andrea, A simple learning strategy
for high-speed quadrocopter multi-flips, in: IEEE International Conference on
Robotics and Automation, Alaska, USA, May 3–8, 2010, pp. 1642–1648.
This paper presents a detailed dynamic modeling approach and [9] D. Mellinger, N. Michael, V. Kumar, Trajectory generation and control for
the design of a flight control scheme with the demands of high- precise aggressive maneuvers with quadrotors, Int. J. Robot. Res. 31 (5) (2012)
664–674.
performance. The nonlinear dynamic model of the quadrotor was
[10] K. Alexis, G. Nikolakopoulos, A. Tzes, Model predictive quadrotor control:
derived from first principles. To improve the model accuracy and attitude, altitude and position experimental studies, IET Control Theory Appl. 6
widen the operating regimes, the influence of various aerodynam- (12) (2012) 1812–1827.
[11] P. Bouffard, A. Aswani, C. Tomlin, Learning-based model predictive control on a
ics and induced momentum effects acting on the quadrotor have
quadrotor: Onboard implementation and experimental results, in: IEEE
been especially considered in the derivation of dynamic equations. International Conference on Robotics and Automation, May 14–18, 2012, pp.
The developed dynamic model of the vehicle covers the aircraft’s 279–284.
dynamics for a large portion of the flight envelope, not only in [12] G. Buskey, G. Wyeth, J. Roberts, Autonomous helicopter hover using an
artificial neural network, in: The IEEE International Conference on Robotics
hover, but also in high-speed and translational motions. The con- and Automation, May 21–26, 2001, pp. 1635–1640.
trol scheme has been designed by employing the hybrid control [13] V. Puttige, S. Anavatti, M. Samal, Real-time validation of a dual neural network
technique in a hierarchical multiple-layer structure. State feedback controller for a low-cost uav, in: IEEE International Conference on Industrial
Technology, Gippsland, Australia, Febrary 10–13, 2009, pp. 1–6.
control law has been applied to the design of the control scheme. [14] J.C. Morris, M. Nieuwstadt, P. Bendotti, Identification and control of a model
The unmeasurable states have been precisely estimated by means helicopter in hover, in: The American Control Conference, Baltimore,
of the reduced-order observer. The steady-state errors of the con- Maryland, USA, June 1994, pp. 1238–1242.
[15] G. Cai, B.M. Chen, K. Peng, M. Dong, T.H. Lee, Design and implementation of a
trol system have been effectively eliminated with the design of nonlinear flight control law for the yaw channel of a uav helicopter, IEEE Trans.
an augmented integral model. The gains of the control and obser- Control Syst. Technol. 55 (9) (2008) 3426–3434.
ver have been optimized by taking the advantage of LQR method. [16] H. Huang, G. Hoffmann, S. Waslander, C. Tomlin, Aerodynamics and control of
autonomous quadrotor helicopters in aggressive maneuvering, in: IEEE
The problems resulting from the unmodeled dynamics, uncertainty
International Conference on Robotics and Automation, Kobe, Japan, May 12–
and external disturbances have been successfully addressed by the 17, 2009, pp. 3277–3282.
integration of sliding mode control approach in the overall control [17] P. Pounds, R. Mahony, P. Corke, Modelling and control of a large quadrotor
robot, Control Eng. Pract. 18 (2010) 691–699.
law design. The simulation results indicate that satisfactory perfor-
[18] G. Hoffmann, H. Huang, S. Waslander, C. Tomlin, Precision flight control for a
mance has been achieved for the stabilization of the quadrotor. The multi-vehicle quadrotor helicopter testbed, Control Eng. Pract. 19 (9) (2011)
designed control scheme has shown substantial improvement and 1023–1036.
superior capabilities over the design based on a single control law. [19] M. Cutler, J. How, Analysis and control of a variable-pitch quadrotor for agile
flight, J. Dyn. Syst. Meas. Control-Trans. ASME 137 (10) (2015), DS-14-1418.
Future work includes the refinement of the developed dynamic [20] P. Bristeau, P. Martin, E. Salaun, N. Petit, The role of propeller aerodynamics in
model of the quadrotor by carrying out specific experiments and the model of a quadrotor uav, in: The European Control Conference, Budapest,
the investigations on the controller design for aggressive Hungary, August 23–26, 2009, pp. 683–688.
[21] R. Mahony, V. Kumar, P. Corke, Multirotor aerial vehicles: modeling,
maneuvers. estimation, and control of quadrotor, IEEE Robot. Autom. Mag. 19 (3) (2012)
20–32.
Acknowledgment [22] J. How, B. Bethke, A. Frank, D. Dale, J. Vian, Real-time indoor autonomous
vehicle test environment, IEEE Control Syst. Mag. 28 (2) (2008) 51–64.
[23] K. Nonami, F. Kendoul, S. Suzuki, W. Wang, Autonomous flying robots:
This work was supported in parts by National Natural Science unmanned aerial vehicles and micro air vehicles, Springer-Verlag, Tokyo, 2010.
Foundation of China (51575544, 51275353), Macao Science and [24] M. Bergerman, O.Amidi, J. Miller, N.Vallidis, T. Dudek, Cascaded position and
heading control of a robotic helicopter, in: The IEEE/RSJ International
Technology development Fund (110/2013/A3, 108/2012/A3), and
Conference on Intelligent Robots and Systems, San Diego, CA, Nov. 2007, pp.
Research Committee of University of Macau (MYRG203(Y1-L4)- 135–140.
FST11-LYM, MYRG2015-00194-FST). [25] K.T. Oner, E. Cetinsoy, E. Sirimoglu, C. Hancer, T. Ayken, M. Unel, Lqr and smc
stabilization of a new unmanned aerial vehicle, in: The World Academy of
Science, Engineering and Technology, 2009, pp. 554–559.
References [26] R. Xu, U. Ozguner, Sliding mode control of a quadrotor helicopter, in: The 45th
IEEE Conference on Decision and Control, San Diego, CA, USA, December 13–
[1] S. Bouabdallah, P. Murrieri, R. Siegwart, Design and control of an indoor 15, 2006, pp. 4957–4962.
quadrotor, in: Proceedings of IEEE International Conference on Robotics and [27] H. Bouadi, M. Bouchoucha, M. Tadjine, Sliding mode control based on
Automation, New Orleans, LA, USA, 2004, pp. 4393–4398, April. backstepping approach for an uav type-quadrotor, Int. J. Mech. Aerospace
[2] D. Mellinger, V. Kumar, Mininum snap trajectory generation and control for Indust. Mechatron. Manuf. Eng. 1 (2) (2007) 39–44.
quadrotors, in: Proceedings of IEEE International Conference on Robotics and [28] B. Walcott, S. Zak, Observation of dynamical systems in the presence of
Automation, Shanghai, China, May 9–13, 2011, pp. 2520–2525. bounded nonlinearities/ uncertainties, in: The 25th Conference on Decision
[3] F. Kendoul, Z. Yu, K. Nonami, Guidance and nonlinear control system for and Control, 1986.
autonomous flight of minirotorcraft unmanned aerial vehicles, J. Field Robot. [29] J. Shin, K. Nonami, D. Fujiwara, K. Hazawa, Model-based optimal attitude and
27 (3) (2010) 311–334. positioning control of small-scale unmanned helicopter, Robotica 23 (2005)
[4] T. Amaral, M. Crisostomo, Automatic helicopter motion control using fuzzy 51–63.
logic, in: Proceedings of the 10th IEEE International Conference on Fuzzy [30] D. Abeywardena, S. Kodagoda, G. Dissanayake, R. Munasinghe, Improved state
Systems, Melbourne, Australia, December 2–5 2001, pp. 860–863. estimation in quadrotor mavs: a novel drift-free velocity estimator, IEEE
[5] S. Salman, V. Puttige, S. Anavatti, Real-time valication and comparison of fuzzy Robot. Autom. Mag. 20 (4) (2013) 32–39.
identification and state-space identification for a uav platform, in: Proceedings [31] J. Leishman, Principles of Helicopter Aerodynamics, Cambridge University
of the IEEE International Conference on Control Applications, Munich, Press, New York, USA, 2000.
Germany, October 4–6, 2006, pp. 2138–2143. [32] R. Prouty, Helicopter Performance, Stability, and Control, Krieger Publishing
[6] O. Purwin, R.D. Andrea, Performing aggressive maneuvers using iterative Company, Florida, USA, 1995.
learning control, in: IEEE International Conference on Robotics and [33] C. Chen, Linear System Theory and Design, Oxford University Press, New York,
Automation, Kobe, Japan, May 2009, pp. 1731–1736. 1995.
[7] P. Abbeel, A. Coates, A.Y. Ng, Autonomous helicopter aerobatics through [34] J. Burl, Linear Optimal Control, Addison Wesley Longman. Inc., 1998.
apprenticeship learning, Int. J. Robot. Res. 29 (13) (2010) 1608–1639.

You might also like