You are on page 1of 6

2017 25th Mediterranean Conference on Control and Automation (MED)

July 3-6, 2017. Valletta, Malta

Simple Adaptive Control of Quadrotor Attitude. Algorithms and


Experimental Results*
Stanislav I. Tomashevich1,2 , Alexander L. Fradkov1,2,3 , Boris Andrievsky1,2,3 ,
Anderey O. Belyavskyi1 , and Konstantin Amelin3

Abstract— In the paper the simple adaptive control approach an autonomous quadrotor helicopter: feedback linearization
is employed to designing the adaptive controllers of quadrotor controller and an adaptive sliding mode controller using
angular motion. The adaptive controllers are synthesized based input augmentation in order to account for the underactuated
on the Implicit Reference Model (IRM) design technique.
The “shunting method” (parallel feedforward compensation) property of the helicopter. It is demonstrated that the sliding
is used to cope with the unmodelled plant dynamics. Analytical mode controller is efficient in the case of ground effects.
justification of system stability is made by employing the Passifi-
cation method. Quality of the closed-loop IRM adaptive control Designing a nonlinear controller for the helicopter vertical
system is studied and compared with that of the proportional- motion is given in [8]. TThe proposed controller combines
derivative (PD) non-adaptive control system experimentally on
two degrees of freedom (2DOF) quadrotor testbed and in-flight adaptive output control and robust stabilization of systems in
tests of the QuadRoy quadrotor for nominal and parametrically feedforward form by means of saturated controlling actions.
perturbed cases. In [9] two-input, two-output nonlinear adaptive model refer-
ence control of a 3-DOF tandem rotor model helicopter is
I. INTRODUCTION describedand the results of the real time implementation of
In the last decade, an interest significantly increased in the control algorithms in a helicopter testbed are presented.
the use of mini and micro UAVs of the helicopter type
(multicopters, quadrocopters, etc.) in a variety of areas, such The aforementioned approaches mainly lead to rather com-
as the surface monitoring, environmental inspection, road plex adaptive control laws. In the present work, the simple
traffic and atmospheric flows investigation, safeguarding, adaptive control (SAC) method combined with “shunting”
cargo transportation, and many others. In many missions, (parallel feedforward compensation) is utilized for attitude
the UAV parameters are subjected to significant variations. control of the quadroror. We deal with the ultralight UAV,
This takes place, for example, due to the load variation which is a part of the research complex QuadRoy (Swarm of
from one flight to another, or may happen during the UAV Quadrotors), see [10]–[12], which has been developed in the
flight; presence of cranes or movable manipulators, which are IPME RAS. The UAV is supplied with autopilot Ardupilot
mounted on the UAV, also leads to a substantial change of the Mega 2.6, the Inertial Navigation System (including gy-
UAV model parameters. As a consequence, a considerable roscope, accelerometer, magnetometer, barometer), and the
number of works devoted to the application of adaptation GPS receiver.
methods for designing the helicopter-type mini-UAVs control
systems have appeared in the recent years. The SAC approach is being intensively developed since
In the series of papers, [1]–[4], the attitude command the 1970s, see [13]–[22], for mentioning a few. Among
system for R-50 helicopter model, based on the approach the advantages of this method with respect to the habitual
of [1] for adaptive output feedback control, is designed model reference adaptive control (MRAC, [23]) technique
and experimentally tested by controlling the pitch axis of a are the low order and computational simplicity of the control
three degrees-of-freedom (DOF) helicopter model. A back- algorithms and relaxation of the restrictions imposed by the
stepping approach is used in [5] for quadrotor controller so-called matching (Erzberger) condition, see [23], [24]. The
design. The Lagrangian formulation is also used in [6] for SAC method has been successfully applied to the various
a 2DOF system, formed from a small-scale helicopter. The adaptive control problems, such as control of power systems
controller is designed by a pole-placement techniques. The [25], robotic systems [26], [27], rigid and flexible spacecrafts
theoretical consideration is experimentally validated. The [28]–[33], atmospheric vehicles [14], [34]–[37], fault tolerant
authors of [7] represent two kinds of nonlinear controllers for control [38], [39], etc.

*This work was performed in the IPME RAS and supported solely by The rest of the paper is organized as follows. Firstly, the
the Russian Science Foundation (grant 14-29-00142).
1 ITMO University, Saint Petersburg, Russia tomashevich. SAC scheme based on the passification method is briefly
stanislav@gmail.com, belyavskyi.a.o@gmail.com recalled in Sec. II. The SAC laws for control of quadrotor
2 Institute of Problems in Mechanical Engineering, the Russian
attitude are derived in Sec. III. The experimental laboratoty
Academy of Sciences, Saint Petersburg, Russia fradkov@mail.ru; and in-flight results of testing the quadrotor attitude control
boris.andrievsky@gmail.com
3 Saint Petersburg State University, Saint Petersburg, Russia system are given in Sec. IV. Concluding remarks and the
konstantinamelin@gmail.com future work intentions finalize the paper.

978-1-5090-4533-4/17/$31.00 ©2017 IEEE 933


II. ADAPTIVE CONTROLLERS WITH IMPLICIT vector G = [g0 , g1 , . . . , 1] ∈ Rl+1 and plant (1) transfer
REFERENCE MODEL function W (s) from input u to output vector [y, ẏ, . . . y (l) ]T ∈
B(s)  T
A simple design tool for SAC design is the Passification Rl+1 as W (s) = 1, s, . . . , sl , s ∈ C in virtue of
A(s)
method, see [13], [17], [40] for more details. The main con- passification theorem [13], [17] with respect to transfer func-
dition to use this method lies in the so-called hyper minimum tion GW (s) one may easily obtain the following stability
phase (HMP) property, imposed on the plant model: a linear conditions of adaptive controller (4), (5):
time invariant system ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t), 1) polynomial B(s) is Hurwitz and b0 > 0;
where x(t) ∈ Rn , u(t) ∈ Rm , y(t) ∈ Rm , A, B, C are 2) l = k−1, where k = n−m is a relative degree of plant
constant real matrices of appropriate dimensions, is called
 model (1).
sIn −A −B
HMP if the polynomial ϕ0 (s) = det (In Algorithm (5) usually ensures vanishing σ(t) essentially
C 0
is (n×n) identity matrix) is Hurwitz and CB = (CB) > 0. T faster than transients in the closed-loop. As a result, changing
Consider some particular cases. the controller (4) gains is stopped and plant (1) output y(t)
obeys the IRM (3).
A. Adaptive Stabilization of LTI SISO Plants The following α-modification of (5) is often used
 
Let LTI SISO plant be modeled as k̇i (t) = −γσ(t)pi y(t) − α ki (t) − ki0 , ki0 = ki (0), (6)
A(p)y(t) = B(p)u(t), t  0, (1) where α  0 is the parametric feedback gain, for avoiding
1 1 unlimited growth of controller (4) gains in the presence of
where u ∈ R , y ∈ R are input and output variables,
external disturbances and measurement errors, [17], [41].
A(p) = pn +an−1 pn−1 +· · ·+a0 , B(p) = bm pm +· · ·+b0 are
polynomials in operator of differentiation on time p ≡ d/dt. B. Adaptive Tracking Systems with IRM
Define k as the relative degree of system (1), k = n−m > 0. Adaptive control law (4), (6) may be straightforwardly
Plant (1) parameters ai , bj (i = 0,. . . , n − 1, j = 1, . . . , m) extended to the solving the tracking problem with the desired
are assumed to be unknown. Desired closed-loop system closed-loop system dynamics, see [17]. To this end let us
behavior may be expressed by means of a “reference” introduce reference signal r(t) and define adaptation error
differential equation. In the classical MRAC this equation is signal σ(t) as
explicitly implemented in the adaptive controller in the form
of a “Reference Model”, cf. [23]. Introduce an adaptation σ(t) = G(p)y(t) − D(p)r(t), (7)
error signal σ(t) as where polynomial G(p) is defined above, and operator poly-
nomial D(p) has the form D(p) = dq pq + · · · + d1 p + d0 .
σ(t) = G(p)y(t), (2)
Evidently, if σ(t) ≡ 0, then closed-loop system is governed
l
where G(p) = p + gl−1 p + · · · + g0 is a given Hurwitz
l−1 by the following “reference model” equation:
polynomial in operator p ≡ d/dt. Coefficients gi are the G(p)y(t) = D(p)r(t), (8)
design parameters; they are chosen based on the desired
dynamics of the closed-loop system. Degree l of polynomial Therefore, (8) may be treated as the IRM for the tracking
G(p) is defined below. Assuming that the adaptation law system case and σ(t) may be treated as the discrepancy in
ensures tendency σ(t) to zero let us notice that as σ ≡ 0, the equation and used for parameters adjustment.
output y(t) satisfies the “reference equation” as By the analogy with (4) let us take the control action in
the form
G(p)y(t) = 0. (3)
   l
 
(3) describes the reference model, but this model is not u(t) = kr (t) D(p)r(t) + ki (t) pi y(t) , (9)
i=0
implemented in the adaptive controller in the form of a
certain dynamical subsystem, but introduced implicitly via where kr (t), ki (t) (i = 0, . . . , l) are tunable parameters, and
its parameters gi (i = 0, 1, . . . , l − 1). Therefore it is called the adaptation law reads as
 
the Implicit Reference Model (IRM). k̇r (t) = γσ(t)D(p)r(t)−α kr (t)−kr0 , kr0 = kr (0),
(10)
Consider the following feedback control law: k̇i (t) = −γσ(t)pi y(t) − α ki (t) − ki0 , ki0 = ki (0),

l
  where γ > 0, α  0 are design parameters; kr0 , ki0 are
u(t) = ki (t) pi y(t) , (4) “guessed” initial controller gain values, i = 0, . . . , l. It worth
i=0
mentioning that both degree q of polynomial D(p) and its
where ki (t) i = 0, . . . , l are adjustable controller parameters. coefficients may be chosen arbitrarily by the designer. One
For the considered case the HMP property leads to the can mention that an only one gain kr may be used despite
following adaptation law, see [13]: using the separate gains for the output y and each of its
derivatives. Potentially, the same may be applicable for r(t)
k̇i (t) = −γσ(t)pi y(t), ki (0) = ki0 , (5)
too, but this seems to be useless, because the closed-loop
where γ > 0 is the adaptation gain, ki0
are given initial system stability in the steady-state mode depends on ki , not
values of the controller gains, i = 0, . . . , l. Introducing row on k̇r .

934
C. Signal-Parametric Adaptive Controllers with IRM A. Quadrotor Model
In the present paper the following model of the quadrotor
Let the regulation goal lim x(t) = 0 for plant
t→∞ rotational motion dynamics is utilized, cf. [22]:


⎪ γ̇ = ωx + Sγ Tϑ ωz + Cγ Tϑ ωy ,
ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t), (11) ⎪


⎪ ϑ̇ = Cγ ωz − Sγ ωy ,

⎨ S C
where x(t) ∈ Rn , u(t) ∈ Rm , y(t) ∈ Rm , A, B, C are ψ̇ = Cγ ωz + Cγ ωy , (15)
ϑ ϑ


constant real matrices of appropriate dimensions, be posed. ⎪ Ix ω̇x = (Iy − Iz )ωy ωz − Ir ωz ωr + τγ ,


⎪ I ω̇ = (Iz − Ix )ωx ωz + τψ ,
Following [17] introduce an auxiliary goal as ensuring the ⎪
⎩ y y
sliding mode motion along the predefined surface (line in the Iz ω̇z = (Ix − Iy )ωx ωy + Ir ωx ωr + τϑ ,
scalar control case), which is given by the equality σ(t) ≡ 0, where γ, ϑ, ψ denote the Euler angles (roll, pitch and
where σ(t) = Gy(t), G is a given (l × n)-matrix. Let us use yaw, respectively); ωx , ωy , ωz are angular rates in body-
the following control law axis frame, Sξ = sin(ξ), Cξ = cos(ξ), Tξ = tan(ξ); g is
the gravity acceleration; Ix , Iy , Iz are rotational moments
u = −γsign σ, σ = Gy, (12) of inertia. Rotating
 T torque with respect to CoG is a vector
τ = τγ , τϑ , τψ with components τγ = l(ω2 − ω4 ), τϑ =
where γ > 0 is a controller parameter. It may be proven l(ω1 −ω3 ), τψ = ar ωR , where ωR = ω1 −ω2 +ω3 −ω4 . Motor
that for system (11), (12) the posed control goal may be rotation velocity is as ω̇i = kl (ω̃i − ωi ), where kl is motor
achieved if there exist matrix P = P T > 0 and vector K∗ parameter, ω̃i input reference value. Rotation velocities are
s.t. P A∗ + A∗ T P < 0, P B = GC, A∗ = A + BK∗ T C. limited by inequalities: 0 < ωmin  ωi  ωmax . Since kl is
As follows from the passification theorem, the mentioned large, the engine inertia is omitted in the sequel.
 iff: transfer function GW (s) is HMP
conditions are fulfilled
B. Adaptive Control Laws for Quadrotor Attitude
(where W (s) = C λIn − A)−1 B); the sign of GCB is
known (we assume that it is positive). Under these conditions Assuming that angular velocities ωx , ωy , ωz are small one
the goal lim x(t) = 0 is achieved for sufficiently large may obtain the linearized model of quadrotor angular motion
t→∞
γ (with respect to the initial conditions and actual plant in the vicinity of zero. In this model the yaw motion may
parameters). be considered separately of the roll and pitch motions and
modeled by the following equation
The following “signal-parametric” (“combined”) control
law may be used instead of (12): ψ̈ = Iy−1 τψ , (16)
 T
or, in the state-space form with state vector x̃ψ = ψ, ψ̇
u = K T (t)y(t) − γ sign σ(y), σ(y) = G y(t)
as
K̇(t) = −σ(y)Γy(t) − α(K(t) − K0 ), K(0) = K0 , (13)      
0 1 0 1 0
x̃˙ ψ = x̃ψ + τ , ỹ = x̃ψ .
0 0 Iy−1 ψ ψ
0 1
(12) where Γ = ΓT > 0, γ > 0, α > 0 are the matrix (17)
and scalar adaptation gains, K0 is some “guessed” value Following the aforementioned IRM design technique, let us
 
of K (design parameters), avoiding dependence of closed- introduce vector Gψ as Gψ = 1, αψ where αψ > 0 is a
loop system stability on initial conditions and plant model design parameter. This leads to the following algorithm for
parameters. A rigorous justification of modification of (12) yaw control:
into (13) may be found in [17], [42].
σ(t) = ψ(t) − ψ ∗ (t) + αψ ψ̇(t),

For the case of scalar control input the following control

law, inspired by [43], [44], may be used instead of (13): uψ (t) = −kψ (t)σ(t)−γσ sign(σ(t))
σ(t)
, (18)
 

k̇ψ (t) = γk σ(t)2 −αk kψ (t)−kψ0 , kψ (0) = kψ0

u = −k(t)σ(y)−γσ sign(σ(y))
σ(y)
, σ(y) = Gy, where ψ ∗ (t) denotes the yaw reference signal.
k̇(t) = γk σ(t)2 . (14) To cope with unmodelled actuator dynamics the shunt
(parallel feedforward compensator) is added to the control
law (18), cf. [15], [45]. This leads to the following control
This law produces more smooth control action than the
law:
“relay” one.
σ(t) = ψ(t) − ψ ∗ (t) + αψ ψ̇(t) − v(t),

uψ (t) = −kψ (t)σ(t) − γσ sign(σ(t))


σ(t)
,
III. SIMPLE ADAPTIVE CONTROL OF   (19)
QUADROTOR ATTITUDE k̇ψ (t) = γk σ(t)2 − αk kψ (t) − kψ0 , kψ (0) = kψ0
v̇(t) = κu(t) − v(t) τ −1 ,
Let us apply the IRM technique for designing the adaptive where κ and τ are the auxiliary feedback gains and κ is
attitude control algorithms for quadrotor. sufficiently small, see [45] for rigorous statements.

935
Linearized pitch (ϑ) and roll (γ) angular motion model angle has been used:
may be written in the following form:
σγ [n] = γ ∗ [n] − γ[n] − αγ ωx [n] − v[n],
⎡ ⎤
0 −Ir Ix−1 ωR 0 0 kγ [n + 1] = (γk σγ2 [n] − αk (kγ [n] − kγ0
))T0 + kγ [n],
⎢ Ir Ix−1 ωR (24)
0 0 0 ⎥ uγ [n] = kγ [n]σγ [n] − γσγ sign(σγ [n]) |σγ [n]|,
ỹ˙ γϑ = ⎢

⎥ ỹ
1 0 0 0 ⎦ γϑ vγ [n + 1] = (κuγ [n] − vγ [n])T0 τ −1 + vγ [n],
⎡ −10 ⎤ 1 0 0 where n = 0, 1, . . . is the step number, T0 denotes the
Ix 0   sampling time, γ ∗ is the roll reference signal, measured
⎢ 0 Ix−1 ⎥
+⎢ ⎥ τγ , output variables γ(t), ωx (t) are sampled at instants nT0 :
⎣ 0 0 ⎦ τϑ γ[n] = γ(nT0 ), ωx [n] = ωx (nT0 ), control action uγ (t),
0 0 zero-order extrapolation is used to form the actuating motor
ỹγϑ = [ωx , ωz , γ, ϑ]T . input signal: uγ (t) = uγ (nT0 ) as nT0  t < (n + 1)T0 . The
This model represents the MIMO system of fourth order similar discrete-time control law is implemented at the pitch
with two control inputs and two outputs. Mutual interaction control loop:
between motions on pitch and roll is caused by presence of σϑ [n] = ϑ∗ [n] − ϑ[n] − αϑ ωz [n] − v[n],
ωR and may considerably vary during the flight. kϑ [n + 1] = (ϑk σϑ2 [n] − αk (kϑ [n] − kϑ0
))T0 + kϑ [n],
Application of the aforementioned approach to the MIMO (25)
uϑ [n] = kϑ [n]σϑ [n] − γσϑ sign(σϑ [n]) |σϑ [n]|,
case, see [40], leads to the following adaptive control law vϑ [n + 1] = (κuϑ [n] − vϑ [n])T0 τ −1 + vϑ [n],
[22]:
where ϑ∗ is the pitch reference signal, ϑ[n] = ϑ(nT0 ),
σ(ỹ) = Gγϑ ỹ,  ωz [n] = ωz (nT0 ), uϑ (t) = uϑ (nT0 ) as nT0  t < (n+1)T0 .
ũ(t) = ξ T (t)ỹ(t) − γM sign (σ(ỹ)) |σ(ỹ)|, (20) Validity of discretized versions of the adaptive control algo-
ξ˙j = −(Gj ỹ)Γỹ − αM (ξj − ξj0 ), ξj0 = ξj (0), rithms with the parametric feedback follows from the method
  of continuous models: analysis of the hybrid nonlinear system
1 0 αγ 0 via analysis of its continuous-time approximate model [46].
with Gγϑ = and j is a number of the
0 1 0 αϑ
corresponding row or column of the matrices. Taking into IV. EXPERIMENTAL RESULTS
account an expression of matrix Gγϑ one could rewrite (20)
A series of experiments have been fulfilled for examining
componentwise:
the closed-loop adaptive control system performance. The
σγ (ỹ) = ωx + αγ γ, following parameters of adaptive controller (24) have been
σϑ (ỹ) = ωz + αϑ ϑ, taken: aφ = 0.3, γσ = 20, γk = 0.5, kψ0 = kγ0 = 30, ku =

ũγ (t) = kγT (t)ỹ(t) − γM sign (σγ (ỹ)) |σγ (ỹ)|, 0.05, αu = 10, κ = 0.05, τ = 10.
(21)
ũϑ (t) = kϑT (t)ỹ(t) − γM sign (σϑ (ỹ)) |σϑ (ỹ)|, A. Experiments on 2-DOF Quadrotor Testbed
ξ˙γ = −σγ (ỹ)Γγ ỹ − αM (kγ − kγ0 ), kγ = ξγ (0),
Several experiments have been accomplished with the help
k̇ϑ = −σϑ (ỹ)Γϑ ỹ − αM (kϑ − kϑ0 ), kϑ = kϑ (0). of the 2DOF Quadrotor Laboratory Setup, described in [22],
The transfer matrix has a form see also an accompanying paper [47] for more details.
1) Design of the setup.: The mechanical part of the 2DOF
Pγϑ (s) = Gγϑ Cγϑ(sI4 − Aγϑ )−1 Bγϑ  laboratory setup consists of two-axis gimbal suspension. The
1 Ix s2 +Ix αγ s −Ir ωR (s+αγ ) (22) square frame is supported by two pillars and freely rotates
= .
s(Ix2 s2 +Ir2 ωR
2
) Ir ωR (s+αϑ ) Ix s2 +Ix αϑ s around the horizontal axis. The pillars are fixed on the
Polynomial δ(s) reads as δ(s) = det (sI4 − Aγϑ ) = square base. The quadrotor is mounted on the bar inside
Ix (Ix s + Ir2 ωR
−2 2 3 2
s)s, threfore polynomial δ(s)detPγϑ = the frame, and may freely rotate on the pitch and roll
Ix (s + (αγ + αϑ )s + αϑ αγ ) is Hurwitz. Let us calculate angles. Quadrotor is based on DJI 450 frame and autopilot
−2 2

 −1  Ardupilot Mega 2.6 with inertial measuring unit (including


Ix 0 gyroscope, accelerometer, and barometer) and GPS satellite
ΓP = lim sPγϑ = GCB = −1 , (23)
|s|→∞ 0 Iy navigation unit. The actuators includes the brushless motors
with external rotor DJI 2213, speed controllers DJI 30A
which fulfills the required condition Γ = ΓT > 0, therefore Opto. The real-time data may be transmitted to the operator’s
matrix GW (s) is HMP and satisfies conditions of passifica- personal computer via the XBee modems.
tion theorem for the signal-parametric adaptive controller. 2) Experimental results on 2DOF Quadrotor setup.:
Experimental evaluation of the proposed adaptive control Experimentally obtained on the 2DOF Quadrotor Laboratory
laws on the laboratory setup is presented in the next Section. Setup performance of the system with adaptive controller
has been compared with that of the time invariant standard
C. Discrete-time adaptive control laws PD controller ui (t) = kp yi∗ (t) − yi (t) − kd ẏi (t), where
For implementation on the on-board controller of Arduino yi∗ , yi denote corresponding reference signal and quadrotor
autopilot, the following discrete-time control law for roll angular variable (i ∈ {ϑ, γ}). The controller gains was

936
calculated for the nominal case as kp = 40, kd = 10 s. planned to make more in-flight experiments for testing the
Time histories of tracking the piecewise constant reference quadrotor adaptive control system.
signal the controlled angles, the tracking errors and the
γ, deg
adaptation gain for nominal and parametrically disturbed
20
cases are plotted in Fig. 1 (the nominal case) and Fig. 2
0
(the perturbed case).
−20
ϑ , ϑ , ϑ*, deg
ad pd 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
50 ϑ, deg
20

0 10

ϑ ϑ ϑ* 0
ad pd
−50
0 5 10 15 20 25 30 35 −10
Δϑ , Δϑ , deg
ad pd
40 −20
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
ψ, deg
20
0 300

−20
Δϑ Δϑ 0 250
−40 ad pd

0 5 10 15 20 25 30 35
t, s 200

60 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
t, s

40 Fig. 3. Time histories of roll, pitch and yaw angles during the flight.
20

0 kγ
0 5 10 15 t, s 20 25 30 35
200
Fig. 1. Nominal case. Upper plot: pitch angle for adaptive – ϑad , non- 150
adaptive PD – ϑpd controllers and pitch reference signal ϑ∗ time histories. 100
Middle plot: pitch tracking error for adaptive – Δϑad , non-adaptive PD –
50
Δϑad controllers. Lower plot: adaptive controller gain kϑ time history.
0
2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7

60

ϑad, ϑpd, ϑ*, deg


40
ϑad ϑpd ϑ*
100 20
50
0
0 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
t, s
−50
5 10 15 20 25 30 35 40 45 50 55 60 Fig. 4. Time histories of adaptive controller gains kγ (t), kϑ (t) during the
Δϑad, Δϑpd, deg flight.
50
0
−50
−100 V. CONCLUSION
Δϑad Δϑpd 0
−150
5 10 15 20 25 30 35 40 45 50 55 60
Simple adaptive control technique based on the passifi-
k
ϑ cation method and the Implicit Reference Model approach
60
is used for design of the adaptive controller for quadrotor
40
attitude. The experimental study has been performed on the
20

0
laboratory setup, created in the IPME RAS. The experiments
5 10 15 20 25 30 t, s 35 40 45 50 55 60
demonstrate good performance quality and high adaptation
Fig. 2. Parametrically perturbed case. Upper plot: pitch angle for adaptive
– ϑad , non-adaptive PD – ϑpd controllers and pitch reference signal ϑ∗
rate of the SAC system. In the future it is planned to extend
time histories. Middle plot: pitch tracking error for adaptive – Δϑad , non- in-flight testing and the comparative analysis of the quadrotor
adaptive PD – Δϑad controllers. Lower plot: adaptive controller gain kϑ attitude control algorithms, to implement on-board adaptive
time history.
control of quadrotor position and to fulfill the experiments
with decentralized adaptive control of a group of quadrotors,
B. In-Flight Quadrotor Experiments flying in the formation.
In-Flight experiments have been carried out with the same R EFERENCES
quadrotor which has been used on the laboratory setup
[1] A. Calise, N. Hovakimyan, and M. Idan, “Adaptive output feedback
and the same adaptive controller parameters. The adaptive control of nonlinear systems using neural networks,” Automatica,
attitude controller saved as an inner part (the attitude control vol. 37, no. 8, pp. 1201–1211, 2001.
loop) of the non-adaptive controller for quadrotor position. [2] N. Hovakimyan, F. Nardi, A. Calise, and N. Kim, “Adaptive output
feedback control of uncertain nonlinear systems using single-hidden-
The desired flight path has been manually assigned on-line layer neural networks,” IEEE Trans. Neural Networks, vol. 13, no. 6,
by the operator. The corresponding time histories of the roll, pp. 1420–1431, 2002.
pitch and yaw angles, obtained from the quadrotor IMU [3] A. Kutay, A. Calise, M. Idan, and N. Hovakimyan, “Experimental
results on adaptive output feedback control using a laboratory model
along with the time histories of the adaptive controller gains helicopter,” in Proc. AIAA Guidance, Navigation, and Control Conf.,
kγ (t), kϑ (t) are depicted in Figs. 3, 4. In the future it is Monterey, California, USA, August 5–8 2002.

937
[4] ——, “Experimental results on adaptive output feedback control using [27] A. Pyrkin, A. Bobtsov, S. Kolyubin, M. Faronov, O. Borisov, V. Gro-
a laboratory model helicopter,” IEEE Trans. Contr. Syst. Technol., mov, S. Vlasov, and N. Nikolaev, “Simple robust and adaptive track-
vol. 13, no. 2, pp. 196–202, 2005. ing control for mobile robots,” IFAC Proceedings Volumes (IFAC-
[5] A. Das, F. Lewis, and K. Subbarao, “Backstepping approach for PapersOnline), vol. 48, no. 11, pp. 143–149, 2015.
controlling a quadrotor using Lagrange form dynamics,” J. Intelligent [28] O. Egeland and J. Godhavn, “Passivity-based adaptive attitude control
and Robotic Systems: Theory and Applications, vol. 56, no. 1–2, pp. of a rigid spacecraft,” IEEE Trans. Automat. Contr., vol. 39, no. 4, pp.
127–151, 2009. 842–846, 1994.
[6] A. Dzul, R. Lozano, and P. Castillo, “Adaptive control for a radio- [29] S. Ulrich, J. Sasiadek, and I. Barkana, “Nonlinear adaptive output feed-
controlled helicopter in a vertical flying stand,” Int. J. Adaptive Control back control of flexible-joint space manipulators with joint stiffness
and Signal Processing, vol. 18, no. 5, pp. 473–485, 2004. uncertainties,” J. Guidance, Control, and Dynamics, vol. 37, no. 6, pp.
[7] D. Lee, H. Kim, and S. Sastry, “Feedback linearization vs. adaptive 1961–1975, 2014.
sliding mode control for a quadrotor helicopter,” Int. J. Control, [30] C. Pittet, A. Luzi, D. Peaucelle, J. Biannic, and J. Mignot, “In-flight
Automation and Systems, vol. 7, no. 3, pp. 419–428, 2009. results of adaptive attitude control law for a microsatellite,” CEAS
[8] A. Isidori, L. Marconi, and A. Serrani, “Robust nonlinear motion Space Journal, vol. 7, no. 2, pp. 291–302, 2015.
control of a helicopter,” IEEE Trans. Automat. Contr., vol. 48, no. 3, [31] S. Ulrich, A. Saenz-Otero, and I. Barkana, “Passivity-based adaptive
pp. 413–426, 2003. control of robotic spacecraft for proximity operations under uncertain-
[9] M. Ishitobi, M. Nishi, and K. Nakasaki, “Nonlinear adaptive model ties,” J. Guidance, Control, and Dynamics, vol. 39, no. 6, pp. 1441–
following control for a 3-dof tandem-rotor model helicopter,” Control 1450, 2016.
Engineering Practice, vol. 18, no. 8, pp. 936 – 943, 2010. [32] I. Barkana, “Parallel feedforward and simple adaptive control of
[10] K. Amelin, S. Tomashevich, and B. Andrievsky, “Recursive identi- flexible structures: First-order pole instead of collocated velocity
fication of motion model parameters for ultralight UAV,” in IFAC sensors?” J. Aerospace Engineering, vol. 29, no. 2, 2016.
Proceedings Volumes (IFAC-PapersOnline). IFAC, 2015, pp. 7–12. [33] H. Leduc, D. Peaucelle, and C. Pittet-Mechin, “LMI-based design of
[11] S. Tomashevich and B. Andrievsky, “Stability and performance of a robust direct adaptive attitude control for a satellite with uncertain
networked control of quadrocopters formation flight,” in Proc. 6th Int. parameters,” in Proc. 20th IFAC Symposium on Automatic Control in
Congress on Ultra Modern Telecommunications and Control Systems Aerospace (ACA 2016). Sherbrooke, Canada: IFAC, Aug. 2016.
and Workshops (ICUMT 2014). St. Petersburg, Russia: IEEE Press, [34] I. Yaesh and U. Shaked, “Simplified adaptive control via improved
October 6–8 2014, pp. 331–336. robust positive real conditions,” Systems & Control Letters, vol. 55,
[12] B. Andrievsky and S. Tomashevich, “Passification based signal- pp. 1033–1040, 2006.
parametric adaptive controller for agents in formation,” IFAC- [35] A. Fradkov, B. Andrievsky, and D. Peaucelle, “Adaptive control design
PapersOnLine, vol. 48, no. 11, pp. 222 – 226, 2015. and experiments for laas ”helicopter” benchmark,” European Journal
[13] A. Fradkov, “Synthesis of adaptive system of stabilization of linear of Control, vol. 14, no. 4, pp. 329–339, 2008.
dynamic plants,” Automation and Remote Control, vol. 35, no. 12, pp. [36] I. Rusnak, H. Weiss, and I. Barkana, “Improving the performance
1960–1966, 1974. of existing missile autopilot using simple adaptive control,” Int. J.
[14] K. Sobel, H. Kaufman, and I. Mabius, “Implicit adaptive control for a Adaptive Control and Signal Processing, vol. 28, no. 7-8, pp. 732–
class of MIMO systems,” IEEE Trans. Aerosp. Electron. Syst., vol. 18, 749, 2014.
pp. 576–589, 1982. [37] F. Chen, Q. Wu, B. Jiang, and G. Tao, “A reconfiguration scheme for
[15] I. Barkana, “Parallel feedforward and simplified adaptive control,” Int quadrotor helicopter via simple adaptive control and quantum logic,”
J. Adaptive Control and Signal Processing, vol. 1, no. 2, pp. 95–109, IEEE Trans. Ind. Electron., vol. 62, no. 7, pp. 4328–4335, 2015.
1987. [38] A. L. Fradkov, B. Andrievsky, and D. Peaucelle, “Adaptive
[16] Z. Iwai and I. Mizumoto, “Robust and simple adaptive control sys- passification-based fault-tolerant flight control,” IFAC Proceedings
tems,” Int. J. Control, vol. 55, pp. 1453–1470, 1992. Volumes, vol. 40, no. 7, pp. 715 – 720, 2007.
[17] B. R. Andrievskii and A. L. Fradkov, “Method of passification in [39] Q. Lam and I. Barkana, “Direct adaptive control treatment to loss of
adaptive control, estimation, and synchronization,” Autom. Remote attitude control actuators,” in Proc. AIAA Guidance, Navigation, and
Control, vol. 67, no. 11, pp. 1699–1731, 2006. Control Conference 2007, vol. 2, Hilton Head, SC, 2007, pp. 1319–
[18] D. Peaucelle and A. Fradkov, “Robust adaptive l2 -gain control of 1332.
polytopic MIMO LTI systems – LMI results,” Systems & Control [40] A. L. Fradkov, “Passification of nonsquare linear systems and
Letters, vol. 57, no. 11, pp. 881–887, 2008. Yakubovich-Kalman-Popov lemma,” Eur. J. Control, vol. 6, pp. 573–
582, 2003.
[19] A. Bobtsov, A. Pyrkin, and S. Kolyubin, “Simple output feedback
[41] P. Ioannou and P. Kokotovic, “Instability analysis and improvement of
adaptive control based on passification principle,” Int J. Adaptive
robustness of adaptive control,” Automatica, vol. 20, no. 5, pp. 583–
Control and Signal Processing, vol. 28, no. 7–8, pp. 620–632, 2014.
594, 1984.
[20] I. Barkana, “Adaptive control? but is so simple!: A tribute to the
[42] B. R. Andrievskii, A. N. Churilov, and A. L. Fradkov, “Feedback
efficiency, simplicity and beauty of adaptive control,” J. Intelligent
Kalman–Yakubovich lemma and its applications to adaptive control,”
and Robotic Systems: Theory and Applications, vol. 83, no. 1, pp.
in Proc. 35th IEEE Conf. Dec. Control. Kobe, Japan: IEEE Press,
3–34, 2016.
Dec. 11–13 1996, pp. 4537–4542.
[21] O. Borisov, A. Bobtsov, A. Pyrkin, and V. Gromov, “Simple adaptive
[43] Y. Shtessel, C. Edwards, L. Fridman, and A. Levant, Sliding Mode
control for quadcopters with saturated actuators,” in AIP Conf. Proc.
Control and Observation. Series: Control Engineering. New York,
World Congress: 11th Int. Conf. Mathematical Problems in Engineer-
Heidelberg, Dordrecht, London: Birkhäuser, 2012.
ing, Aerospace and Sciences (ICNPAA 2016), La Rochelle, France,
[44] A. Levant, A. Pridor, R. Gitizadeh, I. Yaesh, and J. Ben-Asher,
July 5–8, 2016. Melville, NY, USA: AIP, 2017, pp. 020 031–10.
“Aircraft pitch control via second order sliding technique,” AIAA
[22] S. Tomashevich, A. Belyavskyi, and B. Andrievsky, “Passification
Journal of Guidance, Control and Dynamics, vol. 23, no. 4, pp. 586–
based simple adaptive control of quadrotor attitude: algorithms and
594, 2000.
testbed results,” in AIP Conf. Proc. World Congress: 11th Int. Conf.
[45] A. Fradkov, “Adaptive stabilization of the minimum phase plants with
Mathematical Problems in Engineering, Aerospace and Sciences (IC-
vector input without measuring the output derivatives,” Dokl. Rossisk.
NPAA 2016), La Rochelle, France, July 5–8, 2016. Melville, NY,
Akad. Nauk, vol. 337, no. 5, pp. 592–594, 1994.
USA: AIP, 2017, pp. 020 161–13.
[46] D. P. Derevitsky and A. L. Fradkov, “Two models for analyzing
[23] Y. D. Landau, Adaptive control: The model reference approach. New dynamics of adaptation algorithms,” Autom. Remote Control, vol. 35,
York: Marcel Dekker, 1979. no. 1, pp. 59–67, 1974.
[24] H. K. Khalil, Nonlinear Systems. New York: Macmillan, 1992. [47] S. Tomashevich, A. Fradkov, B. Andrievsky, A. Belyavskyi, and
[25] S. Zhang and F. Luo, “An improved simple adaptive control applied to K. Amelin, “Simple adaptive control of quadrotor attitude. Algorithms
power system stabilizer,” IEEE Trans. Power Electron., vol. 24, no. 2, and experimental results,” in Proc. 25th Mediterranean Conf. on
pp. 369–375, 2009. Control and Automation (MED 2017), Valletta, Malta, 3–6 July 2017.
[26] B. Park, S. Yoo, J. Park, and Y. Choi, “A simple adaptive control
approach for trajectory tracking of electrically driven nonholonomic
mobile robots,” IEEE Trans. Contr. Syst. Technol., vol. 18, no. 5, pp.
1199–1206, 2010.

938

You might also like