Proceedings of the 19th World Congress

The International Federation of Automatic Control

Cape Town, South Africa. August 24-29, 2014

Adaptive Sliding Backstepping Control of

Quadrotor UAV Attitude
Tinashe Chingozha ∗ Otis Nyandoro ∗∗
∗
University of the Witwatersrand, Johannesburg, South Africa
(e-mail: [email protected]
∗∗
University of the Witwatersrand, Johannesburg,South
Africa(e-mail:[email protected] )

Abstract: This paper proposes an adaptive sliding backstepping control law for quadcopter
attitude control. By employing adaptive elements in the sliding mode control formulation the
proposed control law avoids a priori knowledge of the upper bounds on the uncertainty. The
controller we propose can be used for systems that are in strict feedback form with matched
uncertainties. Numerical simulations show that this control method is capable of guaranteeing
global asymptotic tracking of the desired attitude trajectory.

Keywords: adaptive control, backstepping control, sliding mode control

1. INTRODUCTION the qUAV easier to maintain and control in comparison to

other rotary UAVs.
The history of unmanned aerial vehicles(UAVs) goes
nearly as far back as the history of manned flight. UAVs It is known that the state of the qUAV evolves in the
can be traced back to 1916 when Elmer Sperry and Special Euclidean 3 space(SE (3) = R3 × SO (3)). Thus
Peter Hewitt successfully demonstrated their Automatic the qUAV can be split into a translational dynamics
Plane dubbed ”the flying bomb”Lt. Kendra L. B. Cook subsystems with configuration space R3 and a rotational
(2007).Research into unmanned flight continued through subsystem with configuration space SO (3). The focus
out World War 1 and 2 with notable successes of this of the work presented in this paper is the control of
era being the German V1 and V2 ”buzz bombs” which the rotational subsystem so as to achieve some desired
could travel at speeds of 650km/hr Lt. Kendra L. B. Cook attitude. It should also be noted that in as much as the
(2007). The Vietnam war heralded the large scale use of focus of the present work is UAV attitude control the
modern UAVs in combat zones with over 3 000 operations theory that is developed in this work can also be applied
being flown by UAVs during this war Lt. Kendra L. B. to other rigid body attitude control problems such as
Cook (2007). In the last two decades major strides have satellites.
been made within the area of UAVs as is witnessed by the Backstepping control is a recursive Lyapunov based con-
huge success of Northrop Grumman’s Global Hawk drone trol technique for systems in strict feedback form. Back-
and General Atomics’ Predator drone which have become stepping control came about from the concerted efforts of
the weapon of choice for the United States Defense Forces. a number on researchers in the 1990s. Cascade integrator
From the brief history that has been given it is evident backstepping appeared in the work of Saberi, Kokotovic
that UAVs have been mostly used in defense related ap- and Sussman A. Saberi et al. (1989) which was devel-
plications. According to Zaka Sarris (2001) by 2000 the oped further by Kanellakopoulos et al I. Kanellakopoulos
the civil UAV market only accounted for 3% of the UAV et al. (1992). Passivity interpretations of the backstepping
market. However over the past decade progress in micro method were given by Lozano, Brogliato and LandauRoge-
electro-mechanical systems(MEMS) and IC miniaturisa- lio Lozano et al. (1992).The backstepping method was ex-
tion has led to a drop in cost of sensors making UAVs tended to cover system with uncertainties for the matched
economical for civilian use. Thus over the past decade case inM.J. Corless and G. Leitmann (1981). Adaptive
there has been a proliferation of civilian applications of control methods and backstepping methods were employed
UAVs such as border interdiction, search and rescue, in Kanellakopoulos, Kokotovic and MorseI. Kanellakopou-
powerline inspection e.t.c.Zaka Sarris (2001). In a major- los et al. (1991) to devise the adaptive backstepping tech-
ity of these civilian applications rotary UAVs especially niques. The important achievement of this techniques was
quadrotor UAVs(qUAVs) are used. Civilian UAV applica- that it could handle the case of extended matching, how-
tions mostly take place in constrained environments(e.g ever this technique had the disadvantage of over parame-
indoors) and hence require UAV platforms that are highly terization i.e required multiple estimates of the same pa-
manueverable. qUAVs possess the required manuevarabil- rameter. The introduction of tuning functions by Kristic,
ity and thus are perfectly suited for civilian applications. Kanellakopoulo and Kokotovic M.Krstic et al. (1992) man-
Additionally qUAVs have a very high thrust to weight aged to remove the over parameterization. Sliding mode
ratio which translates to lighter platforms, the absence of control is a powerful technique that ensures robust system
moving parts in the rotors (i.e.cyclic pitch controls) make performance however it has the drawback of requiring

978-3-902823-62-5/2014 © IFAC 11043

19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

a priori knowledge of the uncertainties. Koshkouei and Ixx −Ixy −Ixz

" #
ZinoberA.J. Koshkouei and A.S.I. Zinober (2000) devised I= −Iyx Iyy −Iyz (3)
a method to combine the adaptive backstepping technique −Izx −Izy Izz
with tuning functions and sliding mode control, this sliding
backstepping technique ensures that the tracking error If we assume the quadrotor to be perfectly symmetrical
moves along the sliding hyperplane. about all of its three axis we now have Ixy = Ixz = Iyz = 0
and the inertia matrix becomes I = diag(Ixx Iyy Izz ). In
Some of the theoretical advances outlined in the preceding equation 2 we are differentiating a body frame vector in
paragraph have been investigated with regards to qUAV the inertia frame. Using the equation of Coriolis we have
control. Madani and BenallegueT. Madani and A. Benal- Grant R. Fowles and George L. Cassiday (1999):
legue (2006) presented a backstepping based controller
for qUAV trajectory tracking. In S. Bouabdallah and R. dHB dHB
= + ωb/i × HB (4)
Seigwart (2007) the authors improved on the backstepping dtI dtB
controller by adding an integral term into the controls Applying this to equation 2 we have for the rotational
to improve steady state errors, A.A Mian and W. Daobo dynamics:
Iω̇ B = −ω × Iω B + τ B


(2008) further employed a full PID backstepping method (5)
for qUAV control. Frazzoli et alE. Frazzoli et al. (2000) h iT
used adaptive backstepping control to design a trajectory where τ B = τφB τθB τψB is the torque acting on the
tracking controller, their results showed that the controller quadrotor expressed in the vehicle frame. In expanded
could even perform aggressive maneuvers(e.g paths were form the rotational dynamics are given by the equations:
the UAV is initially upside down).  Jyy −Jzz  
1
qr Jxx τφ

ṗ
" #
Jxx
In this paper we are going to present an adaptive sliding q̇ =  JzzJ−J
 xx
pr  1
 +  Jyy τθ  (6)
yy
backstepping scheme for qUAV attitude tracking. The next ṙ J −J 1
xx yy
pq τ
Jzz ψ
section outlines the mathematical model for the qUAV Jzz
attitude. In section 3 the main result of this paper is
developed, this section will present the general adaptive 2.3 Full Attitude Dynamics
sliding backstepping scheme. Section 4 presents the simu-
lation results of the adaptive sliding backstepping attitude The full attitude dynamics are given by equations 1 and 5
controller and finally conclusions and final remarks are which are repeated here in a more compact form.
given in section 5.
Θ̇ = Ψ (Θ) ω B (7)
2. MODELING B
Iω̇ = −ω × (Iω) + τ (8)
From this one can clearly see that the attitude dynamics
In describing the attitude of the quadrotor UAV we shall are in strict feedback form which makes them amenable to
use the Z-Y-X Euler angle notation, where the Euler angle backstepping control.
vector Θ = [φ, θ, ψ] denotes the roll, pitch and yaw
respectively. The angular velocity is ω B = [p, q, r] where
3. CONTROLLER DESIGN
the superscript B denotes that the angular velocity is a
body frame vector.
In outlining the proposed adaptive sliding backstepping
method we shall make use of the general system given by:
2.1 Kinematics
ẋ1 = x2 (9)
We state without proof the kinematic equations of the rigid
body however the interested reader can consult I.Raptis ẋ2 = f1 (x1 , x2 ) + g1 (x1 , x2 ) x3 (10)
and K. Valavanis (2011) for a detailed derivation. ẋ3 = f2 (x1 , x2 , x3 ) + g2 (x1 , x2 , x3 ) u (11)
   
φ̇ 1 sinφtanθ cosφtanθ " p # where g1 (0, 0) 6= 0, g2 (0, 0, 0) 6= 0 and f2 (x1 , x2 , x3 ) and
 θ̇  =  0 cosφ −sinφ  q (1) g2 (x1 , x2 , x3 ) are unknown but bounded functions. The
sinφ cosφ
ψ̇ 0 cosθ cosθ
r control task is to ensure that x1 = 0 is asymptotically
| {z } stable.
Ψ(Θ)

3.1 Backstepping Control

2.2 Dynamics
Let us now apply the backstepping procedure to our
For rotational motion Newton’s 2nd law of motion states general system. If we take x2 as a pseudo-control for
that the rate of change of angular momentum is equal to the first equation of our system and if there exists a
the net torque acting on the body. This can be expressed positive definite unbounded function V1 (x1 ) then we can
as : find a function π1 (x1 ) such that the following inequality
dHB is satisfied.
=τ (2) ∂V1
dtI π1 (x1 ) ≤ −W (x1 ) (12)
∂x1
The angular momentum HB = Iω B . With I being the 3×3 where W (x1 ) is a positive definite function. This implies
inertia matrix gien by : that if x2 was an actual control then x2 = π (x1 ) would

11044
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

make equation 9 asymptotically stable. However as this is

 
∂π2 ∂V1 ∂π2
not the case we can define an error variable z1 = x2 − ż2 = − (z1 + π) + λ1 z1 + − z2
∂x1 ∂x1 ∂z1
π1 (x1 ) such that we have the following dynamics:
+ g20 (x1 , x2 , x3 ) ueq (21)
ẋ1 = π1 (x1 ) + z1 (13) Thus the equivalent control(ueq ) is :

π1 ∂π1 1 ∂π2
ż1 = f1 − z1 −π + g1 x3 (14) ueq = (z1 + π1 )
∂x1 ∂x1 g20 (x1 , x2 , x3 ) ∂x1
 
If we construct a new augmented Lyapunov function ∂V1
− λ1 z1 + − z1 (22)
V2 (x1 , z1 ): ∂x1
z2 Before we move on to the design of the switching controller
V2 (x1 , z1 ) = V1 (x1 ) + 1 (15) let us state another important assumption;
2
Now if we take x3 as the pseudo-control for the new Assumption 1. There exists a function β (x1 , z1 , z2 ) such
dynamics given by equations (13) and (14) we choose a that the following inequality is satisfied:
function π2 (x1 , z1 ) such that x3 = π2 (x1 , z1 ) will make the  
 f2 g2 − ĝ2 ueq 
time derivative of the augmented Lyapunov function(15) β (x1 , z1 , z2 ) >  + (23)
g20 g2 
negative definite. Such a function π2 (x1 , z1 ) is given by:
Now if we reconsider the sliding condition:

1 ∂π1 ∂π1
π2 (x1 , z1 ) = −f1 + z1 + π1
1 dz22
  
g1 (x1 , z1 ) ∂x1 ∂x1 ∂π2 ∂V1 ∂π2
 = z2 − (z1 + π) + λ1 z1 + − z2
∂V1 2 dt ∂x1 ∂x1 ∂z
− − λ2 z1 (16)  1
∂x1
+ f2 (x1 , x2 , x3 ) + g2 (x1 , x2 , x3 ) (ueq + usw ) (24)
Again we know that x3 is not an actually control so
we define the error variable z2 = x3 − π2 (x1 , z1 ) such
that now the whole system can be transformed from After substituting the expression of ueq into equation 24
the (x1 , x2 , x3 ) space to the (x1 , z1 , z2 ) space where the we get
1 dz22
 
transformed system dynamics are: ĝ2 ueq
= z2 − + f2 + g2 usw (25)
2 dt g20
x˙1 = π1 (x1 ) + z1 (17) If we choose usw as:
z˙1 = −λ1 z1 + z2 (18) usw = −β (x1 , z1 , z2 ) sign (z2 ) (26)
∂π2

∂V1

∂π2 Thus we now have:
z˙2 = − (z1 + π1 ) + λ1 z1 + − z2 1 dz22
∂x1 ∂x1 ∂z1 ≤ −|z2 |
2 dt
+f2 (x1 , x2 , x3 ) + g2 (x1 , x2 , x3 ) u (19) which satisfies the sliding condition. This means if we start
off the z2 = 0 manifold we will reach this manifold after
3.2 Sliding Backstepping some finite time. The full sliding backstepping control is:
  
1 ∂π2 ∂V1
Up until now the procedure we have followed has been no u= (z1 + π1 ) − λ1z1 + − z1
g (x1 , x2 , x3 ) ∂x1 ∂x1
different from the usual backstepping but now consider the
− β (x1 , z1 , z2 ) sign (z2 ) (27)
z2 equation in which f2 (x1 , x2 , x3 ) and g2 (x1 , x2 .x3 ) are
unknown. If we can make z2 = 0 that would mean that
x3 = π2 (x1 , z1 ) which has been shown in the preceding 3.3 Adaptive Control
section makes the (x1 , z1 ) dynamics asymptotically stable.
Now we can apply sliding mode techniques to ensure that The control described by equation (26) requires the knowl-
we arrive at the z2 = 0 manifold within a finite time edge of the bounds of the functions f2 (x1 , x2 , x3 ) and
and stay there. The control task now becomes finding a g2 (x1 , x2 , x3 ). We are thus going to modify this control
control(u) such that the following condition is met: structure making it adaptive such that the need to know
the bounds of f2 (x1 , x2 , x3 ) and g2 (x1 , x2 , x3 ) is removed.
1 dz22 Now if we assume that there exists some positive constant
≤ −η|z2 | (20)
2 dt Kd such that β (x1 , z1 , z2 ) < Kd for all time, then we
This is the sliding mode condition. Before proceeding we can define K̂d as the estimate for this constant and the
state an important assumption. estimation error K̃d = Kd − K̂d .
Assumption 1. g2 (x1 , x2 , x3 ) can be expressed as
Recall that in the previous section we had chosen our
g2 (x1 , x2 , x3 ) = g20 (x1 , x2 , x3 ) + ĝ2 (x1 , x2 , x3 ) where
control as u = ueq + usw ,for the switching control (usw )
g20 (x1 , x2 , x3 ) is the nominal part and ĝ2 (x1 , x2 , x3 ) is
the uncertain part. let us replace β (x1 , z1 , z2 ) with the estimate K̂d this gives
the new switching control as :
Using the sliding mode technique we divide the control usw = −K̂d sign (z2 ) (28)
u into an equivalent control(ueq ) and a switching control
(usw ). The equivalent control is the control that ensures Consider the candidate Lyapunov function given by
that for the nominal z2 dynamics ż2 is always zero. Now 1 2 K̃d2
V = z + (29)
the nominal z2 dynamics are given by : 2 2 2γ

11045
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

where γ is a positive constant. The time derivative of the x1 = χ f1 = −φ̇d + (sinφtanθ) q + (cosφtanθ) r
candidate Lyapunov function becomes: x2 = eφ
J −J
f2 = yyJxx zz qr
  
∂π2 ∂V1 ∂π2 x3 = p g1 = 1
V̇ = z2 − (z1 + π) + λ1 z1 + − z2 1
∂x1 ∂x1 ∂z g2 = Jxx
 1
+ f2 (x1 , x2 , x3 ) + g2 (x1 , x2 , x3 ) (ueq + usw ) If we choose the Lyapunov function V1 = 21 χ2 following the
backstepping procedure as outlined in the previous section
K̃d K̃˙ d gives the following pseudo-controls and the respective
+ (30) ”tracking” errors.
γ
Substituting for ueq and usw we are left with:
π1 = −λ1 χ
h i K̃ K̃˙
V̇ = z2 f2 (x1 , x2 , x3 ) − K̂d sign (z2 ) +
d d
(31) z1 = eφ + λ1 χ
γ
π2 = φ̇d − (sinφtanθ) q − (cosφtanθ) r
Noting that K̂d = Kd − K̃d we can rewrite the expression
− (λ1 + λ1 ) eφ − (λ1 λ2 + 1) χ
for V̇ as:
z2 = p − φ̇d + (sinφtanθ) q + (cosφtanθ) r
h   i K̃ K̃˙ + (λ1 + λ2 ) eφ + (λ1 λ2 + 1) χ
d d
V̇ = z2 f2 (x1 , x2 , x3 ) − Kd − K̃d sign (z2 ) +
γ
= z2 [f2 (x1 , x2 , x − 3) − Kd sign (z2 )] Thus the control torque in the φ direction is given by:

K̃˙ d
!  
+ K̃d |z2 | + (32) ∂π2 ∂V1
τφ = (z1 + π1 ) − λ1 z1 + − z1
γ ∂χ ∂χ
If we choose the adaptation law given by : − K̂d sign (z2 ) (37)
˙ ˙
K̂d = γ|z2 | (33) K̂d = γ|z2 | (38)
The time derivative of the Lyapunov function will be neg-
ative definite and thus guaranteeing asymptotic conver- 4. RESULTS
gence to the sliding manifold z2 = 0. From the adaptation
law we see that the estimated gain K̂d is always increasing The controller designed in the previous section was sim-
and the rate of increase is proportional to the ”distance” ulated in MATLAB/SIMULINK environment to see its
from the sliding surface. The presented adaptation law performance. The first simulation results show the unit
tends to overestimate the gain since the estimate does not step response of the closed loop system. It should be noted
decrease even if z2 = 0. that in implementing the switching function instead of
using the signum function we used the hyperbolic tangent
function so as to eliminate the chattering caused by the
3.4 Attitude Controller signum function.

Having developed an adaptive sliding backstepping con-

troller for a general system we can now use this result
to formulate a controller for the quadcopter attitude. For
brevity sake we shall develop a controller for only the pitch
dynamics however this can be easily adapted to the roll
and yaw dynamics as they do not differ much from the
pitch dynamics.
We desire the pitch(φ) to track a time varying reference
R e φ = φ − φd
signal φd , thus we define the tracking error
and the integral of the tracking error χ = eφ dt. Thus the
error dynamics are describe by the differential equations:
Fig. 1. φ angle in radians
χ̇ = eφ (34)
Figure 1-3 show that the angles have a settling time of
ėφ = −φ̇d + (sinφtanθ) q + (cosφtanθ) r + p (35)
about 0.5 second with very little overshoot however this
Jyy − Jzz 1 performance is achieved at the cost of large controls. As
ṗ = qr + τφ (36)
Jxx Jxx was stated earlier the adaptation law tends to overestimate
the sliding mode gain, this characteristic of this kind of
From equations 33-35 we can see that the pitch dynam- adaptive sliding mode control is also highlighted in F.
ics are similar in form to the general 3rd order system Plestan et al. (2010). The next set of simulation results
presented in the previous section, the following correspon- show the system performance when the reference signal
dences should be evident between the two systems. for all three signals is a sinusoid of amplitude 1.

11046
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Fig. 2. θ angle in radians Fig. 6. φ angle in radians( green = reference signal, blue
= actual angle)

Fig. 3. ψ angle in radians

Fig. 7. θ angle in radians( green = reference signal, blue
= actual angle)

Fig. 4. Control torques

Fig. 8. ψ angle in radians( green = reference signal, blue

= actual angle)

Fig. 5. Sliding gain estimates (blue = K̂dφ , green = K̂dθ ,

red = K̂dψ )
5. CONCLUSIONS
Fig. 9. Control torques
We have presented an adaptive sliding backstepping con-
trol scheme for attitude tracking for quadrotor UAV. The couple the sliding backstepping controller with an adaptive
sliding mode aspect of the controller ensures that the estimator for the sliding gain which removes the need to
controller is robust against uncertainties however in con- know the upper bounds of the uncertainties. The presented
ventional sliding mode control there is need to know the methodology has the disadvantage of overestimating the
bound of the uncertainty which is difficult to determine sliding gain which results in unnecessarily large controls.
in real life. As such to try to alleviate this problem we Simulations of the adaptive sliding backstepping controller

11047
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Nonlinear Systems. IEEE Transactions on Automatic

Control, 37(9), 1386–1388.
S. Bouabdallah and R. Seigwart (2007). Full Control of a
Quadrotor. In IEEE/RSJ International Conference on
Intelligent Robots and Systems.
T. Madani and A. Benallegue (2006). Backstepping Con-
trol of a Quadrotor Helicopter. In IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems.
Zaka Sarris (2001). A Survey of UAV Applications in
Civil Markets. In 9th IEEE Mediterranean Conference
on Control and Automation.

Fig. 10. Sliding gain estimates (blue = K̂dφ , green = K̂dθ ,

red = K̂dψ )

showed that the controller is able to track constant and

time varying signals almost perfectly.

REFERENCES
A. Saberi, P.V. Kokotovic, and H.J. Sussman (1989).
Global Stabilization of Partially Linear Composite Sys-
tems. In Proceedings of the 28th Conference on Decision
and Control.
A.A Mian and W. Daobo (2008). Modeling and
Backstepping-based Nonlinear Control Strategy for a 6
DOF Quadrotor Helicopter. Chinese Journal of Aero-
nautics, 21, 261–268.
A.J. Koshkouei and A.S.I. Zinober (2000). Adaptive Back-
stepping Control of Nonlinear Systems with Unmatched
Uncertanity. In Conference on Decision and Control.
E. Frazzoli, M A. Dahleh, and E. Feron (2000). Trajectory
Tracking Control Design for Autonomous Helicopters
using a Backstepping Algorithm. In American Control
Conference.
F. Plestan, Y. Shtessel, V. Breǵeault, and A. Pozynak
(2010). New Methodologies for Adaptive Sliding Mode
Control. International Journal of Control, 83(9), 1907–
1919.
Grant R. Fowles and George L. Cassiday (1999). Analytical
Mechanics. Fort Worth: Saunders College Publishing,
6th edition.
I. Kanellakopoulos, P.V. Kokotovic, and A.S Morse (1991).
Systematic Design of Adaptive Controllers for Feedback
Linearizable Systems. IEEE Transactions on Automatic
Control, 36(11), 1241–1253.
I. Kanellakopoulos, P.V. Kokotovic, and A.S Morse (1992).
A toolkit for nonlinear feedback systems. Systems and
Control Letters, 18, 83–92.
I.Raptis and K. Valavanis (2011). Linear and Nonlinear
Control of Small Scale Helicopters. Springer.
Lt. Kendra L. B. Cook (2007). The Silent Force Multiplier:
The Hitory and Role of UAVs in Warfare. In IEEE
Aerospace Conference.
M.J. Corless and G. Leitmann (1981). Continous State
Feedback Guaranteeing Uniform Ultimate Boundedness
for Uncertain Dynamic Systems. IEEE Transaction on
Automatic Control, 26(5), 1139–1144.
M.Krstic, I. Kanellapoulos, and P.V. Kokotovic (1992).
Adaptive nonlinear control without overparameriza-
tions. Systems and Control Letters, 16, 177–185.
Rogelio Lozano, Bernard Brogliato, and I.D. Landau
(1992). Passivity and Global Stabilization of Cascaded

11048