49th IEEE Conference on Decision and Control
December 15-17, 2010
Hilton Atlanta Hotel, Atlanta, GA, USA
Super-twisting Adaptive Sliding Mode Control: a Lyapunov Design
Yuri B. Shtessel, Jaime A. Moreno, Franck Plestan, Leonid M. Fridman, Alexander S. Poznyak
Abstract— A novel super-twisting adaptive sliding mode
controller is proposed. A drift uncertain term is assumed to be
bounded with unknown boundary. The proposed Lyapunov-
based approach consists in using dynamically adapted control
gains that ensure the establishment, in a finite time, of a second
order sliding mode. Finite convergence time is estimated. A
numerical example confirms the efficacy of the proposed
adaptive super-twisting control.
I. INTRODUCTION
S LIDING mode control is one of the best choices for
controlling perturbed systems with matched
disturbances/uncertainties [1,2]. The price for achieving the
robustness/insensitivity to these disturbances is control
chattering [1,2,3]. The traditional ways for avoiding
chattering are as follows:
(a) Replacing the discontinuous control function by
“saturation” or “sigmoid ones” [4,5]. This approach
yields continuous control and chattering elimination.
However, it constrains the sliding system’s trajectories
not to the sliding surface but to its vicinity loosing the
robustness to the disturbances.
(b) Using the higher order sliding mode control techniques
[6-8, 12, 13]. This approach allows driving to zero the
sliding variable and its consecutive derivatives in the
presence of the disturbances/ uncertainties increasing
the accuracy of the sliding variable stabilization.
 ∂
One of the most powerful second order continuous sliding
mode control algorithms is the super-twisting control law
(STW) that handles a relative degree equal to one. It
generates the continuous control function that drives the
sliding variable and its derivative to zero in finite time in the
presence of the smooth matched disturbances with bounded
gradient, when this boundary is known. Since STW
algorithm contains a discontinuous function under the
integral, chattering is not eliminated but attenuated.
The main drawback of STW control algorithm is the
requirements to know the boundaries of the disturbance
gradient. In many practical cases this boundary cannot be
easily estimated. The overestimating of the disturbance
Y. B. Shtessel is with the University of Alabama in Huntsville,
Huntsville, AL 35899 USA (tel: +(256) 824-6164; fax: +(256)824-6803; e-
mail: shtessel@eng.uah.edu).
J. A. Moreno is with Electrica Y Computacion, Instituto de Ingeneria,
UNAM, D.F., Mexico, (e-mail: jmorenop@iingen.unam.mx)
F. Plestan.is with Ecole Centrale de Nantes-IRCCyN, Nantes, France
(e-mail: Franck.Plestan@irccyn.ec-nantes.fr).
L. M. Fridman is with Division de Ingenieria Electrica, Facultad de
Ingenieria, UNAM, 04510, D.F., Mexico, (e-mail: lfridman@unam.mx).
A. S. Poznyak is with CINVESTAV-IPN, DF, Mexico
978-1-4244-7746-3/10/$26.00 ©2010 IEEE
boundary yields to larger than necessary control gains, while
designing the STW control law.
Contribution. In this work we propose the novel adaptive
STW control law that continuously drives the sliding
variable and its derivative to zero in the presence of the
bounded disturbance with the unknown boundary. The
finite convergence time is estimated. The proof is based on
recently proposed Lyapunov function [9, 10] that is used for
the derivation of the novel adaptive STW control algorithm.
II. PROBLEM FORMULATION
Consider a single-input uncertain nonlinear system
x = f ( x, t ) + h( x, t )u (1)
n
where x ∈ \ is a state vector, u ∈ \ is a control function,
f ( x, t ) ∈ \ n is a differentiable, partially known vector-field.
Assume that
(A1) A sliding variable σ = σ ( x, t ) ∈ \ is designed so
that the system’s (1) desirable compensated dynamics
are achieved in the sliding mode σ = σ ( x, t ) = 0 .
(A2) The system’s (1) input-output ( u → σ ) dynamics are
of a relative degree one, and the internal dynamics are
stable.
Therefore, the input-output dynamics can be presented
∂σ ∂σ ∂σ
σ = + f ( x, t ) + h( x ) u →
∂t
 ∂ 
x
x 
ϕ ( x ,t ) b ( x ,t )
σ = ϕ ( x, t ) + b( x, t )u → (2)
−1
σ = ϕ ( x, t ) + ω , ω = b( x, t )u ↔ u = b ( x, t )ω
The solution of system (2) is understood in the sense of
Filippov [11].
Assume that
(A3) the function b( x, t ) ∈ \ is known and not equal to
zero ∀x and t ∈ [ 0, ∞ )
(A4) the function ϕ ( x, t ) ∈ \ is bounded
1/ 2
ϕ ( x, t ) ≤ δ σ (3)
where the finite boundary δ > 0 exists but is not known.
The problem is to drive the sliding variable σ and its
derivative σ to zero in finite time in the presence of the
bounded perturbation with the unknown boundary by means
of continuous control.
The classical SMC and the second order sliding mode
(SOSM) controllers, including the continuous STW control
algorithm, can robustly handle such problem if the boundary
of the perturbation is known. The main disadvantage of the
classical SMC is introducing control chattering, while
5109
 ( )
SOSM controllers are able to attenuate it. Classical sliding T
z = ( z1 , z2 ) = σ sign (σ ) , v
T 1/ 2
mode control with gain adaptation for a class of uncertain (8)
systems with unknown bounding function are proposed in and system (6) can be rewritten as
works [14, 15]. In the work [14] in order to avoid chattering
⎧ 1 ⎛ −α 1 1 ⎞
the discontinuous control is replaced by a continuous ⎪ z1 = ⎜ z1 + z2 + ϕ ( x, t ) ⎟
approximation. The price is a loss of robustness. The gain ⎪ z1 ⎝ 2 2 2 ⎠
⎨ (9)
adaptation laws proposed in work [15] also guarantee ⎪ z = − β z
convergence to a small domain but without overestimating ⎪ 2 z1
1
⎩
the control gains.
Equation (9) can be rewritten in a vector-matrix format
In this work we are looking for an adaptive-gain STW
algorithm that is able to address this problem via generating ⎡ −α 1 ⎤ ⎡1⎤
⎡ z1 ⎤ 1 ⎢ ⎥ ⎡ z1 ⎤ 1 ⎢ ⎥
continuous control function (chattering attenuation) so that ⎢ z ⎥ = 2 2 ⎢ ⎥ + 2 ϕ ( x, t ) ⇒
its gains are adapted to the unknown perturbation with the ⎣ 2 ⎦ z1 ⎢⎢ − β 0 ⎥⎥ ⎣ z2 ⎦ z1 ⎢⎢ 0 ⎥⎥ (10)
⎣ ⎦ ⎣ ⎦
unknown boundary. z = A( z1 ) z + g ( z1 )ϕ ( x, t )
III. CONTROL STRUCTURE where
⎡ −α 1 ⎤ ⎡1⎤
The following STW control is proposed 1 ⎢ 1 ⎢ ⎥
A( z1 ) = 2 2 ⎥ , g ( z1 ) = 2 (11)
⎧⎪ω = −α σ 1/ 2 sign (σ ) + v z1 ⎢ ⎥ z1 ⎢ ⎥
⎣⎢ − β 0 ⎦⎥ ⎣⎢ ⎦⎥
0
⎨ (4)
⎪⎩v = − β sign (σ ) It can be observed that
where the adaptive gains (a) If z1 , z2 → 0 in finite time then σ , σ → 0 in finite
α = α (σ , σ , t ) time;
(5)
and sign ( z1 ) = sign (σ )
1/ 2
β = β (σ , σ , t ) (b) z1 = σ
are to be defined. In the second step of the proof, the stability analysis of
Control system given by eqs. (2) and (4) is presented in a system given by eqs. (10) and (11) is performed. In order to
form do it, the following Lyapunov function candidate is
⎧⎪σ = −α σ 1/ 2 sign (σ ) + v + ϕ ( x, t ) introduced

( ) ( )
⎨ (6) 1 2 1 2
⎪⎩v = − β sign (σ ) V ( z1 , z2 , α , β ) = V0 + α −α* + β − β* (12)
2γ 1 2γ 2
The control design problem is reduced to designing adaptive where
STW control (4), (5) that drives σ , σ → 0 in finite time in
the presence of the bounded perturbation with the unknown ( )
V0 ( z ) = λ + 4ε 2 z12 + z22 − 4ε z1 z2 = zT Pz (13)
boundary. ⎡ λ + 4ε 2 −2ε ⎤
P=⎢ ⎥ , λ > 0, ε > 0 . (14)
IV. MAIN RESULTS ⎢⎣ −2ε 1 ⎥⎦
The main result of the paper is formulated in the and α * > 0, β * > 0 are some constants. It is worth noting
following theorem. that the matrix P in eq. (14) is positive definite if λ > 0
and ε is any real number.
Theorem: Consider system (6). Suppose that the The derivative of the Lyapunov function candidate (13) is
perturbation ϕ ( x, t ) satisfies Assumption (A4) for some presented
unknown constant δ > 0 . Then for any initial conditions V ( z , α , β ) = zT Pz + zT Pz +
x(0), σ (0) the sliding surface σ = 0 will be reached in
( ) ( )
1 1 (15)
finite time via STW control (4) with the adaptive gains α − α * α + β − β * β
γ1 γ2
⎧ γ1
⎪ω , if σ ≠ 0 The first two terms of eq. (15) are computed taking into
α = ⎨ 1 2 , account eqs. (10) and (11)
⎪0, (7)
⎩ if σ = 0 1 T 
V0 = zT Pz + zT Pz ≤ − z Qz (16)
β = 2εα + λ + 4ε 2 z1
where ε , λ , γ 1 , ω1 are arbitrary positive constants. The symmetric matrix Q is computed taking into account
Proof. The proof is split into two steps. In the first step, we inequality (3):
will present system (6) in a form convenient for the
Lyapunov analysis. In order to do this a new state vector is
introduced

5110
 
⎡2λα + 4ε ( 2εα − β ) − 2 λ + 4ε 2 δ
⎢
Q − 2ε I = ⎢
*⎤
⎥ ( ) (x 2
+ y2 + z2 )
1/ 2
≤ x+ y+ z (26)
⎥ (17)
⎢⎣ (
β − 2εα − λ − 4ε 2 + 2εδ 4ε ⎥
⎦ ) and in view of eq. (12), we can derive
ω1 ω
In order to guarantee the positive definiteness of the matrix −rV01/ 2 − α −α* − 2 β − β* ≤
2γ 1 2γ 2 (27)
Q we enforce
−η V ( z , α , β )
β = 2εα + λ + 4ε 2 (18)
where η = min ( r , ω1 , ω2 ) .
The matrix Q will be positive definite with a minimal
Taking into account eq. (27), we can rewrite eq. (25) as
eigenvaue λ (Q ) ≥ 2ε if
min 1/ 2 1
V ( z1, z2 , α , β ) ≤ −η [V ( z1, z2 ,α , β )] + ( )
α − α * α +
α>
(
εδ 2 + λ + 4ε 2 ( 2ε + δ ) + ε ) (19)
γ1
(28)
λ
In view of eq. (16) and assuming that eqs. (18), (19) hold, it
1
γ2
( ) ω
β − β * β + + 1 α − α * + 2 β − β *
2γ1
ω
2γ 2
is easy to show that Now we assume that the adaptation law (7) makes the
V ≤ −rV 1/ 2
0 0 (20) adaptive gains α (t ), β (t ) bounded (this assumption will be
where proven later). Then there exist positive constants α * , β * that
1/ 2
2ελmin ( P) α (t ) − α * < 0 and β (t ) − β * < 0 ∀t ≥ 0 .
r= (21)
λmax ( P) In view of the above assumption, eq. (28) can be reduced to
the following:
Indeed, since
V ( z1 , z2 , α , β ) ≤ −η [V ( z1 , z2 , α , β )] −
1/ 2
1 T  ε T ε 2
V0 ( z ) ≤ − z Qz ≤ − z z=− z (22)
z1 z1 z1 ⎛1 ω ⎞
α −α* ⎜ α − 1 ⎟ −
and ⎜ γ1 2γ 1 ⎟⎠
⎝ (29)
λmin { P} z ≤ zT Pz ≤ λmax { P} z
2 2
(23) ⎛ 1 ω ⎞
β − β * ⎜ β − 2 ⎟ ⇒
2 ⎜ γ2 2γ 2 ⎟⎠
where z = z12 + z22 = σ + z22 and ⎝
V ( z , z , α , β ) ≤ −η [V ( z , z , α , β )] + ξ
1/ 2
1/ 2 2 V01/ 2 ( z ) 1 2 1 2
z1 = σ ≤ z ≤ (24)
1/ 2
λmim {P} where
⎛ 1 ω ⎞
then ξ = − α −α* ⎜ α − 1 ⎟−
1/ 2 ⎜ γ1 2γ 1 ⎟
2ελmin ( P) ⎝ ⎠
V0 ( z ) ≤ −rV01/ 2 , r = (30)
λmax ( P ) ⎛ 1 ω ⎞
β − β ⎜ β −* 2 ⎟
Now, in view of eqs. (20), eq. (15) can be rewritten ⎜ γ2 2γ ⎟
⎝ 2 ⎠
V ( z , α , β ) = zT Pz + zT Pz +
1
α − α * α
γ1
( ) For the finite time convergence we have to assure ξ = 0 that
is supposed to be achieved via adaptation of the gains α , β .
+
γ2
1
( β − β ) β ≤ − z1
*
1
 + 1 α − α * α +
zT Qz
γ1
( ) α = ω1
γ1
(31)
2

γ
1
( β − β ) β ≤ −rV
2
* 1/ 2
0
γ1
+
1
(α − α )α + *
β = ω2
γ2
(32)
2

γ
1
( β − β ) β = −rV
2
* 1/ 2
0
ω
− 1 α −α* −
2γ 1
(25)
After selecting ε =
ω2 γ2
eq. (18) and (32) coincide,
2ω1 γ1
ω2
2γ 2
β − β* +
1
γ1
α − α * α + ( ) since
β = 2εα + λ + 4ε 2 →
1
γ2
(β − β ) * ω ω
β + + 1 α − α * + 2 β − β *
2γ 1 2γ 2
β = 2εα → β = εω1 2γ 1
(33)

It is worth noting that for the finite time convergence α (t )

Taking into account a well-known inequality must satisfy inequality (19). It means that α (t ) is supposed
to increase in accordance with eq. (31) until eq. (19) is met

5111
that guarantees the positive definiteness of the matrix Q .
After that the finite convergence is guaranteed according to
eq, (29). Also, as soon as the sliding variable σ and its
derivative converges to zero it does not make sense to
increase α (t ) and β (t ) by making α = 0 as σ = 0 .
Therefore, we obtain the gain-adaptation law (7). Theorem
is proven.
Now we can prove the assumption about boundedness of
α (t ) and β (t ) .
Proposition 1. The adaptive gains α (t ) and β (t ) are
bounded.
Proof. A solution to eq. (7) can be constructed as
⎧ γ1
⎪α (0) + ω1 t , 0 ≤ t ≤ tr
⎪ 2
α =⎨ (34)
⎪ γ1
⎪⎩α (0) + ω1 2 tr , t > tr
Where tr is finite reaching time. Therefore α (t ) is
bounded. The adaptive gain β (t ) is also bounded, since
β (t ) = 2εα (t ) + λ + 4ε 2 . The proposition is proven.
Now we can easily estimate finite reaching time.
⎧⎪ω = −α (σ , t ) σ 1/ 2 sign (σ ) + v
⎨ (38)
⎪⎩v = − β (σ , t ) sign (σ )
The adaptive gain α (σ , t ) and β (σ , t ) dynamics follow eq.
(7). The values of the parameters of the adaptive gain law
(7) have been taken as ω1 = 6, γ 1 = 1, λ = 1, ε = 1 . The
parameter tuning has been made without “optimal”
objectives but in order to get sufficiently accurate and fast
convergence. Note, that, firstly, inequality (19) is to be
fulfilled in finite time, while the gains are increasing, and
only then the σ (t ) − trajectory starts finite-time converging.
Finally, the adaptive-gain law (7) becomes
⎧ ω1
⎪ , if σ ≠ 0
α = ⎨ 2 ,
⎪0, (39)
⎩ if σ = 0
β = 2α + 5
Figures 1 and 2 show the efficacy of the proposed adaptive
STW control algorithm. Another parameter tuning has been
made by increasing the parameter ω1 up to ω1=15. The
simulations displayed by Figures 3 and 4 show that, as the
gain is increasing faster, the perturbation is faster
compensated: the convergence of σ (t ) to 0 is obtained in a
shorter time.

Proposition 2. As soon as inequality (19) is fulfilled in finite

time the adaptive-gain STW control law (7) drives the
sliding variable σ and its derivative to zero in finite time
that is estimated as:
2V 1/ 2 (0)
tr ≤ (35)
η
where η = min ( r , ω1 , ω2 ) .
Proof. Inequality (19) is fulfilled in finite time, since its Figure 1 – Sliding variable σ(t) versus time (sec)
right hand side is bounded and the adaptive gain α (t ) is
increasing linearly with respect to time in accordance with
eq. (7). Inequality (35) is obtained by a direct integration of
inequality (29) bearing in mind that ξ = 0 due the
adaptation law (7).

V. EXAMPLE
Consider the following uncertain nonlinear system
σ = ϕ (σ , t ) + ω (36)
The disturbance with the unknown gain-boundary is Figure 2 – Gain α(t) versus time (sec)
simulated
⎧10sin t σ , if t ≤ 5sec
⎪
ϕ (σ , t ) = ⎨ (37)
⎪⎩35sin t σ , if t > 5sec
The initial value of the sliding variable is taken as
σ (0) = −10 .
The adaptive-gain STW control is designed

5112
 [9] J. A. Moreno, and M. Osorio, “A Lyapunov approach to second order
sliding mode controller and observers,” 47th Conference on Decision
and Control, pp. 2856-2861, 2008.
[10] A. Polyakov and A. Poznyak, “Reaching time estimation for “super-
twisting” second order sliding mode controller via Lyapunov function
designing,” IEEE Transactions on Automatic Control, Vol. 54, No. 8,
pp. 1951-1955, 2009.
[11] A. F. Filippov, Differential equations with discontinuous right hand
side, Kluwer, Dordrecht, The Netherlands, 1988.
[12] S. Laghrouche, F. Plestan, and A. Glumineau, “Higher order sliding
mode control based on integral sliding surface,” Automatica, Vol. 43,
No. 3, pp. 531-537, 2007.
[13] F. Plestan, A. Glumineau, and S. Laghrouche, “A new algorithm for
high order sliding mode control,” International Journal of Robust and
Nonlinear Control, Vol. 18, No. 4-5, pp. 441-453, 2008.
Figure 3 – Sliding variable σ(t) versus time (sec)
[14] W. Chen and M. Saif, “A variable structure controller for a class of
uncertain systems with unknown uncertainty bounding function,”
Proceedings of American Control Conference, Minneapolis, MN,
USA, 2006, pp. 4700-4705.
[15] F. Plestan, Y. Shtessel, V. Brégeault, and A. Poznyak, “New
methodologies for adaptive sliding mode control,” International
Journal of Control, Vol. 83, No. 9, 2010, pp. 1907–1919.

Figure 4 – Gain α(t) versus time (sec)

It is worth noting that the STW control gains adapt their-

selves while the unknown boundary of the disturbance
changes during the system’s performance.

VI. CONCLUSIONS
A novel adaptive-gain super-twisting sliding mode
controller is proposed. A drift uncertain term is assumed to
be bounded with unknown boundary. The proposed
Lyapunov-based approach consists in using dynamically
adapted control gains that ensure the establishment, in a
finite time, of a second order sliding mode. Finite
convergence time is estimated. A numerical example
confirms the efficacy of the proposed adaptive super-
twisting control. The future work will be dedicated to the
study of adaptive STW control with a broader clases of
possible bounded disturbances with unknown bounds.

VII. REFERENCES
[1] C. Edwards and S. Spurgeon, Sliding Mode Control: Theory and
Applications, Taylor & Francis, Bristol, 1998.
[2] V. Utkin, J. Guldner, and J. Shi, Sliding Modes in Electromechanical
Systems, Taylor and Francis, London, 1999.
[3] I. Boiko, Discontinuous Systems, Birkhauser, Boston, 2008.
[4] J.-J. Slotine and W. Li., Applied Nonlinear Control, Prentice, 1991.
[5] J. A. Burton and A. S. I. Zinober, “Continuous approximation of
VSC,” International Journal of Systems Sciences, Vol. 17, pp. 875-
885, 1986.
[6] Levant, A., “Higher-order sliding modes, differentiation and output-
feedback control,” International Journal of Control, 76, 9/10, 2003,
924-941.
[7]. A. Levant, “Homogeneity approach to high-order sliding mode
design,” Automatica, Vol. 41, No. 5, 2005, pp. 823-830.
[8] Y. Shtessel, I. Shkolnikov and A. Levant, “Smooth Second Order
Sliding Modes: Missile Guidance Application,” Automatica, Vol. 43,
No.8, 2007, pp. 1470-1476.

5113