1
Higher Order Sliding Mode controllers with optimal
reaching
Francesco Dinuzzo, Antonella Ferrara, Senior Member
Abstract—Higher order sliding mode (HOSM) control design is
considered for systems with a known permanent relative degree.
In this paper, we introduce the Robust Fuller’s Problem that is a
robust generalization of the Fuller’s problem, a standard optimal
control problem for a chain of integrators with bounded control.
By solving the Robust Fuller’s Problem it is possible to obtain
feedback laws that are HOSM algorithms of generic order and,
in addition, provide optimal finite-time reaching of the sliding
manifold. A common difficulty in the use of existing HOSM
algorithms is the tuning of design parameters: our methodology
proves useful for the tuning of HOSM controller parameters in
order to assure desired performances and prevent instabilities.
The convergence and stability properties of the proposed family
of controllers are theoretically analyzed. Simulation evidence
demonstrates their effectiveness.
Index Terms—Uncertain systems, Higher order sliding mode,
Optimal control, Variable Structure Systems
I. I NTRODUCTION
Sliding mode controllers [1]–[3] are powerful tools to
robustly control uncertain systems. They are able to achieve
finite time reaching and exact keeping of a suitably chosen
sliding manifold in the state space by means of a discontin-
uous control. The choice of the sliding manifold is strictly
related to the control objective to be attained. In standard
formulations, the controlled system state is first steered to the
sliding manifold in finite time, and then maintained confined
to the manifold itself, giving rise to the so-called sliding
mode, so that the equilibrium point corresponding to the
fulfillment of the control objective is made asymptotically
stable. Indeed, standard sliding mode controllers can only
deal with systems having relative degree equal to one. In
addition, they can induce undesirable chattering effects, i.e.
high frequency oscillations of the controlled variable, which
cannot be tolerated in some practical applications, especially
mechanical and electro-mechanical ones [4].
Higher order sliding mode (HOSM) controllers are effective
in extending the good properties of standard sliding mode
controllers to systems with higher relative degree, and can
be used also to reduce the chattering effect [5], [6]. Other
methods, such as the so-called “dynamic sliding modes” have
been proposed to improve the performances of standard sliding
mode controllers. A joint analysis of higher order sliding
controller and conventional sliding mode with dynamics can
be found in [7]. HOSM controllers have the capability of
stabilizing to zero in finite time not only the sliding variable,
but also a number of its time derivatives. Second order sliding
F. Dinuzzo is with the Department of Mathematics, Università di Pavia,
Via Ferrata, 1, 27100 Pavia, Italy, e-mail: francesco.dinuzzo@gmail.com.
A. Ferrara is with the Dipartimento di Informatica e Sistemistica, Università
di Pavia, Via Ferrata, 1, 27100 Pavia, Italy, e-mail: antonella.ferrara@unipv.it.
mode controllers have been widely studied, leading to well-
established algorithms like the super-twisting controller [8],
and the sub-optimal controller [9], [10], also proposed for
the multi-input case [11]. However, very few methods have
been proposed for the third [12] and generic order case [13],
[14]. See also [15], where the so-called “quasi-continuous
controller” is presented. Recently, an arbitrary order sliding
mode algorithm has been proposed that relies on the idea of
integral sliding modes, and exploits optimal linear quadratic
control of the auxiliary system with a final state constraint
[16]. The key element in the study of HOSM algorithms is
the so-called auxiliary system for an uncertain system, that is
a perturbed chain of integrators built relying on the output
variable and its time derivatives. The goal of any HOSM
control problem is to stabilize the auxiliary system state to
the origin in a finite time, see e.g. [17].
In this paper, the connection between the construction of
HOSM algorithms and the problem of stabilizing in finite
time a chain of integrators with bounded control is studied.
It is shown that the solution to the so-called Fuller’s problem
can be adapted to generate HOSM algorithms that guarantee
optimal reaching of the sliding manifold. First, the simple
second order case is considered, where it is possible to obtain
an interesting geometrical interpretation. Then, the attention
is focused on third order algorithms and an explicit formula
for the controller is provided. Finally, an HOSM algorithm of
arbitrary order is derived. Several already known controllers
are shown to belong to this general framework. Within our
methodology, it is easy to tune HOSM controller parameters in
order to assure desired performances and prevent instabilities.
The performances of the proposed algorithms are theoretically
analyzed and verified relying on simulation.
The paper is organized as follows. In Section II, higher
order sliding modes is reviewed with reference to a dynamical
system affected by bounded uncertainty. It is shown that finite-
time stabilization of an uncertain system can be obtained by
solving a rather general class of homogeneous problems for
a chain of integrators. Then, the “Robust Fuller Problem” is
introduced, that is the main tool used in this paper for design of
HOSM controllers. In Section III, second, third and generic
order algorithms are derived. First, a complete convergence
analysis for a class of second order controllers that switch
over branch of parabolas is carried out. Then, a third order
algorithm that is based on the solution to a specific version of
the Robust Fuller Problem is proposed. Finally, the procedure
is generalized in order to obtain an arbitrary order sliding
mode algorithm. In Section IV, the behavior of the proposed
algorithms is tested relying on a simulation benchmark. Some
conclusions (Section V) end the paper.
 II. P RELIMINARIES
Consider a SISO dynamic system affine in the control
variable
ż(t) = a(z, t) + b(z, t)u(t),

(1)
y(t) = σ(z, t),
where z ∈ Rn , u ∈ R, σ : Rn ×R → R is a sufficiently smooth
output function. We say that (1) is an uncertain system if the
system order n and functions a(z, t), b(z, t) are unknown.
However, we assume that the relative degree of the system
[18], [19] is globally well defined, uniform, time-invariant,
and equal to r. The relative degree of the non-linear system
(1) will be a crucial factor in the subsequent analysis. For
autonomous systems, r is the minimum order i of the time
derivatives σ (i) in which the control u appears explicitly. For
suitable functions h(z, t) and g(z, t), one has
y (r) (t) = h(z(t), t) + g(z(t), t)u(t).
Functions h(z(t), t) and g(z(t), t) are assumed to be bounded.
More precisely, it is assumed that there exist positive constants
C, Km , KM , such that
−C ≤ h(z(t), t) ≤ C,
0 < Km ≤ g(z(t), t) ≤ KM .
The design of arbitrary order sliding mode controllers can
be reduced to the design of controllers stabilizing the so-called
auxiliary system (5) to the origin in finite time. Once the origin
is reached, the controlled system is said to exhibit an r-th order
sliding mode. The order of a sliding mode is the number,
increased by one, of derivatives of the sliding variable which
are steered to zero or, in other words, the dimension of vector
σ.
Stabilization of a chain of integrators to the origin has been
widely studied, see e.g. [22], [23]. Herein, we are interested
in controllers that are able to obtain finite-time stabilization
robustly with respect to bounded perturbations of the control.
In the following, all the functions are assumed to be defined
on R+ , L∞ denotes the Banach space of essentially bounded
functions whose norm is denoted by k · k∞ .
The next Theorem is relative to a rather general class of
optimal control problems, hereafter denoted as Problem 1,
formulated on the unperturbed chain of integrators
0 1 ··· 0 0
   
 .. .. . . .   .. 
ż =  . .
 . ..  z +  .  , z ∈ Rr , (6)
 
(2)  0 0 ··· 1   0 
(3) 0 0 ··· 0 v

Let with bounded control kvk∞ ≤ 1. Let

di
σ (i) := i y. c1 = 1, ci ≥ 0, i = 2, . . . , r,
dt r
The goal of any r-th order sliding mode control is to attain
X
L(z) := ci |zi |νr/(r+1−i) , ν > 0.
and keep the manifold i=1

σ (0) (z, t) = σ (1) (z, t) = · · · = σ (r−1) (z, t) = 0, (4) Problem 1:

Z +∞
in finite time. By introducing min L(z(t))dt,
 T kvk∞ ≤1 0
σ := σ (0) , σ (1) , . . . , σ (r−1) , subject to (6) and z(0) = z0 .
condition (4) can be rewritten as σ = 0.
Since the only available information about h(z, t) and In the following, the pair (v ∗ , z∗ ) denotes a generic optimal
g(z, t) are the bounds (2) and (3), the original dynamical solution (control, state trajectory).
system (1) implies the differential inclusion Theorem 1: For all the initial conditions z0 ∈ Rr , Problem
0 1 ··· 0 0 1 admits at least one optimal solution.
   
 .. .. . . .  .. Proof: The proof of this Theorem can be found in [24].
. .. 
 
σ̇ =  . . σ +  .  (5)
  
 0 0 ··· 1   0  There are two limiting cases for Problem 1. For ν → +∞
0 0 ··· 0 f1 + f2 u we obtain r
f1 (t) ∈ [−C, C] , f2 (t) ∈ [Km , KM ] , a.e. X
min ci kzi (t)k∞ ,
kvk∞ ≤1
in which the dependence of σ (i) on the original state variable i=1
z has disappeared. For a background on differential inclusions subject to (6), z(0) = z0 . For ν → 0+ , it can be shown that
see [20]. the solution of Problem 1 converges to the solution of the
In the following, we will need to consider control inputs u minimal-time problem defined as
obtained by discontinuous feedback laws u = U (σ). Solutions Problem 2: Z T
of differential inclusions are to be understood in the Filippov
min 1dt,
sense: they are absolutely continuous functions whose time kvk∞ ≤1 0
derivative satisfy almost everywhere (a.e.) the inclusion (5)
subject to (6), z(0) = z0 , and z(T ) = 0.
with a properly-enlarged right hand side [21]. More precisely,
at each discontinuity point, the right hand-side of (5) is Notice that, while Problem 1 is formulated over an infinite
replaced by the convex hull of the set of velocity vectors horizon, Problem 2 is formulated on the compact interval [0, T ]
obtained by approaching σ from all the directions in Rr , while and have the additional final constraint z(T ) = 0. Therefore,
avoiding zero-measure sets. Problem 2 is not obtained just by letting ν = 0 in Problem

2
1, but is the result of a limiting process. More precisely, if uncertainty reduces the control amplitude αf2 to the smallest
(vν∗ , z∗ν ) denotes a generic optimal solution of Problem 1 with possible value αKm .
ν > 0, and (vT∗ , z∗T ) denotes a generic optimal solution of Thanks to the following lemma, it is possible to reduce
Problem 2, there exists a sequence νk → 0+ such that vν∗k Problem 4 to the standard form of Problem 3. It is useful to
converges to vT∗ in the weak* topology [25] of L∞ , and z∗νk introduce the reduced control amplitude
is point-wise convergent to z∗T , convergence being uniform in
all the compact subsets of R. This is a consequence of weak* αr := αKm − C.
compactness of the the unit ball in L∞ (Banach-Alaouglu Lemma 1: Suppose that αr > 0. Then, if v ∗ is an optimal
Theorem) and Theorem 2 in section III, which shows that, control for Problem 3 with z0 = αr−1 σ 0 , an optimal control
for any ν > 0, the optimal trajectory z∗νk vanish in finite u∗ for Problem 4 is given by
time. This implies that the integral in the definition of the
cost function of Problem 1 can be restricted to a finite interval. u∗ (t) = αv ∗ (t). (7)
For an example of convergence of the corresponding feedback
Relation (7) still holds when u∗ and v ∗ are the limiting optimal
laws, the reader can refer to the case of second order sliding
controls for ν → 0+ of Problems 3 and 4, respectively.
modes in subsection III-A, in which explicit expressions are
Proof: Let
available (notice that parameter β converges to 1/2 when ν
tends to zero). z := αr−1 σ. (8)
By taking ci = 0, ∀i = 2, . . . , n the so-called Fuller’s Apply the Pontryagin’s Maximum Principle [28]. The Hamil-
problem [26] is obtained: tonian function is
Problem 3 (Fuller’s Problem): r−1
X
H(z, λ, f1 , f2 , u) = |z1 |ν + λi zi+1 + λr (f1 + f2 u).
Z +∞
min |z1 (t)|ν dt, i=1
kvk∞ ≤1 0

subject to (6) and z(0) = z0 . The adjoint system is

 λ̇∗1 = −ν|z
 ∗ ∗ ν−2 ∗
By the bang-bang principle [27], the optimal feedback law  1| z1 ,
 λ̇ = −λ∗ ,

for Problem 3 is a discontinuous function of the system 2 1
..
state vector assuming values −1 and +1 almost everywhere. 
 .

However, while in the minimal-time case the optimal state  ∗
λ̇r = −λ∗r−1 .
trajectory can be simply computed by integrating backward
in time the system dynamics starting from the origin, in the First, we show that there are no singular arcs, that is λ∗r = 0 on
general Fuller’s case this is not possible. This is because a a finite time interval cannot occur. Indeed, note that λ∗r = 0 on
certain kind of vibrations occurs near the origin: a detailed a finite time interval would imply λ∗i = 0, (i = 1, . . . , r − 1)
description of this phenomenon is given in [24]. on the same time interval. For 0 ≤ ν ≤ 1 (observe that the
Theorem 1 is relative to an unperturbed chain of integrators. limiting case ν = 0 is included), λ∗1 = 0 is not possible. For
However, in order to derive HOSM algorithms, it is neces- ν > 1, the first equation of the adjoint system yields z1∗ = 0
sary to achieve finite-time stabilization to the origin of the so that zi∗ = 0, (i = 1, . . . , r) as well. Hence, λ∗r 6= 0 almost
perturbed chain (5). Hence, we introduce a robust version of everywhere. Maximization of H with respect to f1 and f2
the Fuller’s problem, which can be named “Robust Fuller’s yields
Problem” associated with the control amplitude α > 0 and
f1∗ = Csgn(λ∗r ),
the output function σ.
Km , uλ∗r < 0,

Problem 4 (Robust Fuller’s Problem): f2∗ =
Z +∞ KM , uλ∗r > 0.
min max |σ (0) (t)|ν dt, Minimization with respect to u yields u∗ = −αsgn(λ∗r ). Thus,
kuk∞ ≤α kf1 k∞ ≤C 0
Km ≤f2 (t)≤KM
a.e. f1∗ = −Csgn(u∗ ),
subject to (5) and σ(0) = σ 0 . f2∗ = Km .
Let v := αr−1 (f1 + f2 u). Then, (5) reduces to (6). Since
The underlying idea is the following: under the assumption
of matched bounded disturbance, we seek the control law v ∗ = αr−1 (−Csgn(u∗ ) + Km αsgn(u∗ )) = sgn(u∗ ),
that guarantees optimality for the worst-case trajectory, that
is the trajectory of the differential inclusion (5) such that the the optimization can be restricted to kvk∞ ≤ 1. Finally,
max with respect to f1 and f2 is attained in Problem 4. It u∗ = |u∗ |sgn(u∗ ) = αv ∗ .
turns out that, independently of ν, the worst-case trajectory is
such that f1 = −Csgn(u) and f2 = Km almost everywhere
(see the proof of Lemma 1). In other words, in the worst- We conclude the section with the following.
case trajectory, the additive uncertainty f1 always opposes the Corollary 1: Assume that αr > 0. For all the initial
control u with maximum amplitude, while the multiplicative conditions, Problem 4 admits at least one optimal solution.

3
Moreover, if V (z) is an optimal feedback law for Problem 3,
then an optimal feedback law for Problem 4 is given by
 
σ
U (σ) = αV . (9)
αr
Proof: By Lemma 1, optimal solutions to Problem 4
are associated with optimal solutions to Problem 3 via the
correspondence (7)-(8). As a consequence, (9) holds. Existence
of optimal solutions is given by Theorem 1.
• W (Tc (σ 0 )) = c(νr+1) W (σ 0 ).
The last two properties also imply that W is coercive. Since
αr > 0, the system (5) is locally finite-time controllable to
the origin. Then, for any σ 0 in a neighborhood of the origin,
there exists a control function such that the integral at the right
hand side of the definition of W can be restricted to a finite
interval [0, T ]. In that interval, σ (0) is bounded so that W is
bounded in a neighborhood of the origin. Recalling that W is
homogeneous,

III. HOSM CONTROLLERS WITH OPTIMAL REACHING |W (σ 0 )| ≤ c−(νr+1) |W (Tc (σ 0 ))|, ∀c > 0.
In this Section, it is first noticed that solutions to Problem 4
Since W is bounded in a neighborhood of the origin, the last
provide control laws that make the auxiliary system (5) finite-
inequality implies that W is bounded on every compact set.
time stable. Then, solutions to the Robust Fuller Problem for (k)
We now show that W is lower semi-continuous. Let σ 0
second order systems are provided. A complete characteriza-
denote a sequence of initial conditions converging to σ 0 and
tion for a class of second order sliding modes control laws (k) (k)
that switch over branch of parabolas is carried out. Next, a u(k) , f1 , f2 denote a corresponding sequence of optimal
generic r-th order algorithm is derived by solving the limiting solutions. Since W is positive and bounded, we can apply the
case ν → 0+ for the Robust Fuller Problem. In the third order Fatou Lemma [29] to obtain
case it is possible to obtain the solution in explicit form, while Z +∞
(k) (k) (k)
in the general case, the controller is characterized by a system lim inf |σ (0) (t; σ 0 , f1 , f2 , u(k) )|ν dt ≥ W (σ 0 ).
k→+∞ 0
of polynomial equations and inequalities. Observe that, in our
approach to the HOSM design, the control amplitude α is This proves lower semi-continuity.
considered a physical constraint and is not to be chosen: this Now, denote by U (σ) an optimal feedback law obtained by
feature greatly simplifies the tuning of design parameters. solving Problem 4 and let
In the next result (Theorem 2), the finite-time stability
property is proven exploiting homogeneity of the system (5) Lk := {σ ∈ Rr : W (σ) = 1/2k }.
and the loss function of Problem 4. Our proof also provides
a specific homogeneous Lyapunov function (that is the value
function associated with Problem 4). Tk (σ 0 , f1 , f2 ) :=
Theorem 2: Suppose that αr > 0. Then, the origin is a
uniformly global finite-time stable equilibrium of system (5) inf {T : ∀t ≥ T, σ(t; σ 0 , f1 , f2 , U (σ)) ∈ Lk+1 } .
with the feedback law obtained by solving Problem 4.
Proof: Denote by σ(t; σ 0 , f1 , f2 , u) a trajectory of (5)
starting from σ 0 , associated with measurable selections f1 (·), τk := sup sup Tk (σ 0 , f1 , f2 ).
σ 0 ∈Lk kf1 k∞ ≤C
f2 (·) of the right hand side, and the control function u(·). Let Km ≤f2 (t)≤KM
c denote a positive number. For any function f ∈ L∞ , fc a.e.

denotes a function such that Above, Lk are the levels sets of W , which are compact
fc (t) = f (ct), a.e. since W is lower semi-continuous and coercive. Scalars Tk
are settling times within which the solution of (5) with a
Introduce also the family of transformations specific selection f1 , f2 reaches Lk+1 starting from σ 0 , while
τk is the maximum time within which all the solutions of the
Tc : Rr → Rr
differential inclusion (5) reach Lk+1 starting from any point
Tc (x1 , . . . , xr ) = (cr x1 , . . . , cxr ) . in Lk .
From (5), it is easy to see that Let c0 = 21/(1+νr) , σ 1 := Tc0 (σ 0 ), and observe that

σ(t; Tc−1 (σ 0 ), f1 , f2 , u) = Tc−1 (σ(ct; σ 0 , (f1 )c , (f2 )c , uc )) . σ 1 ∈ Lk ⇔ σ 0 ∈ Lk+1 .

Consider the value function associated with Problem 4,

Z +∞ W (σ(t; σ 0 , f1 , f2 , U (σ)))
W (σ 0 ) := min max |σ (0) (t)|ν dt = W (σ(t; Tc−1 Tc0 σ 0 , f1 , f2 , U (σ)))
kuk∞ ≤α kf1 k∞ ≤C 0
0
Km ≤f2 (t)≤KM = W (Tc−1 (σ(c0 t; σ 1 , (f1 )c0 , (f2 )c0 , U (σ c0 ))))
a.e. 0
1
subject to (5) and σ(0) = σ 0 . = W (σ(c0 t; σ 1 , (f1 )c0 , (f2 )c0 , U (σ c0 ))).
2
Apparently, we have Hence,
• W (σ 0 ) ≥ 0,
• W (σ 0 ) = 0 ⇔ σ 0 = 0, Tk+1 (σ 0 , f1 , f2 ) = c−1
0 Tk (σ 1 , f1 , f2 ).

4
By definition of τk , Now, it is possible to select 0 for the right-hand side, so that
the origin is the only uniformly asymptotically stable point.
τk+1 = sup sup 0 Tk (σ 1 , f1 , f2 )
c−1
σ 0 ∈Lk+1 kf1 k∞ ≤C
Km ≤f2 (t)≤KM
a.e.
An alternative way to conclude the proof of Theorem 2
can be devised by invoking a recent result [30]. However,
= c−1
0 sup sup Tk (σ 1 , f1 , f2 )
σ 1 ∈Lk kf1 k∞ ≤C this would require to prove that the feedback law obtained by
Km ≤f2 (t)≤KM
a.e.
solving Problem 4 is homogeneous. A discussion on the role
of homogeneity in higher order sliding modes can be found
= c−1
0 τk .
in [31], where robustness and asymptotic accuracy issues are
Now, it is sufficient to show that the control law obtained by analyzed.
solving Problem 4 makes the origin a globally asymptotically
stable equilibrium of system (5). Indeed, this would imply
A. Second Order Sliding Mode Algorithms
τ0 < +∞, so that
∞ Consider Problem 3 in a two dimensional state space. It
τ0
can be shown [26] that an optimal feedback law with ν > 0
X
τk = < +∞.
i=0
1 − c−1
0 is given by
Now, we show that, by the definition of W and the optimal 
1 1

feedback law, W (σ 0 ) < +∞ implies uniform asymptotic sta- V (x1 , x2 ) = −sgn (x1 + β(ν)x2 |x2 |) , β(ν) ∈ , .
4 2
bility. By Corollary 1 and the bang-bang principle [27], U (σ)
is equal to ±α almost everywhere on Rr and is discontinuous Different values for the continuous function β(ν) are tabulated
on a zero-measure set. Thus, we can write Rr = Σ1 ∪ Σ2 , in [32] where it can be seen that the curve β(ν) is decreasing
with Σ1 ∩ Σ2 = Ø, where Σ2 is a zero measure set in which from 1/2 to 1/4 as ν tends to +∞. For ν → +∞, we have
U is discontinuous. Recalling that W is bounded, assume, β(ν) → 1/4, while for ν → 0+ (minimal time problem) the
by contradiction, that for some initial value σ 0 and for some optimal feedback law [27] is given by
f1 , f2 , we have limt→+∞ σ (0) (t; σ 0 , f1 , f2 , U (σ)) 6= 0. This
 0, (x1 , x2 ) ∈ M0 ,

implies divergence of the integral in the definition of W (the
V (x1 , x2 ) = − sgn(x2 ),
(x1 , x2 ) ∈ M1 \ M0 ,
integrand is positive) contradicting W (σ 0 ) < +∞. Hence, we
sgn x1 + 21 x2 |x2 | , else

have
lim σ (0) (t; σ 0 , f1 , f2 , U (σ)) = 0, where
t→+∞

:= (x1 , x2 ) ∈ R2 : x1 = x2 = 0 ,

for any σ 0 and any f1 , f2 . M0
Letting Λ1 := {σ ∈ Rr : σ (0) = 0}, we have 
1

M1 := (x1 , x2 ) ∈ R : x1 + x2 |x2 | = 0 .
2
lim σ(t; σ 0 , f1 , f2 , U (σ)) ∈ Λ1 . 2
t→+∞

Thus, Λ1 contains an asymptotically stable set. Assume σ ∈ By Corollary 1, an optimal family of second order switching
Λ1 . For σ (1) 6= 0, any selection of the velocity vector (the right curves can be derived from the solution to the Robust Fuller’s
hand side) of system (5) strictly points outside the manifold Problem (Problem 4):
Λ1 . Thus, the asymptotically stable set is entirely contained 1 1
   
β(ν)
in Λ2 := {σ ∈ Rr : σ (0) = 0, σ (1) = 0}. By applying this U (σ, σ̇) = −αsgn σ + σ̇|σ̇| , β(ν) ∈ , .
αr 4 2
argument r − 1 times, we conclude that the asymptotically (10)
stable set reduces to the origin Theorem 2 ensures that these laws are indeed second-order
sliding mode algorithms. Note that, when the value of σ̇ is not
Λr := {σ ∈ Rr : σ (i) = 0, i = 0, . . . , r − 1}.
available, one can rely on Levant’s exact differentiator [33] to
Now, assuming 0 ∈ Σ1 , by (5) we have either (U (0) = −α) retrieve σ̇ in finite time, so that the proposed second order
T sliding mode control law can be applied by simply measuring
σ̇ ∈ (0, 0, . . . , [−C − αKM , C − αKm ]) ,
the sliding variable σ.
or (U (0) = α) Remarkably, it can be shown that stabilization in finite time
T is possible with all the values of β > 1/4. In order to see
σ̇ ∈ (0, 0, . . . , [−C + αKm , C + αKM ]) .
that stabilization in finite time is guaranteed also for β > 1/2,
In both cases, since αr > 0, any selection of the velocity let us rewrite the control law (10) in a slightly different form.
vector points outside the origin. Thus, 0 ∈ Σ2 . Over Σ2 , First, observe that, for any x, y ∈ R, γ > 0, p > 1,
U (σ) is discontinuous so that, according to the notion of 1 1 1
Filippov solution, the right hand-side of (5) must be replaced x = −γy|y|p−2 ⇒ y = − x|x|q−2 , + = 1.
(γ)q−1 p q
by the convex hull of the set of velocity vectors obtained by
approaching σ from all the directions in Rr . For system (5), In particular, for p = 3, we have
the convexification leads to
1 1
r r
− 12
x = −γy|y| ⇒ y = − =−
p
σ̇ ∈ (0, 0, . . . , [−(C + αKM ), C + αKM ]) .
T x|x| sgn(x) |x|.
γ γ

5
Hence, we can write B. Third Order Sliding Mode Algorithm
Solution to the general Robust Fuller Problem in the third-
 
αr
r
U (σ, σ̇) = −αsgn σ̇ +
p
sgn(σ) |σ| . (11) order case has a very complicated form. Hence, we restrict the
β(ν)
analysis to the case ν → 0+ , corresponding to the minimal
Equation (11) coincides with the controller with a “prescribed time problem for the worst-case trajectory. The idea of using
convergence law” [8], [34]: time-optimal control in the third order case is present also in
 p  [12]. It is possible to analyze the general case ν > 0 by adapt-
Up (σ, σ̇) = −αsgn σ̇ + λsgn(σ) |σ| , ing the results of [35] and [36] to the Robust Fuller Problem.
The minimal time case is probably the most interesting one, in
provided that
that it provides a bound for the worst-case reaching time while
αr
r
λ= . it is sufficient for the finite-time stabilization of the auxiliary
β(ν) system.
For these controllers, it is easy to see that the condition The solution to Problem 3 with ν → 0+ for a chain of three
integrators is known since [37]. By applying Corollary 1, the
α > 0, λ > 0, αr > λ2 /2, solution to the Robust Fuller Problem 4 with ν → 0+ in the
third order case is obtained.
implies finite-time stabilization of the auxiliary system via a
u0 := 0,

sliding mode on the switching curve (see, for example, [14]).  σ ∈ M0 ,
 u1 := sgn(σ̈),

 σ ∈ M 1 \ M0 ,

This last condition is equivalent to β > 1/2. In fact, Theorem
U (σ) = −α

σ̈ 2 u1
2 shows that finite-time convergence is guaranteed under the 
 u2 := sgn σ̇ + 2αr , σ ∈ M2 \ M1 ,
u3 := sgn(s(σ, σ̇, σ̈)),

less strict condition 
else,
(12)
α > 0, λ > 0, αr > λ2 /4. where
" 3 #
By putting together the previous observations, we can con- σ̈ 3 1 σ̈ 2 2

σ̇σ̈
clude that, for all β > 1/4, the controller (10) is a second order s(σ, σ̇, σ̈) := σ + 2 + u2 √ u2 σ̇ + + .
3αr αr 2αr αr
sliding mode algorithm. However, the behavior of the auxiliary
system trajectories depends on β. For 1/4 ≤ β < 1/2, the and the manifolds M1 , M2 , M3 are defined by
system state may converge twisting around the origin with
M0 := (σ, σ̇, σ̈) ∈ R3 : σ = σ̇ = σ̈ = 0 ,

an infinite number of turns in a finite time-interval. Also,
σ̈ 3
 
σ̈|σ̈|
the control variable may have an infinite number of switches M1 := (σ, σ̇, σ̈) ∈ R : σ − 2 = 0, σ̇ +
3
=0 ,
in a finite time interval. This phenomenon has been called 6αr 2αr
M2 := (σ, σ̇, σ̈) ∈ R3 : s(σ, σ̇, σ̈) = 0 .

“chattering” in the optimal control literature [24], although it
is different from that arising in sliding mode control. We have
It is interesting to observe that the proposed controller
already seen that the values 1/4 ≤ β ≤ 1/2 are particularly
has no free parameters to be tuned (except for the control
interesting being those associated with the Robust Fuller’s
amplitude α).
Problem. For β > 1/2, the optimal control interpretation is lost
and all the trajectories reach the origin in finite time featuring
a sliding mode on the switching curve. The three different
switching regions are illustrated in Fig. 1.
1

0.8

0.6

0.4

0.2

0
σ̈

−0.2

−0.4

−0.6

−0.8

−1 1
−3
−2 0.5
−1
0
0
1 −0.5
2
3 −1
σ̇
σ

Figure 2. Switching surface for the third order controller (12).

Figure 1. The three switching regions: the sliding-mode region is associated
with the controller with a “prescribed convergence law” (β > 1/2), the
Fuller’s region is associated with the controller (10) obtained by solving
the Robust Fuller’s Problem (1/4 ≤ β ≤ 1/2), and, finally, the case C. Generic Order Sliding Mode Algorithms
0 < β < 1/2. The construction of the previous sections can be general-
ized to obtain an r-th order HOSM algorithm with optimal

6
reaching. In order to solve the Robust Fuller’s Problem with
ν → 0+ , the computation of the switching hyper-surfaces
associated with the standard time-optimal control is required.
Again, the time-optimal control has a nested structure similar
to (12). However, for practical purposes it is sufficient to eval-
uate the “outer” switching surface. This can be accomplished
by a rather standard procedure that consists in the integration
of the system equations (6) backward in time, changing the
control sign in r time instants, starting from either +1 or
−1. By parameterizing the state trajectory with the switching
instants, two systems of polynomial equations are obtained,
the first associated with the control sequence {+1, −1, . . .},
the second associated with {−1, +1, . . .}. If the system state
is not exactly on the switching surface, only one of these two
systems admits a solution satisfying ti ≤ 0, i = 1, . . . , r.
By knowing which of these two systems is solvable in the
real negative orthant, it is possible to decide the sign of the Figure 3. Kinematic car model.
optimal control.
Via a suitable change of variable (the procedure is summa-
15
rized in the Appendix), it is possible to show that the solution
to the Robust Fuller Problem with ν → 0+ is given by the
following r-th order HOSM algorithm :
10

U (σ) = α(−1)r+1 Ũ (σ), (13)

where Ũ ∈ {−1, +1} is such that the system of equation and
5
y

inequalities
r
X (r − i + 1)! (i−1)
z1r−i+1 +2 (−1)n+1 znr−i+1 = Ũ σ , (14)
n=2
αr 0
OR

L
i = 1, . . . , r LQC

Reference

z1 ≤ . . . ≤ zr ≤ 0. (15) −5
0 20 40 60 80 100 120 140 160 180 200
x
admits solutions.
The complexity of the switching surface grows very fast
with r. It would be an interesting issue to determine if it is Figure 4. Car trajectory.
possible to obtain exact tests for the switching hyper-surface
exploiting recent algorithms from computational algebra [38]. 5

Alternatively, numerical methods can be employed in order 0

to check whether (14)-(15) admit solutions. It is still an open

σ

−5

−10
issue the derivation of efficient methods (numerical or exact-
−15
algebraic) to deal with this problem. 0 2 4 6 8 10
t
12 14 16 18 20

10

IV. S IMULATION RESULTS 5

σ̇

In this section, we illustrate the behavior of the algorithms 0

proposed in the previous sections, by means of a simulated −5

0 2 4 6 8 10 12 14 16 18 20
t
example. 15

Nonhonolomic car motion planning [39] is a typical appli- 10

5
cation area for testing the performances of HOSM algorithms.
σ̈

0
OR
We use a benchmark problem [13], [15], [16] that gives us the −5

−10
L
LQC
0 2 4 6 8 10 12 14 16 18 20
possibility to test the behavior of our algorithms and obtain a t

comparison with existing approaches. The system is described

by
Figure 5. Sliding variable and sliding derivatives versus time [s].
v
ẋ = v cos ϕ, ẏ = v sin ϕ, ϕ̇ = tan θ, θ̇ = u,
`
where (x, y) are the cartesian coordinates of the rear-axle v = 10 m/s, ` = 5 m represents the longitudinal velocity
middle point, ϕ is the orientation angle, θ is the steering and the distance between the two axles, respectively. The
angle and u is the control variable, see Fig. 3. Parameters goal of the control system is to steer the car from a given

7
 0.6
switching surface is a zero-measure set, it is sufficient to use
the “outer” switching surface when implementing (12). The
0.4 controller laws are denoted by UL3 (Levant 3-th order), UQC 3
3
(Quasi-Continuous 3-th order) and UOR (Optimal Reaching
0.2
3-th order):
h 1 2
i
0
UL3 (σ) = −αsgn σ̈+β2 (|σ̇|3 +σ 2 ) 6 sgn(σ̇+β1 |σ| 3 sgn(σ)) ,
θ

−0.2
 2 1
 2
σ̈ + β2 (|σ̇| + |σ| 3 )− 2 (σ̇ + β1 |σ| 3 sgn(σ))
−0.4 3
UQC (σ) = −α ,
|σ̈| + β2 (|σ̇| + β1 |σ| 3 )− 2
2 1

−0.6

OR
L
" 3 #
σ̈ 3 σ̈ 2 2

LQC
u2 u2 σ̇σ̈
−0.8
0 2 4 6 8 10 12 14 16 18 20
3
UOR (σ) = −αsgn σ + 2 + √ u2 σ̇ + + ,
t
3αr αr 2αr αr

 
Figure 6. Steering angle [rad] versus time [s]. σ̈|σ̈|
u2 := sgn σ̇ + .
2αr
initial position (x0 , y0 , ϕ0 , θ0 ) = (0, 0, 0, 0) to the trajectory For the sake of a fair comparison, the amplitude α is
y = 10 sin(x/20) + 5. Thus, a sliding variable σ is defined as considered a physical constraint and thus fixed for all the
controllers to 20. Moreover, to make the comparison depend
σ = y − 10 sin(x/20) − 5.
only on the control laws, the use of state observers has
It is easy to check that the relative degree of this system is 3: been avoided by directly calculating derivatives of the sliding

1 x  variable. For what concerns the choice of design parameters
σ̇ = v sin ϕ − cos cos ϕ , in UL and UQC , the author suggests setting β1 = 1, β2 = 2,
2 20 3
[13], [15]. Note that in UOR the only needed parameter is the
v2 1
  x 
σ̈ = tan θ cos ϕ + cos sin ϕ reduced control amplitude αr = αKm − C = 10.
` 2 20 Simulations have been carried out in the time horizon 0 −
20 seconds. The controllers have been enabled only at t =

` x
+ sin cos2 ϕ ,
40 20 0.5s (in the first half-second are set to zero) in order to better
v2 1 compare the results with [13], [15]. Results are summarized
 x 
...
σ = u sec2 θ cos ϕ + cos sin ϕ in Fig. 4-6: dotted lines are relative to UL , continuous lines
` 2 20
to UQC and thick-continuous lines to UOR . In Fig. 4, the car
v3 1
 x 
+ tan 2
θ − sin ϕ + cos cos ϕ trajectory obtained with the three controllers is shown together
`2 2 20
with the reference trajectory y = 10sin(x/20)+5. In Fig. 5 the
v3 x
auxiliary system state (σ, σ̇, σ̈) is plotted against time for all
− sin cos ϕ sin ϕ(2 tan θ + 1)
40` 20 the controllers. Finally, in Fig. 6 the dynamics of the steering
v3  x 
angle θ is reported.
+ cos cos3 ϕ.
20 20 From Fig. 5 we see that UOR performs very well in terms
Note that, for small angles ϕ ≈ 0, θ ≈ 0, we have of reaching time: in this respect, it consistently outperforms
the other two controllers. From the analysis of Fig. 6, it can
... v2 v3 x x
σ ≈u + cos = 20u + 50 cos . be seen that UOR produces quite effective steering angles,
` 20 20 20
while avoiding dangerous vibration (differently from UL ).
Hence, under the assumption of small angles The controller UQC produces a quite smooth evolution of
... the steering angle and also avoid vibrations, but is definitely
σ ∈ [−50, 50] + 20u,
slower than UOR and a bit more unpredictable. It is interesting
so that Km ≈ KM ≈ 20, C ≈ 50. However, since moderately to observe that the speed of convergence of UQC is rather
ample angles can manifest in the transient, we conservatively independent of the control amplitude: in [15], the author
let Km = 5, KM = 5000, C = 90. Thus, we consider the obtained a reaching time similar to that of this experiment (for
larger inclusion UQC ) by using only α = 1. This also shows that our bounds
... on C and Km are very conservative. In our setting, using
σ ∈ [−90, 90] + [5, 5000] u.
α = 1 corresponds to C < Km (feasibility condition). Notice
The feasibility condition αr = αKm − C > 0 yields α > 18. that, to compare the controllers with such value of the control
Three different 3-th order controllers are considered: the amplitude, parameter αr should be reduced accordingly in
third order algorithm proposed in [13], the recently proposed UOR . We have made additional simulations by fixing α = 1
quasi-continuous third order sliding mode algorithm [15], and and trying several values of αr in the interval [1, 4]. UOR still
our third order law (12). As previously observed, since the obtains the best reaching times.

8
 V. C ONCLUSIONS Then, we obtain
The connection between the design of high order sliding x̃k+1 = Ak+1 x̃k + bk+1 ,
modes algorithms and the solution to some optimal control
x̃0 = 0,
problems for a perturbed chain of integrators has been inves-
tigated in the paper. The Robust Fuller Problem, which is a with
trk
 
suitable generalization of the so-called Fuller’s problems for r!
r−1
 tk 
a chain of integrators, has been formulated. It is shown that
bk = Ũ (−1) k+1
 (r−1)!

 .
the optimal control law solving the Robust Fuller’s Problem is  .. 
able to attain finite-time stabilization of the auxiliary system
 . 
associated with arbitrary order sliding modes. In this way, a tk
rather general family of HOSM controllers is obtained which Now we can compute the state variables as a function of the
guarantees optimal reaching of the sliding manifold with switching instants ti :
respect to a suitable optimality index. This family includes k−1

k

k−1
several already known HOSM control algorithms. Special X Y X A Pk tj
x̃k =  Aj  bn + bk = e j=n+1 bn + bk .
cases, namely second and third order algorithms are analyzed n=1 j=n+1 n=1
in the paper. The proposed controllers have simple rules
to select the design parameters and exhibit state of the art We have
 Pk 
performances. A
e j=n+1
tj
bn ei =
Pk
A PPENDIX r
X ( j=n+1 tj )s−i tnr−s+1
= Ũ (−1) n+1
Denote by ei the i-th vector of the canonical basis of R r
s=i
(s − i)! (r − s + 1)!
(all zeros, with 1 in position i): Pk
r−i
X ( j=n+1 tj )s tr−i+1−s
ei = 0 . . . 1 . . . 0

T
. = Ũ (−1)n+1 n

s=0
s! (r − i + 1 − s)!
s
System (6) can be written as Pr−i r−i+1

Pk
tnr−i+1−s
s=0 s j=n+1 tj
ẋ = Ax + Bv = Ũ (−1)n+1
(r − i + 1)!
P r−i+1 P r−i+1
with k k
j=n tj − j=n+1 tj
0 1 0 0 = Ũ (−1)n+1 .
 
···
(r − i + 1)!
 0 0 1 ··· 0 

.. .. .. .. ..
 Introduce the variables
A= , B = er .
 
. . . . . k
 0 0 0 1 
  X
... znk := tj , zn := znr ,
0 0 0 ... 0 j=n

We have so that
k r−i+1

k

r−i+1
t2 tr−1
n+1 zn − zn+1
 Pk 
1
 
t 2 ··· (r−1)! e
A
j=n+1
tj
bn ei = Ũ (−1) .
 0

1 t ··· tr−2  (r − i + 1)!
+∞
An n  (r−2)!

X
eAt = t = Now,

 ... .. .. ..
.. .
n! . . . . 
k r−i+1
n=1 k−1
k

r−i+1
 0 0 0 n+1 zn − zn+1
 
... t
X
(x̃k )i = Ũ (−1)

0 0 0 ... 1 n=1
(r − i + 1)!
For the considered system, it is possible to reach the origin (zkk )r−i+1
+ Ũ (−1)k+1
from any initial state with at most r − 1 switches of the (r − i + 1)!
control variable v. Since linear systems are reversible, the (z1 )
k r−i+1
Pk
+ 2 n=2 (−1)n+1 (znk )r−i+1
characterization of the optimal state trajectories can be made = Ũ .
(r − i + 1)!
using a standard techniques from optimal control of linear
systems [27], that is backward integration in time from the In particular, letting k = r, we have
Pr
final state (in this case, the origin). More precisely, we z1r−i+1 + 2 n=2 (−1)n+1 znr−i+1
integrate the system equations backward in time in r adjacent xi = Ũ .
(r − i + 1)!
time intervals, starting from the origin, changing the control Since we are integrating backward in time, we have
sign at each step. Let Ũ = ±1 = Ũ −1 denote the first control
sign, and ti ≤ 0, i = 1, . . . , r,

x̃k := x(t1 , . . . , tk ), Ak := eAtk , which implies

Z tk z1 ≤ . . . ≤ zr ≤ 0.
bk := Ũ (−1)k+1 eAtk e−Aτ er dτ.
0
By applying Corollary 1, controller (13)-(14)-(15) is obtained.

9
 R EFERENCES
[1] V. I. Utkin, Sliding Modes in Optimization and Control Problems. New
York: Springer Verlag, 1992.
[2] A. S. I. Zinober, Ed., Variable Structure and Lyapunov Control. Berlin:
Springer Verlag, 1994.
[3] C. Edwards and S. K. Spurgeon, Sliding Mode Control: Theory and
Applications. London: Taylor and Francis, 1998.
[4] V. I. Utkin, J. Guldner, and J. Shi, Sliding mode control in electrome-
chanical systems. London: Taylor and Francis, 1999.
[5] L. M. Fridman, “Singularly perturbed analysis of chattering in relay
control systems,” IEEE Transactions on Automatic Control, vol. 47,
no. 12, pp. 2079–2084, 2002.
[6] I. Boiko and L. M. Fridman, “Analysis of chattering in continuous
sliding-mode controllers,” IEEE Transactions on Automatic Control,
vol. 50, no. 9, 2005.
[7] A. Koshkouei, K. Burnham, and A. Zinober, “Dynamic sliding mode
control design,” IEE Proceedings - Control Theory and Applications,
vol. 152, no. 4, pp. 392–396, 2005.
[8] A. Levant, “Sliding order and sliding accuracy in sliding mode control,”
International Journal of Control, vol. 58, no. 6, pp. 1247–1263, 1993.
[9] G. Bartolini, A. Ferrara, and E. Usai, “Output tracking control of
uncertain non linear second order systems,” Automatica, vol. 33, no. 12,
pp. 2203–2212, 1997.
[10] ——, “Chattering avoidance by second order sliding mode control,”
IEEE Transaction on Automatic Control, vol. 43, no. 2, pp. 241–246,
1998.
[11] G. Bartolini, A. Ferrara, E. Usai, and V. I. Utkin, “On multi-input
chattering-free second order sliding mode control,” IEEE Transaction
on Automatic Control, vol. 45, no. 9, pp. 1711–1717, 2000.
[12] G. Bartolini, S. Pillosu, A. Pisano, and E. Usai, “Time-optimal stabiliza-
tion for a third-order integrator: a robust state-feedback implementation,”
in Dynamics, Bifurcations and Control, ser. Lecture Notes in Control and
Information Sciences, F. Colonius and L. Gruene, Eds. Springer Verlag,
2002, vol. 273, pp. 131–144.
[13] A. Levant, “Universal single-input–single-output (SISO) sliding-mode
controllers with finite-time convergence,” IEEE Transactions on Auto-
matic Control, vol. 46, no. 9, pp. 1447–1451, 2001.
[14] ——, “Higher-order sliding modes, differentiation and output-feedback
control,” International Journal of Control, vol. 76, no. 9/10, pp. 924–
941, 2003.
[15] ——, “Quasi-continuous high-order sliding-mode controllers,” IEEE
Transactions on Automatic Control, vol. 41, no. 5, pp. 823–830, 2005.
[16] A. Laghrouche, F. Plestan, and A. Glumineau, “Higher order sliding
mode control based on integral sliding mode,” Automatica, vol. 43, no. 3,
pp. 531–537, 2007.
[17] G. Bartolini, A. Ferrara, A. Levant, and E. Usai, “On second order
sliding mode controllers,” in Variable Structure Systems, Sliding Mode
and Nonlinear Control, ser. Lecture Notes in Control and Information
Series, K. D. Young and U. Ozguner, Eds. London: Springer Verlag,
1999, pp. 329–350.
[18] A. Isidori, Nonlinear control systems, 3rd ed. Berlin: Springer Verlag,
1995.
[19] D. Liberzon, A. S. Morse, and E. D. Sontag, “Output-input stability and
minimum-phase nonlinear systems,” IEEE Transactions on Automatic
Control, vol. 47, no. 3, pp. 422–436, March 2002.
[20] J. P. Aubin and A. Cellina, Differential Inclusions, Set-Valued Maps and
Viability Theory. Berlin: Springer Verlag, 1984.
[21] A. F. Filippov, Differential Equations with Discontinuous Right-Hand
Side. Dordrecht, The Netherlands: Kluwer, 1988.
[22] H. J. Sussman and Y. Yang, “On the stabilizability of multiple integrators
by means of bounded feedback,” in Proc. IEEE conference on Decision
and Control, vol. 1, 1991, pp. 70–73.
[23] A. R. Teel, “Global stabilization and restricted tracking for multiple
integrators with bounded controls,” Systems and Control Letters, vol. 18,
no. 3, pp. 165–171, 1992.
[24] M. I. Zelikin and V. F. Borisov, Theory of Chattering Control with
applications to Astronautics, Robotics, Economics, and Engineering, ser.
Systems and Control: Foundations and Application. Boston: Birkhäuser,
1994.
[25] K. Yosida, Functional analysis, 6th ed. Springer, 1980.
[26] A. T. Fuller, “Relay control systems optimized for various performance
criteria,” in Automatic and Remote Control (Proc. First World Congress
IFAC, Moscow, 1960), vol. 1, London, 1961, pp. 510–519.
[27] M. Athans and P. L. Falb, Optimal Control. New York: McGraw Hill,
1966.
10
[28] L. S. Pontryagin, V. G. Boltyanskii, R. V. Gamkrelidze, and E. F.
Mishchenko, The mathematical theory of optimal processes, 2nd ed.
Wiley, 1962.
[29] W. Rudin, Real and complex analysis. McGraw-Hill, 1987.
[30] Y. Orlov, “Finite time stability and robust control synthesis of uncertain
switched systems,” SIAM Journal of Control and Optimization, vol. 43,
no. 4, pp. 1253–1271, 2005.
[31] A. Levant, “Homogeneity approach to high-order sliding mode design,”
Automatica, vol. 50, no. 11, pp. 1812–1816, 2005.
[32] A. T. Fuller, “Minimization of various performance indices for a system
with bounded control,” International Journal of Control, vol. 41, no. 1,
pp. 1–37, 1985.
[33] A. Levant, “Robust exact differentiation via sliding mode technique,”
Automatica, vol. 34, no. 3, pp. 379–384, 1998.
[34] S. V. Emelyanov, S. K. Korovin, and L. V. Levantovsky, “Higher order
sliding modes in the binary control systems,” Soviet Physics, Doklady,
vol. 31, no. 4, pp. 291–293, 1986.
[35] P. E. Grensted and A. T. Fuller, “Minimization of integral-square error
for non-linear control system of third and higher order,” International
Journal of Control, vol. 2, no. 1, pp. 33–73, 1965.
[36] E. P. Ryan and C. M. Dorling, “Minimization of a non-quadratic
cost functional for third-order saturating control systems,” International
Journal of Control, vol. 34, no. 2, pp. 231–258, 1981.
[37] A. A. Feldbaum, “On synthesis of optimal systems with the help of
phase space,” Avtomatika i Telemehanika, vol. 16, no. 2, pp. 129–149,
1955.
[38] U. Walther, T. T. Georgiou, and A. Tannenbaum, “On the computation
of switching surfaces in optimal control: a Gröbner basis approach,”
IEEE Transactions on Automatic Control, vol. 46, no. 4, pp. 534–540,
2001.
[39] R. Murray and S. Sastry, “Nonhonolomic motion planning,” IEEE
Transactions on Automatic Control, vol. 38, no. 5, pp. 700–716, 1993.
Francesco Dinuzzo Francesco Dinuzzo was born in
Tricarico (MT), Italy on July, 4, 1984. He received
the Laurea degree (with honors) in 2005 and the
Laurea Magistrale degree (with honors) 2007, both
in Computer Science Engineering from the Univer-
sity of Pavia. In 2007, he also received the “Corsi
Ordinari” degree from IUSS (Istituto Universitario
di Studi Superiori - Institute for Advanced Studies),
Pavia. On July 5, 2008 he received the “2006/2007
Best Student Award” (Faculty of Engineering, Uni-
versity of Pavia). Currently, he is a Visiting Stu-
dent at CBLC (Center for Biological and Computational Learning), MIT
(Massachusets Institute of Technology). He is pursuing the Ph.D. degree
in Mathematics and Statistics at the University of Pavia. His research
interests include machine learning, system identification, non-linear control,
regularization methods, distributed computation.
 Antonella Ferrara Antonella Ferrara was born in
Genova, Italy. As a student at the Faculty of Engi-
neering of the University of Genova, she got the
”IEEE North Italy Section Electrical Engineering
Student Award” in 1986. She received the Laurea
Degree with honors in Electronic Engineering in
1987 and the Ph.D. in Computer Science and Elec-
tronics in 1992 from the University of Genova. She
has been Assistant Professor in the Department of
Communication, Computer and System Sciences of
the University of Genova since 1992. In November
1998 she became Associate Professor of Automatic Control at the University
of Pavia. Since January 2005 she is Full Professor of Automatic Control in the
Department of Computer Engineering and Systems Science of the University
of Pavia. Her research activities are mainly in the area of sliding mode control,
with application to automotive control, process control, and robotics. She
has authored or coauthored more than 200 papers, including more than 65
international journal papers. From 2000 to 2004 she has been Associate Editor
of the IEEE Transactions on Control Systems Technology. She is Associate
Editor of the IEEE Transactions on Automatic Control for the triennium 2007-
2009. She currently is and has been member of the International Program
Committee of numerous international conferences and events. She is Senior
Member of the IEEE Control Systems Society, member of the IEEE Technical
Committee on Variable Structure and Sliding Mode Control, member of
the IEEE Robotics and Automation’s Technical Committee on Autonomous
Ground Vehicles and Intelligent Transportation Systems, and member of the
IFAC Technical Committee on Transportation Systems.

11