Finite-time controller design for nonholonomic mobile robot using the

heisenberg form
Michael Defoort and Mohamed Djemaı̈

Abstract— This paper considers the finite-time controller nonholonomic systems [11], [12]), the derived control laws
design for wheeled mobile robots. After the mobile robot are of discontinuous type and can lead to discontinuous
dynamics have been transformed into an advantageous form, velocities in practice. Some recent works proposed first order
called the heisenberg form, a finite-time controller is designed
in order to promulgate the practical stability of the system. The sliding mode controllers for the stabilization of the perturbed
desired stability property of the closed-loop system is provided Heisenberg system [9], [13], [14], [15]. Nevertheless, in
by applying a homogeneous sliding mode controller, driving [13], high gains are needed in the design of the sliding
the system to zero dynamics in a known finite-time. Finally, variable. This fact could yield difficulties to tune the control
simulation results demonstrate the effectiveness of the proposed parameters. In [9], [14], an integral sliding mode controller is
strategy.
Index Terms— Heisenberg system, Nonholonomic mobile proposed in order to remove such a problem. However, these
robot, sliding mode control, Lyapunov method. approaches only enable an asymptotic convergence of the
system. In some applicative fields, finite-time convergence
I. I NTRODUCTION is needed.
In [16], a distributed receding horizon planner is proposed
Control of systems which fail necessary conditions for
to solve the motion navigation of multi-robot systems. In
the existence of smooth static state feedback that guarantees
order to apply this scheme, a controller should be added to
asymptotic stability of the equilibrium [1], has been an
track the planned trajectory in spite of external disturbances
active field of research for two decades. One reason is that
and inherent discrepancies between the model and the real
nonholonomic systems fall into this class [2]. Work on the
process. If the controller design is not well done, the tracking
stabilization problem for such systems has mainly focussed
errors may accumulate over each time segment over which
on the design of time-varying or discontinuous feedback
planned trajectories are computed. This fact could yield col-
controllers. Thus, many control strategies such as smooth
lisions between robots and obstacles. That is why, it should
time-varying feedbacks [3], sinusoidal and polynomial con-
be desirable to achieve the stabilization of the tracking errors
trols [4], controls based on backstepping approaches [5], and
by the end of the time interval over which the trajectories
nonsmooth feedbacks [6], [7] have been investigated.
are computed using the receding horizon framework.
An important class of such systems is the Heisenberg
system, or nonholonomic integrator: The discussed claim for real-time requirements can be
 effectively accomplished using finite-time controller. There
 ẋ1 = u1 are several methods achieving finite-time convergence, e.g.
ẋ2 = u2 (1) sliding mode controllers [17], [18], homogeneous controller

ẋ3 = x1 u2 − x2 u1 [19], [20], [21], but some of these are not smooth like first
It belongs to the general class of driftless systems where the order sliding mode controllers.
dimension of the control input is smaller than the dimension This paper focusses on the real-time stabilizing controller
of the state. The problem of stabilizing this class of system is design for a nonholonomic mobile robot. First, the system
a nontrivial task since the linearized part is not controllable. is transformed into the heisenberg form using an appropriate
In spite of its inherent difficulties, there are many practical diffeomorphism. Then, a controller based on homogeneity is
reasons for focusing on this particular system since many designed in order to guarantee the finite-time stability of the
mechanical systems like wheeled mobile robots are closely system. An estimation of the settling time is provided.
related to the nonholonomic integrator [8], [9]. The problem statement is reported in Section II. Prelimi-
In many works about the stabilization of the Heisen- nary results are given in Section III. Then, the control laws
berg system [10] (or, in general, about the stabilization of are designed in Section IV. Finally, simulation results for the
stabilization of a wheeled mobile robot are given.
Michael Defoort and Mohamed Djemaı̈ are with Univ. Lille
Nord de France, F-59000 Lille, France; CNRS, FRE 3304, F-
59313 Valenciennes, France; UVHC, LAMIH, F-59313 Valenciennes,
II. P ROBLEM STATEMENT
France [email protected], mohamed.djemai@univ-
valenciennes.fr. It is known, that a third order nonholonomic system of
This work has been supported by International Campus on Safety and the form ẋ = B(x)u can be locally written as the so-called
Intermodality in Transportation, the European Community, the Delega- Heisenberg system (1). For example, the link between the
tion Regionale a la Recherche et a la Technologie, the Ministere de
l’Enseignement suprieur et de la Recherche, the Region Nord Pas de Calais wheeled mobile robot and the nonholonomic integrator has
and the Centre National de la Recherche Scientifique. been highlighted in [9].
 Let us consider a unicycle-type mobile robot, see Fig. System (6) can be seen as the interconnection of two sub-
1, with two driving wheels mounted on the same axis and systems: a 2−dimensional integrator and a scalar nonlinear
independently controlled by two actuators (DC motors). This system whose dynamics is entirely driven by ż = U T JX.
robot is fully described by a three dimensional vector of One of the main difficulty is the fact that the stabilization
generalized coordinates q constituted by the coordinates of X leads to zeroing right-hand side of (6) and therefore,
(x, y) of the midpoint between the two driving wheels and the variable z cannot be steered to zero anymore. This
by the orientation angle θ with respect to a fixed frame: simple observation implies that X must not vanish before z
T in order to stabilize system (6). Therefore, it is critical to
q = [x, y, θ] (2)
select an appropriate sliding surface.
Under the pure rolling and non slipping condition, the ideal
kinematic equations are: In this paper, the objective is to design a finite-time
  controller for system (6).
cos θ 0  
wl
q̇ =  sin θ 0  (3)
wa III. P RELIMINARY RESULTS
0 1
where wl and wa are the linear and angular velocities, A. Notations
respectively. • We denote by t → x(t, x0 ) a solution of the system:

ẋ(t) = f (x(t))
y− axis (8)
x(t0 ) = x0
where x ∈ R, f : D 7→ R is continuous on an open
θ
neighborhood D of the origin x = 0.
y • We define the continuous function ⌈a⌋b as the function
described by:
∀a ∈ R, ∀b ≥ 0, ⌈a⌋b = |a|b sign (a) (9)

B. Finite-time stability
O x x−axis Let us recall the main definition and properties for finite-
time stability.
Fig. 1. Unicycle-type mobile robot. D EFINITION 1: [19] The origin of system (8) is finite-
time convergent if there is an open neighborhood U ⊂ D of
In order to express the model in the Heisenberg form, the the origin and a function T : U \ {0} 7→ R+ such that for
following invertible transformation is used: all x0 ∈ U , the solution x(t, x0 ) of system (8) is defined
T
[z, X] = T [x, y, θ]
T
(4) (and unique) and x(t, x0 ) ∈ U \ {0} for t ∈ [0, T (x0 )) and
limt→T (x0 ) x(t, x0 ) = 0. T (x0 ) is called the settling time.
where the invertible matrix T ∈ R3×3 is defined as: If the origin of (8) is finite-time convergent, the set of point
 
−θ cos θ + 2 sin θ −θ sin θ − 2 cos θ 0 x0 such that x(t, x0 ) → 0 is called the domain of attraction
T = 0 0 1  (5) of the solution.
cos θ sin θ 0 The origin of (8) is finite-time stable if it is Lyapunov
T
stable and finite-time convergent.
where z ∈ R and X = [x1 , x2 ] ∈ R2 are the new state If U = D = R, the origin is said to be globally finite-time
vector. stable.
Differentiating z and X with respect to time and using
equations (3)-(5) yield the following differential equations: The following result gives a sufficient condition for system
 (8) to be finite time stable:
ż = U T JX
(6) L EMMA 1: [19] Let the origin be an equilibrium point for
Ẋ = U
system (8), and f be continuous on an open neighborhood U
where J is a constant, skew symmetric matrix: of the origin. If there is a Lyapunov function1 V : U → R+
  such that the time derivative along the solution of (8) satisfies
0 1
J=
−1 0 V̇ ≤ −cV a (10)
T
The control input U = [u1 , u2 ] introduced in (6) is defined with a ∈ (0, 1) and c > 0. Then, the origin is finite-time
as follows: stable.

u 1 = wa 1 V is a continuously differentiable function defined on U such that V is
(7)
u2 = wl − wa (x sin θ − y cos θ) positive definite and V̇ is negative definite.
C. Preliminary results as in the conventional way (see [21] for further details).
Let us consider the following system: Since
  
α|x1 |α−1 (x2 − K1 ⌈x1 ⌋α )
ẋ(t) = u
(11) ξ˙ = 2α−1
x(0) = x0  2 ⌈x1 ⌋
−K 
α−1 α (x 2 − K1 ⌈x1 ⌋α )
= |x1 |
−K2 ⌈x1 ⌋α
L EMMA 2: Under the controller = |x1 | α−1
A0 ξ
u = −⌈x(t)⌋α (12) the time derivative of V along the solutions of system (14)
with α ∈ (0, 1), system (11) is continuous everywhere and is given by:
locally Lipschitzian everywhere except at the origin. The

V̇ = |x1 |α−1 ξ T AT0 P + P A0 ξ
solutions are described by: x(t, x0 ) = 1
( = − ξ T Qξ

1 1−α |x1 |1−α
sign (x0 ) |x0 |1−α − (1 − α)t 1−α if 0 < t < |x1−α
0|
1
0 else ≤ − λmin {Q}||ξ||2
|x1 |1−α
and thus converge to zero in finite-time.
where ||ξ|| is the Euclidean norm on R2 . Therefore, V̇ is
negative definite since Q > 0.
R EMARK 1: The drawbacks of this controller are that the
From the standard inequality for quadratic forms, one can
uncertainty limit defines the boundary layer, and even in the
deduce:
absence of uncertainty, the domain of attraction to x = 0 is
proportional to the discrete interval τ under the discrete-time λmin {P }||ξ||2 ≤ V (ξ) ≤ λmax {P }||ξ||2 (17)
control.
1−α
Using (17) and 0 < , one can obtain:
T HEOREM 1: Let us consider A0 : 2α
  1−α
−αK1 α 1−α 1−α V 2α (ξ)
A0 = |x1 | ≤ ||ξ|| α ≤ 1−α (18)
−K2 0 λmin {P }
2α

where K1 > 0, K2 > 0 and 31 < α < 1 are chosen such As α > 13 , one can deduce that V (ξ) is a strong Lyapunov
that matrix A0 is Hurwitz. Then, system (11) is globally function for system (14) since
finite-time stable under the smooth control law:
Z 3α−1
V̇ ≤ −γV 2α
u = −K1 ⌈x(t)⌋α − K2 ⌈x(t)⌋2α−1 dt (13)
with 1−α

Proof: The system (11) and (13) can be equivalently λmin {Q}λmin
2α
{P }
γ=
presented by the system of two first order equations: λmax {P }

ẋ1 = x2 − K1 ⌈x1 ⌋α Since the transformation (15) is continuous, it follows that a
(14)
ẋ2 = −K2 ⌈x1 ⌋2α−1 trajectory of system (14) starting from x0 converges to the
R origin in finite-time and reaches that point at most after:
with x1 = x and x2 = −K2 ⌈x(t)⌋2α−1 dt.
1−α
Let us consider the following vector: 2α V 2α (x0 )
    T = (19)
ξ1 ⌈x1 ⌋α 1−α γ
ξ= = (15)
ξ2 x2 The reaching time depends on parameters α, K1 , K2 and
Define the candidate Lyapunov function as: x0 . One can tune these parameters in order to settle down
the system arbitrarily fast.
V (ξ) = ξ T P ξ (16)
 
p1 p3
P = PT = > 0 is the solution of the following The proposed controller provides a finite-time convergence
p3 p2
algebraic Lyapunov equation: to the domain |x| ∼ τ 2 in a digital implementation. It also
provides smoothness of control derivative and good robust-
AT0 P + P A0 = −Q
ness properties. Hence, the proposed finite-time controller
where Q = QT > 0 is a positive definite matrix. design will be incorporated in order to guarantee the finite-
One can note that V (ξ) is continuous, positive definite and time stabilization of system (6).
radially unbounded in R2 . Furthermore, it is differentiable
everywhere except on the manifold {x1 = 0}. Since the R EMARK 2: It is interesting to note that the case α =
1
trajectory of system (14) cannot stay on this set, before 2 corresponds to the classical super-twisting controller. It
reaching the origin, the derivative of V can be calculated enables a finite time stabilization of the tracking errors.
 IV. F INITE - TIME CONTROLLER DESIGN on the manifold X T X = 0. Nevertheless, the choice of the
Since X must not vanish before z, it appears interesting sliding vector (20) enables to avoid such control singularities.
T
to use the following sliding vector S = [s1 , s2 ] ∈ R2 with
 Based on Theorem 1, the control law W which ensures the
s1 = z finite time stabilization of the sliding vector S is designed
(20)
s2 = 21 X T X − ϕ(z) as:
where ϕ : R → R+ is a differentiable positive definite   R 
function, i.e. −x2 + x1 z x1 −K1 ⌈s1 ⌋α − K2 R ⌈s1 ⌋2α−1 dt
U=
 x1 + x2 z x2 −K1 ⌈s2 ⌋α − K2 ⌈s2 ⌋2α−1 dt
ϕ(z) > 0 if z 6= 0 (22)
ϕ(z) = 0 if z = 0 where the coefficients K1 > 0, K > 0 and 1
< α < 1
2 3
It is chosen as a radially unbounded convex function. One −αK1 α
are chosen such that the matrix is Hurwitz.
can choose for instance: ϕ(z) = dz 2m where d ∈ R+ and −K2 0
m ∈ N are suitable constants. Note that this function implies Here, it is necessary to point out that the proposed finite-time
a degree of correspondence between the convergence rate controller (22) ensures that X does not vanish before z. The
of kXk and the one of z. settling time can be arbitrarily tuned. An upper estimation
of the settling time is given in equation (19).
Here, the control input U is designed such that the state
S, defined in (20), is stabilized toward zero in finite time. In R EMARK 3: The continuity of the resulting finite-time
such a case, one gets the following desired properties: kinematic controller allows the extension of the robot control
• first, the convergence of s1 implies the finite-time at the dynamic level by having the dynamic subsystem
convergence of z toward zero, outputs track the desired kinematic control inputs of the
• then, after the stabilization of z, the convergence of kinematic subsystem (see [23] for details about the dynamic
s2 implies the finite-time convergence of kXk2 toward level).
zero since ϕ is positive definite,
T T
• at last, the stabilization of Ξ = [z, X ] implies the V. S IMULATION RESULTS
stabilization of q = [x, y, θ]T .
This section demonstrates the performance of the finite-
Note that the relative degree of system (6) with respect to time controller introduced in (22). The following simulations
S is [1, 1]T . This vector is related to the number of times showcase two different scenarios. The robot is represented
that s1 and s2 have to be differentiated before the control by the nonlinear model (3). The main parameters of the
input explicitly appears (see [22] for further details about the proposed finite-time controller are ϕ(z) = 0.005z 2 and
notion of relative degree). Indeed, the time derivative of S α = 34 . Data acquisition and control implementation are
along the trajectories of (6) is given by: performed at a sampling period of τ = 0.001s.
Ṡ = M (Ξ)U (21)
A. Stabilization problem
with  T

−X J Here, the control objective is the stabilization of the
M (Ξ) =
X T + ∂ϕ(z) T
∂z X J nonholonomic mobile robot towards the origin. The sliding
The system is strongly coupled by the matrix M (Ξ) ∈ R2×2 . vector S has been chosen according to (20). Using the
In order to circumvent this problem, let us define the static controller defined in (22), the finite-time stabilization of
feedback: [s1 , s2 ]T is guaranteed. Hence, one gets the following desired
h i properties:
U = JX X − ∂ϕ(z) ∂z JX
W
• first, the convergence of s1 implies the finite-time
where W = [w1 , w2 ] ∈ R2 . Equation (21) becomes: convergence of z toward zero,
• then, after the stabilization of z, the convergence of
Ṡ = N (Ξ)W s2 implies the finite-time convergence of kXk2 toward
with zero since ϕ is positive definite,
 h i • at last, the stabilization of [z, X T ]T implies the stabi-
−X T J ∂ϕ(z) lization of [x, y, θ]T .
N (Ξ) = T ∂ϕ(z) T JX X − ∂z JX
 X + ∂z X J  The trajectory of the mobile robot is depicted in Fig. 2. As
2 ∂ϕ(z) 2
kJXk − ∂z kJXk
=  ∂ϕ(z) 2 2
  2
2
 expected, the proposed controller ensures the stabilization
− ∂z kJXk kXk + ∂ϕ(z) ∂z kJXk of the mobile robot towards the origin. One can see in
Fig. 3 the linear and angular velocities of the robot. One
2 2
Note that det N = kJXk kXk . Since J is a nonsingular should highlight that the robot goes backward since its initial
matrix, the decoupling matrix N is always invertible except orientation is 170◦ .
 and
2.5

 x1r = θr
2 x2r = xr cos θr + yr sin θr

zr = xr (−θr cos θr + 2 sin θr ) − y(θr sin θr + 2 cos θr )
1.5
The trajectory tracking problem is equivalent to stabilize
e2 ]T and ze, i.e. the difference between the actual
y [m]

1
X = [ex1 , x
and reference trajectories:
0.5

 x e1 = x1 − x1r
xe2 = x2 − x2r
0 
ze = z − zr

−0.5 Through these state transformations, one can obtain the

−1.5 −1 −0.5 0 0.5 1 following differential equations:
x [m]

zė = U T JX + Π
Fig. 2. Stabilization of the nonholonomic mobile robot.
Ẋ = U

2 with Π = 2 (wa,r x2 − wl,r sin x1 ). One can note that this

model is similar to (6) with a term Π which is consider as
a perturbation.
wl [m/s]

−2 Here, the control objective is to steer the nonholonomic

mobile robot from a given initial position to a neighborhood
−4 of the reference time-varying trajectory given by the follow-
0 2 4 6 8 10
ing reference linear and angular velocities:
1

wl,r = 10(1 − exp(−t))
wa [rad/s]

0 (23)
wa,r = sin(t)
−1 The resulting reference time-varying Cartesian position and
orientation are depicted in Fig. 4. The linear velocity is
−2
0 2 4 6 8 10 related to a first order dynamic system with appropriate pa-
t [s] rameters in order to take into account the robot specifications.
The angular velocity is chosen as a sinusoidal function.
Fig. 3. Linear and angular velocity of the robot Without retuning the controller (i.e using the same control
parameters), the proposed strategy is implemented. The ac-
tual and the reference trajectories of the mobile robot are
B. Tracking problem depicted in Fig. 4-5. As expected, the stabilization of the
In order to show the robustness of the proposed strategy, tracking errors of the mobile robot is guaranteed in spite of
one could apply the controller (22) to solve the trajectory the presence of perturbation Π because of the robustness of
tracking problem. the proposed controller. Furthermore, the proposed controller
To formulate this problem, we define the following refer- is also robust with respect to noise measurements thanks to
ence model: its homogeneous properties.
 
cos θr 0  
wl,r
q̇r =  sin θr 0 
wa,r
0 1 VI. C ONCLUSION
T
where qr = [xr , yr , θr ] denotes the reference trajectory, and This paper considers the finite-time controller design for
wl,r and wa,r are the reference linear and angular velocities. wheeled mobile robots using the heisenberg transformation.
Similarly to Section II, the following transformations are The desired stability property of the closed-loop system is
used: provided by applying a sliding mode controller, driving the

 x1 = θ system to zero dynamics in a known finite-time. Finally, sim-
x2 = x cos θ + y sin θ ulation results demonstrate the effectiveness of the proposed

z = x(−θ cos θ + 2 sin θ) − y(θ sin θ + 2 cos θ) strategy.
 30
refence refence
30 actual actual

wl [m/s]
20
25
10
20
0
0 1 2 3 4 5
y [m]

15
20
10 refence

w [rad/s]
10 actual
5
0

a
0
−10
−20 −10 0 10 20 0 1 2 3 4 5
x [m] t [s]

Fig. 4. Actual and reference trajectories for trajectory tracking. Fig. 6. Linear and angular velocity of the robot.

setpoint control of wheeled mobile robots: a Lyapunov approach”,

1 refence Automatica, 36, pp. 1741–1746, 2000.
actual [9] Defoort, M., Floquet, T., Perruquetti, W., and Drakunov, S.V., “Integral
sliding mode control of an extended Heisenberg system”, IET Control
0 Theory and Applications, 3, pp. 1409–1424, 2009.
[10] Khennouf, H., and Canudas de Wit, C., “On the construction of
stabilizing discontinuous controllers for nonholonomic systems”, Proc.
−1 IFAC Nonlinear Control Systems Design Symposium, Tahoe City, CA,
y [m]

pp. 747–752, 1995.

−2 [11] Marchand, N., and Alamir, M., “Discontinuous exponential stabiliza-
tion of chained form systems”, Automatica, 39, pp. 343–348, 2003.
[12] Prieur, C., and Astolfi, A., “Robust stabilization of chained systems via
−3 hybrid control”, IEEE Trans. on Automat. Cont., 48, pp. 1768–1772,
2003.
[13] Drakunov, S. V., Floquet, T., and Perruquetti, W., “Stabilization and
−4 tracking control for an extended Heisenberg system with a drift”, Syst.
Cont. Let., 54, pp. 435–445, 2005.
[14] Defoort, M., Palos J., Floquet, T., Kokosy, A. and Perruquetti, W.,
−4 −3 −2 −1 0 1 2 “Practical Stabilization and Tracking of a Wheeled Mobile Robot With
x [m] Integral Sliding Mode Controller”, Proc. IEEE Int. Conf. On Decision
and Control, New Orleans, USA, 2007.
[15] Barbot, J. P., Djemai, M., Floquet T., and Perruquetti, W., “Practical
Fig. 5. Zoom. stabilization of a unicycle-type mobile robot system using higher
order sliding mode control”, Proc. European Control Conference,
Cambridge, UK, 2003.
[16] Defoort, M., Kokosy, A., Floquet, T., Perruquetti, W., and Palos J.,
R EFERENCES “Motion planning for cooperative unicycle-type mobile robots with
limited sensing ranges: A distributed receding horizon approach”,
Robotics and autonomous systems, 57, pp. 1094-1106, 2009.
[1] Brockett, R. W. , “Asymptotic stability and feedback stabilization”, in [17] Utkin, V. I., Guldner, J. , and Shi, J., “Sliding modes control in
Differential Geometric Control Theory, R. W. Brockett, R. Millman electromechanical systems”, London, Taylor and Francis, 1999.
and H. Sussmann, eds., Boston: Birkhauser, pp. 181–191, 1983. [18] Perruquetti, W., and Barbot, J. P., “Sliding mode control in engineer-
[2] Kolmanovsky, I., and McClamroch, N. H., “Developments in nonholo- ing”, Control Eng. Series, Marcel Dekker, 2002.
nomic control problems”, IEEE Control Syst. Mag., 15, pp. 20–36, [19] Bhat, S., and Bernstein, D., “Continuous finite-time stabilization of
1995. the translational and rotational double integrators”, IEEE Trans. on
[3] Pomet, J. B., “Explicit design of time-varying stabilizing control laws Automat. Cont., 43, pp. 678–682, 1998.
for a class of controllable systems without drift”, Syst. Cont. Let., 18, [20] Bhat, S., and Bernstein, D., “Geometric homogeneity with applications
pp. 147–158, 1992. to finite time stability”, Mathematics of Control, Signals and Systems,
[4] Murray, R., and Sastry, S., “Nonholonomic Motion Planning: Steering 17, pp. 101–127, 2005.
Using Sinusoids”, IEEE Trans. on Automat. Cont., 38, pp. 700–716, [21] Hong, Y., “Finite time stabilization and stabilizability of a class of
1993. controllable systems”, Syst. Cont. Let., 46, pp. 231–236, 2002.
[5] Jiang, Z. P., and Nijmeijer, H., “Tracking control of mobile robots: a [22] Isidori, A., “Nonlinear control systems (2nd ed.)”, New York: Springer,
case study in backsteeping”, Automatica, 33, pp. 1393–1399, 1997. 1989.
[6] Astolfi, A., “Discontinuous control of nonholonomic systems”, Syst. [23] Zhang, J., and Jayasuriya, S., “Finite time settling real time control
Cont. Let., 27, pp. 37–45, 1996. for multi robot formation”, Proc. IEEE Int. Conf. On Decision and
[7] Defoort, M., Floquet, T., Kokosy, A., and Perruquetti., “Integral sliding Control, China, pp. 2990–2995, 2009.
mode control for trajectory tracking of a unicycle type mobile robot”,
Integrated Computer Aided Engineering, 13, pp. 277–288, 2006.
[8] Dixon, W. E., Jiang, Z. P., and Dawson, D. M., “Global exponential