Proceedings of the 42nd IEEE

Conference an k i s i a n and Control

Maui, Hawaii USA,December 2003

TuP08-3

Nonlinear Path Following with Applications to the Control of Autonomous Underwater Vehicles
L. Lapierre, D. Soetanto, A. Pascoal
Institute for Systems and Robotics and
Dept. Electrical Engineering and Computers
Instituto Superior Tkcnico
AV. Rovisco Pais, 1, 1049-001 Lisboa, Portugal
E-mail: {lpl, didik, antonio} @isr.ist.utl.pt
Abslracl-This paper derives a control law to steer the
dynamic model of an autonomous underwater vehicle (AUV)
along a desired path. The methodology adopted for path
following deals explicitly with vehicle dynamics. Furthermore, it
overcomes stringent initial condition constraintsthat are present
in a number of path follo-ing control strategies described in
the literature. Controller design builds on Lypaonov theory
and backstepping techniques. The multing nonlinear feedback
control law yields convergence of the path follo-ing error trajectories to zero. Simulation results illustrate the performance
of the control system proposed.

I. INTRODUCTION
Fast paced progresses in marine robotics are steadily
affording scientists advanced tools for ocean exploration and
exploitation. However, much work remains to be done before
marine robots can roam the oceans freely, acquiring scientific
data at the temporal and spatial scales that are naturally
imposed by the phenomena under study. To meet these goals,
robots must be equipped with systems to steer them accurately and reliably in the harsh marine environment. For this
reason, there has been over the last few years considerable
interest in the development of advanced methods for marine
vehicle motion control. Namely, point stabilization, trajectory
tracking, and path following control.
Point stabilization refers to the problem of steering a vehicle to a final target point, with a desired orientation. Trajectory tracking requires a vehicle to track a time-parameterized
reference curve. Finally, path following control aims at
forcing a vehicle to converge to and follow a desired spatial
path, without any temporal specifications. The latter objective
occurs for example when it is required that an AUV examine
an area by performing a lawn mowing maneuver along
desired tracks with great accuracy, at speeds determined
by a scientific end-user. The underlying assumption in path
following control is that the vehicles forward speed tracks a
desired speed profile, while the controller acts on the vehicles orientation to drive it to the path. Typically, smoother
convergence to a path is achieved when path following
strategies are used instead of trajectory tracking control laws,
and the control signals are less likely pushed to saturation.
This paper proposes a new methodology for the design
of path following systems for AUVs. The reader is referred
to [8], [9] and the references therein for related ground
breaking work in the field of land robots, where powerful
nonlinear path following control structures were introduced.

0-7803-7924-1103/$17.00 02003 IEEE

It is important to remark that even though the problem of

path following is essentially solved for land vehicles, the
same does not hold true for marine craft. This is due to the
fact that dynamics play a key role in the motion of the latter,
thus requiring the development of methodologies for accurate
path following that take explicitly into account the presence
of possibly complex, nonlinear hydrodynamic terms. This is
in striking contrast with land vehicles, where methodologies
that build on pure vehicle kinematics are often adequate for
control.
The present paper builds on previous results obtained in
[4], [5], where the work of [SI, [9] was extended to deal with
the control of marine vehicles in three dimensional space and
to address explicitly the presence of non-negligible marine
vehicle dynamics. The methodology for path following proposed in [5] can be easily understood by recalling that the
total velocity vector of an AUV is not necessarily aligned
with the vehicles main axis, as in the case of wheeled robots
( A W s sideslip). However, by drawing a simple analogy
between the problems of path following for wheeled robots
and A W s , the latter can be cast as the equivalent problem of
aligning the total AUV velocity vector with the tangent to the
path by manipulating the vehicles yaw rate. It is important to
remark that in spite of its broader scope of applications, the
results in [51 inherit the major shortcoming already present
in the path following control strategy for wheeled robots
described for example in [9] : the initial position of the
vehicle is restricted to lie inside a tube around the path,
the radius of which must be snialler than the sniallest radius
of curvature that is present in that path.
The work reported in this paper lifts this restriction
entirely. This is done by controlling explicitly the rate of
progression of a virtual target to be tracked along the path,
thus bypassing the problems that arise when the position
of the virtual target is simply defined by the projection of
the actual vehicle on that path. See [ l l ] where a similar
technique was first proposed for wheeled robots. This design
procedure effectively creates an extra degree of freedom that
can then be explored to avoid the .singularities that occur
when the distance to path is not well defined (this occurs
for example when the vehicle is located exactly at the center
of curvature of a circular path). Controller design starts at
a kinematic level and evolves to a dynamic setting using
backstepping techniques. The resulting control strategy yields

1256

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.

global convergence of the actual path of the vehicle to the

desired path.
As remarked in 1111, the idea of exploring the extra
degree of freedom that comes from controlling the motion
of a virtual target along a path seems to have appeared
for the first time in [l] for the control of wheeled robots.
This idea was later extended to the control of marine craft
in [ 2 ] . However. none of these references addresses the
issues of vehicle dynamics. Furthermore, the methodologies
adopted in [l], [2] for control system design build on an
entirely different technique that requires the introduction
of a nonsingular transformation in the original error space.
Interestingly enough, a very recent publication explores the
same concept of a virtual target for path following of wheeled
robots 131.
The paper is organized as follows. Section I1 formulates
the problem of path following control for an A W . Section
I11 develops a nonlinear path following control law that
deals explicilty with vehicle dynamics. The performance of
the control system proposed in illustrated in simulation in
Section IV. Finally, Section V contains the conclusions and
describes some problems that warrant research.
11. PROBLEM FORMULATION

This section introduces the simplified dynamic model of

an AUV in the horizontal plane and provides a rigorous

formulation of the problem of steering it along a desired
path.
The type of AUV considered in this paper is equipped
with two identical hack thrusters mounted symetrically with
respect to its longitudinal axis. Thus, the vehicle is underactuated since it lacks a lateral thruster. The common
and differential modes of the thrusters generate a force F
along the vehicle's longitudinal axis and a torque r about its
vertical axis, respectively. In this study, a full dynamic model
of the INFANTE AUV operated by the Institute for Systems
and Robotics (ISR) of Lisbon will be used, see Table I . It is
assumed that only the hack thrusters are used to maneuver
in the horizontal plane.

A. Vehicle Modeling. Kinematics and Dynamics.

where U and U are the longitudinal (surge) and transverse

(sway) velocities, respectively.
With this notation, the kinematic equations of the AUV
can he written as
X = UCOS($B)- vsin($B)
y = us in($.^) + a c o s ( + ~ )
(1)
$JB = r
where T is the vehicle's angular speed (yaw rate). Assuming
is never equal to zero, define the sideslip angle 0 =
arctan(v/a) and consider the reference frame {W)that is
obtained from { B )by rotating it around the Z B axis through
angle p in the positive direction . The above equations can
then be re-written to yield
U

X = tJtCOS($Mr)

Y = ut sin(+w)

(2)

$w=r+p

where $Jw7 = +B p and ut is the x w component of the

tot? vehicle velocity expressed in {W).Clearly, ut = (U'
U ' ) 12. In the aircraft literature { W }is called the wind frame
and will henceforth be called the flow frame. Notice how
the choice of a new frame simplified the first two kinematic
equations and brought out their similarities with those of a
wheeled robot. See 191, [ill.
Neglecting the equations in heave, roll, and pitch the
simplified equations for surge, sway and yaw can be written
as [61, U01
F = m,u +d,
0 = m,it m,,ur
d,
(3)
l- = m,? d,

where
mu

m,
m,

m-XC

= m-Y,
= I, - N+
mu? = m - Y ,

-X =,U2 - X""W2
-Y,uu - Y"l"l"l"/
= -N,,uu - N,~&J
-Nrur

d,

d,
d,

The symbols m and I , denote the mass and moment of

inertia of the AUV respectively, X(.i,Y(.},and N{.) are
classical hydrodynamic derivatives, and [ F I?]= defines the
input vector of force and torque that are applied to the AUV.
The model presented in this paper is based on the model of
the INFANTE A W described in [lo], to which the reader
is referred for complete details.

The following notation will he used in the sequel. The

symbol { A } := { Z A , ya, Z A } denotes a reference frame
with origin OA and unit vectors Z A , ~ A ,and ZA. Given
two reference frames { A } and { B ) , :R is the rotation B. Path Following. Error Coordinates.
The solution to the problem of path following proposed
matrix from { B } to { A } . Following standard practice, the
general kinematic and dynamics equations of a vehicle can here builds on the following intuitive explanation [9], [ 5 ]
be developed using a global coordinate frame { U } and a (see Figure 1): a simple path following controller should
body-fixed coordinate frame { B } ,as depicted in Figure 1. compute i) the distance between the vehicle's center of mass
Q and the closest point P on the path and ii) the angle
Let Q denote the center of mass of the vehicle, which we
assume is coincident with OB, and let q = [z, y, OIT he the between the vehicle's total velocity vector wt and the tangent
position of Q in { U ) . Further let +rg denote the yaw angle to the path at P , and reduce both to zero. This motivates
that parametrizes the rotation matrix from { B } to { U } . Let the development of the "kinematic" model of the vehicle in
W I = [ U , u, 0IT he the velocity of Q in { U } expressed in { B } , terms of a Serret-Frenet frame {F} that moves along the

1257

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.

It is also straightfomard to compute the velocity of Q in

{U} as
(f),+R-'($)
F +R-'(wFx~)

(s),=

where d is the vector from P to Q. Multiplying the above

equation on the left by R gives the velocity of Q in {I}
expressed in {F} as

(g),

= ($)F

+ (g)

F +WF

Using the relations

Fig. I .

path; {F} plays the role of the body axis of a "virtual target
vehicle" that should he tracked by the "real vehicle". Using
this set-up, the abovementioned distance and angle become
the coordinates of the ermr space where the control problem
is formulated and solved. In this paper, however, a Frenet
frame {F} that moves along the path to be followed is used
with a significant difference: the Frener frame is not attached
to the point on the path that is closest to the vehicle.,Instead,
the origin OF = P of {F} along the path is made to evolve
according to a conveniently defined control law, effectively
yielding an extra controller design parameter. As will be seen,
this seemingly simple procedure is instrumental in lifting the
stringent initial condition constraints that are present in the
work of [9] for path following of wheeled robots and in that
of [51 for marine vehicles. The notation that follows is by
now standard.
Consider Figure 1, where P is an arbitrary point on the
path to be followed. Associated with P, consider the corresponding Serret-Frenet frame {F}. The signed curvilinear
abscissa of P along the path is denoted s. Clearly, Q can
either be expressed as q = [z, y, 0IT in {U} or as [SI, y1, 0IT
in {F}. Stated equivalently, Q can be given in ( q y ) or
(SI,y1) coordinates. Let

E=

cos$^

sin$F

cosGF 0
0
1
be the rotation matrix from {U} to {F}, parameterized
-sin$F
0

locally by the angle @ F . Define

W F = +F.

= 4 F = C&)S
CdS) = gc(s)j.

WF

[a]

($)"=

Path Following: Reference Frames

[ ;]

($)F=

(5)

equation (5) can be rewritten as

R
Solving for

[] [
=

i ( 1- C&)Yl)
y 1 +c,(s)Ss~
0

s1

SI and y1 yields

Finally, replacing the top two equations of (2) in (6) and

introducing the variable $ = ?,6w- $JF gives the "kinematic"
model of the A W in (s, y) coordinates as

= - B ( 1 - C&)
w,cos$
yl = -c&l
ut sin@
$ = W W - c,B
s1

Then,

(7)

(4)

where ce(s) and (gc(s) =

denote the path curvature
and its derivative, respectively. The velocity of P in {U} can
he expressed in {F} to yield

where WI.V = ?i),y = r p.

At this point it is important to notice that in [5] and [91
s1 = 0 for all t, since the location of point P is defined
by the projection of Q on the path, assuming the projection
is well defined. One is then forced to solve for S in the
equation above. However, by doing so 1 - c,yl appears in
the denominator, thus creating a singularity at y1 =
As
a result, the control laws derived in [9] and [5] require that
the initial position of Q be restricted to a tube around the

1258

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.

2.

path, the radius of which must he less than -,

where
c,,,,,
denoted the maximum curvature of the path. Clearly,
this constraint is very conservative since the occurrence of a
large c,,,,,
in just a small section of the path will impose a
rather strict constraint on the initial vehicles position, even
if it happens to start in a region that is far away from the
problematic section. By making s1 not necessarily equal to
zero, a virtual target that is not coincident with the projection
of the vehicle on the path is created, thus introducing an
extra degree of freedom for controller design. By specifying
how fast the newly defined target moves, the occurrence of
a singularity at y1 = $ is removed.

limt-, vt(t) is different fnmi zem. S u p p m the path to be

fidlowed is parametrized by its cunilinear abscissa s arid
~ cc are
assume that for each s the variables dJ, SI, y 1 and
well defined. Then the kinematic conrml law

(where kl and k2 are arbitrary positise conctants) drives y1.

$ asymptofically to zero.

SI. and

Indication of Proof Consider the candidate Lyapunov

- 6). The control law
function VI =

h($

C. Problem Formulation

With the above notation, the problem under study can be

formulated as follows:
Consider the AUV model with kinematic and dynamic
equations given b)) ( I ) and (3), respectively Given a path
to be followed and a desired pnifile u d ( t ) > u,in > 0 for
the surge speed U , derive feedback control laws for the force
F , torque I?, and rate of evolution S qfthe curvilinear abcissa
s of the virtual target point P along the path so tlwt y1.
SI, $, and U - u d tend to zem asyntptotically.
111. NONLINEAR PATH FOLLOWING
CONTROLLER DESIGN

~ = 6 - P - k1(11,-6)+cc(s)S
makes VI = -kl(O - 6) 5 0. Since VI is positive and
VI ( t ) is well
monotonically decreasing function, limt,,
defined. Moreover, since VI is uniformly continuous, an,
application of,Barbalats lemma allows for the conclusion
V, = 0. A simple argument shows that with
that limt,,
the control laws proposed the variables y ~ ? s l and
,
d~ are
hounded and approach the set E defined by V, = 0. Examine
now the motion of the feedback control system restricted
to E. To do this, consider the candidate Lyapunov function
candidate V , = $(s:
y:) and compute its derivative
VE = ylut sinJ-K.& 5 0. Since VE is bounded, Barbalats
lemma allows for the conclusion that limt-,
V , = 0. This
in turn implies that all trajectories in E satisfy 1imf-- y1 =
0 and limt4, SI = 0. An argument of continuity can now
be used to conclude the proof of the theorem.

This section introduces a nonlinear closed loop control law

to steer the dynamic model of an AUV described by (1)-(3)
along a desired path. Controller design builds on previous
work by [5] and [9] on path following control and relies
heavily on backstepping techniques. The reader will find in
[7] a lucid exposition of interesting theoretical and practical
issues involved in backstepping. Controller design is done
in two steps. The first step yields a kinematic controller
by adopting the yaw rate T = @B as a virtual control
input and assuming that the actual surge speed equals the
desired speed u d ( t ) . The second step addresses the vehicle
dynamics, builds on the kinematic controller derived, and
uses backstepping techniques to obtain control laws for the
input variables F and I?.

A. Controller Design using the Kinematic Model

This section derives a kinematic controller for the AUV.
As in [9], we let

be a desired approach angle parameterized by k,j > 0 and

0<
< T / Z , satisfying ylut6(y1,ut) I 0 fur all y1 and
ut. The approach angle is instrumental in shaping transient
maneuven during the path approach phase.
Proposition 1: Consider the kinematic model of an AUV
described in ( I ) and the corresponding path following error
model (7). Let the appmach angle 6(yl, u t ) be dejined as
in (8). Assume the total velociry of the vehicle is such that

B. The Dynamic controller

The above feedback control law applies to the kinematic
model of the A W only. However, using backstepping techniques this control law can be extended to deal with the
vehicle dynamics. In the kinematic design the total velocity
ut(t) of the vehicle was left free but implicitly dependent on
a desired profile.ud(t) for surge speed u(t). In the dynamic
design the variable U will be brought explicitly into the
picture and a control law will he derived so that u ( t )- u d ( t )
tends to zero. Notice also that the robots angular speed r
was assumed to be a control input. This assumption is lifted
by taking into account the vehicle dynamics. The following
result holds.
Proposition 2: Consider the kinematic and dynamic
models of an AUV described in ( I ) and (3) respectiwly,
and rhe corresponding path fiillowing ermr model in (7).
Let the appmach angle 6( y1, u t ) be defined as in (8)and let
a desired speed pmfle u d ( t ) for u(t)be given. Assume that
u d ( t ) is a G function and tlwt limt-m u d ( t ) is different
from zero. Suppose the path to be followed is paranietrized
by its curvilinear abscissa s and assume rhut for each s the
are well defined. Then the
variables 8, SI, yl,and c,
dynamic contnil law

1259

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.

where

a, = 8 - j - (kl+ k3)($ - 8)
- ( k 5 + k 1 k 3 ) ( $ - 6 ) + c c S + %i,

k l through k5 are arbitrury positive gains, and d, and d,

are cunis .fhydrodynamic coeficienrs, drives y1, SI, and $
asynptorically to zero.

Indication of Proof Define the virtual control law for

(desired behaviour of r in (9)) as

C=s-p-kl($-s)+c,(s)S

and let e = T - C be the difference between actual and desired

values of T. Set T = z C and consider the total candidate
Lyapunov function

1
v,= kSV, + 21 , + -(u
2
--t

- Ud),

(11)

with k5 positive. Tedious but straightfonvard computations

will show that with the control law proposed
Vz

- k i k s ( $ - 6)'

- k3ez

-kq(U - ud)'

50

-'

The proof that all relevant state variables go to zero follows

from successive applications of Barbalat's lemma.
Note. The dynamic control law in (10) requires the comSince this variable cannot be measured in
putation of
practice, one must resort to the dynamic model of the A W
to compute it. Using simple algebraic manipulations it can
be shown that the problem of computing fi is indeed well
posed because the vehicle is stem dominant.

a.

IV. SIMULATION RESULTS

This section illustrates the performance of the path following control law derived in simulation. Table I summarizes
the key parameters of the AUV model used. This is a simple
modification of the Infante A W model described in [IO] to
account for the fact that only thrusters are used to maneuver
in the horizontal plane.
The reference and actual robot paths are shown in the top
diagram of Figure 2. The desired surge speed U d ( t ) was set
to lms-'. The controller design parameters are displayed in
Table I1
Figure 2 shows the results of the simulation. The top
diagram shows the envelope defined by the initial condition
constraint imposed in [9]. Notice how the coordinate sl and
y1 tend to zero asymptotically. This is equivalent to stating
that: i) the position of the virtual target (origin of frame
{ F } ( s )along the path) approaches the projection of the A W
on that path, and ii) the lateral distance of the AUV to the
path is driven to zero. Notice also that the actual surge speed
u ( t ) converges to u d = 1ms-'.
Fig. 2.

Simulation ~ S U I I S

1260

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.

VI. ACKNOWLEDGEMENTS
This work was supported by the Portuguese FCT POSI
Programme under framework QCA I11 and by projects
MAROV of the FCT and MAYA of the A d . The first author
benefited from a grant of the EC, under project FREESUB.
The second author was partially supported by a post doctoral
grant from the FCT.

P
.

VII. REFERENCES

TABLE I
THE INFANTE AUV. MOOEI. PARAMETERS

V. CONCLUSIONS
A nonlinear control law was developed for accurate path
following of autonomous undenvater vehicles (AWs). The
key idea behind the new control law developed was to control
explicitly the rate of progression of a virtual target to be
tracked along the path, thus overcoming the singularity
problems that arise when the position of the virtual target is
simply defined by the projection of the actual vehicle on that
path. Controller design relied on backstepping techniques.
The paper offered a formal proof of convergence of the
vehicles trajectory to the path. Simulation results illustrated
the performance of the control system proposed.
The controller derived relies heavily on accurate knowledge of the vehicle dynamics. Future work will address the
problems of reducing the complexity of the controller as well
as evaluating its robustness against parameter uncertainty.
The problem of precise path following in the presence of
unknown sea currents warrants also further consideration.

[I] M. Aicardi, G. Casalino, A. Bicchi, and A. Balestlino,

Closed Loop Steering of Unicyle-Like Vehicles via
Lyapunov Techniques, IEEE Robotics and Automation
Magazine, pp. 21-35, March 1995.
[21 M. Aicardi, G. Casalino, G. Indiveri, P. Aguiar,
P. EncarnaGlo, and A. Pascoal, A Planar Path Following Controller for Underactuated Marine Vehicles,
Pmc. MED200l , Dubrovnik, Croatia, June 2001.
[31 del Rio, F. et al., A New Method for Tracking Memorized Paths: Applications to Unicycle Robots, PNX.
MED2002, Lisbon, Portugal, July 2002.
[4] P. EncarnaGlo, A. Pascoal and M. Arcak, Path Following for Marine Vehicles in the Presence of Unknown
Currents, Pmc. SYROCO2000, 6h IFAC Syniposium
on Robot Control, Vienna, Austria, 2000.
[5] P. Encama$lo and A. Pascoal, 3D Path Following for
Autonomous Underwater Vehicles, Pmc. 3gth IEEE
Conference on Decision and Control, Sydney, Australia,
2000.
[61 Fossen, T. (1994). Guidance and Control of Ocean
Vehicles. John Willey & Sons, Inc., New York, 1994.
[71 M. KrstiC, I. Kanellakopoulos, and P. Kokotovic, Nonlinear and Adaptive Contml Design, John Willey &
Sons, Inc., New York, 1995.
[8] A. Micaelli and C. Samson, 3D Path Following and
Time-Varying Feedback Stabilization of A Wheeled
Robot, Pmc. Inremarional Conference ICARCV92,
RO-13.1, Singapore, September 1992.
191 A. Micaelli and C. Samson, Trajectory - Tracking for
Unicycle - Type and Two - Steering - Wheels Mobile
Robots, Technical Repor! No. 2097, INRIA, SophiaAntipolis, Nov. 1993.
[lo] C. Silvestre. Multi-Objective Optimization Theory with
Application tu the Integrated Design of Controllers
/ Plants for Autonomous Vehicles. PhD Thesis (in
English). IST, Lisbon, Portugal, 2000.
[ l l ] D. Soetanto, L. Lapiem, and A. Pascoal, Adpative Nonsingular Path Following Control of Dynamic
Wheeled Robots, Proc. 42nd IEEE Conference on
Decision and Conrml, Hawai, 2003.

1261

Authorized licensed use limited to: UNIVERSIDADE DE BRASILIA. Downloaded on November 26, 2008 at 14:52 from IEEE Xplore. Restrictions apply.