Proceeding of the 2004 American Control Conference
4
Boston. MassachusettsJune 30 ~ J u l 2y , 2 0 0 4
Directional Control of a Streamlined
Underwater Vehicle by Feedback Passivation
Hye-Young Kim and Craig A. Woolsey
Absmocl-This paper describes a model-based technique for
stabilizing the directional motion of a streamlined underwater
vehicle controlled in surge, roll, pitch, and yaw. The six
degree of freedom vehicle model allows for a broad range
of viscous force and moment representations. The control law,
which is derived using feedback passivation, makes steady,
streamlined translation in a desired inertial direction globally
asymptotically stable.
I. INTRODUCTION
Conventional streamlined underwater vehicles are under-
actuated by design. Typically, they include a propulsor and
three or more torque actuators, such as tail fins, which pro-
vide control in four degrees of freedom: surge, roll, pitch,
and yaw. Linear methods can provide good directional con-
trol performance over a restricted range of motion, however
these approaches are not always well justified, particularly
for maneuverable vehicles in dynamic environments. In this
paper, we propose a nonlinear control law which globally
asymptotically stabilizes streamline translation in a desired
direction. The control law is derived using the theory of
passive systems and, in particular, the notion of feedback
passivation described in [9].
Nonlinear stabilization for streamlined, underactuated un-
derwater vehicles has attracted considerable interest. While
results on dynamic stabilization, trajectory tracking, and
path following are abundant for vehicles in planar motion,
results for motion in three dimensional space are more
scarce. In [I], a kinetic energy shaping technique known
as interconnection and damping assignment was applied
to stabilize a streamlined vehicle using only surge, pitch,
and yaw inputs. Internal rotor actuators were proposed
for this purpose in [12]; the stabilization result relied on
a similar energy shaping technique, the method of con-
trolled Lagrangians. These treatments focused solely on
stabilization of the vehicle dynamics, however, ignoring
the rotational kinematics. The problem of stabilizing an
underactuated vehicle to a particularpath in inertial space is
not a trivial extension of the dynamic stabilization problem.
In [3], a nonlinear path following strategy is proposed for
an underwater vehicle with surge, pitch, and yaw control.
The approach begins with Lyapunov-based control design
for the feedback linearized kinematic equations. The full
(force and moment) control law is then derived through
H.-Y. Kim is is a former post-doctoral research assistant with the
Department of Aerospace and Ocean Engineering, Virginia Polytechnic
Institute and State University, Blackburg. VA 24061, USA.
C. A. Woolsey is wilh the Department of Aerospace and Ocean Engi-
neering, Virginia Polytechnic Institute and Slate University, Blackburg,
VA 24061, USA. cwoolaey@vf.edu
0-7803-8335-4/04/$17.0002004 AACC
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.
ThM12.
T''
Fig. 1. Spheroidal underwater vehicle.
backstepping. Simulations suggest that the approach is quite
effective, although the results stop short of a formal proof
of convergence. Another path following strategy is proposed
in [2] for a vehicle with surge, roll, pitch, and yaw control.
This approach makes use of backstepping and includes an
adaptive element to compensate for model uncertainty. The
results of [2] and [3], however, only hold locally.
In this paper, we derive a control law which globally
asymptotically stabilizes steady translation of a streamlined
underwater vehicle in a particular inertial direction. The
technique is based on the method of feedback passivation
described in [9]. In this approach, the system is transformed
into a feedback interconnection of two passive subsystems.
It follows that the interconnection is passive and one may
then use output feedback to asymptotically stabilize the
system. Intrinsic properties, such as passivity, allow one to
derive control algorithms which are globally effective and
which work with the natural dynamics rather than supplant
them.
11. UNDERWATER VEHICLE EQUATIONS
OF MOTION
The underwater vehicle is modeled as a rigid, spheroidal
hull, a shape which adequately represents streamlined un-
derwater vehicles. It is assumed that the vehicle mass m is
equal to the mass of the fluid displaced by the vehicle; that
is, the vehicle is neutrally buoyant. The vehicle is equipped
with a single propulsor, which is aligned with the axis of
symmetry, and with torque actuators (e.g., tail fins) that
provide control moments about three axes. The dynamic
model places few restrictions on the expressions for viscous
force and moment, allowing viscous models that extend over
the full range of vehicle motion.
Vehicle kinematics. The principal axes of the spheroid
are taken as the axes of a body-fixed reference frame;
these are represented by the unit vectors bl, b2. and bS
in Figure 1. Assuming uniform fluid density, the origin of
2998
 the body frame is the vehicle’s center of buoyancy, i.e., frame vector rcmand let M l z = -?Gem. The generalized
the center of mass of the displaced fluid. Another reference momentum is
frame, denoted by the unit vectors il, iz, and is, is fixed in p=Mv
inertial space. The location of the body frame with respect
to the inertial frame is given by the inertial vector I .
Let R denote the proper rotation matrix which transforms
free vectors from the body frame to the inertial frame.
From Euler’s theorem on rotations, R may be represented In expression (2), P is the translational momentum of the
as a rotation through an angle B E [O,Z?r) about a fixed hody/fluid system and II is the angular momentum about
axis defined by a unit vector x [8]. The matrix R may be the center of buoyancy.
parameterized using unit quaternions as follows. Define The dynamic equations are
Mi. + C(V)V+g(v) = Text (3)
where
Then
R(q)= n - 2 (qon - 4,) B,, C(V)= -C(v)T =
( 03?
-P -p
-fl j
where 1 is the identity matrix and where the operator is the “Coriolis and centripetal matrix’’ in [4] and where
denotes the 3 x 3 skew-symmetric matrix satisfying bb =
0
a x b for vectors a and b. Note that the identity rotation
corresponds to the unit quaternion q = [l,0, 0, 01T . (’) = ( -rem x mg (R(q)Ti3)
The unit quaternions are a redundant parameterization. is the generalized gravitational force. While there is no
Generically, any two quaternions q and - q describe the true gravity force, because the vehicle is neutrally buoyant,
same rotation. To avoid this difficulty, one might choose there is a gravity torque about the center of buoyancy.
a three-parameter representation, such as the modified Ro- Equation (3) may be rewritten in terms of the generalized
drigues parameters described in [IO]. Any three-parameter momentum:
representation, however, is necessarily singular at some
point. While this may or may not be a meaningful obstruc- /j= -C (M-~P )
M-’P - g(q)+Text (4)
tion, in practice, we proceed using unit quatemions. The generalized force rex$ in (3) and (4) represents ex-
T
Let the body l?ame vector v = [U],ziZ, us] represent temal forces and moments which do not derive from scalar
the translational velocity of the body frame origin with potential functions. These forces and moments include, for
T
respect to inertial space. Let w = [UI, W Z , w3j represent
example, control torques, forces due to lift and drag, and
the body angular velocity. Following [4], we define propulsive forces. We decompose rextinto control inputs
and viscous terms:

The kinematic equations are

Let pi represent the ith basis vector
for W3;for example,
=[l,0, 0IT. Because the vehicle’s single thruster is
aligned with the axis of symmetry, F , = Fc PI, where F,
and where
represents (scalar) thrust.
The terms F , and T , are generally quite complicated.
We make some simple and relatively general modeling as-
In this paper, we consider only directional control. We sumptions. For example, we make the following assumption
therefore consider only the attitude kinematics about the viscous moment.
Assumption 2.1:
T , . w < O when w#O
Once the global directional control problem is addressed,
one may consider path following as a natural extension.
T , = 0 when ( w , v ) = ( 0 ,U&) V ud E W
Vehicle dynamics. Let the diagonal matrix MI1 =
diag(ml,mz,m3) represent the sum of mn and the added In words, we assume that the viscous moment opposes
mass matrix, which is diagonal for a spheroid. For aprohte angular rate, in general, and that it vanishes for pure
spheroid, whose 1-axis is the axis of revolution, ml < translation along the symmetry axis.
mz = m3. Let the matrix Mzz represent the sum of the We impose a bit more structure on the model of the
rigid body inertia and the added inertia matrix. Finally, viscous force. By definition, drag opposes vehicle velocity
suppose that the center of mass is located by the body and lift acts orthogonally to the velocity vector. Considering
2999

Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.
 To transform the viscous force from the current frame to
the body frame, we define the proper rotation matrix
- ^

RBC(~,C
= Ye-cP1,-mPz
)

1
cos CY 0 - sin CY

V
Then
=( sinpsina
cos p sin oi
cosp
- sin p
sinwcosoi
cos p cos CY
.

(5)
Fig. 2. Hydrodynamic angles for an axisymmetnc body.

axial symmetry, we assume that the lift force acts in the
plane determined by the velocity vector and the vehicle’s
longitudinal axis. This assumption fails to capture out-of-
plane forces due to asymmetric fluid flow, however it is
consistent with standard modeling assumptions.
To express F,, we first introduce some hydrodynamic
angles. Let
arctan (2) zr2 #o and/or 213
212 = 2)s = 0
The “4-quadrant’’ arctangent is used in the definition of p;
identifying T with -n, we have p E (-.,I Rotating
through the I.1 about the bl axis defines an inter-
mediate reference frame in which the velocity vector has
components only in the “1-3” plane. Moreover, the 3-axis
component of velocity, in this intermediate frame, is non-
.~
negative. Let
where, once again, the 4-quadrant arctangent is used. Note
that CY E [O,.]. Rotating through the angle a about the
intermediate Z-axis defines a new reference frame in which
the 1-axis is aligned with the velocity vector, as shown in
Figure 2. Following the terminology of [4], we refer to this
reference frame as the “current” frame.
It is standard practice to express lift and drag in terms
of non-dimensional coefficients. To non-dimensionalize the
forces, we define Fo as the product of dynamic pressure
and’a reference area S:
1
& = pIIvlI2S.
Fv = -RBc(P,~)
= -FoRBc(P,a ) (2:;)’
We make the following assumptions about the form of
the drag and lift
Assumption 2.2:
+o CD(CY)
is a continuous, even function which is positive
(5)
- for all a.
C,(a) is a continuous, odd function which is positive
(negative) when eim lies in the first or third (second or
fourth) quadrant of the complex plane.
For a simple, prolate spheroid, the assumption that C,
is even and positive is an empirical fact, The assumption
regarding the form of CL is consistent with intuition for
a prolate spheroid in a
flow. These assumotions are
satisfied, for example, by the drag and lift coefficients
CD(Q) = CD, + C D , (1- cos(zCY))
1
CL(CY)= ZCL,,sin(2~)
Constant parameters CL-, CD,, and CD, can be approx-
imated from steady aerodynamic data as presented, for
example, in [ 5 ] and [6].
In reality, a spheroid moving at a large angle of attack is
subject to complicated, unsteady forces [ 111. While these
effects certainly impact the dynamics, they are difficult
to model accurately. Such effects are typically ignored in
control design with the understanding that well-designed
model-based feedback can provide suitable system perfor-
mance even when the model is imperfect.
Error dynamics. Our goal is to stabilize the motion

Here, p is the fluid density. In the current frame, the viscous qe qd, we = 0, v e = ‘ud PI
force takes the form where qd is the unit quatemion representing some constant
CD(4 desired attitude and vd > 0 is a constant desired speed. Let

(<)=-.( c z ) ) l
tid represent the quaternion conjugate of qd and define the
attitude error quaternion
In general, the dimensionless coefficients C , and CL e = Gd * q?
depend on Reynolds number as well as angle of attack.
where * denotes quatemion multiplication; see 8 Thus,
For convenience, we ignore Reynolds number effects in 41. .
when q = qd, we have e = e d = [l, 0, 0, 01 , To shift
the remaining discussion; these effects can be incorporated,
the equilibrium to the origin, define the attitude error vector
if necessary, with little impact on the assumptions and
conclusions. -
e = e - ed.

3000

Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.
 Next, define the generalized velocity error
~ = ( ~ ) = v - Y , where ue=( 2)
Correspondingly, the generalized momentum error is
The error dynamics are
1
2: = - Q ( E + e d ) O
2
where it is understood that the appropriate shift to error
coordinates is to be made in the rotation matrix R and in
the generalized viscous force 7”.Our goal is to stabilize
the equilibrium at the origin for equations (7) and (8).
111. FEEDBACK
PASSIVATION OF NONLINEAR
CASCADE
SYSTEMS
With an appropriate choice of T , and F,, the system
described in Section II can be transformed into a nonlinear
cascade which can be treated using the method of “feedback
passivation” in [9].In this section, we review some basic
concepts from the theory of passive systems.
Consider a system H described by the state and output
equations
k = f(z,u)
Y = h(l,u),
where U is a vector of inputs and y is a vector of outputs
of the same dimension. We assume that f ( 0 , O ) = 0,so
that the system has an equilibrium at the origin. We also
assume that h(0,O)= 0.
Definition 3.1: The system (9) is passive if there exists a
positive semi-definite function S(z), with S(0) = 0, such
that for every input u ( t )
S(z(7)) - S(z(0)) 5 /‘y(t) , u(t) dt
0
for all 7 2 0. The function S(z) is called a storugefirnction.
Note that if S is C’, then condition (IO) may be written
S 5 y . U.Making an analogy between the storage function
S and a physical system’s total energy, the condition for
passivity requires that the rate of increase of stored energy
does not exceed the input power. A simple but important
observation is that the negative feedback interconnection
o f two passive systems is also passive. The sum of the
two systems’ storage functions is a storage function for the
feedback interconnected system.
If a storage function S(z) for the system (9) is positive
definite, then it is a Lyapunov function when U = 0.Thus,
the equilibrium at the origin is stable in the absence of any
Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.
input. Choosing U = - K y with K > 0 makes 50
and one may examine asymptotic stability using Lasalle’s
invariance principle.
Now, consider a nonlinear cascade system of the form
2 = f,(z)+i(z,Q (11)
i = f<(O+dE)U (12)
and suppose that this system has an equilibrium at the
origin. Following [9],we make two assumptions.
Assumpfion 3.2: The equilibrium z = 0 of z = f,(z)
(7) is globally stable and a Cz, positive definite function S,(z)
is known which satisfies
Assumption 3.3: There exists an output y = h(E)such
that
.the system (12) is passive with respect to input U
and output y. with a C’, positive definite, radially
.unbounded storage function St(.$), and
one may write
I I
(9)
Fig. 3. Passive feedback interconnection
The following theorem, given in [9], says that the com-
plete system (11) and (12) may be transformed into a
feedback interconnection of two passive subsystems.
Theorem 3.4: The feedback transformation
renders the system (11) and (12) passive from the input w
(10) to the output y = h(E).A storage function is
S(€,.) = SdE) + S&). (14)
Figure 3 illustrates Theorem 3.4. Because the feedback
interconnected system is passive (with input tu, output
y, and storage function S ( [ , z ) ) , the equilibrium at the
origin is stable when w = 0. Moreover, under a zero-state
detectability assumption described in [9], choosing
w = -Ky
with K > 0 makes the origin globally asymptotically sta-
ble. Altematively, rather than require zero-state detectabil-
ity, one may use Lasalle’s invariance principle to investigate
asymptotic stability.
3001
 IV. CONTROL DESIGN The modified equations of motion, under the feedback
control laws (15) and (16), are
By defining the thrust and control moments appropriately,
equations (7) and (8) may he written in the nonlinear . 1
7: = 5Q(6 + e & (18)
cascade form (11) and (12).
Pmposition 4. I : ' The feedback control law
F,.=
and
E := Fo (COS~CD(CY)
- sinaCL(a)) - kuL% (15) (g) = ((.+.dl
(&Pd) xW+Fv+FP,
X r j + ( P + P d ) X6+Tv+U
(19)

Tc = - (rem X mg (@is) + P X (VdP1)) +U, (16)

Define
z=E and E=ji
with k,, > 0, renders the subsystem (8) passive with input
U,output y = w , and storage function Then equations (18) and (19) are in the nonlinear cascade
form (11) and (12) with
SE= -pTM-'ji
1-
(17) 1
2 .
Pmof: Under the given choice of feedback, f,(z)= 0 and $(z,E) = ,&(E + ed)rj.
s, = l7.F
= 6 ' (Fc +Fv) (VdP1) : ( P x w )
+U ( - P X (Vdp1) + U )+ W ' T v
'

= + +
6 . ( F , F , ) + U .U w . T , .
The last term is non-positive by Assumption 2.1. The first s(E) = ( )
term satisfies
Equation (18) satisfies Assumption 3.2 with

S,(z) = - e . 7:
ke - =
1
-k, [(l-eo)' + e , . e,]
2 2
= k,(l - e o )
and k , > 0. The choice of S, is inspired by [7], where
the problem of rigid body attitude control is considered.
(sinaCo(a)+cosaC~(a))Fo
= -(~~sinp++Zlgcosp) Equations (18) and (19) also satisfy Assumption 3.3 with
+cl ( F -~ Fo ( C O S C Y C D-( ~sinaCL(a))).
) y = 5 = w , +(r,E) 7 $Q(6 ed), and with S, given+
in (17). By Theorem 3.4, the feedback control law
Consider the first term. By definition of LL,
U = w - (L+S,)T
irzsin~+ir3cos~=\/ir22+i)32LO
Also, given the assumptions on the form of CD and CL
and the fact that CY E [O,T], we have I
= w--k,e,
2
sinaCD(a) + c o s a C ~ ( a2) 0 renders the system passive from the new input w to the
(and strictly positive for a E ( 0 , ~ ) ) .Therefore, the first output y = w with respect to the storage function
term is non-positive. Now consider the second term and let
F, = F, as stated in the proposition. The rate of change of S(E>2) = Se(€)+ S&).
S, becomes The rate of change of the storage function is

s = - m ( s i n a C D ( a ) +cosa~r.(a))~o
-k,,$ +W .U + U .T, -kU,$ +T , .w +y . w
5 W.U.
5 Y'W.
This inequality proves the proposition. 0
It follows by Lyapunov's direct method that the origin
Suppose we choose w = - K y with K > 0. Then
of (8) is stable when U = 0. We have completed the
first step of a two-step design procedure in which we first
S ,'-J-2-2 u2 + U , (sinaCo(a) + c o s a C ~ ( a ) Fo
)

stabilize the vehicle dynamics and then the attitude. -k,,8: + T , .w - w T K w 5 0.

3002

Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.
 Because S is radially unbounded, this proves that the origin important to develop control schemes which perform well
is globally stable. To conclude global asymptotic stability, over a broad operating envelope.
we apply Lasalle’s invariance principle. Define the set Acknowledgements. The authors thank the reviewers for
their careful consideration and helpful suggestions. This
work was supported in part by NSF grant CMS-0133210
and ONR grant N00014-01-1-0588.
and let M be the largest invariant set contained in E. By
Lasalle’s principle, trajectories converge to the set M as REFERENCES
t + CO. If M = IO}, then the equilibrium is globally [ I ] A. Astolfi, D. Chhabra, and R. Onega, ‘Xsymptotic stabilization o f
asymptotically stable. Now, S 0 implies that ij 5 0. selected equilibria of the underachmted Klrehhoffs equations,” Pmc.
Americon Conlml Conz, pp. 48624867,Arlington, VA, 2001.
Thus, [2] K. D. Do and 1. Pan, “Robust and adaptive path following for
ij=o =+ p=o. underactuated autonomous underwater vehicles:’ Pmc. Americon
Conlml Conr, pp. 199k1999, Denver, CO, 2003.
Within the set M , [3] P. E n c a m a ~ band A. Pascoal, “3D path following for autonomous

’= (
underwater vehicle:’ Pmc. Con$ Decision and Conlml, pp. 2971-
0 2982, Sydney, Australia, 2000.
-ik.e, ) 141 T. I. Fosse”, Marine Conrml Sysrems, Marine Cybernetics, Trond-
heim, Norway, 2002.
[5] S. F. Hoemer, FluidDynomic D r q . Published by the author, Midland
Thus, e , = 0 so eo = 1 and Z = 0. Global asymptotic Park, NI, 1965.
stability follows from Lasalle’s invariance principle. [6] S . F. Hoemer and H. V. Borst, Fluid Dynamic Lifr, Published by L.
A. Hoemer, Brick T o w , NI, 1975.
[7] E Lizanalde and 1. T. Wen, “Attilude control without angular velocity
V. CONCLUSIONS AND FUTURE WORK measurement: a passivity approach:’ lEEE R ~ L X
Auromoric Conlml,
Vo. 41, No. 3, pp. 46&472, 1996.
This paper has presented a control strategy which glob- [8] R. M. Munay. 2. Li, and S. S. Sastry, A Malhemvlieol Intmducrion
ally asymptotically stabilizes translation of a streamlined lo Robotic Manipulotian, CRC Press, Ann Arbor, MI, 1994.
underwater vehicle in a desired inertial direction. The [9] R. Sepulchre, M.Jankovic and P. Kokotovic, ComrrucriveNonlineor
Contml, Springer-Verlag. Landon, 1997.
approach relies on the technique of feedback passivation, [IO] P. Tsiotras, “Stsbilization and optimality results for the attitude
in which the system is transformed through feedback into control problem:’ J Guidance. Contml. and Dynamics, Vo. 19, No.
a passive system. Using known results from the theory 4,pp. 712-119, 1996.
[ I l l T. G. Wetzel and R. L. Simpron, “Unsteady crossflow separation
of passive systems, one may then define output feedback location measurements On a maneuvering 6 1 prolate spheroid:’ A I M
to provide asymptotic stability. The main result uses very Joumol, Vo. 38,No. 11, pp. 2063-2071, 1998.
general modeling assumptions for the viscous force. [I21 C. A. Woolsey and N. E. Leonard. “Stabilizing underwater vehicle
motion using intemal rotors:’ Aulomorico, Vo. 38, pp. 2053-2062,
The problem of directional control is a simplification 2002.
of a more practical problem: nonlinear path following.
Intuitively, having globally asymptotically stabilized the di-
rection of travel, one might close an “outer loop” to stabilize
motion along a straight path. The more challenging problem
of following a parameterized curve in space remains as
future work. This problem was partially addressed in [3],
where the error kinematics were expressed in terms of
the Serret-Frenet frame associated with the desired path.
The hydrodynamic model was limited, however, because
it neglected added mass and assumed small angles of
attack and sideslip. Our hope is that the present results on
feedback passivation will serve as a preliminary step toward
global nonlinear path following for streamlined underwater
vehicles.
Several other issues remain to be addressed, including
the effect of non-zero mean disturbances, such as a net
buoyant force or cross currents. There are also the issues
of actuator dynamics and actuator limits. Regarding the
actuators, it may be feasible to eliminate roll control for
a vehicle with a low center of gravity. Some streamlined
underwater vehicles use control planes that are slaved to
provide only pitch and yaw actuation. Thus, the problem
has practical relevance. An increasing number of vehicles
use vectored thrusters, which offer improved agility but do
not provide roll control. For agile vehicles, it is especially
3003

Authorized licensed use limited to: Central Mechanical Engineering Research Institute. Downloaded on October 21,2022 at 07:48:25 UTC from IEEE Xplore. Restrictions apply.