IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO.
2, MARCH 2017
Trajectory Tracking With Prescribed Performance
for Underactuated Underwater Vehicles Under
Model Uncertainties and External Disturbances
Charalampos P. Bechlioulis, George C. Karras, Shahab Heshmati-Alamdari, and Kostas J. Kyriakopoulos
Abstract— This paper addresses the tracking control problem
of 3-D trajectories for underactuated underwater robotic vehicles.
Our recent theoretical results on the prescribed performance
control of fully actuated nonlinear systems are innovatively
extended on the control of the most common types of
underactuated underwater vehicles, namely, the torpedo-like
(i.e., vehicles actuated only in surge, pitch, and yaw) and the
unicycle-like (i.e., vehicles actuated only in surge, heave, and
yaw). The main contributions of this paper concentrate on: 1) the
reduced design complexity; 2) the increased robustness against
system uncertainties; 3) the prescribed transient and steady-
state performance; and 4) the minimal tracking information
requirements. A comparative simulation study points out the
intriguing performance properties of the proposed method, while
its applicability is experimentally verified using a small unicycle-
like underactuated underwater vehicle in a test tank.
Index Terms— Prescribed performance control, robust
tracking control, underactuated underwater vehicles.
I. I NTRODUCTION
U NDERWATER activities have steadily grown in the
last 50 years, imposing new challenges on technical
and engineering researchers supporting the offshore indus-
try growth. Complex inspections of oil/gas subsea pipelines,
risers, exploitation rings, and novel operations in geological
exploration are now required, with the ultimate objective of
educing efficiently the plentiful underwater resources.
In addition, a multitude of other nonprofitable underwater
applications in the fields of oceanography, environmental
monitoring, and marine archeology have recently emerged.
In the recent wake of rapid progress in marine robotics
that affords numerous advanced tools for offshore activities,
underwater vehicles have particularly received considerable
attention. However, further work remains to be done
before underwater robots roam the ocean freely. In that
sense, the motion control problem of underwater vehicles
Manuscript received February 9, 2015; revised December 23, 2015; accepted
March 27, 2016. Date of publication May 2, 2016; date of current version
February 8, 2017. Manuscript received in final form April 14, 2016. This
work was supported by the EU through the Project entitled PANDORA: Per-
sistent Autonomy through learNing, aDaptation, Observation and ReplAnning
under Grant FP7-288273 and Grant 2012-2014. Recommended by Associate
Editor J. Sun.
The authors are with the Control Systems Laboratory, School of Mechanical
Engineering, National Technical University of Athens, Athens 15780, Greece
(e-mail: chmpechl@mail.ntua.gr; karrasg@mail.ntua.gr; shahab@mail.
ntua.gr; kkyria@mail.ntua.gr).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TCST.2016.2555247
1063-6536 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
429
continues to pose considerable challenges to system designers
in view of the increasingly demanding missions envisioned
for underwater robots, especially in the presence of underac-
tuated dynamics, stringent environmental constraints, and large
model uncertainties.
A typical motion control problem is trajectory tracking that
is concerned with the design of control laws that force a
vehicle to reach and follow a reference trajectory. Classi-
cal approaches, such as local linearization and input–output
decoupling, have been used in the past to design the track-
ing controllers for underactuated vehicles [3]. Nevertheless,
the aforementioned methods yielded poor closed-loop perfor-
mance, and the results were local, around only certain selected
operating points. An alternative approach involves output
feedback linearization [4], [5], which, however, is not always
possible. Moreover, based on a combined approach involving
Lyapunov theory and backstepping, various nonlinear model-
based trajectory tracking controllers have been reported during
the last two decades [6]–[12]. However, these schemes demand
a very accurate knowledge of the vehicle dynamic parameters,
which in most cases is quite difficult to obtain.
Uncertainties in the dynamic model of underactuated
underwater vehicles have been mainly compensated by the
adaptive control techniques. In [13], a switching adaptive law
is combined with a nonlinear control scheme. Do et al. [14]
presented a robust adaptive control strategy for path following
for an underactuated vehicle in the presence of external distur-
bances and model uncertainties. However, the application of
the aforementioned control schemes in a real-time experiment
is questionable, owing to their sensitivity to unknown parame-
ters. On the other hand, the experimental results for an adaptive
control scheme were presented in [15]. Nevertheless, partial
a priori knowledge of the dynamic parameters was requested.
In addition, a hybrid parameter adaptation law, based on
switching control theory, was adopted in [16]. However,
environmental disturbances and unmodeled dynamics were
not considered. Alternatively, the sliding mode control theory
was also employed in [17]–[21] to deal with model uncer-
tainty in the vehicle’s dynamic model. Nonetheless, the main
disadvantage of the aforementioned control schemes is the
inherent control input chattering that is energy intensive and
may result in high-frequency dynamics, which is undesirable
for underwater applications. Finally, adaptive neural network
and fuzzy control schemes that deal with model uncertainties
have also been proposed in [22]–[24], exploiting the universal
430
approximation capabilities of neural network and fuzzy system
structures. Unfortunately, the aforementioned schemes inher-
ently introduce certain issues affecting closed-loop stability
and robustness. Specifically, even though the existence of a
closed-loop initialization set as well as of control gain values
that guarantee closed-loop stability can be proved, the problem
of proposing an explicit constructive methodology capable
of a priori imposing the required stability properties is not
addressed. As a consequence, the produced control schemes
yield inevitably reduced levels of robustness against modeling
imperfections. Moreover, the results are restricted to be local,
as they are valid only within the compact set, where the
capabilities of the universal approximators hold. Finally, the
introduction of approximating/estimating structures increases
the complexity of the proposed control schemes in the sense
that extra adaptive parameters have to be updated (i.e., extra
nonlinear differential equations have to be solved numerically)
and extra calculations have to be conducted to output the
control signal, thus making their implementation on embedded
control systems difficult.
Despite the recent progress in the tracking control for under-
actuated underwater vehicles, certain issues still remain open.
First, even in the case of accurately known vehicle model,
external disturbances affect the tracking performance severely,
thus making the problem of guaranteeing prescribed transient
and steady-state performance difficult or impossible in certain
situations. Furthermore, the tracking performance deteriora-
tion becomes more intense when uncertainty in the vehicle
model (which is inevitable) is also present. Moreover, all the
aforementioned developments require accurate measurements
of the vehicle velocities in all degrees of freedom (DoFs) as
well as of the velocity (and in some cases, the acceleration)
of the desired trajectory, which significantly increases the cost
of the vehicle’s sensor suite and the complexity of the sensor
fusion algorithms, thus reducing their applicability.
Contrary to the current state of the art in the underwater
vehicle control literature, the proposed control schemes that
are designed for the most common types of underactuated
underwater vehicles, namely, torpedo-like and unicycle-like
vehicles, achieve tracking with prescribed performance despite
the presence of external disturbances representing ocean cur-
rents and waves and without requiring prior information of the
vehicle’s dynamic model parameters. The main contributions
of this paper concentrate on: 1) the reduced design complexity;
2) the increased robustness against system uncertainties; 3) the
prescribed transient and steady-state performance; and 4) the
minimal tracking information requirements. More specifically,
only the desired trajectory, and none of its higher order
derivatives, is employed in the control schemes, involving
thus applications where the desired trajectory is not a priori
known for all time, but it is evaluated online, and hence, its
time derivatives are not available. Moreover, the proposed
control schemes are of low complexity and can be easily
integrated on embedded control platforms with limited power
and computational resources [e.g., autonomous underwater
vehicles (AUVs)]. In addition, the stability of the unactu-
ated DoFs is assured without incorporating the corresponding
velocity measurements in the control schemes, thus increasing
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017
greatly the robustness against noises that corrupt the specific
measurements and simplifying further the implementation.
Finally, through the appropriate selection of certain designer-
specified performance functions, the proposed schemes
efficiently avoid both controllability and representation
singularities that inherently arise during the kinematic control
design of underactuated vehicles.
II. P ROBLEM S TATEMENT
In this section, the kinematics and the dynamics of a
six DoF underwater vehicle model are initially presented.
Subsequently, we introduce two of the most common classes
of underactuated vehicles that are considered in this paper
(i.e., torpedo-like and unicycle-like underactuated vehicles),
and finally, the trajectory tracking problem is rigorously
formulated.
A. Vehicle Kinematics and Dynamics
Consider a neutrally buoyant underwater vehicle modeled as
a rigid body subject to external forces and torques. Let {I } be
an inertial coordinate frame and {B}, a body-fixed coordi-
nate frame with orthonormal axes n = [n x , n y , n z ]T , o =
[ox , o y , oz ]T , and t = [tx , t y , tz ]T relative to {I }, whose origin
O B is located at the center of mass of the vehicle. Furthermore,
let p = [x, y, z]T ∈ 3 be the position of O B in {I } and
R = [n, o, t] ∈ S O(3), the rotation matrix that describes the
orientation of the vehicle. Let v = [u, v, w]T be the linear
velocity (u, v, and w are the longitudinal-surge, transverse-
sway, and vertical-heave velocities along the body frame axis
n, o, and t, respectively) and w = [ p, q, r ]T be the angular
velocity ( p, q, and r are the angular velocities around the
longitudinal-roll, the transverse-pitch axis, and vertical-yaw
axis, respectively) of O B with respect to {I } expressed in {B}.
Hence, the kinematic equations of motion for the considered
vehicle may be written as
ṗ = Rv + δc (t) (1)
Ṙ = RS (w) (2)
where δc (t) = [δx (t) , δ y (t), δz (t)]T denotes bounded and
slowly varying ocean currents [3] and
⎡ ⎤
0 −r q
S (w) = ⎣ r 0 − p ⎦.
−q p 0
Alternatively, the orientation of the vehicle may be described
by the Euler angles’ parameterization η = [φ, θ, ψ] T , where
φ, θ , and ψ are the roll, pitch, and yaw angles, respectively,
that is mainly employed owing to its physical meaning (i.e., the
inertial coordinate frame {I } after three successive rotations of
ψ, θ , and φ angles about its z, y, and x axes, respectively, ends
up parallel to the body-fixed coordinate frame {B}). Thus, the
rotation matrix R may be expressed via the roll, pitch, and
yaw angles as follows:
⎡ ⎤
cψ cθ cψ sθ sφ − sψ cφ cψ sθ cφ + sψ sφ
R (η) = ⎣ sψ cθ sψ sθ sφ + cψ cφ sψ sθ cφ − cψ sφ ⎦ (3)
−sθ cθ sφ cθ cφ
BECHLIOULIS et al.: TRAJECTORY TRACKING WITH PRESCRIBED PERFORMANCE
where s = sin () and c = cos (). In addition, the body-
fixed angular velocity w and the Euler angles’ rate η̇ are related
through η̇ = J (η) w where
⎡ ⎤
1 tθ sφ tθ cφ
J (η) = ⎣ 0 cφ − sφ ⎦ (4)
0 sφ /cθ cφ /cθ
is a transformation matrix that does not belong to S O (3)
[i.e., J T (η) = J −1 (η)] with t = tan (). In this way, the
kinematic equations of the vehicle may be written in the Euler
angles’ representation as follows:
ṗ = R (η) v + δc (t) (5)
η̇ = J (η) w (6)
where R (η) and J (η) are defined in (3) and (4), respectively.
Under standard simplifications owing to symmetries in the
mass configuration [3], the dynamic equations of motion of
underwater vehicles may be written as
m u u̇ = m v vr − m w wq + X u u + X |u|u |u| u + X + δu (t)
(7)
m v v̇ = m w wp − m u ur + Yv v + Y|v|v |v| v + Y + δv (t) (8)
m w ẇ = m u uq − m v v p + Z w w + Z |w|w |w| w + Z + δw (t)
(9)
m p ṗ = m vw vw + m qr qr + K p p + K | p| p | p| p
+ z B W cθ sφ + K + δ p (t) (10)
m q q̇ = m wu wu + m r p r p + Mq q + M|q|q |q| q + z B W sθ
+ M + δq (t) (11)
m r ṙ = m uv uv + m pq pq + Nr r + N|r|r |r | r + N + δr (t)
(12)
where m u , m v , m w , m p , m q , and m r denote the vehicle’s
mass/moment of inertia and added mass/moment of inertia
with m vw = m v − m w , m wu = m w − m u , m uv = m u − m v ,
m qr = m q − m r , m r p = m r − m p , and m pq = m p − m q , and
X u , X |u|u , Yv , Y|v|v , Z w , Z |w|w , K p , K | p| p , Mq , M|q|q , Nr ,
and N|r|r are the negative hydrodynamic damping coefficients.
It is also assumed that the vehicle’s center of buoyancy lies on
the body-fixed z-axis (i.e., heave) above its center of gravity
(i.e., z B < 0, where z B defines the position of the center of
buoyancy with respect to O B ). Moreover, δu (t), δv (t), δw (t),
δ p (t), δq (t), and δr (t) denote bounded exogenous forces and
torques acting on surge, sway, heave, roll, pitch, and yaw
owing to ocean waves [3], and X, Y , Z , K , M, and N denote
(depending on the vehicle’s degrees of actuation as discussed
in the sequel) control input forces and torques that are applied
in order to produce the desired motion of the body-fixed frame.
B. Classes of Underactuated Vehicles
In this paper, we consider the most common types of
underactuated underwater vehicles, namely, the torpedo-like
and unicycle-like vehicles.
1) Torpedo-Like Underactuated Vehicles: The vehicles con-
sidered in this class are actuated by a force X along their
longitudinal (surge) axis as well as by torques M and N about
their transverse (pitch) and vertical (yaw) axes, respectively.
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
431
The aforementioned force X and torques M and N define
the input control variables of the corresponding dynamic
system (7)–(12), which in this case is unactuated in sway,
heave, and roll DoFs (i.e., Y = 0, Z = 0, and K = 0).
2) Unicycle-Like Underactuated Vehicles: The vehicles
considered in this class are actuated by forces X and Z along
the longitudinal (surge) and vertical (heave) axes, respectively,
and a torque N about the vertical (yaw) axis. The aforemen-
tioned forces X and Z , and torque N define the input con-
trol variables of the corresponding dynamic system (7)–(12),
which in this case is unactuated in sway, roll, and pitch DoFs
(i.e., Y = 0, K = 0, and M = 0).
Remark 1: Despite its physical meaning, Euler angles’
parameterization introduces a representation singularity in (6)
when the pitch angle θ approaches ±90°. In this respect, since
torpedo-like vehicles may operate close to this singularity,
our approach for this class will be based on the rotation
matrix parameterization that does not suffer from geometric
singularities. On the other hand, for the class of unicycle-
like underactuated vehicles, since the unactuated pitch DoF is
passive (i.e., the pitch angle will be proved to remain bounded
close to zero), the Euler angles’ parameterization will be
adopted for the clarity of presentation.
C. Control Objective
Let pd (t) = [x d (t), yd (t), z d (t)]T denote a smooth desired
trajectory with bounded time derivatives (i.e., a trajectory
with bounded velocity, acceleration, and so on). The objec-
tive of this paper is to propose robust control laws for the
aforementioned classes of underactuated vehicles that track
the desired trajectory pd (t) with prescribed performance,
regarding the steady-state error and the speed of convergence,
despite the presence of exogenous disturbances representing
ocean currents and waves. Hence, the problem treated in this
paper reads as follows.
Problem: Design approximation-free control schemes for
torpedo-like and unicycle-like underactuated underwater vehi-
cles with dynamic model uncertainty, such that all signals in
the closed-loop system remain bounded, and moreover, the
desired trajectory pd (t) is tracked with prescribed transient
and steady-state performance.
Remark 2: In this paper and contrary to what is common
practice in the related literature, no prior knowledge of the
vehicle dynamic parameters and external disturbances or even
of some corresponding upper bounding constants will be
incorporated in the control design. Furthermore, no estimation
(i.e., adaptive control techniques) will be employed to acquire
such knowledge, thus reducing the computational complexity
significantly, and hence making implementation straightfor-
ward and efficient on embedded control platforms of AUVs,
endowed with limited power and computational resources.
III. C ONTROL D ESIGN
In this section, we shall present two model-free control
schemes of low complexity, for torpedo-like and unicycle-
like underactuated underwater vehicles, respectively, that lead
to the solution of the robust prescribed performance tracking
control problem stated in Section II.
432 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

A. Torpedo-Like Underactuated Vehicles where the following equality:

Given the desired trajectory pd (t) = [x d (t), yd (t), z d (t)] , T ex ey ez
n x + n y + n z  cθ n
let us define the distance error ed ed ed

ed = e2x + e2y + ez2 (13) has been utilized, with θn denoting the angle measured from
the vector ed to the longitudinal n = [n x , n y , n z ]T axis of the
where
vehicle. Finally, to solve the robust prescribed performance
ex = x d (t) − x, e y = yd (t) − y, ez = z d (t) − z (14) tracking control problem, we pose Assumption 1.
as well as the orientation errors Assumption 1: The initial position error ed (0) and heading
ex ey ez angle θn (0) satisfy: 1) ed (0) > ρ d and 2) |θn (0)| < (π/2) for
et = t x + t y + t z  cθ t (15) an arbitrarily small positive design constant ρ d .
ed ed ed
ex ey ez Control Scheme: The proposed control scheme is first
eo = o x + o y + o z  cθ o (16) derived at the kinematic level, assuming that the control signals
ed ed ed
are the surge velocity u as well as the pitch and yaw angular
where θt and θo are the angles measured from the unit error
velocities q and r . Subsequently, the kinematic controller is
vector ed  [(ex /ed ), (e y /ed ), (ez /ed )]T to the transverse
extended to the dynamic model, considering the actual control
o = [ox , o y , oz ]T and vertical t = [tx , t y , tz ]T axes of the
input signals X, M, and N (i.e., a force X in surge and torques
vehicle, respectively. Equations (15) and (16) can be easily
M and N in pitch and yaw, respectively). Hence, given a
verified if we consider the inner products of the unit vector ed
smooth desired trajectory pd (t) = [x d (t), yd (t), z d (t)]T with
with the unit vectors of the transverse o and vertical t axes,
bounded time derivatives, and any initial vehicle configuration
respectively, which equal to the cosine of the angles defined
satisfying Assumption 1 for an arbitrarily small positive design
by the corresponding vectors. It should be noted that the
constant ρ d , the proposed control design is as follows.
aforementioned error transformations (13)–(16) are employed
Kinematic Controller: Select position/orientation perfor-
to display the vehicle kinematics in a form that greatly helps
mance functions ρd (t), ρt (t), and ρo (t) that: 1) satisfy
motivate the structure of the controller derived in the sequel.
Hence, the tracking control problem is solved if the distance
error ed and the orientation errors et , eo reduce to zero (i.e., the
vehicle is heading to the desired trajectory, since the unit
vector ed tends to be normal to both the transverse o and
vertical t axes of the vehicle and consequently aligned to its
longitudinal n axis, when et → 0 and eo → 0). However, for a positive constant ρ̄ < 1 and 2) incorporate the desired
notice from (15) and (16) that the orientation errors et and eo performance specifications regarding the steady-state error and
are well defined only for nonzero values of ed , since the the speed of convergence. Subsequently, design the desired
angles θt and θo are unidentified when ed = 0. Thus, the velocities
proposed control scheme will be designed to further guarantee ⎛ ρ (t)+ρ
⎞
that ed (t) > ρ d > 0, ∀t ≥ 0, for an arbitrarily small positive ed − d 2 d
⎜ 1 + ρd (t)−ρd ⎟
design constant ρ d , in order to avoid the aforementioned ⎜ ⎟
u d = kd ln ⎜ ρd (t)+ρ d ⎟
2
(20)
singularity issue when ed → 0. ⎝ e − ⎠
1 − d ρd (t)−ρ2
Differentiating (13)–(16) and employing (1) and (2) to d
2
obtain the kinematics of the vehicle in the aforementioned et 
error coordinates, we arrive at 1+ ρt (t )
qd = −kt ln et (21)
1−
ėd = −ucθn − vcθo − wcθt ρt (t )
eo 
ex (ẋ d − δx ) + e y ( ẏd − δ y ) + ez (ż d − δz ) 1+
+ (17) rd = ko ln
ρo (t )
(22)
ed 1− eo
ρo (t )
ucθn cθt + vcθo cθt − wsθ2t
ėt = qcθn − pcθo + with positive control gains kd , kt , and ko .
ed
(ẋ d − δx )(ed tx + ex cθt ) ( ẏd − δ y )(ed t y + e y cθt ) Dynamic Controller: Select velocity performance functions
+ + ρu (t), ρq (t), and ρr (t) that satisfy
ed2 ed2
(ż d − δz )(ed tz + ez cθt )
+ (18)
ed2
ucθn cθo − vsθ2o + wcθt cθo
ėo = −r cθn + pcθt +
ed and design the force in the surge as well as the torques in
(ẋ d − δx )(ed ox + ex cθo ) ( ẏd − δ y )(ed o y + e y cθo )
+ + pitch and yaw as
ed2 ed2 
(ż d − δz )(ed oz + ez cθo ) 1 + u−u d
ρu (t )
+ (19) X = −ku ln (23)
ed2 1 − u−u d
ρu (t )

Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
BECHLIOULIS et al.: TRAJECTORY TRACKING WITH PRESCRIBED PERFORMANCE
⎛ ⎞
q−qd
1+
ρq (t )
M = −kq ln ⎝ ⎠ (24)
1 − q−q d
ρq (t )
 by the vector [cψ , sψ , 0]T . Equation (28) can be easily verified
1 + r−r d
ρr (t )
N = −kr ln
1 − r−r d
ρr (t )
with positive control gains ku , kq , and kr .
We now summarize the main results of this section
in Theorem 1.
Theorem 1: Consider any smooth desired trajectory
pd (t) = [x d (t), yd (t), z d (t)]T with bounded derivatives and
a torpedo-like underactuated underwater vehicle modeled
by (1), (2), and (7)–(12) in any initial configuration satisfying
Assumption 1, for an arbitrarily small positive design con-
stant ρ d . The proposed control scheme (20)–(25) guarantees
⎫ the orientation error eo is well defined only for nonzero values
ρ d < ed (t) < ρd (t) ⎬
−ρt (t) < et (t) < ρt (t) ∀t ≥ 0
⎭
−ρo (t) < eo (t) < ρo (t)
for appropriately selected performance functions ρd (t),
ρt (t), and ρo (t) that incorporate the desired transient and
steady-state performance specifications, and avoids the
singularities introduced by the error transformation (13)–(16)
with bounded closed-loop signals, thus leading to the
solution of the robust prescribed performance tracking control
problem, as stated in Section II-C.
Proof: See the Appendix.
Remark 3: Initially, based on Assumption 1, the orientation
errors et and eo are well defined and (17)–(19) are control-
lable. Subsequently, the performance functions ρd (t), ρt (t),
and ρo (t) are selected, such that the orientation errors are
retained well defined and the controllability of (17)–(19) is
preserved as long as the prescribed performance is guaranteed,
that is ρ d < ed (t) < ρd (t), −ρo (t) < eo (t) < ρo (t),
−ρt (t) < et (t) < ρt (t), ∀t ≥ 0. Due to property-dkin
[i.e., ρo2 (t) + ρt2 (t) ≤ ρ̄ < 1], the prescribed performance
guarantees lead to eo2 (t) + et2 (t) = cθ2o (t )+cθ2t (t ) = 1−cθ2n (t ) ≤
ρo2 (t) +
√ ρt (t) ≤ ρ̄ and consequently to |θn (t)| < θ̄n =
2
−1
cos ( 1 − ρ̄) < (π/2), ∀t ≥ 0. Hence, the satisfaction of
prescribed performance constitutes a sufficient condition for
bypassing the aforementioned singularity issues.
B. Unicycle-Like Underactuated Vehicles
Given the desired trajectory pd (t) = [x d (t), yd (t), z d (t)]T ,
let us define the position errors
ex = x d (t) − x, e y = yd (t) − y, ez = z d (t) − z
the projected on the horizontal plane distance error
 tees that the vehicle initializes away from the representation
ed = e2x + e2y
as well as the projected on the horizontal plane orientation
error
ex ey
eo = sψ − cψ = sψe (28)
ed ed
where ψ is the yaw angle and ψe is the angle measured
from the normalized error vector ed := [(ex /ed ), (e y /ed ), 0]T
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
433
on the horizontal plane to the normalized projection of the
longitudinal axis of the vehicle on the horizontal plane, defined
if we consider the cross product of the unit vectors ed
(25) and [cψ , sψ , 0]T , which equals to the sine of the angle the
aforementioned vectors define. It should be noted that the
aforementioned error transformations (26)–(28) are employed
to display the vehicle kinematics in a form that greatly helps
motivate the structure of the controller derived in the sequel.
Hence, the tracking control problem is solved if the projected
on the horizontal plane distance error ed , the vertical error ez ,
and the orientation error eo reduce to zero (i.e., the vehicle
is heading to the desired trajectory, since the unit vector ed
tends to be aligned with the longitudinal axis of the vehicle,
when ez → 0 and eo → 0). However, notice from (28) that
of ed , since the angle ψe is unidentified when ed = 0. Thus, the
proposed control scheme will be designed to further guarantee
that ed (t) > ρ d > 0, ∀t ≥ 0, for an arbitrarily small positive
design constant ρ d , in order to avoid the aforementioned
singularity issue when ed → 0.
Differentiating (26)–(28) and employing (3)–(6) to obtain
the kinematics of the vehicle in the aforementioned error
coordinates, we arrive at
ėd = −ucψe cθ − v(cψe sθ sφ − sψe cφ ) − w(cψe sθ cφ + sψe sφ )
+(ẋ d (t) − δx (t))cψ−ψe + ( ẏd (t) − δ y (t))sψ−ψe (29)
ėz = usθ − vcθ sφ − wcθ cφ + ż d (t) − δz (t) (30)
s c s s +c 2 c
sφ cφ sψ cψ cθ ψe ψe θ φ ψe φ
ėo = qcψe + r cψe +u e e +v
cθ cθ ed ed
sψe cψe cφ sθ − cψ2 s
e φ (ẋ d − δx )
+w + (sψe cψ−ψe + sψ )
ed ed
( ẏd − δ y )
+ (sψe sψ−ψe − cψ ). (31)
ed
Finally, to solve the robust prescribed performance tracking
control problem, we pose Assumption 2.
Assumption 2: The vehicle initializes in such configuration
that: 1) ed (0) > ρ d ; 2) |ψe (0)| ≤ ψ̄e < (π/2); 3) |φ(0)| <
φ̄ < (π/2); and 4) |θ (0)| < θ̄ < (π/2) for an arbitrarily small
positive design constant ρ d and some positive constants ψ̄e ,
φ̄, and θ̄ satisfying cot ψ̄e >tφ̄ sθ̄ .
Remark 4: Assumption 2 guarantees via: 1) that the
orientation error eo is initially well defined and via 2)–4)
that (29)–(31) are initially controllable with respect to u, w,
and r . First, notice that |ψe (0)| ≤ ψ̄e < (π/2), |φ(0)| <
(26) φ̄ < (π/2) and |θ (0)| < θ̄ < (π/2) render (31) controllable
with respect to r . In addition, |θ (0)| < θ̄ < (π/2) guaran-
(27) singularity [i.e., θ = ±(π/2)] of the Euler angles’ parame-
terization. Moreover, extracting the input gain matrix [25]
of (29) and (30)
 
−cψe cθ −(cψe sθ cφ + sψe sφ )
G (φ, θ, ψe ) = (32)
sθ −cθ cφ
considering as inputs the velocities u and w, we can
easily derive a sufficient controllability condition [i.e., whether
434 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

G (φ, θ, ψe ) is invertible] by observing the determinant of Dynamic Controller: Select velocity performance functions
G (φ, θ, ψe ) ρu (t), ρw (t), and ρr (t) that satisfy

det (G (φ, θ, ψe )) = cψe cφ + sψe sφ sθ

which is initially strictly positive owing to Assumption 2

[i.e., cot(|ψe (0)|) ≥ cot ψ̄e > tφ̄ sθ̄ > t|φ(0)| s|θ(0)| leads
straightforwardly to det(G(φ(0), θ (0), ψe (0))) > cψ̄e cφ̄ − and design the forces in surge and heave as well as the torque
sψ̄e sφ̄ sθ̄ > 0]. In this respect, we shall design a controller in yaw as

that, besides guaranteeing ed (t) > ρ d > 0, ∀t ≥ 0, further 1 + u−u d
ρu (t )
forces the heading angle ψe to satisfy |ψe (t)| ≤ ψ e < (π/2), X = −ku ln (35)
1 − u−u d
∀t ≥ 0. Furthermore, since the unactuated roll and pitch ρu (t )

DoFs are passive, the corresponding angles will be proved 1 + w−w
ρw (t )
d

to remain bounded close to the origin, satisfying |φ(t)| < φ̄ Z = −kw ln (36)
1 − w−w
ρw (t )
d

and |θ (t)| < θ̄ , ∀t ≥ 0. Hence, (29)–(31) will remain 

controllable [i.e., |ψe (t)| ≤ ψ̄e , |φ(t)| < φ̄, |θ (t)| < θ̄ and 1 + r−r d
ρr (t )
N = −kr ln (37)
det(G(φ(t), θ (t), ψe (t))) > 0, ∀t ≥ 0] and will evolve away 1 − r−r d
ρr (t )
from the representation singularity. Finally, notice from a prac-
with positive control gains ku , kw , and kr .
tical point of view that Assumption 2 can be easily enforced
We now summarize the main results of this section
by simply placing the vehicle in an initial configuration that is
in Theorem 2.
heading toward the desired trajectory (target), which is quite
Theorem 2: Consider any smooth desired trajectory
reasonable in underwater applications.
pd (t) = [x d (t), yd (t), z d (t)]T with bounded derivatives and
Control Scheme: Similarly to the torpedo case, the proposed
a unicycle-like underactuated underwater vehicle modeled
control scheme is first derived at the kinematic level assuming
by (5), (6), and (7)–(12), in any initial configuration satisfying
that the control signals are the surge, heave, and yaw veloc-
Assumption 2, for an arbitrarily small positive design con-
ities u, w, and r . Subsequently, the kinematic controller is
stant ρ d . The proposed control scheme (33)–(37) guarantees
extended to the dynamic model, considering the actual control ⎫
input signals X, Z , and N (i.e., forces X and Z in surge ρ d < ed (t) < ρd (t) ⎬
and heave as well as a torque N in yaw). Hence, given −ρz (t) < ez (t) < ρz (t) ∀t ≥ 0
a smooth desired trajectory pd (t) = [x d (t), yd (t), z d (t)]T ⎭
−ρo (t) < eo (t) < ρo (t)
with bounded time derivatives, and any initial vehicle con-
for appropriately selected performance functions ρd (t), ρz (t),
figuration satisfying Assumption 2 for an arbitrarily small
and ρo (t) that incorporate the desired transient and steady-
positive design constant ρ d , the control design proceeds as
state performance specifications, and avoids the singularities
follows.
introduced by the error transformations (27) and (28) with
Kinematic Controller: Select position/orientation perfor-
bounded closed-loop signals, thus leading to the solution of
mance functions ρd (t), ρz (t), and ρo (t) that: 1) satisfy
the robust prescribed performance tracking control problem,
as stated in Section II-C.
Proof: The proof proceeds similarly to the proof
of Theorem 1.
C. Commentary
for a positive constant ρ̄ < 1 and 2) incorporate the desired 1) Structural Complexity: The proposed control schemes
performance specifications regarding the steady-state error and do not incorporate any prior knowledge of the external
the speed of convergence. Subsequently, design the desired disturbances and the vehicle’s dynamic model parameters
velocities or even of some corresponding upper bounding constants.
⎡ ⎛ ρ (t)+ρ ⎞
⎤
ed −
d d Furthermore, no estimation (i.e., adaptive control) has been
2
⎢ 1+
⎟⎥
⎢ k ln ⎜
ρd (t)−ρ d employed to acquire such knowledge. Moreover, compared
  ⎢ d ⎜
2 ⎟⎥
⎝ ⎠⎥
ρd (t)+ρ d with the traditional backstepping-like approaches, the pro-
ud −1 ⎢ ed − ⎥
= − [G (φ, θ, ψe )] ⎢ 1− ρ (t)−ρ 2
⎥ posed schemes prove significantly less complex. Notice that no
wd ⎢ d d ⎥
⎢  2
 ⎥ hard calculations are required to output the proposed control
⎣ ez
1+ ρz (t) ⎦
k z ln e
signals, thus making implementation straightforward. Finally,
1− ρz z(t)
it should be noted that no velocity measurements of the
(33) unactuated DoFs (i.e., sway–heave–roll and sway–roll–pitch
eo 
1+ ρo (t ) velocities for the torpedo-like and unicycle-like underactuated
rd = −ko ln eo (34) vehicles, respectively) are incorporated in the control schemes,
1− ρo (t )
thus greatly increasing the robustness against noises that
where G (φ, θ, ψe ) is the input gain matrix defined in (32) corrupt the specific measurements and simplifying further the
and kd , k z , and ko are positive control gains. implementation.

Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
BECHLIOULIS et al.: TRAJECTORY TRACKING WITH PRESCRIBED PERFORMANCE
2) Robust Prescribed Performance: It is worth noticing
that the proposed control schemes achieve their goals without
residing to the need of rendering εi (t), i ∈ {d, t, o, u, q, r }
(or i ∈ {d, z, o, u, w, r }) arbitrarily small, through extreme
values of the control gains ki , i ∈ {d, t, o, u, q, r } (or i ∈
{d, z, o, u, w, r }). More specifically, notice that (48)–(52) hold
no matter how large the finite bounds ε̄i , i ∈ {d, t, o, u, q, r }
and i ∈ {d, z, o, u, w, r } are. In the same spirit, large model
uncertainties, either in the vehicle parameters or the external
disturbances, can be compensated, as they affect only the
size of ε̄i , i ∈ {d, t, o, u, q, r } and i ∈ {d, z, o, u, w, r },
but leave the achieved stability properties unaltered, as shown
in (48)–(52). Hence, the actual tracking performance given
in Theorems 1 and 2, which is solely determined by the
designer-specified performance functions ρi (t), i ∈ {d, t, o}
and i ∈ {d, z, o}, becomes isolated against model uncertainties,
thus greatly extending the robustness of the proposed control
schemes.
3) Control Parameter Selection: Unlike what is common
practice in the related literature, the performance of the closed-
loop system is explicitly and solely determined by appropri-
ately selecting the parameter ρ d and the position/orientation
performance functions ρi (t), i ∈ {d, t, o,} (or i ∈ {d, z, r }).
In particular, the decreasing rate of ρi (t), i ∈ {d, t, o,}
(or i ∈ {d, z, r }) introduces directly a lower bound on the
speed of convergence of the corresponding position/orientation
errors. Furthermore, ρ d and limt →∞ ρi (t), i ∈ {d, t, o,}
(or i ∈ {d, z, r }) regulate the maximum allowable size of
the position/orientation errors at steady state. In that respect,
the performance attributes of the proposed control schemes
are selected a priori, in accordance with the desired transient
and steady-state performance specifications. In this way, the
selection of the control gains ki , i ∈ {d, t, o, u, q, r } (or
i ∈ {d, z, o, u, w, r }), that has been isolated from the actual
control performance, is significantly simplified to adopting
those values that lead to reasonable control effort. Nonethe-
less, it should be noted that their selection affects both the
quality of evolution of the position/orientation errors inside
the corresponding performance envelopes as well as the
control input characteristics (i.e., decreasing the gain values
leads to increased oscillatory behavior within the prescribed
performance envelopes, which is improved when adopting
higher values, enlarging, however, the control effort both in
magnitude and rate). In addition, fine tuning might be needed
in real-time scenarios, to retain the required control input
signals within the feasible range that can be implemented by
the actuators. Similarly, the control input constraints impose an
upper bound on the required speed of convergence of the posi-
tion/orientation performance functions as well as on the uncer-
tainty level, either from the model or the external disturbances,
that can be handled by the proposed control schemes. Hence,
the selection of the control gains ki , i ∈ {d, t, o, u, q, r }
(or i ∈ {d, z, o, u, w, r }) can have positive influence on
the overall closed-loop system response. In the same spirit,
although performance specifications are not required for the
velocity errors, the selection of the corresponding velocity
performance functions ρi (t), i ∈ {u, q, r } (or i ∈ {u, w, r })
affects both the evolution of the position errors within the
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
435
corresponding performance envelopes as well as the control
input characteristics (i.e., relaxing the convergence rate and the
steady-state limit of the velocity performance functions leads
to increased oscillatory behavior, which is improved when
considering tighter performance functions, enlarging, however,
the control effort both in magnitude and rate). Nevertheless,
the only hard constraint attached to their definition is related
to their initial values.
4) Minimal Tracking Information: Interestingly, the pro-
posed control schemes are independent of the time derivatives
of pd (t) = [x d (t), yd (t), z d (t)]T . Regarding the torpedo-like
(or unicycle-like) underactuated vehicles, the desired veloc-
ities u d , qd , and rd (or u d , wd , and rd ) depend only on
pd (t) = [x d (t), yd (t), z d (t)]T . However, u̇ d , q̇d , and ṙd (or u̇ d ,
ẇd , and ṙd ), which involve ṗd (t) = [ẋ d (t), ẏd (t), ż d (t)]T ,
are proven bounded, and therefore, we do not compensate
for them through the design of the control signals X, M,
and N (or X, Z , and N). Thus, applications where the
desired trajectory is not a priori known for all time, but it
is computed online (e.g., tracking a moving target where the
desired trajectory—target position—is obtained at each time
instant via a measuring device and is unknown beforehand),
and thus, its time derivatives are not available (numerical
differentiation typically does not help in this direction, since it
may introduce large errors), can be successfully and efficiently
tackled.
IV. S IMULATIONS
To demonstrate the proposed approach and point out its
intriguing performance properties with respect to the existing
results in the related literature, a comparative simulation study
with the well-established PID control law was conducted
for an underwater pipeline inspection task under external
disturbances representing ocean currents and waves. More
specifically, we considered the tracking control problem of
a multisector trajectory along a complex pipeline structure,
involving line and curved segments as well as vertical and
horizontal helixes, for a torpedo-like underactuated underwater
vehicle.
The dynamics of a six DoF model (the parameters of the
dynamic model were found in [26]) and the motion control
scheme were simulated in UWSim, a realistic simulation
environment developed in the robot operating system frame-
work, using a fourth order Runge–Kutta integration method
with 1-ms time step. The vehicle initialized at rest from the
configuration x(0) = −38 m, y(0) = −9.1 m, z(0) = 27.8 m,
and R(0) = diag([−1, −1, 1]) and was requested to track
a multisector trajectory along the pipeline with a maximum
steady-state position error of 0.3 m and minimum convergence
rate as obtained by the exponential exp(−0.1t). In addition, we
considered that the vehicle’s motion was affected by external
disturbances representing a constant ocean current of the form
δx (t) = −0.25 m/s, δ y (t) = 0.25 m/s, and δz (t) = 0 m/s
(the direction of the ocean stream is shown in Fig. 1 with
green arrows) and harmonic waves modeled by three random
sinusoids with amplitude and frequency equally distributed in
[0.1, 1] N (for the translational dynamics), [0.1, 1] Nm (for
the rotational dynamics) and [0.01, 0.1] rad/s, respectively.
436 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017
Fig. 1. Simulation: trace of the vehicle (blue line) along with the desired
trajectory (green dotted line).
Fig. 2. Simulation: tracking error evolution. Red dashed lines: desired
performance bounds. Blue solid line: evolution of ed (t) for the proposed
scheme. Green solid line: evolution of ed (t) for the PID control scheme.
Given the initial configuration of the vehicle as well
as the required performance specifications and following
Sections III-A and III-C, we selected the performance
functions: ρd (t) = (6.0−0.3) exp(−0.1t)+0.3, ρt (t) = (0.6−
0.1) exp(−t) + 0.1, ρo (t) = (0.6 − 0.1) exp(−t) + 0.1, ρu (t) =
(2−0.1) exp(−1.5t)+0.1, ρq (t) = (2−0.1) exp(−1.5t)+0.1,
and ρr (t) = (2 − 0.1) exp(−1.5t) + 0.1 and the control
gains kd = 10, ko = 2, kt = 2, ku = 100, kq = 20,
and kr = 20, such that (20)–(25) yield reasonable control
effort. Furthermore, it should be noticed that Assumption 1
is satisfied by the aforementioned initial configuration for
ρ d = 0.1. Finally, a fair comparison with the PID controller
was achieved by setting appropriately its gains, following
the pole placement method, in order to achieve the desired
performance specifications, based on a realistic case, where
the model parameters, that were adopted in the approximate
linearization technique, deviated up to 10% from their actual
values.
The trajectory tracking of the proposed scheme is shown
in Fig. 1 for four consecutive time instants. The distance error
of both the proposed scheme and the PID controller along
with the desired performance bounds are shown in Fig. 2.
The evolution of the orientation errors under the proposed
control law along with the corresponding performance bounds
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
Fig. 3. Orientation tracking error evolution.
Fig. 4. Control commands in surge (X), pitch (M), and yaw (N ).
is shown in Fig. 3. Finally, the requested control effort under
the proposed scheme is shown in Fig. 4. Notice that as it
was predicted by the theoretical analysis, tracking with pre-
scribed performance and bounded control signals is achieved
despite the presence of external disturbances and the lack
of knowledge of the vehicle’s dynamic model parameters.
On the contrary, although the response of the PID controller,
as observed in Fig. 2, was quite satisfactory during the
transient and along the line segments of the desired trajectory,
where the linearization is rather accurate, it should be noted
that the error increased up to almost 1.5 m with a quite
unsatisfactory response during the helical segments, where the
linearization was not adequate, owing to the large operating
envelope and the system uncertainty. In this way, it is verified
that the proposed scheme outperforms the well-established
PID controller in challenging operating conditions that are
commonly met in underwater applications. An accompanying
video demonstrating the aforementioned simulation study may
be found in https://youtu.be/2cNN3ksPjd4.
V. E XPERIMENT
To verify the tracking performance and robustness of the
proposed scheme, an experimental procedure was carried out
inside a water tank using a small remotely operated underwater
vehicle. The deployed vehicle VideoRay PRO is equipped with
three thrusters, affecting surge–heave–yaw motions (i.e., it
falls within the class of unicycle-like underactuated vehicles)
and its control unit is connected with a host computer through
BECHLIOULIS et al.: TRAJECTORY TRACKING WITH PRESCRIBED PERFORMANCE 437

Fig. 5. Experiment: tracking error evolution. Red dashed lines: desired Fig. 6. Control commands in surge (X), heave (Z ), and yaw (N ).
performance bounds. Blue solid lines: evolution of ed (t), ez (t), and eo (t).
The subplots give details at the steady state.

The existing setup did not allow us to create external

a serial communication interface (RS-232). In addition, a disturbances in the form of waves and currents. However,
Polhemus-Isotrack device, interfaced to the host computer via the power and communication tether attached to the vehicle
RS-232 serial communication at 30 Hz, was employed as a was creating nontrivial external disturbances in the form of
pose feedback sensor for the motion control scheme. The random forces and torques, as the vehicle was constantly
Isotrak tracking system consists of a transmitter and a receiver changing configuration during the experiment. As it can be
(tracker) and uses electromagnetic fields to determine the seen from Fig. 5, and the overall performance of the con-
tracker’s position/orientation. In our case, the transmitter was troller, the aforementioned external disturbances were rejected
placed at a fixed spot outside the water tank and the tracker successfully, thus verifying the efficiency and robustness of
was mounted on the vehicle. Finally, the overall software the proposed control scheme. An accompanying video demon-
architecture was developed in C++ on a Linux operating strating the aforementioned simulation study may be found in
system. https://youtu.be/knaK-BcSwiM.
The vehicle initialized at rest from the configuration
x(0) = 0.22 m, y(0) = 0.76 m, z(0) = 0.27 m, φ(0) = 0.1°,
A PPENDIX
θ (0) = 1.2°, and ψ(0) = −94.59° and was requested to track
a helical trajectory toward the bottom of the tank described A. Proof of Theorem 1
by x d (t) = 0.4 cos(0.1πt) m, yd (t) = −0.4 sin(0.1πt) m, Let us define the normalized errors
and z d (t) = 0.3 + 0.015t m, with a maximum steady-state ρd (t )+ρd
error of 0.2 m and minimum convergence rate as obtained by ed − et eo
ξd = 2
ρd (t )−ρ d
, ξt = , ξo = (38)
the exponential exp(−0.25t). Notice that the aforementioned ρt (t) ρo (t)
2
initial configuration satisfied Assumption 2 for ρ d = 0.05. u − ud q − qd r − rd
Hence, following Section III-B, we selected the perfor- ξu = , ξq = , ξr = . (39)
ρu (t) ρq (t) ρr (t)
mance functions: ρd (t) = (2.0 − 0.15) exp(−0.25t) + 0.15,
ρz (t) = (0.5 − 0.05) exp(−0.25t) + 0.05, ρo (t) = (0.8 − In this respect, the desired velocities (20)–(22) and the control
0.08) exp(−0.25t)+0.08, ρu (t) = (2−0.5) exp(−0.25t)+0.5, law (23)–(25) may be written as the functions of the normal-
ρw (t) = (2 − 0.5) exp(−0.25t) + 0.5, and ρr (t) = (2 − ized errors ξi , i ∈ {d, t, o, u, q, r } as follows:
0.5) exp(−0.25t)+0.5 and the control gains kd = 0.3, k z = 1,    
1 + ξd 1 + ξt
ko = 1, ku = 7.5, kw = 5, and kr = 0.25, such that (33)–(37) u d = kd ln , qd = −kt ln
1 − ξd 1 − ξt
yield reasonable control effort.  
As it was predicted by the theoretical analysis and is 1 + ξo
rd = ko ln (40)
actually shown in Fig. 5, tracking with prescribed performance 1 − ξo
   
is successfully achieved, with the errors evolving strictly 1 + ξu 1 + ξq
within the predefined performance bounds, despite the lack of X = −ku ln , M = −kq ln
1 − ξu 1 − ξq
knowledge regarding the vehicle’s dynamic model parameters.  
In addition, as it is shown in Fig. 6, the required control effort 1 + ξr
N = −kr ln . (41)
was satisfactorily smooth and did not impose any saturation 1 − ξr
to the thrusters (the limits of the considered vehicle for Let us now define the overall state vector
the X, Z , and N degrees of actuation are 15 N, 7.5 N,
and 1 Nm, respectively). ξ = [ξd , ξo , ξt , ξu , ξq , ξr , s T ]T

Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
438 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017

where s = [v, w, p]T denotes the unactuated velocities and |s(t)| < s̄ for all t ∈ [0, τmax ). Therefore, the signals
(i.e., sway, heave, and roll velocities). Differentiating the  
normalized errors (38) and (39) with respect to time and 1 + ξi (t)
εi (t) = ln , i ∈ {d, t, o, u, q, r } (44)
substituting (7)–(12), (17)–(19), (40), and (41), we obtain in 1 − ξi (t)
a compact form, the closed-loop dynamical system
are well defined for all t ∈ [0, τmax ). Consider now the positive
ξ̇ = h(t, ξ ) (42) definite and radially unbounded function Vd = (1/2)εd2 .
where the function h(t, ξ ) includes all terms found in the right- Differentiating with respect to time and substituting (17),
hand side, after the differentiation of ξ . Let us also define the we obtain (45), as shown at the bottom of the page. Incor-
set porating u = u d + ξu ρu (t) from (39) and substituting u d from
(40) and εd from (44), V̇d becomes the equation shown at the
ξ = (−1, 1) × · · · × (−1, 1) ×s , bottom of the page.
  
6-times Furthermore, utilizing: 1) (43); 2) the fact that ρ̇d (t), ρu (t),
where s = {s ∈ 3 : s < s̄} is an open set with ẋ d (t), ẏd (t), ż d (t), δx (t), δ y (t), and δz (t) are bounded by
s̄ denoting a positive constant to be specified later, for the construction and by assumption; and 3) s(t) < s̄, we
analysis purposes only. arrive at

In the sequel, we proceed in two phases. First, the existence 
and uniqueness of a maximal solution ξ (t) of (42) over the  − ξu ρu (t)cθ − vcθ − wcθ
 n o t
set ξ for a time interval [0, τmax ) is ensured [i.e., ξ (t) ∈ ξ , 
ex (ẋ d − δx ) + e y ( ẏd − δ y ) + ez (ż d − δz ) (1+ξd )ρ̇d (t) 
∀t ∈ [0, τmax )]. Then, we prove that the proposed control + − 
scheme guarantees, for all t ∈ [0, τmax ): 1) the boundedness of e 2
⎧d ⎛ ⎞ ⎫
all closed-loop signals of (42) and 2) that ξ (t) remains strictly ⎨ ẋ d (t) − δx (t) ⎬
within a compact subset of ξ , which subsequently will ≤ ρu (0) + s̄ + sup ⎝ ẏd (t) − δ y (t) ⎠ + |ρ̇d (t)| := F̄d
t ≥0 ⎩ ż d (t) − δz (t)
⎭
lead to τmax = ∞ by contradiction and consequently to the
solution of the tracking control problem stated in Section II-C. (46)
Moreover, notice that the design parameter ρ d and the
performance functions ρd (t), ρt (t), and ρo (t) were selected for an unknown positive constant F̄d . Moreover,
(see Section III-A), such that no singular point lies in ξ (1/(1 − ξd2 )) > 1 and as mentioned earlier cθn ≥ cθ̄n > 0 and
(i.e., ed > ρ d > 0 and |θn | ≤ θ̄n < π/2, ∀ξ ∈ ξ ). ρd (t) − ρ d > 0. Therefore, we conclude that V̇d is negative
In this respect, the proposed scheme avoids the singularity when |εd (t)| > ( F̄d /kd cθ̄n ) and consequently that
issues mentioned in Section III-A, which have been the main
! "
drawbacks in similar control approaches for underactuated F̄d
underwater vehicles, via guaranteeing that ξ (t) remains strictly |εd (t)| ≤ ε̄d = max |εd (0)|, (47)
kd cθ̄n
within a compact subset of ξ .
Phase A: The set ξ is nonempty and open. Moreover,
for all t ∈ [0, τmax ), which from (44), by applying the inverse
owing to the selection of the performance functions ρi (t),
logarithmic function, leads to
i ∈ {d, t, o, u, q, r } as well as to Assumption 1, we conclude
that ξ (0) ∈ ξ for a positive constant s̄, satisfying s(0) < s̄. exp(−ε̄d ) − 1 exp(ε̄d ) − 1
In addition, due to the smoothness of: 1) the system nonlin- −1 < = ξd ≤ ξd (t) ≤ ξd = <1
exp(−ε̄d ) + 1 exp(ε̄d ) + 1
earities; 2) the desired trajectory; and 3) the proposed control (48)
scheme over ξ , it can be easily verified that h(t, ξ ) is
continuous on t and continuous for all ξ ∈ ξ . Therefore, for all t ∈ [0, τmax ). Hence, the desired velocity u d also
the hypotheses of [27, p. 476, Th. 54] hold and the existence remains bounded [i.e., |u d (t)| ≤ kd ε̄d ] for all t ∈ [0, τmax ).
and uniqueness of a maximal solution ξ (t) of (42) on a time Following the similar analysis for (18) and (19) with
interval [0, τmax ) such that ξ (t) ∈ ξ , ∀t ∈ [0, τmax ) is the positive definite and radially unbounded functions
ensured. Vt = (1/2)εt2 and Vo = (1/2)εo2 , we arrive at
Phase B: We have proven in Phase A that ξ (t) ∈ ξ , |εt (t)| ≤ ε̄t = max{|εt (0)|, ( F̄t /kt cθ̄n )}, |εo (t)| ≤ ε̄o =
∀t ∈ [0, τmax ) or equivalently that max{|εo (0)|, ( F̄o /ko cθ̄n )}, ∀t ∈ [0, τmax ) for some positive
ξi (t) ∈ (−1, 1), i ∈ {d, t, o, u, q, r } (43) constants F̄o and F̄t satisfying similar inequalities to (46),

 
4εd ex (ẋ d − δx ) + e y ( ẏd − δ y ) + ez (ż d − δz ) (1 + ξd )ρ̇d (t)
V̇d = # $ − ucθn − vcθo − wcθt + − . (45)
1−ξd2 (ρd (t)−ρ d ) ed 2
 
4εd ex (ẋ d −δx )+e y ( ẏd −δ y )+ez (ż d −δz ) (1+ξd )ρ̇d (t)
V̇d = # $ − ξu ρu (t)cθn − vcθo − wcθt + − − k d ε d cθ n .
1−ξd2 (ρd (t)−ρ d ) ed 2

Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
BECHLIOULIS et al.: TRAJECTORY TRACKING WITH PRESCRIBED PERFORMANCE 439

which further lead to: where α  min{|Y|v|v |, |Z |w|w |, |K | p| p |}, β 

min{|Yv |, |Z w |, |K p |}, and
exp(−ε̄d ) − 1 exp(ε̄d ) − 1
−1 < = ξt ≤ ξt (t) ≤ ξt = <1 (m u ū r̄ + sup{|δv (t)|})
exp(−ε̄d ) + 1 exp(ε̄d ) + 1
t ≥0
(49)
Fvwp := (m u ū q̄ + sup{|δw (t)|})
exp(−ε̄d ) − 1 exp(ε̄d ) − 1 t ≥0
−1 < = ξo ≤ ξo (t) ≤ ξo = < 1. (m qr q̄ r̄ + |z B |W + supt ≥0 {|δ p (t)|})
exp(−ε̄d ) + 1 exp(ε̄d ) + 1
(50) (−m u ur + δv (t))
≥ (m u uq + δw (t)) .
Moreover, the desired velocities qd and rd remain bounded as (m qr qr + z B W cθ sφ + δ p (t))
well [i.e., |qd (t)| ≤ kt ε̄t , |rd (t)| ≤ ko ε̄o , ∀t ∈ [0, τmax )]. Thus,
invoking (39), it is straightforward to deduce the boundedness Therefore, we conclude that V̇vwp is negative when
of u(t), q(t), and r (t) [i.e., |u(t)| ≤ ū := ρu (0) + kd ε̄d , '
(
√ 2 √ √
|q(t)| ≤ q̄ := ρq (0) + kt ε̄t , |r (t)| ≤ r̄ := ρr (0) + ko ε̄o , ( 3Fvwp
s > S := )
3β 3β
+ −
∀t ∈ [0, τmax )], and consequently, by differentiating (40) 2α α 2α
with respect to time and after some straightforward algebraic
manipulations, the boundedness of u̇ d (t), q̇d (t), and ṙd (t), and consequently that s(t) ∈ s , ∀t ∈ [0, τmax ), where
∀t ∈ [0, τmax ). % % &&
  S
Applying the aforementioned line of proof for the dynamics s = s ∈  : s ≤ s̄ := max s(0),
3
min{m v , m w , m p }
of the velocity errors εu , εq , and εr defined in (44) that
involve the dynamic model of the vehicle (7)–(12), considering (53)
Vu = (1/2)m u εu2 , Vq = (1/2)m q εq2 , and Vr = (1/2)m r εr2 is a compact set the size of which depends on: 1) the
and the proposed control law (23)–(25), we conclude control gain values kd , kt , and ko ; 2) the control parameters
that ρu (0), ρq (0), and ρr (0); 3) the parameters of the dynamic
% & model (7)–(12); 4) the magnitude of the external disturbances;
F̄i
|εi (t)| ≤ ε̄i = max |εi (0)|, ∀t ∈ [0, τmax ) (51) and 5) implicitly on the desired trajectory and the transient
ki performance specifications.
Up to this point, what remains to be shown is that tmax = ∞.
for some positive constants F̄i , i ∈ {u, q, r } that satisfy similar Notice that (48)–(50), (52), and (53) imply that ξ (t) ∈ ξ ,
inequalities to (46) relating the performance specifications, the ∀t ∈ [0, τmax ), where
unknown parameters of the dynamic model and the external  
wave disturbances. Accordingly, we arrive at * exp(−ε̄i ) − 1 exp(ε̄i ) − 1

ξ = , × s
exp(−ε̄i ) + 1 exp(ε̄i ) + 1
exp(−ε̄i ) − 1 exp(ε̄i ) − 1 i∈{d,t,o,u,q,r}
−1 < = ξi ≤ ξi (t) ≤ ξi = <1
exp(−ε̄i ) + 1 exp(ε̄i ) + 1 is a nonempty and compact set. Moreover, it can be easily
(52) verified that ξ ⊂ ξ for certain, sufficient small control
gain values kd , kt , and ko and parameters ρu (0), ρq (0), and
for all t ∈ [0, τmax ) and i ∈ {u, q, r }, as well as at the ρr (0) satisfying s̄  < s̄. Hence, assuming τmax < ∞, and
boundedness of the control law (23)–(25) for all t ∈ [0, τmax ) since ξ ⊂ ξ , [27, p. 481, Proposition C.3.6] dictates the
[i.e., |X (t)| ≤ ku ε̄u , |M(t)| ≤ kq ε̄q , |N(t)| ≤ kr ε̄r ]. existence of a time instant t  ∈ [0, τmax ), such that ξ (t  ) ∈
/ ξ ,
In the sequel, special attention will be paid on the sta- which is a clear contradiction. Therefore, τmax = ∞. As a
bility of the unactuated velocity vector s = [v, w, p]T . result, all closed-loop signals remain bounded, and moreover,
In this way, let us define the positive definite and radially
unbounded function Vvwp = (1/2)s T diag([m v , m w , m p ])s,
where m v , m w , and m p denote the vehicle’s mass/moment
of inertia and added mass/moment of inertia of the corre-
sponding DoFs. Differentiating Vvwp with respect to time and
substituting (8)–(10), we obtain
ξ (t) ∈ ξ ⊂ ξ , ∀t ≥ 0. Finally, from (38) and (48)–(50), we
conclude that ρ d < ed (t) < ρd (t), −ρz (t) < ez (t) < ρz (t),
and −ρo (t) < eo (t) < ρo (t) for all t ≥ 0, and consequently,
the solution of the robust prescribed performance tracking
control problem, as stated in Section II-C, which completes
the proof.

V̇vwp = s T diag([Yv + Y|v|v |v|, Z w + Z |w|w |w|, R EFERENCES

⎡ ⎤
−m u ur + δv (t) [1] C. P. Bechlioulis and G. A. Rovithakis, “Robust partial-state feedback
⎢ ⎥
K p + K | p| p | p|])s + s T ⎣ m u uq + δw (t) ⎦ prescribed performance control of cascade systems with unknown non-
linearities,” IEEE Trans. Autom. Control, vol. 56, no. 9, pp. 2224–2230,
m qr qr + z B W cθ sφ + δ p (t) Sep. 2011.
[2] C. P. Bechlioulis and G. A. Rovithakis, “A low-complexity global
which, after straightforward algebraic manipulations, leads to approximation-free control scheme with prescribed performance for
unknown pure feedback systems,” Automatica, vol. 50, no. 4,
α pp. 1217–1226, Apr. 2014.
V̇vwp ≤ − √ s3 − βs2 + Fvwp s [3] T. I. Fossen, Guidance and Control of Ocean Vehicles. Chichester, U.K.:
3 Wiley, 1994.

Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
440 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 25, NO. 2, MARCH 2017
[4] S. A. Al-Hiddabi and N. H. McClamroch, “Tracking and maneuver reg-
ulation control for nonlinear nonminimum phase systems: Application
to flight control,” IEEE Trans. Control Syst. Technol., vol. 10, no. 6,
pp. 780–792, Nov. 2002.
[5] J. R. T. Lawton, R. W. Beard, and B. J. Young, “A decentralized
approach to formation maneuvers,” IEEE Trans. Robot. Autom., vol. 19,
no. 6, pp. 933–941, Dec. 2003.
[6] F. Alonge, F. D’Ippolito, and F. M. Raimondi, “Trajectory tracking of
underactuated underwater vehicles,” in Proc. 40th IEEE Conf. Decision
Control, vol. 5. Dec. 2001, pp. 4421–4426.
[7] A. P. Aguiar and A. M. Pascoal, “Dynamic positioning and way-point
tracking of underactuated AUVs in the presence of ocean currents,” Int.
J. Control, vol. 80, no. 7, pp. 1092–1108, 2007.
[8] A. Baviskar, M. Feemster, D. Dawson, and B. Xian, “Tracking control of
an underactuated unmanned underwater vehicle,” in Proc. Amer. Control
Conf., vol. 6. Jun. 2005, pp. 4321–4326.
[9] F. Repoulias and E. Papadopoulos, “Planar trajectory planning and
tracking control design for underactuated AUVs,” Ocean Eng., vol. 34,
nos. 11–12, pp. 1650–1667, 2007.
[10] J. E. Refsnes, A. J. Sorensen, and K. Y. Pettersen, “Model-based
output feedback control of slender-body underactuated AUVs: Theory
and experiments,” IEEE Trans. Control Syst. Technol., vol. 16, no. 5,
pp. 930–946, Sep. 2008.
[11] F. Y. Bi, Y. J. Wei, J. Z. Zhang, and W. Cao, “Position-tracking control
of underactuated autonomous underwater vehicles in the presence of
unknown ocean currents,” IET Control Theory Appl., vol. 4, no. 11,
pp. 2369–2380, Nov. 2010.
[12] K. D. Do, “Global tracking control of underactuated ODINs in three-
dimensional space,” Int. J. Control, vol. 86, no. 2, pp. 183–196, 2013.
[13] A. P. Aguiar and J. P. Hespanha, “Trajectory-tracking and path-following
of underactuated autonomous vehicles with parametric modeling uncer-
tainty,” IEEE Trans. Autom. Control, vol. 52, no. 8, pp. 1362–1379,
Aug. 2007.
[14] K. D. Do, J. Pan, and Z. P. Jiang, “Robust and adaptive path following for
underactuated autonomous underwater vehicles,” Ocean Eng., vol. 31,
no. 16, pp. 1967–1997, 2004.
[15] G. Antonelli, S. Chiaverini, N. Sarkar, and M. West, “Adaptive con-
trol of an autonomous underwater vehicle: Experimental results on
ODIN,” IEEE Trans. Control Syst. Technol., vol. 9, no. 5, pp. 756–765,
Sep. 2001.
[16] L. Lapierre, “Robust diving control of an AUV,” Ocean Eng., vol. 36,
no. 1, pp. 92–104, 2009.
[17] R. Cristi, F. A. Papoulias, and A. J. Healey, “Adaptive sliding mode
control of autonomous underwater vehicles in the dive plane,” IEEE J.
Ocean. Eng., vol. 15, no. 3, pp. 152–160, Jul. 1990.
[18] M. Santhakumar and T. Asokan, “Investigations on the hybrid tracking
control of an underactuated autonomous underwater robot,” Adv. Robot.,
vol. 24, no. 11, pp. 1529–1556, 2010.
[19] H. Ashrafiuon, K. R. Muske, L. C. McNinch, and R. A. Soltan, “Sliding-
mode tracking control of surface vessels,” IEEE Trans. Ind. Electron.,
vol. 55, no. 11, pp. 4004–4012, Nov. 2008.
[20] D. Zhu, X. Hua, and B. Sun, “A neurodynamics control strategy for real-
time tracking control of autonomous underwater vehicles,” J. Navigat.,
vol. 67, no. 1, pp. 113–127, 2014.
[21] B. Taheri and E. Richer, “Equidistance target-following controller for
underactuated autonomous underwater vehicles,” Int. J. Intell. Comput.
Cybern., vol. 6, no. 2, pp. 108–125, 2013.
[22] J. H. Li and P. M. Lee, “A neural network adaptive controller design for
free-pitch-angle diving behavior of an autonomous underwater vehicle,”
Robot. Auto. Syst., vol. 52, nos. 2–3, pp. 132–147, 2005.
[23] K. Ishii, T. Fujii, and T. Ura, “An on-line adaptation method in a neural
network based control system for AUVs,” IEEE J. Ocean. Eng., vol. 20,
no. 3, pp. 221–228, Jul. 1995.
[24] J. Guo, F.-C. Chiu, and C.-C. Huang, “Design of a sliding mode fuzzy
controller for the guidance and control of an autonomous underwater
vehicle,” Ocean Eng., vol. 30, no. 16, pp. 2137–2155, 2003.
[25] H. Xu and P. A. Ioannou, “Robust adaptive control for a class of MIMO
nonlinear systems with guaranteed error bounds,” IEEE Trans. Autom.
Control, vol. 48, no. 5, pp. 728–742, May 2003.
[26] K. Y. Pettersen and O. Egeland, “Time-varying exponential stabilization
of the position and attitude of an underactuated autonomous underwater
vehicle,” IEEE Trans. Autom. Control, vol. 44, no. 1, pp. 112–115,
Jan. 1999.
[27] E. D. Sontag, Mathematical Control Theory. London, U.K.: Springer,
1998.
Authorized licensed use limited to: University of Illinois. Downloaded on May 28,2020 at 06:44:41 UTC from IEEE Xplore. Restrictions apply.
Charalampos P. Bechlioulis was born in Arta,
Greece, in 1983. He received the Diploma (Hons.)
degree in electrical and computer engineering, the
B.Sc. (Hons.) degree in mathematics, and the
Ph.D. degree in electrical and computer engineer-
ing from the Aristotle University of Thessaloniki,
Thessaloniki, Greece, in 2006, 2011, and 2011,
respectively.
He is currently a Post-Doctoral Researcher with
the Control Systems Laboratory, School of Mechan-
ical Engineering, National Technical University of
Athens, Athens, Greece. He has authored over 55 papers in scientific journals,
conference proceedings, and book chapters. His current research interests
include nonlinear control with prescribed performance, system identification,
control of robotic vehicles, multiagent systems, and object grasping.
George C. Karras received the Diploma degree in
mechanical engineering, the M.S. degree in robotics
and automatic control systems, and the Ph.D. degree
in mechanical engineering from the National Tech-
nical University of Athens (NTUA), Athens, Greece,
in 2003, 2006, and 2011, respectively.
He was a Research and Development Engi-
neer with the Electronics Department, Hellenic
Aerospace Industry, Tanagra, Greece. He has over
ten years of experience in robotics and systems
engineering. He is currently a Research Associate
with the Control Systems Laboratory, NTUA.
Shahab Heshmati-Alamdari received the
Diploma degree in mechanical engineering and the
M.Sc. degree in control systems from the National
Technical University of Athens, Athens, Greece, in
2009 and 2011, respectively, where he is currently
pursuing the Ph.D. degree with the Control Systems
Laboratory, School of Mechanical Engineering.
His current research interests include underwater
robotics, nonlinear control of underwater vehicle
manipulator systems, multiagent systems, nonlinear
control, visual servoing, model predictive control,
and event-based control.
Mr. Heshmati-Alamdari has served as a Reviewer to a number of journals
and conferences and has been involved in EC projects.
Kostas J. Kyriakopoulos received the Diploma
(Hons.) degree in mechanical engineering from the
National Technical University of Athens (NTUA),
Athens, Greece, in 1985, and the M.S. and Ph.D.
degrees in electrical, computer, and systems engi-
neering from the Rensselaer Polytechnic Insti-
tute (RPI), Troy, NY, USA, in 1987 and 1991,
respectively.
He was involved in research with the NASA
Center for Intelligent Robotic Systems for Space
Exploration, Troy, from 1988 to 1991. From 1991
to 1993, he was an Assistant Professor with the Department of Electrical,
Computer and Systems Engineering, RPI, and the NY State Center for
Advanced Technology in Automation and Robotics, Troy. Since 1994, he
has been with the Control Systems Laboratory, Mechanical Engineering
Department, NTUA, where he currently serves as a Professor and the Director.
He has authored nearly 280 papers in journals and refereed conferences, and
has contributed to a large number of projects funded by the EC and the Greek
Secretariat for Research and Technology. His current research interests include
nonlinear control systems applications in sensor-based motion planning and
control of multirobotic systems, such as manipulators and vehicles (mobile,
underwater, and aerial), and micro and biomechatronics.