European Journal of Control 58 (2021) 357–372

Contents lists available at ScienceDirect

European Journal of Control

journal homepage: www.elsevier.com/locate/ejcon

L1 adaptive backstepping control for path-following of underactuated

marine surface ships
Haitong Xu a, P. Oliveira b, C. Guedes Soares a,∗
a
Centre of Marine Technology and Ocean Engineering (CENTEC), Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1049-001 Lisboa,
Portugal
b
Associated Laboratory for Energy, Transports and Aeronautics, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1049-001 Lisboa,
Portugal

a r t i c l e i n f o a b s t r a c t

Article history: A L1 adaptive backstepping controller is proposed for path-following control of an underactuated ship
Received 16 January 2020 based on a nonlinear steering model, and the control law is derived using Lypunov control function. L1
Revised 13 July 2020
adaptive controller is a novel technology considering both the robustness and fast adaptation. The closed-
Accepted 8 August 2020
loop control system is proved to be uniformly global asymptotically stable. A novel guidance law, time-
Available online 14 August 2020
varying vector field, is proposed for the guidance system. The kinematic analysis of path-following shows
Recommended by Prof. T. Parisini that the guidance and control systems are coupled, and the guidance system is affected by the perfor-
mance of the controller. The Lyapunov method is employed to show that the equilibrium point of the
Keywords:
L1 adaptive control whole system is uniformly global asymptotically stable. Path-following simulations are carried out to val-
Vector field idate the performance of the proposed control law and vector field guidance law using a fully nonlinear
Underactuated ship manoeuvring mode of an underactuated tanker in the presence of stochastic ocean current.
Path-following © 2020 European Control Association. Published by Elsevier Ltd. All rights reserved.
Stability analysis

1. Introduction technology. In the following year, Norway opens an official test

bed for autonomous shipping [35]. In 2018, IMO defined degrees
Marine autonomous surface ships (MASS) have been drawing of autonomy for autonomous surface ships, and the preliminary
much attention in the maritime industry, and this is reflected in research on the regulation of autonomous ships was carried out.
the recent guidelines for the conduct of autonomous ship trials For an autonomous ship, the most important and fundamental
approved by the Maritime Safety Committee (MSC) of the Interna- system is the guidance and control system as it directly affects the
tional Maritime Organisation (IMO) in 2019. The main reasons for whole performance of autonomous ships.
the rapid development of autonomous ships are safety and eco- A robust and reliable guidance and control system are closely
nomic benefit. The direct causes of ship losses are of different na- related to the safety of autonomous ships. The research on the
ture [17], but different studies have concluded that about 75% to guidance and control system of autonomous ships is becoming im-
96% of marine accidents can be traced back to human errors [1,18]. portant and urgent, especially with the fast development of au-
Autonomous shipping can significantly improve safety by reducing tonomous ships.
human errors. Another driving force for the development of au- A review in the area of nonlinear controllers for ships can be
tonomous or unmanned ships is an economic benefit. Autonomous found in [11], which briefly introduced the application of the non-
ships can significantly reduce the cost of onboard personnel and linear controllers, such as linear quadratic regulator (LQR), state
fuel consumption, although initial investment in equipment and feedback linearization, backstepping, sliding mode control, etc. In
shore based support maybe higher [46]. Meanwhile, without the recent years, model predictive control was used for surface ships
Life support system, the ship can carry more cargos. by Oh and Sun [39], where the rudder dynamic model was consid-
These important benefits have motivated recent research on ered. A sliding mode controller was used for trajectory tracking of
autonomous ships. For example, Rolls-Royce [45] led a project an ROV by Guerrero et al. [20]. Liu et al. [33] used the model pre-
that addressed the various factors related to the development dictive control for the path following control of a marine surface
of autonomous ships, such as economic, social, regulatory and ship considering the roll motion effect. Meanwhile, experiments
were carried to validate the proposed control law. Fuzzy logic con-
∗ trol was widely used in autonomous systems due to the simple
Corresponding author.
E-mail address: [email protected] (C. Guedes Soares). control structure. Li et al. [31] used an adaptive neural network

https://doi.org/10.1016/j.ejcon.2020.08.003
0947-3580/© 2020 European Control Association. Published by Elsevier Ltd. All rights reserved.
358 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

fast power reaching for the path following control of underactu- most desired stability for a closed-loop nonlinear system is global
ated marine surface vessels. A review on the fuzzy logic control for exponential stable (GES) [2,25,34], because the convergence of
marine underwater vehicles can be found in Xiang et al. [51]. Wang the error dynamic equations is bounded by an exponential decay.
et al. [50] proposed a fuzzy observer-based controller for surface However, unfortunately, it cannot be achieved for path-following
ships, and globally asymptotically stable was proved. control of autonomous ships. The kinematics of the path-following
The adaptive control method is one promising technology for a introduces saturation through trigonometric functions [13].
dynamic system, especially for autonomous ships because the ma- Global κ -exponential stability of LOS guidance law was firstly
rine surface ships usually travel in a complex environmental with proved by Fredriksen and Pettersen [15]. Fossen and Pettersen
waves, wind, and currents. The classical control system is designed [13] gave the proof of uniform semiglobal exponential stability
using a static steering model, for example, the Nomoto model, (USGES) for LOS guidance law. Do et al. [7] used a robust adap-
which is the most widely used. The static steering model does not tive controller for path-following control of underactuated ships.
consider the time-varying environmental disturbance and thus it The following works can found in [5,8,9],
usually will fail to describe the ship’s dynamic response when the In this paper, L1 adaptive backstepping controller is proposed
environment changes [42]. For example, when ships access a har- for path-following control, and the performance is evaluated with
bour from the ocean, the pilots can feel that the ship is becoming a fully nonlinear manoeuvring model of an underactuated au-
hard to manoeuvre. In fact, research shows that the shallow water tonomous ship. The key feature of L1 adaptive-control architectures
has a significant effect on the manoeuvrability of ships [26,48]. Un- is the decoupling of estimation and control, which enables the use
fortunately, the classical controller based on static steering models of arbitrarily fast estimation rates without sacrificing robustness.
does not consider the effect of shallow water, which changes the A nonlinear steering model considering the environmental distur-
manoeuvring model (Xu and Guedes Soares, 2019). bance is employed for the control system design. The Lyapunov
Recently, a novel adaptive control architecture, denoted as L1 control function is used to derive the control law, where the pa-
adaptive control, was proposed. It can guarantee both fast adap- rameter update law is also given. The stability proof shows that the
tation and transient performance, as described in [21]. Guerreiro, L1 adaptive backstepping controller is a uniform global asymptotic
Silvestre, Cunha, Cao, and Hovakimyan [19] used L1 adaptive con- stable (UGAS). The guidance system is designed based on the vec-
troller for an inner loop controller for autonomous rotorcraft. tor field method, and a time-varying vector field guidance law is
Kaminer et al. [23] employed the L1 adaptive controller for the proposed. The coupled guidance and control system are proved to
path-following of unmanned aerial vehicles. Lee et al. [28] pro- be uniform global asymptotic stable (UGAS) using a cascaded sys-
posed an L1 adaptive backstepping controller for autopilot of an tem theory. The proposed control law and guidance law are vali-
agile missile. Sørensen and Breivik [47] compared the performance dated with a fully nonlinear manoeuvring mode in the presence of
of L1 adaptive backstepping controller with the classical adaptive stochastic ocean current.
control. The results show that the L1 adaptive control has a bet- This paper is organized as follows. Sections 2 describes the
ter performance. Some published papers demonstrate that the L1 kinematic of path-following control for an underactuated marine
adaptive control greatly improved the performance of unmanned surface ship. Section 3 proposed the L1 adaptive backstepping con-
aerial vehicles, such as, [16,19,24,28]. In this paper, L1 adaptive troller and stability proof is also outlined. The parameters update
backstepping controller is employed for the path-following control law is also given in this section. In Section 4, the error dynamic
problem of an underactuated ship. equations of the guidance and control system are summarised, and
Guidance and control system are the basic component for an the theorem and proof for UGAS are given using cascaded sys-
autonomous ship. The guidance portion of the system is used to tem theory. Section 5 presented the path-following simulations.
develop the variable input signal, which represents the desired The performance of the proposed guidance and control system is
course to the target. There are many guidance laws for an au- discussed. Conclusions are given in Section 6.
tonomous system, such as pure pursuit (PP), vector field (VF) [57],
constant bearing (CB) and line of sight (LOS). LOS is one popular
method, and widely used for autonomous ships [3,14,15,30,36,49]. 2. Problem formulation
Lekkas and Fossen [30], proposed a revised version of LOS, look-
ahead LOS, for the autonomous ships, where the look-ahead dis- Path-following control of marine surface ships is briefly in-
tance is a time-varying function of the cross-track error. Follow-up troduced in this section. As presented in Fig. 1, a typical path-
work can be found in [37]. An integral LOS (ILOS) was proposed following control system consists of path generation (waypoints),
to compensate the drift forces due to environmental disturbance guidance law, controller and the marine craft. Here, several as-
[3,12,29]. LOS guidance law is a geometry method, because the de- sumptions are given. First, the drift angle β can be directly mea-
sired heading angle is determined with a triangle, which consists sured. As discussed in [29], the drift angle can be measured using
of the location of the ship, path, and look-ahead distance. a global navigation satellite system an inertial navigation system
Vector field guidance law is a mathematical method that was and inertial measurement unit. Second, the velocities of the ship
proposed by borrowing the concept of potential flow. Vector field can be measured correctly.
generates the vectors near the path, and the vectors indicate the This paper considers the straight path, where the path tangen-
desired course angle. When the ship follows the vectors, she will tial angle is constant. Fig. 2 shows the geometric information of
converge to the desired path. The core part of the vector field guid- two-dimensional path-following control for a marine craft. The red
ance law is the mathematical function that defines the vectors. It line is the predefined path. Two coordinate frames are defined in
was used for autonomous air vehicles [27,38]. Recently, the vec- this figure. The straight path, which connects the predefined way
tor field method has been used for path-following control of au- points (x j , y j ), j = 1 · · · N, is defined in the North-East-Down (NED)
tonomous ships [52,53], and shows a good performance. One ad- coordinate frame. The body-fixed frame is a moving coordinate
vantage of the vector field is its flexible structure. It calculates the frame that is fixed to the craft. The ship motions, such as surge
desired heading angle using a vector generation function, which and sway speed (u,v), are measured in the body-fixed frame. The
can be defined with the requirement of the tasks, for example, ar- path-tangential frame is a moving coordinate frame, whose origin
rival angle [32], trajectory-tracking [27], obstacles avoidance. is the projection of the ship’s centre of gravity (x p , y p ). γ p is the
The guidance and control system are a typical coupled nonlin- path tangential angle. The yaw angle and position are measured
ear system, and the stability properties need to be analysed. The relative to the NED frame.
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 359

Fig. 1. Marine control system for path following, where the path is defined by waypoints; β is the drift angle.

Fig. 2. Geometry description of two-dimensional path-following.

2.1. Path-following control objective 2.2. Kinematic-error equations of path-following

As presented in Fig. 2, the path connects the N way- Considering a ship moving in a horizontal plane, the kinematic
points(x j , y j )for j = 1, · · · , N. The cross-track errorye is the distance equations of a marine surface vessel moving in 3DOF (surge, sway,
between the ship and the predefined path. From Fig. 2, the equa- and yaw) can be described:
tions of cross-track error is given:
    x˙ = u cos (ψ ) − v sin (ψ )
0 x(t ) − x p (t ) y˙ = u sin (ψ ) + v cos (ψ ) (5)
= R (γ p ) (1)
ye y(t ) − y p (t ) ψ˙ = r
where, (x(t ), y(t )) is the ship’s location in real-time, (x p (t ), y p (t ) ) Differentiation of the Eq. (3), gives:
is the projection of the centre of the ship. γ p is the path tangential y˙ e = −x˙ sin (γ p (θ ) ) + y˙ cos (γ p (θ ) )
angle. R(γ p ) is the rotation matrix, and given as: = U sin (ψ − γ p (θ ) + β )
(6)
 
cos(γ p ) − sin(γ p )
R (γ p ) = ∈ SO(2 ) (2) It can be further simplified by substituting Eq. (5) into Eq. (6),
sin(γ p ) cos(γ p ) resulting:
Expanding (1), leads to the cross-track error ye : y˙ e = U sin (ψ − γ p + β ) (7)
ye (t ) = −(x(t ) − x p (t ) ) sin (γ p ) + (y(t ) − y p (t ) ) cos (γ p ) (3) 
where, U is the ground speed of a ship, (U = u2 + v2 ). The β
Obviously, the control object of the path-following is to make is the drift angle β = atan2(v, u). Finally, the time differential, ye ,
cross-track error converge to zero. It is given as: becomes:
lim ye (t ) = 0 (4) y˙ e = U sin (χ − γ p ) (8)
t→+∞
360 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 3. Vectors around the straight path.

where the χ = ψ + β is the course angle of the ship. Expanding the formula gives:
The principle of vector field method is to generate vectors to-   θ (t,ye ) 
  |ye |
wards the predefined path. The vectors denote the desired travel- y˙ e = U sinψ˜ cos tan−1
ling direction for the autonomous vessels, as presented in Fig. 3. 
  θ (t,ye ) 
A guidance law is used to generate the vector around the desired   |ye |
path (red line). The vectors indicate the reference heading (course) − sgn(ye )U cos ψ˜ sin tan−1 (12)

angle for the control system. Obviously, if the ship tracks the vec-
tors successfully, it will converge to the path finally. In this paper, It can be further simplified as:
vector field method is chosen as the guidance law. Here, a time-
varying vector field guidance law is proposed: U |ye |θ (t,ye )  
y˙ e = −sgn(ye )  + U φ t, ye , ψ˜ ψ˜ (13)
 θ (t,ye )
 2 + |ye | 2θ (t,ye )
|y |
χd = γ p − tan−1 sgn(ye ) e g(t,ye ,ψ˜ )
 f1 (t,ye )
 θ (t,ye ) 
|ye | where φ (t, ye , ψ˜ ) is defined as:
= γ p − sgn(ye )tan−1 (9)  
   sin ψ˜ 
φ t, ye , ψ˜ = 
where θ (t, ye ) is a time-varying function and ࢞ > 0 is a pre- χ˜  + |ye |2θ (t,ye )
2
defined constant and 0 < ࢞min < ࢞ < ࢞max . The function, θ (t, ye ),  
cos ψ˜ − 1 |ye |θ (t,ye )
is defined as: − sgn(ye )  (14)
χ˜ 2 + |ye |2θ (t,ye )
Definition 1. The function θ (t, ye ) is a non-decreasing pos-
itive semi-definite function and θ (t, ye = 0 ) ≥ 1. Furthermore,
3. L1 adaptive controller for heading
θ  (t, ye ) ≥ 0.
Remark. . The function, θ (t, ye ) plays an important role in the pro- In this section, the L1 adaptive controller is proposed for the
posed guidance law. It controls the convergence rate. In order to heading control of autonomous ships. The classical autopilot for
make sure the errors converge fast, the function θ (t, ye ) should in- ships is usually designed based on Nomoto model. It is a linearized
crease with the variable, ye . It means that the system has a higher manoeuvring model, and describes the yaw dynamic response to
convergence rate when ye is large. It can also be chosen as a con- the rudder steering. The oversimplified structure decreases the ac-
stant value. For example, if θ (t, ye ) = 1, then the vector field guid- curacy of the yaw motion’s prediction. So, in this paper, a non-
ance law degenerate to the classical LOS guidance law. linear Nomoto model with bounded environmental disturbance is
chosen:
The heading tracking error satisfies:
ψ˙ = r (15)
ψ˜ = ψ − ψd = χ − χd == χ˜ (10) T ∗ r˙ + H ∗ (r )r = K ∗ δ + b∗

With the Eqs. (9) - (10), the Eq. (7) can be simplified: where, T ∗ and K ∗ are the Nomoto constants. They can be esti-
  θ (t,ye )  mated using a system identification method resorting for instance
|ye | to zigzag manoeuvring test [43,44,56]. H ∗ (r ) = 1 + a2 r 2 , is the pos-
y˙ e = U sin ψ˜ − sgn(ye )tan−1 (11)
 itive nonlinear damping term. δ is the rudder angle. b∗ ≤ bmax
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 361

Fig. 4. Block diagram of the L1 adaptive controller.

represent the environmental disturbance. Assume there are uncer- of the parameter ση . Then, consider the positive defined Control
tainties associated with the parameters. The relationship between Lyapunov Function (CLF),
the real and estimated parameter is assumed as:  
1 1 1 1 ˜∗ T ˜∗ 1 2 1 ˜2
T = ηT
∗ Vpred = ρ˜η2 + σ˜ η2 + b η bη + T r˜ + ψ (22)
2 γρη γση γbη 2 2
H ∗ (r ) = σ H (r ) (16)
K∗ = ρK where, γρη ,γση and γbη are the adaptation gains. Differentiation of
(22) yields:
where, η ∈ R+ , σ ∈ R+ , ρ
∈ R+ are the uncertainties associated
with the T, H(r) and K, respectively. Additionally, it is assumed that 1 1 1 ˜ ∗ ˆ˙ ∗
V˙ pred = ρ˜η ρˆ˙ η + σ˜ η σˆ˙ η + bη b η + ψ˜ ψ˜˙ + r˜T r˜˙
the uncertainty and environmental disturbance, b∗ , is slowly vary- γρη γση γbη
ing relative to the ship dynamics, the derivative of the uncertainty ∗
1 1 ˙ 1
is negligible, (i.e. η˙ = 0, σ˙ = 0, ρ˙ = 0, and, b˙ ∗ = 0). Substituting (16) = ρ˜ ρˆ˙ + σ˜ σˆ˙ + b˜ ∗ bˆ − 1 ψ˜ 2 − T 2 r˜2
into (15) gives:
γρη η η γση η η γbη η η
 
ψ˙ = r + r˜ ρ˜η K δ + b˜ ∗η − σ˜ η H (r )r
(17)    
ηT r˙ + σ H (r )r = ρ K δ + b∗ 1 ˙ 1 ˙
= ρ˜η ρˆ η + r˜K δ + σ˜ η σˆ η − r˜H (r )r
γρη γση
3.1. State predictor with adaptation laws  
1 ˆ˙ ∗
+ b˜ ∗ b + r˜ − 1 ψ˜ 2 − T 2 r˜2 (23)
Fig. 4 shows the block diagram of L1 adaptive controller. It con- γbη η
sists mainly of a state predictor, parameters adaptation, and con-
trol law with low-pass filter. In the following part, a L1 adaptive From the Eq. (28), the following adaption laws are given:
backstepping controller is proposed, and the control law is given.
ρˆ˙ η = −γρη r˜K δ
The parameter adaption laws are derived first. The stability proof
for the closed-loop control system is also presented and UGAS is
σˆ˙ η = γση r˜H (r )r (24)
∗
˙
proved. The control law is derived using the backstepping method bˆ η = −γbη r˜
[10,28].
In order to prevent parameter drift in adaptation schemes, the
Given a state predictor, the prediction errors are:
projection-based adaptation laws are given:
ψ˜ = ψˆ − ψ ; r˜ = rˆ − r (18)
ρˆ˙ η = γρη Proj(ρη , −˜rK δ )
where, ψˆ and rˆ are the estimated value, respectively. The ideal pre- σˆ˙ η = γση Proj(ση , r˜H (r )r ) (25)
diction errors are defined as: ˙
∗
bˆ η = γbη Proj(b∗η , −˜r )
ψ˜˙ ideal = −1 ψ˜ ; r˜˙ ideal = −2 r˜ (19)
Since the structure of the control law and the definition of the
where, 1 and 2 are positive constants, which render their equi- reference model do not change, the tracking error dynamics will
librium point exponentially stable. Then the state prediction is de- be same. The advantage of using projection-type adaptation is that
fined as: one ensures boundedness of the adaptive parameters [21].
ψˆ˙ = −1 ψ˜ + r 
Substituting (24) into (23) gives:
(20)
rˆ˙ = −2 r˜ + T −1 ρˆη K δ + bˆ ∗η − σˆ η H (r )r V˙ pred = −1 ψ˜ 2 − T 2 r˜2 ≤ 0 (26)

where, ρˆη = ρˆ /ηˆ , σˆ η = σˆ /ηˆ , bˆ ∗η = bˆ ∗ /ηˆ , and the superscript repre- 3.2. Adaptive backstepping control law
sents the estimated value, and ηˆ is assumed positive, i.e. ηˆ ∈ R+ .
A candidate Lyapunov function is used to design the adaption laws In this part, the control law of the adaptive backstepping
for the uncertainties. The prediction errors are defined as: controller is derived using Lyapunov control function (CLF). For
the heading control of autonomous ships, the object is to make
ψ˜˙ = −1 ψ˜ (21) |ψ − ψd | → 0, where ψd is specified by the vector field guidance
r˜˙ = −2 r˜ + T −1 (ρ˜η K δ + b˜ ∗η − σ˜ η H (r )r ) law. Defining the error variables, z1 and z2 as:
where, ρ˜η = ρˆη − ρη is the error of the parameter ρη . b˜ ∗η = bˆ ∗η − z1 = ψ − ψd
∗ ∗
bη is the error of the parameter, bη . σ˜ η = σˆ η − ση is (27)
the error z2 = r − α
362 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 5. Closed-loop system of the L1 adaptive backstepping controller for an underactuated marine surface ship.

Fig. 6. Error dynamics equations of the guidance and control system.

where, α is a stabilising function. A positive defined CLF is defined The control law is given:
as:
ρˆη K δ = T α˙ + σˆ η H (r )α − bˆ ∗η − z1 + σˆ η H (r )z2 − K2 z2 (35)
1
V1 = z1 2 (28) where, K2 >0. Submitting Eq. (35) into Eq. (34), resulting:
2
The time differential of V1 is given: V˙ 2 = −K2 z2 2 − K1 z2 1 ≤ 0 (36)
    As discussed above, the parameters ρˆη , σˆ η and η are adaptive bˆ ∗
V˙ 1 = z1 z˙ 1 = z1 r − ψ˙ d = z1 z2 + z1 α − ψ˙ d (29)
updating using Eq. (24). The CLF considering the parameter uncer-
Choose the stabilising function, α = ψ˙ d − K1 z1 , and submit intro tainty can be expanded to:
(29), gives:
 
1 1 1 1 ˜∗ T ˜∗
Vctrl = Vpred + V2 = ρ˜η +
2
σ˜ η +
2
b η bη
V˙ 1 = −K1 z2 1 + z1 z2 (30) 2 γρη γση γbη
1 2 1 ˜2 1 2 1
The time differential z2 is given: + T r˜ + ψ + z1 + z2 T −1 z2 (37)
2 2 2 2
T z˙ 2 = T r˙ − T α˙ = ρˆη K δ + bη − σˆ η H (r )r − T α˙
ˆ∗ (31) The derivative of Vctrl becomes:
where, α˙ = ψ̈d − K1 z˙ 1 . Considering the new variables, z1 and z2 ,
 
V˙ ctrl = V˙ pred + V˙ 2 = −L1 ψ˜ 2 − T L2 r˜2 − K2 + σˆ η H (r ) z2 2 − K1 z2 1 ≤ 0
the CLF is modified to
(38)
1
V2 = z2 T −1 z2 + V1 (32) From the above discussion, the origin of the error system,
2
(ψ˜ , r˜, z1 , z2 , ρ˜η , σ˜ η , b˜ ∗η ), is uniformly globally asymptotically stable
The derivative of Vctrl is given:
(UGAS), [10]. The adaption of uncertainties of parameters may con-

tain high-frequency signals. A low pass filer is applied to control
V˙ 2 = z2 T −1
z˙ 2 + V˙ 1 = z2 ρˆη K δ + bη − σˆ η H (r )r − T α˙
ˆ∗
signals [21,47]. It is given as:
− z1 K1 z1 + z1 z2 ω2
 δctrl = C (s )δ = δ (39)
s2 + 2ζ ω s + ω 2
= z2 ρˆη K δ + bˆ ∗η − σˆ η H (r )r − T α˙ + z1 − z1 K1 z1 (33)
where, C (s ) is a second-order low-pass filter, ζ > 0 is the damp-
Submitting, r = z2 + α , into the Eq. (33), resulting, ing ratio. ω > 0 is the natural frequency. Fig. 5 shows the closed-
 loop system of the C (s ) is a second-order low-pass filter, ζ > 0 is
V˙ 2 = z2 ρˆη K δ + bˆ ∗η − σˆ η H (r )α − T α˙ + z1 − σˆ η H (r ) z2 − K1 z2 1 (34) the damping ratio. ω > 0 is the natural frequency. Fig. 5 shows
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 363

Fig. 7. The geometrical information of a straight path in the NED frame.

Fig. 8. The magnitude of 2D irrotational stochastic ocean current.

the closed-loop system of the L1 adaptive backstepping controller.  

the control law, parameter adaption laws and state estimator are 2 : ψ˜˙ = f2 t, ψ˜ (41)
presented in this figure. The underactuated nonlinear manoeuvring
model is given the following part. where f2 (t, ψ˜ ) is the heading tracking error dynamics discussed
in Section 3, corresponding to Eqs. (20)–(24), (35). From the above
4. Stability analysis of guidance and control system discussion, the error dynamic equation of the heading controller,
 2, is the driving system. The driving system output the heading
The L1 adaptive backstepping heading controller proposed in errors, which is feed-forward to the driven system,  1. The driven
the above can be used in cascade with the vector field for path- system is described by Eq. (40), and its performance is affected by
following control of underactuated ships. If the guidance law in the input, ψ˜ . In the following phase, the stability property of the
Eq. (9) is chosen, the cross-track error in Eq (13) forms a cascade proposed guidance and control system is proved by using the cas-
system, as described in Eqs. (40) and (41). Fig. 6 shows the error cade system theory.
dynamics of the cascaded system [4,40,41]. As can be observed, the
Property 1. For a typical marine surface ship, the ground speed is
errors of the heading control system are fed into the error dynamic
positive and up bounded,
of the guidance system. It means that the two sub-systems are
coupled together. The performance of the controller will directly 0 < Umin < U < Umax . (43)
affect the guidance system.
The error dynamics of the coupled guidance and control system Based on the above discussion, a theorem is presented and it
is given: guarantees that the cascade system Eqs. (40) and (41) of the vec-
tor field guidance law and L1 adaptive backstepping controller is
U |ye |θ (t,ye )   uniformly global asymptotically stable (UGAS).
1 : y˙ e = −sgn(ye )  + U φ t, ye , ψ˜ ψ˜ (40)
2θ (t,ye )
2 + |ye | Theorem 1. Assume the system  2 are stabilized using the adaptive
(
g t,ye ,ψ˜ ) control law in Eq. (35), where the parameters are dynamically up-
f1 (t,ye )
364 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 9. The trajectory of the ship model in the simulations.

dated using Eq. (25), the guidance law given by Eq. (9) defines the ref- the motion of the marine surface ship. This model was validated
erence heading angle, ψd . Then the equilibrium point (ye , ψ˜ ) = (0, 0 ) with sea-trials of full-scale ship test, is quite comprehensive, and
of the coupled guidance and control system (40) - (41) is UGAS, if the it gives highly realistic results [22,37,53,55].
function, θ (t, ye ) satisfies the Definition 1.
5.1. Nonlinear manoeuvring model
Proof. See Appendix A.
During the simulations, the current force will be considered as
5. Simulation the main environmental disturbance for manoeuvring modelling.
The effect of the current can be directly included in the mathemat-
Path-following simulations for an underactuated marine surface ical model. The relative forward velocity ur and transverse velocity
ship model will be carried out to evaluate the performance of the vr are given by:
proposed guidance and control system. The robustness of the L1
ur = u − uc cos(ψ − α )
adaptive backstepping controller with environmental disturbance (44)
is emphatically discussed. A 3-DOF (surge, sway and yaw) nonlin-
vr = v + uc sin(ψ − α )
ear mathematical model of the “Esso Osaka” is chosen to describe where, uc is the magnitude of the current, α is the direction of the
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 365

Fig. 10. The motions (yaw angle, surge speed, sway speed) measured in the simulations.

current. The time derivatives of u and v are given: where, f1 , f2 and f3 are the nondimensionalized hydrodynamic
forces and moment. They are defined as:

u˙ = u˙ r − uc r sin(ψ − α )
v˙ = v˙ r − uc r cos(ψ − α )
(45) ⎧     
⎪ f 1 = η 1 ur2 + η 2 n u r + η 3 n 2 − C  R + Xv2r vr2 + Xe2 e2
⎪
⎪     
⎪
⎪   
+ X r2 + m x G r 2 + X  vr r + m v r r  + X  v2r r2 vr2 r 2
⎪
⎪ 
⎪    
⎪ f  = Y  + Y  v + Y  (c − c )v + Y  − m u r− − Yδ (c − c )r
where u˙ r and v˙ r are defined by: ⎨ 2 0 vr r δ 0 r r r 2 0

⎪ +Yδ δ + Yr2 vr r 2 v r + Ye3 e3
⎪
⎪       
⎧ ⎪
⎪ f 3 = N0 + Nvr v r − Nδ (c − c0 )v r + Nr − m xG ur r+ 12 Nδ (c − c0 )r 
 
⎪ u˙  = f 1 ⎪
⎪
⎨  r m −X  u˙ r  ⎪
⎪  

+Nδ δ + Nr2 vr r 2 v r + Ne 3 e3
   
⎩
v˙ r = f1 (I z − N r˙ ) f  2 − (m xG − Yr˙  ) f  3 (46) f  4 = m − Yv˙r Iz − Nr˙ − m xG − Nv˙ r m xG − Yr˙
⎩r˙  = 1 (m − Y  ) f  − (m xG − N ) f  
⎪ 4

(47)
f 4 v˙ r 3 v˙ r 2
366 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 11. The measured course angle and drift angle.

The detailed description of this mathematical model can be Table 1

found in the [53,54]. The hydrodynamic coefficients in Eq. (47) are Nondimensional hydrodynamic coefficients of “Esso Os-
presented in Table 1. aka” ship model (scale factor = 100).

Coefficient Value Coefficient Value

5.2. Path-following study (m − Yv˙ ) 0.0352 Y0 1.90∗ e-6

(Iz − Nr˙ ) 0.00222 Yvr −0.0261
(m − Xu˙ r ) 0.0116 Yδ 0.00508
As discussed above, the control objective for path following is η1 −0.962∗ e-5 Yr 0.00365
to follow a predefined path. In this paper, straight-line is used to η2 −0.446∗ e-5 Yrrvr −0.0450
connect the waypoints. The waypoints are given: wpt1 = (40, 20), η3 0.0309∗ e-5 
Yeee −0.00185
wpt2 = (120, 25), wpt3 = (160, 18), wpt4 = (200, 22), wpt5 = (280, CR 0.00226 N0 −0.00028
Xv2 −0.006 Nv r −0.0105
5), wpt6 = (360, 20), and wpt7 = (440, 18), where the units are 
r
Xee −0.00224 Nδ −0.00283
meters. The geometrical information of the predefined paths is Xrr 0.00515 Nr −0.00480
presented in Fig. 7. There are 8 waypoints. Without loss of (Xvr r + m ) 0.0266 Nrrvr 0.00611
generality, they are chosen randomly in a Cartesian coordinate Xrrvv −0.00715 
Neee 0.00116
frame.
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 367

Fig. 12. The measured cross-track error and rudder angle.

The ship is assumed to start from the origin point. 2D irrota- The parameters of the L1 adaptive backstepping controller are
tional stochastic ocean current with constant direction (β c = 180̊) given based on experience. The positive constants, K1 and K2 , in
is considered in the simulation, as shown in Fig. 8. The stochas- the control law are given as, K1 = 0.02, K2 = 15. The parameters,
tic ocean current is divided into two phases. A weak ocean cur- 1 and 2 , in the state prediction, are given as: 1 = 100∗ IZ , 2
rent is assumed in the first phase, where the magnitude oscillates =10∗ IZ , where IZ is the inertia moment of the scaled ship model.
around 0.1 m/s. In the second phase, a stronger current is consid- The initial values of the uncertainties, ρˆη , σˆ η , bˆ ∗η , are ρˆη = 1, σˆ η = 1
ered, where the mean magnitude is 0.2 m/s. and bˆ ∗ = 0. The adaptation gains in Eq. (24) are γρ = 20, γσ =
η η η
A saturation is used for rudder dynamics, where the maximum 20, γbη = 10.
rudder angle is set as 35◦ . During the simulation, the ship’s speed As discussed above, the time-varying function, θ (t, ye ), plays
is kept constant, U0 = 0.41 m/s. The ship starts from the initial an important role in guidance law, and affects the overall perfor-
point, and the initial heading angle is set ψ 0 = 26◦ , which equals mance. Here, the time-varying function as chosen as: θ (t, ye ) =
to the path-tangent angle of predefined path. The initial yaw rate 0.05|ye | + 1. As can be observed, it increases with the variable ye ,
is zero, r0 = 0, and initial rudder angle is zero. The surge speed is and it is continuous, positive. The minimum of the function is 1. So
controlled using a classical PID controller. this time-varying function fulfils the Definition 1. The vector field
368 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 13. The predictions of the heading angle and yaw rate versus the real values in the simulations.

guidance law is given: damping with ζ =1. The following simulations will show the ef-
 0.05|ye |+1
 fect of the low-pass filter. If the low pass filter is turned off, the
(|ye | )
ψd = γ p − sgn(ye )tan−1 −β (48) proposed controller is a typical adaptive backstepping controller.
 Fig. 9 shows the predefined path and the trajectory of the ship
The predefined parameter is chosen as twice the ship length, model during the simulations. The low pass filler was used in
࢞ = 2Lpp [10]. The adaptation of the uncertainties usually con- Fig. 9(b), and it is turned off in Fig. 9(a). From the figure, the ship
tains high-frequency signals. These high-frequency signals will eas- can follow the straight-line path successfully in both cases. It can
ily cause frequently back-and-forth steering of the rudder. High- be concluded that the proposed vector field guidance law and L1
frequency steering is not desired and impossible to implement adaptive backstepping controller works properly.
in real ships, because, on the one hand, it will make more fuel Fig. 10, shows the performance of the L1 adaptive backstepping
consumption, and on another hand, the ship rudder is usually controller during the path-following simulations. Fig. 10(a) shows
steered using a hydraulic system, it cannot respond in time. Here, the results when the low pass filter was turned off. The desired
a second-order low pass filter is employed to eliminate the high- and true values of the heading angle are presented in Fig. 10. The
frequency signals. The parameters are ω = 0.25 rad/s and critical autopilot can track the desired heading angles. The surge velocities
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 369

Fig. 14. The evaluations of the parameters in the simulations.

in simulations are also presented in Fig. 10. The ship can keep a by the adaption of the uncertainty. In both cases, the cross-track
constant surge speed during the simulations. The sway speed in errors are nearly same. This is because the cross-track error dy-
both cases is also given in Fig. 10. As can be observed from the namics in Eq. (40) are directly affected by the guidance law and
figure, the fluctuation of sway speed is larger when the low pass heading error. In both cases, the ship can track the predefined
filter is turned off, as presented in Fig. 10(a). This results from path, but, the rudder angle is more stable when the low pass filter
the high-frequency rudder steering when the low pass filter was is turned on.
turned off. It can also be observed in Fig. 11, which shows the Fig. 13 shows the state predictor of the yaw angle and yaw
course angle (desired versus true) and drift angle for both cases. In rate in the simulations. The estimations of the yaw angle (ψˆ ) and
Fig. 11(a), the drift angle has a large oscillation without the low- yaw rate (rˆ) agrees well with the real values. The evaluations of
pass filter. The sway speed is related to the drift angle. Usually, the the parameters, ρˆη , σˆ η , bˆ ∗η , are presented in Fig. 14. When the low
large fluctuation of sway speed or drift angle indicated that the pass filter was turned off, as presented in Fig. 13(a), the param-
ship was steering a lot to stabilize the system. This high-frequency eters ρˆη , σˆ η , bˆ ∗η contains high-frequency signals, especially the bˆ ∗η ,
steering output is due to the adaptation of the uncertainties. which represented the environmental disturbance and unmodeled
Fig. 12 shows the cross-track errors and rudder angles in the dynamics.
simulations. Obviously, when the low pass filter is turned off, the In order to show its effective performance of the L1 adap-
rudder angle contains the high-frequency signals, which induced tive backstepping controller, the classical PID controller was used
370 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

Fig. 15. The cross-track error (above) and rudder angles (below) compared with the classical PID controller for path-following.

in the previous simulations. The predefined parameters and way- water, confined waterway etc. The proposed system can also be
points are same with the previous simulations. The parameters for used for other autonomous vehicles.
the PID controller are chosen as: K p = ωn2 T /K; Kd = (2ζ ωn T − 1 )/K;
Ki = ωn3 T /(10K ) [10], where ωn is the natural frequency and ζ is Declaration of Competing Interest
the relative damping ratio of the first order system. In this case,
ωn = 1 rad/s and critical damping with ζ =1. The cross-track error There are no conflicts of interest.
and rudder angles are plotted in Fig. 15. As can be observed, the
PID controller takes more time to converge. The L1 adaptive back-
Acknowledgement
stepping controller response fast and have a better performance.

This work contributes to M&MSHIPS -“manoeuvring & Moored

SHIPS in ports” (PTDC/EMSTRA/5628/2014), funded by the Por-
6. Conclusions tuguese Foundation for Science and Technology. This work also
contributes to the Strategic Research Plan of the Centre for Ma-
In this paper, L1 adaptive backstepping controller was proposed rine Technology and Ocean Engineering (CENTEC), which is fi-
for path-following control of autonomous ships. L1 adaptive back- nanced by the Portuguese Foundation for Science and Technol-
stepping was designed based on nonlinear steering model. The pa- ogy (Fundação para a Ciência e Tecnologia - FCT) under contract
rameter update laws and control law were derived using Lyapunov UIDB/UIDP/00134/2020.
functions. The stability proof shows that the equilibrium point
of the error system, (ψ˜ , r˜, z1 , z2 , ρ˜η , σ˜ η , b˜ ∗η ), is uniformly globally
Appendix A. Proof of Theorem 1
asymptotically stable. A novel time-varying vector field guidance
law was proposed in this paper. It formed a cascade system with
the control system. The whole guidance and control system were Proof. . As presented in Eq. (14), it can be observed, the sub-
proved to be uniformly global asymptotically stable using cascade function, | sinx(x ) | ≤ 1, and | cos(xx )−1 | ≤ 0.73, are upper bounded,
system theory. A theorem and proof were given to guarantee sta- and,
bility performance.    
   
The effectiveness of the proposed control law and vector field  |ye |θ (t,ye )    
  ≤ 1; and   
guidance law was assessed with a fully nonlinear manoeuvring  2 + |ye |2θ (t,ye )   2 + |ye |2θ (t,ye ) 
model of an underactuated tanker in the presence of stochastic
ocean current. This manoeuvring model has been validated with ≤ 1, where, 0 <min <  < max (A1)
experiments and it gave confidence to the results. The results
obtained indicated that the proposed heading controller and Then, the function, φ (t, ye , ψ˜ ), is upper bounded, φ (t, ye , ψ˜ ) ≤
guidance law can control the underactuated ship to follow the c for all ye andψ˜ .
   
path, while rejecting the environmental disturbance. The low pass    
 |ye |θ (t,ye )    
filter can eliminate the unnecessary high-frequency signals, which   ≤ 1; and   ≤1 (A2)
were caused by the adaptation of the uncertainties, in the control  2 + |ye |2θ (t,ye )   2 + |ye |2θ (t,ye ) 
input. This paper investigates the important system, guidance and
control, of autonomous ships. The proposed time-varying guidance Then, the function,φ (t, ye , ψ˜ ), is upper bounded [12].
law and the adaptive controller can improve the performance of As discussed in Section 3, the equilibrium point of the heading
the autonomous ship in a complex environment, such as shallow autopilot system  2 given by Eq. (40) is UGAS. The nominal system
 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372 371

of the driven system  1 ( 1 system with ψ˜ = 0) is: [19] B.J. Guerreiro, C. Silvestre, R. Cunha, C. Cao, N. Hovakimyan, L1adaptive con-
trol for autonomous rotorcraft, in: Proceedings of the American Control Con-
U |ye |θ (t,ye ) ference, 2009, doi:10.1109/ACC.2009.5159940.
y˙ e = −sgn(ye )  (A3) [20] J. Guerrero, J. Torres, V. Creuze, A. Chemori, Trajectory tracking for autonomous
2 + |ye |2θ (t,ye ) underwater vehicle: an adaptive approach, Ocean Eng 172 (2019) 511–522,
doi:10.1016/J.OCEANENG.2018.12.027.
The function θ (t, ye ) is time-varying, so the nominal system in [21] N. Hovakimyan, C. Cao, L1 adaptive control theory : guaranteed robustness
with fast adaptation, Soc. Ind. Appl. Math. (2010).
Eq. (A3) is nonautonomous. Consider the CLF,
[22] W. Hwang, Application of System Identification to Ship Maneuvering, Institute
1 2 of Technology, Massachusetts, 1980.
V1 (t, ye ) = y (A4) [23] I. Kaminer, A. Pascoal, E. Xargay, N. Hovakimyan, C. Cao, V. Dobrokhodov, Path
2 e following for small unmanned aerial vehicles using L1 adaptive augmentation
where V1 (t, ye ) > 0 if ye = 0. The time derivative of V1 (t, ye ) is of commercial autopilots, J. Guid. Control. Dyn. (2010a), doi:10.2514/1.42056.
[24] I. Kaminer, A. Pascoal, E. Xargay, N. Hovakimyan, C. Cao, V. Dobrokhodov, Path
given: following for unmanned aerial vehicles using L1 adaptive augmentation of
commercial autopilots, J. Guid. Control. Dyn. (2010b), doi:10.2514/1.42056.
U |ye |θ (t,ye ) U |ye |θ (t,ye )+1 [25] H.K. Khalil, Nonlinear Systems, Prentice Hall, 2002.
V˙ 1 (t, ye ) = −sgn(ye )ye  = − [26] E. Lataire, M. Vantorre, G. Delefortrie, M. Candries, Mathematical modelling of
2 + |ye |2θ (t,ye ) 2 + |ye |2θ (t,ye ) forces acting on ships during lightering operations, Ocean Eng. 55 (2012) 101–
115, doi:10.1016/J.OCEANENG.2012.07.029.
≤0 (A5) [27] D.A. Lawrence, E.W. Frew, W.J. Pisano, Lyapunov vector fields for autonomous
unmanned aircraft flight control, J. Guid. Control. Dyn. 31 (2008) 1220–1229,
since V1 (t, ye ) > 0 and V˙ 1 (t, ye ) ≤ 0, according to the Theorem 4.9 doi:10.2514/1.34896.
by Khalil [25], the equilibrium point, ye = 0, is globally uniformly [28] C. Lee, M. Tahk, B. Jun, Autopilot design for an agile missile using L1 adaptive
asymptotically stable (UGAS). Moreover, every trajectory starting in backstepping control, in: Proceedings of the 28th Congress of the International
Council of the Aeronautical Sciences, Brisbane, Australia, 2012.
{x ∈ Br } and Br = { x ≤ r}, satisfies [29] A.M. Lekkas, T.I. Fossen, Integral LOS path following for curved paths based
on a monotone cubic hermite spline parametrization, IEEE Trans. Control Syst.
|ye (t )| ≤ β ( ye (t0 ) ,t − t0 )∀t ≥ t0 ≥ 0 (A6) Technol. 22 (2014) 2287–2301, doi:10.1109/TCST.2014.2306774.
[30] A.M. Lekkas, T.I. Fossen, A time-varying lookahead distance guidance law
Then, the cross tracked error is globally bounded and it will for path following, in: IFAC Proceedings Volumes (IFAC-PapersOnline), 2012,
globally asymptotically converge to zero. Finally, the equilibrium pp. 398–403, doi:10.3182/20120919- 3- IT- 2046.0 0 068.
point (ye = 0, ψ˜ = 0) of the cascaded system described by (39)-(40) [31] M. Li, T. Li, X. Gao, Q. Shan, C.L.P. Chen, Y. Xiao, Adaptive NN event-triggered
control for path following of underactuated vessels with finite-time con-
is UGAS [6,12,13,34]. vergence, Neurocomputing 379 (2020) 203–213, doi:10.1016/j.neucom.2019.10.
044.
References [32] S. Lim, W. Jung, H. Bang, Vector field guidance for path following and ar-
rival angle control, in: Proceedings of the International Conference on Un-
[1] Allianz Global Corporate and SpecialitySafety and Shipping 1912–2012, Cardiff manned Aircraft Systems, ICUAS, IEEE, 2014, pp. 329–338, doi:10.1109/ICUAS.
University, 2012. 2014.6842271.
[2] P. Batista, C. Silvestre, P. Oliveira, Globally exponentially stable attitude ob- [33] C. Liu, J. Sun, Z. Zou, Integrated line of sight and model predictive control for
server with Earth velocity estimation, Asian J. Control (2019) asjc.2056, doi:10. path following and roll motion control using rudder, J. Sh. Res. 59 (2015) 99–
1002/asjc.2056. 112, doi:10.5957/JOSR.59.2.140057.
[3] W. Caharija, K.Y. Pettersen, P. Calado, J. Braga, A comparison between the [34] A. Loría, E. Panteley, Cascaded nonlinear time-varying systems: analysis and
ILOS guidance and the vector field guidance, in: Proceedings of the IFAC- design, in: F. Lamnabhi-Lagarrigue, A. Loría, E. Panteley (Eds.), Advanced Topics
PapersOnLine, Elsevier, 2015, pp. 89–94, doi:10.1016/j.ifacol.2015.10.263. in Control Systems Theory: Lecture Notes from FAP 2004, Springer London,
[4] A. Chaillet, A. Loría, Uniform global practical asymptotic stability for time- London, 2005, pp. 23–64, doi:10.1007/11334774_2.
varying cascaded systems, Eur. J. Control 12 (2006) 595–605, doi:10.3166/ejc. [35] K. Maritime (2016). World’s first official test bed for autonomous ship-
12.595-605. ping opens in Norway [WWW Document]. URL https://www.km.kongsberg.
[5] K. Do, J. Pan, Control of Ships and Underwater Vehicles, Springer, Australia, com/ks/web/nokbg0238.nsf/AllWeb/166309633F206651C125804200250951?
2009, doi:10.1007/978- 1- 84882- 730- 1. OpenDocument (accessed 9.6.18).
[6] K.D. Do, Global robust adaptive path-tracking control of underactuated ships [36] S. Moe, K.Y. Pettersen, Set-based Line-of-Sight (LOS) path following with col-
under stochastic disturbances, Ocean Eng. 111 (2016) 267–278, doi:10.1016/j. lision avoidance for underactuated unmanned surface vessel, in: Proceedings
oceaneng.2015.10.038. of the 24th Mediterranean Conference on Control and Automation, MED 2016,
[7] K.D. Do, Z.P. Jiang, J. Pan, Robust adaptive path following of underactuated 2016, pp. 402–409, doi:10.1109/MED.2016.7535964.
ships, Automatica (2004), doi:10.1016/j.automatica.2004.01.021. [37] L. Moreira, T.I. Fossen, C. Guedes Soares, Path following control system for a
[8] K.D. Do, J. Pan, Global robust adaptive path following of underactuated ships, tanker ship model, Ocean Eng. 34 (2007) 2074–2085, doi:10.1016/j.oceaneng.
Automatica (2006a), doi:10.1016/j.automatica.2006.04.026. 20 07.02.0 05.
[9] K.D. Do, J. Pan, Underactuated ships follow smooth paths with integral actions [38] D.R. Nelson, D.B. Barber, T.W. McLain, R.W. Beard, Vector field path following
and without velocity measurements for feedback: theory and experiments, for small unmanned air vehicles, in: Proceedings of the American Control Con-
IEEE Trans. Control Syst. Technol. (2006b), doi:10.1109/TCST.2005.863665. ference, 2006, pp. 5788–5794, doi:10.1109/ACC.2006.1657648.
[10] T.I. Fossen, Handbook of Marine Craft Hydrodynamics and Motion Control, [39] S.-R. Oh, J. Sun, Path following of underactuated marine surface vessels using
John Wiley & Sons, Ltd, Chichester, UK, 2011, doi:10.1002/9781119994138. line-of-sight based model predictive control, Ocean Eng. 37 (2010) 289–295,
[11] T.I. Fossen, A survey on Nonlinear Ship Control: from Theory to Practice, IFAC doi:10.1016/j.oceaneng.2009.10.004.
Proc. 33 (20 0 0) 1–16, doi:10.1016/S1474-6670(17)37044-1. [40] E. Panteley, A. Loria, On global uniform asymptotic stability of nonlinear time-
[12] T.I. Fossen, A.M. Lekkas, Direct and indirect adaptive integral line-of-sight varying systems in cascade, Syst. Control Lett. 33 (1998) 131–138, doi:10.1016/
path-following controllers for marine craft exposed to ocean currents, Int. J. S0167-6911(97)00119-9.
Adapt. Control Signal Process. 31 (2015) 445–463, doi:10.1002/acs.2550. [41] E. Panteley, A. Loría, A. Sokolov, Global uniform asymptotic stability of cas-
[13] T.I. Fossen, K.Y. Pettersen, On uniform semiglobal exponential stability (USGES) caded non-autonomous non-linear systems: application to stabilisation of a
of proportional line-of-sight guidance laws, Automatica 50 (2014) 2912–2917, diesel engine, Eur. J. Control 5 (1999) 107–115, doi:10.1016/S0947-3580(99)
doi:10.1016/J.AUTOMATICA.2014.10.018. 70145-7.
[14] T.I. Fossen, K.Y. Pettersen, R. Galeazzi, Line-of-sight path following for dubins [42] L.P. Perera, C. Guedes Soares, Pre-filtered sliding mode control for nonlinear
paths with adaptive sideslip compensation of drift forces, IEEE Trans. Control ship steering associated with disturbances, Ocean Eng. 51 (2012) 49–62.
Syst. Technol. 23 (2015) 820–827, doi:10.1109/TCST.2014.2338354. [43] L.P. Perera, P. Oliveira, C. Guedes Soares, System identification of nonlinear
[15] E. Fredriksen, K.Y. Pettersen, Global κ -exponential way-point maneuvering of vessel steering, J. Offshore Mech. Arct. Eng. 137 (2015) 031302, doi:10.1115/
ships: theory and experiments, Automatica 42 (2006) 677–687, doi:10.1016/J. 1.4029826.
AUTOMATICA.2005.12.020. [44] L.P. Perera, P. Oliveira, C. Guedes Soares, System identification of vessel steer-
[16] I.M. Gregory, C. Cao, E. Xargay, N. Hovakimyan, X. Zou, L1 adaptive control de- ing with unstructured uncertainties by persistent excitation maneuvers, IEEE J.
sign for NASA AirSTAR flight test vehicle, in: Proceedings of the AIAA Guid- Ocean. Eng. 41 (3) (2016) 515–528.
ance, Navigation, and Control Conference and Exhibit, 2009, doi:10.2514/6. [45] [45]Rolls-Royce, (2015). Rolls-Royce to Lead Autonomous
2009-5738. Ship Research Project [WWW Document]. URL https://
[17] C. Guedes Soares, A.P. Teixeira, Risk assessment in maritime transportation, Re- www.rolls-royce.com/media/our-stories/press-releases/2015/
liab. Eng. Syst. Saf. 74 (3) (2001) 299–309. pr- 02- 07- 15- rolls- royce- to- lead- autonomous- ship- research- project.aspx
[18] C. Guedes Soares, A.P. Teixeira, P. Antao, Accounting for human factors in the (accessed 9.6.18).
analysis of maritime accidents, in: M. Cottam, D. Harvey, R. Pape, J. Tait (Eds.), [46] T.A. Santos, C. Guedes Soares, Economic feasibility of an autonomous con-
Foresight and Precaution, Rotterdam, Balkema, 20 0 0, pp. 521–528. tainer ship, in: C. Guedes Soares, A.P. Teixeira (Eds.), Maritime Transporta-
372 H. Xu, P. Oliveira and C. Guedes Soares / European Journal of Control 58 (2021) 357–372

tion and Harvesting of Sea Resources, Taylor & Francis, London, 2018, pp. [52] H. Xu, T.I. Fossen, C. Guedes Soares, Uniformly semiglobally exponential stabil-
861–870. ity of vector field guidance law and autopilot for path-following, Eur. J. Control
[47] M.E.N. Sørensen, M. Breivik, Comparing nonlinear adaptive motion controllers 53 (2019a) 88–97, doi:10.1016/j.ejcon.2019.09.007.
for marine surface vessels, in: Proceedings of the IFAC-PapersOnLine, Elsevier, [53] H. Xu, C. Guedes Soares, Vector field path following for surface marine vessel
2015, pp. 291–298, doi:10.1016/j.ifacol.2015.10.295. and parameter identification based on LS-SVM, Ocean Eng 113 (2016) 151–161,
[48] M. Vantorre, Manoeuvring coefficients for a container carrier in shallow water: doi:10.1016/j.oceaneng.2015.12.037.
an evaluation of semi-empirical formulae, in: Proceedings of the Mini Sympo- [54] H. Xu, V. Hassani, C. Guedes Soares, Uncertainty analysis of the hydrody-
sium on Prediction of Ship Manoeuvring Performance, Tokyo, Japan, 2001 18 namic coefficients estimation of a nonlinear manoeuvring model based on pla-
October 2001. nar motion mechanism tests, Ocean Eng. 173 (2019b) 450–459, doi:10.1016/j.
[49] N. Wang, Z. Sun, J. Yin, Z. Zheng, Surge-varying LOS based path following oceaneng.2018.12.075.
control of underactuated marine vehicles with accurate disturbance observa- [55] H. Xu, M.A. Hinostroza, C. Guedes Soares, Estimation of hydrodynamic coeffi-
tion, in: Proceedings of the IEEE 7th International Conference on Underwa- cients of a nonlinear manoeuvring mathematical model with free-running ship
ter System Technology: Theory and Applications (USYS), IEEE, 2017, pp. 1–6, model tests, Int. J. Marit. Eng. 160 (2018) A-213-A-226, doi:10.3940/rina.ijme.
doi:10.1109/USYS.2017.8309464. a3.2018.448.
[50] N. Wang, Z. Sun, J. Yin, Z. Zou, S.F. Su, Fuzzy unknown observer-based robust [56] H. Xu, M.A. Hinostroza, V. Hassani, C. Guedes Soares, Real-time parameter es-
adaptive path following control of underactuated surface vehicles subject to timation of a nonlinear vessel steering model using a support vector machine,
multiple unknowns, Ocean Eng. (2019) 57–64, doi:10.1016/j.oceaneng.2019.02. J. Offshore Mech. Arct. Eng. 141 (2019c) 061606, doi:10.1115/1.4043806.
017. [57] H. Xu, H. Rong, C. Guedes Soares, Use of AIS data for guidance and control
[51] X. Xiang, C. Yu, L. Lapierre, J. Zhang, Q. Zhang, Survey on fuzzy-logic-based of path-following autonomous vessels, Ocean Eng. 194 (2019d) 106635, doi:10.
guidance and control of marine surface vehicles and underwater vehicles, Int. 1016/j.oceaneng.2019.106635.
J. Fuzzy Syst. 20 (2018) 572–586, doi:10.1007/s40815- 017- 0401- 3.