This action might not be possible to undo. Are you sure you want to continue?
analysis of an underactuated underwater vehicle
Cheng Siong Chin,Micheal Wai Shing Lau,
Eicher Low and Gerald Gim Lee Seet
Robotic Research Centre,
Department of Mechanical and Aerospace Engineering,
Nanyang Technological University,
Singapore 639798.
The problem of designing a stabilizing feedback controller for underactuated sys
tem is a challenging one since a nonlinear system is not stabilizable by a smooth
static state feedback law. A necessary condition for asymptotically stabilization of an
underactuated vehicle to a single equilibrium is that his gravitational ﬁeld has non
zero elements corresponding to unactuated dynamics. However, global asymptotical
stability (GAS) cannot be guaranteed. In this paper, a robust proportionalintegral
derivative (PID) controller on the actuated dynamic is proposed and unactuated
dynamic is shown to be global exponentially bounded by Sordalen lemma. This gives
a necessary and suﬃcient condition to guarantee global asymptotic stability (GAS)
of the URV system. The proposed method is ﬁrst adopted on a remotelyoperated
vehicle RRC ROV II designed by Robotic Research Centre in Nanyang Technological
University (NTU). Through the simulation using ROV Design and Analysis toolbox
(RDA) written by NTU in MATLAB/SIMULINK environment, the ROV RRC II is
robust against parameter perturbation.
1
Keywords: underwater vehicle, underactuated, stabilizable,robust controller, sim
ulation
1 Introduction
In this paper, a nonlinear system consisting of actuated and unactuated dynamics are
studied. The problem of designing a stabilizing feedback controller for underactuated
systems is a challenging one since the system is not stabilizable by a smooth static
state feedback law [Brockett,1993]. [Fossen, 1994, Yuh, 1990] showed that a fully ac
tuated vehicle (a vehicle where the control and conﬁguration vector have the same
dimension) can be asymptotically stabilized in position and velocity by a smooth
feedback law. [Byrnes et al., 1991] explained the underactuated vehicles having zero
gravitational ﬁeld are not asymptotically stabilizable to a single equilibrium. On the
other hand, [Wichlund et al.,1995] stated the vehicle with gravitational and restor
ing terms in the unactuated dynamic is stabilizable to a single equilibrium point.
However, it is a necessary but not suﬃcient condition to state that the vehicle is
asymptotically stabilizable. The closedloop asymptotical stability of the vehicle in
the earthﬁxed frame needs to be further examined.
A diﬀerent method than those proposed in [Wichlund et al.,1995] is used to show
the actuated dynamic in earthﬁxed frame are exponentially decaying under the non
linear controller. The method described in [Wichlund et al.,1995] uses a nonlinear
dynamic control law to achieve a neat closed loop actuated subsystem that yields
exponential decaying solution, but gives low ﬂexibility in designing the control law
since the nonlinear vehicle dynamic has to be known exactly for nonlinear dynamic
cancellation.
In this paper, a robust ProportionalIntegralDerivative (PID) controller is cho
sen due to its simplicity in implementation and it is commonly used in the indus
trial. The PID is designed only for the actuated dynamic, such that it provides a
necessary and also suﬃcient condition for asymptotic stability of the unactuated
dynamics. In this method, the unactuated dynamic is selfstabilizable and con
2
verged exponentially to zero. With the controller, the actuated dynamics become
asymptotical stable while the actuated states in unactuated dynamic equation dimin
ishes. Applying the asymptotic stability lemma by Sordalen [Sordalen et al., 1993a,
Sordalen et al., 1993b, Sordalen et al., 1995] to the unactuated dynamic equation pro
vides a validation for the initial argument of convergence to zero. Thus the vehicle in
the earthﬁxed frame, which is selfstabilzable in the unactuated dynamics, is asymp
totically stable.
The paper is organized as follows: A nonlinear model of underactuated ROV, de
veloped by Robotic Research Centre in Nanyang Technological University [Micheal et al., 2003,
Koh, et al., 2002b], is presented in Section 2. Section 3 describes a necessary condi
tion for vehicle with gravitational and restoring terms in the unactuated dynamic to
be stabilizable. In Section 4 and 5, a robust PID controller is proposed for asymptotic
stability of the vehicle in the earthﬁxed frame which is selfstabilzable in the unac
tuated dynamic. Results of computer simulations using ROV Design and Analysis
(RDA) toolbox written in MATLAB/SIMULINK environment by NTU are presented
in Section 6.
2 Nonlinear model of an underactuated ROV
The dynamic behavior of an underwater vehicle is designed through Newton’s laws of
linear and angular momentum. The equations of motion of such vehicles are highly
nonlinear[Fossen, 1994] and coupled due to hydrodynamic forces, which are acting
on the vehicle. Usually, the ROV model can be described in either bodyﬁxed or
earthﬁxed frame.
3
2.1 Bodyﬁxed model of an underactuated ROV
It is convenient to write the general dynamic (1) and kinematic (2) equations for ROV
in the bodyﬁxed frame.
M
v
˙ v + C
v
(v)v + D
v
(v)v + g(η) = B
v
u
v
(1)
˙ η = J(η)v (2)
where B
v
∈ '
6×4
is a thruster conﬁguration matrix(deﬁned by the thruster layout as
shown in Fig. 1), u
v
∈ '
4
is an input vector,v = [u, v, w, p, q, r]
T
∈ '
6
is a velocity
vector, η = [x, y, z, φ, θ, ψ]
T
∈ '
3
S
3
is a position and orientation vector, M
v
∈ '
6×6
is a mass inertia matrix with added mass coeﬃcients, C
v
(v) ∈ '
6×6
is a centripetal
and coriolis matrix with added mass coeﬃcients, D
v
(v) ∈ '
6×6
is a diagonal hydro
dynamic damping matrix and g(η) ∈ '
6
is a buoyancy and gravitational forces and
moments vector. The ROV path relative to the earthﬁxed reference frame is given
by kinematic equation (2), where J(η) = J(η
2
) ∈ '
6×6
, η
2
= [φ, θ, ψ]
T
is an Euler
transformation matrix.
2.2 Earthﬁxed model of an underactuated ROV
Sometimes, we need to express the ROV model from body coordinate to earthﬁxed
coordinates [Fossen, 1994] by performing the coordinate transformation (η, v)
µ
→(η, ˙ η)
deﬁned by:
_
¸
_
η
˙ η
_
¸
_ =
_
¸
_
I 0
0 J(η)
_
¸
_
_
¸
_
η
v
_
¸
_ (3)
where the transformation matrix, J has the following form
J(η) =
_
¸
_
c(ψ)c(θ) −s(ψ)c(φ)+c(ψ)s(θ)s(φ) s(ψ)s(φ)+c(ψ)c(φ)s(θ)
s(ψ)c(θ) c(ψ)c(θ)+s(φ)s(θ)s(ψ) −c(ψ)s(φ)+s(θ)s(ψ)c(φ)
−s(θ) c(θ)s(φ) c(θ)c(φ)
0 0 0
0 0 0
0 0 0
0 0 0
0 0 0
0 0 0
1 s(φ)t(θ) c(φ)t(θ)
0 c(φ) −s(φ)
0 s(φ)/c(θ) c(φ)/c(θ)
_
¸
_ (4)
4
where c(.) = cos(.), t(.) = tan(.) and s(.) = sin(.). The coordinate transformation µ, a
nonsingular matrix is a global diﬀeomorphism function which is analogues to similarity
transformation in linear system. This transformation is undeﬁned for θ = ±90
o
. To
overcome this singularity, a quaternion approach must be considered. However, in
the study this problem does not exist because the vehicle is not required to operate
at θ = ±90
o
. Moreover, the vehicle is completely stable in roll and pitch, and the
thruster actuation is not enough to move the vehicle to operate at this angle. The
ROV model in earth ﬁxedcoordinates becomes:
M
η
( ˙ η, η)¨ η + C
η
( ˙ η, η) ˙ η + D
η
( ˙ η, η) ˙ η + g
η
(η) = B
η
u
η
(5)
where M
η
( ˙ η, η) = J
−T
M
v
J
−1
, C
η
( ˙ η, η) = J
−T
(C −M
v
J
−1
˙
J)J
−1
, D
η
( ˙ η, η) = J
−T
D
v
J
−1
, g
η
(η) = J
−T
g
v
and B
η
u
η
= J
−T
B
v
u
v
.
3 Stabilizability
A method to test system stabilizability is presented in this section. In general, vector
g
(
η) can be further decomposed into the elements corresponding to the actuated
dynamics (ﬁrst to third and sixth element), g
a
(η), and the element corresponding to
the unactuated dynamic (fourth and ﬁfth element), g
u
(η). The proof of Theorem 3
of [Wichlund et al.,1995] considered the system given in equations (1) and (2), and
suppose that (η, v) = (0, 0) is an equilibrium point of the system. If g
u
(η) is zero then
there exists no continuous and discontinuous state feedback law [Byrnes et al., 1991],
k(η, v) : '
6
⇒'
4
, that makes (0, 0) asymptotically stable equilibrium.
However, RRC ROV II (2) has a gravitational ﬁeld at the unactuated dynamic,
g
u
(η) = [29.61 cos θ sinφ 29.61 sinθ]
T
,= 0 but no gravitational ﬁeld at g
a
(η) = 0.
Therefore, the RRC ROV II may be stabilizable at the equilibrium point. However,
it is a necessary condition but not suﬃcient condition to said that the ROV is asymp
totical stabilizable at the equilibrium point. Inevitably, it gives rise to the need of
ﬁnding a control law to stabilize the ROV at the equilibrium point.
5
4 Asymptotic stabilization of an actuated dynamic
by smooth state feedback
In sections 4 and 5, the concepts in the asymptotic stabilization of an actuated
dynamic are as follows: (i)by observation, the unactuated dynamic in (7) is self
stabilizable and exponentially decaying. (ii)use a robust PID controller to globally
asymptotically stabilize the actuated dynamic in(6), (iii)with actuated dynamic, f
1
is globally asymptotically stable(GAS) implies that the actuated dynamic, h
2
in the
unactuated equation (7)becomes zero, (iv) ﬁnally, step (i) is veriﬁed by the Sordalen
lemma. For clarity, this section is divided into two parts, Problem deﬁnition is given in
Part I, perturbation on ROV’s parameters and the controller design and the stability
analysis on the actuated dynamic are provided in Part II.
4.1 Problem Deﬁnition
Separation of the entire system into actuated and unactuated subsystems, as described
in [Micheal et al., 2003], yields
¨ η
a
= f
1
( ˙ η
a
, η
a
, t) + h
1
( ˙ η
u
, η
u
, t) + B
η
a
u
a
(6)
¨ η
u
= f
2
( ˙ η
u
, η
u
, t) + h
2
( ˙ η
a
, η
a
, t) +B
η
u
u
u
(7)
where
f
1
( ˙ η
a
, η
a
, t) = −
1
det(M
η
)
M
η22
[C
η11
(v, η) + D
η11
(v, η)] ˙ η
a
;
h
1
( ˙ η
u
, η
u
, t) = −
1
det(M
η
)
M
η12
g
ηu
+M
η12
C
η21
(v, η) ˙ η
u
;
f
2
( ˙ η
u
, η
u
, t) = −
1
det(M
η
)
M
η11
[C
η22
(v, η) + D
η22
(v, η)] ˙ η
u
;
h
2
( ˙ η
a
, η
a
, t) = −
1
det(M
η
)
M
η11
g
ηu
−M
η12
C
η12
(v, η) ˙ η
a
;
det(M
η
) = M
η22
M
η11
−M
η12
2
6
η = [z x y ψ [ φ θ]
T
= [η
a
[ η
u
]; η
a
∈ '
3
S, η
u
∈ '
2
, B
η
= [B
η
a
B
η
u
] are the input
matrices for the actuated and unactuated dynamic in (5). Note that subscript ’a’ and
’u’ refer to actuated and unactuated dynamic respectively.
Let η
d
denotes desired set points (position and orientation) in the earthﬁxed
frame. Error of this actuated position about the hovering or stationkeeping condition
can be designated as:
e = η
a
−η
d
⇒ ˙ e = ˙ η
a
, ¨ e = ¨ η
a
(8)
Substitute the preceding equation into (6) and (7) yields,
¨ e = f
1
( ˙ e, e, t) + h
1
( ˙ η
u
, η
u
, t) + B
ηa
u
a
(9)
¨ η
u
= f
2
( ˙ η
u
, η
u
, t) + h
2
( ˙ e, e, t) + B
ηu
u
u
(10)
Considering h
1
( ˙ η
u
, η
u
, t) as a perturbation on (9) and assume that it could converge
(or exponentially decay) to zero as time increases. The (9) becomes:
¨ e = f
1
( ˙ e, e, t) + B
ηa
u
a
(11)
The h
1
( ˙ η
u
, η
u
, t) diminishes from (9) as the restoring forces (based on ROV design
intention) in, η
u
= ¦φ, θ¦ directions enable these two motions to stabilize themselves
eﬀectively instead of destabilizing the system. Furthermore if η
a
can be proven to
be asymptotical stable,i.e. e → 0 as t → 0, the h
2
( ˙ e, e, t) in (10) dimmishes and it
becomes:
¨ η
u
= f
2
( ˙ η
u
, η
u
, t) + B
ηu
u
u
. (12)
Applying the asymptotic stability proof for η
u
, validates the initial assumption of
h
1
( ˙ η
u
, η
u
, t) →0 as t →∞. The following will illustrate the above mentioned method.
Assuming that the perturbation, h
1
( ˙ η
u
, η
u
, t) is bounded by decaying exponential
function and u
a
= u
PID
= −B
−1
η
a
(K
p
e+K
i
_
τ
0
edt+K
d
de
dt
) for the actuated subsystem
exists, (9) becomes ¨ e = f
1
( ˙ e, e, t) + B
ηa
u
PID
. The asymptotical stability of η
a
, i.e.
e →0 as t →0 is proven in the following.
7
4.2 Perturbation on ROV’s Parameters
To test the robustness of the PID control schemes, the ROV’s mass inertia, centripetal
and coriolis matrix and diagonal hydrodynamic damping matrix are allowed to vary
within a limit speciﬁed in (14) and (16). These variations can be contributed by the
inaccuracy in modelling and possible change in the mass distribution in the ROV.
The limits were obtained using a computeraided design (CAD) software, ProE.
By changing the mass properties of each thrusters, pods, stainlesssteel frame as
shown in Fig 2, a diﬀerent mass inertia matrix was obtained. By evaluating the
diﬀerences between the nominal and new mass inertia matrix, the following limits
can be determined.
The m
η
1
and m
η
1
are obtained by ﬁrst evaluating the inverse of bodyﬁxed mass
inertia matrix, M
−1
v
in (5) obtained from CAD software, ProE:
M
η
( ˙ η, η) = J
−T
M
v
J
−1
M
η
( ˙ η, η)
−1
= JM
−1
v
J
T
(13)
Substituting the M
−1
v
that ranges from −0.001 to 0.01 at J = I(as the Euler angles
are small) in (4) results in m
η
1
= −0.001 and m
η
1
= 0.01. Hence, the bound of the
M
η
−1
1
becomes
m
η
1
I ≤ M
η
−1
1
≤ m
η
1
I (14)
where
M
η
−1
1
= M
η22
_
M
η22
M
η11
−M
η12
2
¸
−1
(15)
While the upper limit bound on centripetal and coriolis matrix and diagonal hydro
dynamic damping matrix in C
η1
are set as:
C
η1
≤ K
A
L + K
B
(16)
8
where K
A
> 0(= 0.001) and K
B
> 0(= 0.001) are constant obtained indirectly from
the CAD software, ProE.,
C
η1
= [C
η11
(v, η) + D
η11
(v, η)] (17)
,. is the Euclidean norm and L = [e ˙ e]
T
.
4.3 Controller Design and Stability Analysis on Actuated Dy
namic
Implement the control law into (11) yields:
¨ e = M
η
−1
1
C
η1
˙ e −M
η
−1
1
[K
p
e + K
i
_
τ
0
edt + K
d
de
dt
] (18)
Statespace equations become,
x
1
=
_
τ
0
e
T
dt, ˙ x
1
= e
T
;
x
2
= e
T
, ˙ x
2
= ˙ e
T
;
x
3
= ˙ e
T
,
˙ x
3
= ¨ e
T
= M
η
−1
1
C
η1
x
3
−M
η
−1
1
K
d
x
3
−M
η
−1
1
K
p
x
2
−M
η
−1
1
K
i
x
1
] (19)
where superscript T in e
T
indicates the transpose of e. Equation (19) in matrix form
becomes
_
˙ x
1
˙ x
2
˙ x
3
_
=
A
¸ .. ¸
_
0 I
n
0
0 0 I
n
−Mη
−1
1
Ki −Mη
−1
1
Kp −Mη
−1
1
(K
d
+Cη1)
_
_
x
1
x
2
x3
_
(20)
9
To analyze the system robust stability, consider the following Lyapunov function:
V (x) = x
T
Px
=
1
2
[α
2
_
t
0
e(τ)dτ + α
1
e + ˙ e]
T
M
η1
[α
2
_
t
0
e(τ)dτ + α
1
e + ˙ e] +
T
P
1
(21)
where
=
_
¸
_
_
t
0
e(τ)dτ
e
_
¸
_; P
1
=
1
2
_
¸
_
α
2
K
p
+ α
1
K
i
α
2
K
d
+ K
i
α
2
K
d
+ K
i
α
1
K
d
+ K
p
_
¸
_ (22)
Hence,
P =
1
2
_
α
2
K
p
+α
1
K
i
+α
2
2
M
η1
α
2
K
d
+K
i
+α
1
α
2
M
η1
α
2
M
η1
α
2
K
d
+K
i
+α
1
α
2
M
η1
α
1
K
d
+K
p
+α
2
1
M
η1
α
1
M
η1
α
2
M
η1
α
1
M
η1
M
η1
_
(23)
Since M
η1
is a positive deﬁnite matrix, P is positive deﬁnite, if only if P
1
is positive
deﬁnite. Now choose, K
p
= k
p
I, K
d
= k
d
Iand K
i
= k
i
I such that,P in (23):
_
¸
_
α
2
k
p
+ α
1
k
i
α
2
k
d
+ k
i
α
2
k
d
+ k
i
α
1
k
d
+ k
p
_
¸
_ (24)
becomes positive deﬁnite. The following Lemma gives the conditions where V (x) can
become positive deﬁnite, bounded from upper and lower side.
Lemma 1: Assume the following inequalities hold:
α
1
> 0, α
2
> 0, α
1
+ α
2
< 1
s
1
= α
2
(k
p
−k
d
) −(1 −α
1
)k
i
−α
2
(1 +α
1
−α
2
)m
η
1
> 0 (25)
s
2
= k
p
+ (α
1
−α
2
)k
d
−k
i
−α
1
(1
+α
2
−α
1
)m
η
1
> 0 (26)
10
Then P is positive deﬁnite and satisﬁes the following inequality(RayleighRitz):
λ(P)x
2
≤ V (x) ≤ λ(P)x
2
(27)
in which,
λ(P) = min¦
1 −α
1
−α
2
2
m
η1
,
s
1
2
,
s
2
2
¦ (28)
λ(P) = max¦
1 + α
1
+ α
2
2
m
η
1
,
s
3
2
,
s
4
2
¦ (29)
and
s
3
= α
2
(k
p
+ k
d
) + (1 + α
1
)k
i
+ (1 + α
1
+α
2
)α
2
m
η
1
(30)
s
4
= α
1
m
η
1
(1 + α
1
+α
2
) + (α
1
+ α
2
)k
d
+ k
p
+ k
i
(31)
Since P is positive deﬁnite then
˙
V (x) = x
T
(A
T
P + PA +
˙
P)x
= −x
T
Qx +
1
2
x
T
_
α
2
I
α
1
I
I
_
˙
M
η1
[α
2
I α
1
I I] x
+
1
2
x
T
_
0 α
2
2
I α
1
α
2
I
α
2
2
I 2α
1
α
2
I (α
2
1
+α
2
)I
α
1
α
2
I (α
2
1
+α
2
)I α
1
I
_
_
Mη1 0 0
0 Mη1 0
0 0 M
η1
_
x (32)
since
˙
M
η1
= 0, (32) yields
˙
V (x) ≤ −γx
2
+ ζ
2
m
η1
x
2
(33)
and
γ = min¦α
2
k
i
, α
1
k
p
−α
2
k
d
−k
i
, k
d
¦ (34)
11
Let L ≤ x. Denote by λ
2
= λ(R
2
) the largest eigenvalue of R
2
,
R
2
=
_
¸
¸
¸
¸
_
0 α
2
2
I α
1
α
2
I
α
2
2
I 2α
1
α
2
I (α
2
1
+ α
2
)I
α
1
α
2
I (α
2
1
+ α
2
)I α
1
I
_
¸
¸
¸
¸
_
(35)
As a result, the error system of ROV RRC II, (20) is rendered GAS, if λ
2
is chosen
small enough and the control gains K
p
, K
d
and K
i
are large enough. Next step is to
show that the unactuated dynamic is exponentially bounded.
5 Exponentially stability of the unactuated dynamic
using Sordalen lemma
As shown in Section 4, the term h
2
( ˙ e, e, t) that consists of ˙ e and e converged exponen
tially to zero, i.e. ˙ e, e →0 as t →0 yielding ¨ η
u
= f
2
( ˙ η
u
, η
u
, t) +B
ηu
u
u
. The solution
of the tracking error, e can be approximated as e = C
e
e
−γ
e
t
⇒ ˙ e = C
˙ e
e
−γ
˙ e
t
and by
substituting it into h
2
( ˙ e, e, t), yields:
h
2
( ˙ e, e, t) = −
M
η12
C
η12
Mη22
Mη11
−Mη12
2 C
e
e
−γet
+ M
η11
g
ηu
+M
η11
B
η2
u
u
→0 (36)
for u
u
= −B
η
−1
2
g
ηu
. The initial assumption h
1
( ˙ η
u
, η
u
, t) → 0 as t → ∞ can be
validated by checking the asymptotic stability of the η
u
. First, decompose η
u
into
two part as follows
¨ η
u
=
_
¨
φ
¨
θ
_
=
_
f ˙
φ
(
˙
φ,φ,t)+d ˙
φ
(t)
f ˙
θ
(
˙
θ,θ,t)+d ˙
θ
(t)
_
(37)
where d
˙
φ
(t), d
˙
θ
(t) are considered as a perturbation on the
˙
φ and
˙
θ respectively. The
proof for exponential bound of the unactuated subsystems can be obtained as shown
12
below. The deﬁnite integral of f
˙
φ
(
˙
φ, φ, t) from time 0 to t becomes
[
_
t
0
f
˙
φ
(
˙
φ, φ, τ)dτ [ ≤
_
t
0
[ f
˙
φ
I1
∂ψ
∂τ
/k
a1
[ + [ f
˙
φ
I2
∂φ
∂τ
/k
a1
[
+ [ f
˙
φ
I3
∂θ
∂τ
/k
a1
[ + [ f
˙
φ
I4
[ dτ (38)
where k
a1
f
˙
φ
I1
, f
˙
φ
I2
, f
˙
φ
I3
, f
˙
φ
I4
can be found in Appendix. Substituting, I
xx
,
I
xy
, I
xz
, I
yz
,
˙
ψ,
˙
φ and
˙
θ into the preceding equation gives,
[
_
t
0
f
˙
φ
(
˙
φ, φ, τ)dτ [ ≤
_
t
0
[ β
1
˙
θ [ + [ β
2
˙
ψ [
+ [ β
3
˙
φ [ dτ
≤
_
t
0
β
1
[ C
˙
θ
e
−α˙
θ
τ
[ +β
2
[ C
˙
ψ
e
−α ˙
ψ
τ
[ +β
3
[ C
˙
φ
e
−α ˙
φ
τ
[
+β
4
dτ
[
_
t
0
f
˙
φ
(
˙
φ, φ, τ) +
˙
φ
dτ [ ≤ β
1
C
˙
θ
α
˙
θ
+ β
2
C
˙
ψ
α
˙
ψ
+β
3
C
˙
φ
α
˙
φ
(39)
where β
1
, β
2
, β
3
, β
4
> 0. In ROV RRC II,β
1
= 15603, β
2
= 15470, β
3
= 2.7, β
4
=
1650.Then,d
˙
φ
(t) in (37) becomes:
[ d
˙
φ
(t) [ = [ f
˙ z1
˙ z + f
˙ x1
˙ x + f
˙ y1
˙ y + f
˙
ψ
1
˙
ψ + f
˙
θ
1
˙
θ [
≤ [ f
˙ z1
˙ z [ + [ f
˙ x1
˙ x [ + [ f
˙ y1
˙ y [ + [ f
˙
ψ
1
˙
ψ [
+ [ f
˙
θ
1
˙
θ [ (40)
Deﬁne
γ
˙
φ
= min¦α
˙ z
+ α
˙ x
, α
˙ z
+ α
˙ y
, α
˙ x
+α
˙ y
, α
˙
ψ
+ α
˙
θ
, α
˙
ψ
+α
˙
φ
, α
˙
θ
+ α
˙
φ
, α
˙ z
, α
˙ x
, α
˙ y
, α
˙
ψ
, α
˙
φ
, α
˙
θ
¦ (41)
13
and ˙ z = C
˙ z
e
−γ
˙ z
, ˙ x = C
˙ x
e
−γ
˙ x
, ˙ y = C
˙ y
e
−γ
˙ y
,
˙
ψ = C
˙
ψ
e
−γ ˙
ψ
, ∀α
˙ z
, α
˙ x
, α
˙ y
, α
˙
ψ
where
C
˙ z
, C
˙ x
, C
˙ y
, C
˙
ψ
> 0[Kreyszig, 1998] for an exponentially stable system. Then
[ d
˙
φ
(t) [ ≤ De
−γ ˙
φ
t
(42)
Solution of the
¨
φ(t) becomes
[
˙
φ(t) [ = [ e
−[f ˙
φ
(
˙
φ,φ,t)+ ˙
φ
]t
˙
φ(0)
+
_
t
0
e
−[f ˙
φ
(
˙
φ,φ,t)+ ˙
φ
]τ
d
˙
φ
(τ)dτ [
≤ e
−P ˙
φ
t
[
˙
φ(0) [ + [
_
t
0
e
−P ˙
φ
τ
d
˙
φ
(τ)dτ [
≤ e
−P ˙
φ
t
[
˙
φ(0) [ +
[ D[e
−(P ˙
φ
+γ ˙
φ
)t
−1] [
[ P
˙
φ
+ γ
˙
φ
[
(43)
where P
˙
φ
=
_
t
0
[f
˙
φ
(
˙
φ, φ, τ) +
˙
φ
]dτ. Thus
˙
φ(t) is bounded ∀t ≥ 0. Also
˙
φ(t) →0 since
1
P ˙
φ
+γ ˙
φ

is small.
Next, to show that φ(t) is bounded, consider
[ φ(t) [ = [
_
t
0
˙
φ(t)dτ + φ(0) [
≤
_
t
0
[
˙
φ(t) [ dτ+ [ φ(0) [ (44)
Using (43),
[ φ(t) [ ≤
_
t
0
e
−P ˙
φ
τ
dτ [
˙
φ(0) [
+
D
[ P
˙
φ
+ γ
˙
φ
[
_
t
0
(e
−(P ˙
φ
+γ ˙
φ
)τ
−1)dτ+ [ φ(0) [
≤ −
e
−P ˙
φ
τ
+ 1
P
˙
φ
[
˙
φ(0) [ −
D
[ P
˙
φ
+ γ
˙
φ
[
2
e
−(P ˙
φ
+γ ˙
φ
)t
−
D
[ P
˙
φ
+ γ
˙
φ
[
(t −1)+ [ φ(0) [ (45)
Thus, φ(t) is bounded ∀t ≥ 0 as
1
P ˙
φ
+γ ˙
φ

is small for the RRC ROV II.
14
Repeat the same procedure from (38) to (45) for f
˙
θ
(
˙
θ, θ, τ). By substituting,
I
xx
, I
xy
, I
xz
, I
yz
, in [Koh, et al., 2002a],
˙
ψ,
˙
φ and
˙
θ into the preceding equation gives
the deﬁnite integral of f
˙
φ
(x, t)
[
_
t
0
f
˙
θ
(
˙
θ, θ, τ)dτ [ = [
_
t
0
f
˙
θ
1
dτ [ (46)
≤
_
t
0
[ α
1
˙
θ [ + [ α
2
˙
ψ [
+ [ α
3
˙
φ [ + [ α
4
[ dτ
≤
_
t
0
α
1
C
˙
θ
e
−α˙
θ
τ
+α
2
C
˙
ψ
e
−α ˙
ψ
τ
+ α
3
C
˙
φ
e
−α ˙
φ
τ
+ α
4
dτ
[
_
t
0
[f
˙
θ
(
˙
θ, θ, τ) +
˙
θ
]dτ [ ≤ α
1
C
˙
θ
α
˙
θ
+ α
2
C
˙
ψ
α
˙
ψ
+ α
4
C
˙
φ
α
˙
φ
(47)
where α
1
, α
2
, α
3
, α
4
> 0. In ROV RRC II, α
1
= 25.8, α
2
= 12270.8, α
3
= 12529, α
4
=
260378. Then,d
˙
θ
(t) in (37) becomes:
[ d
˙
θ
(t) [ = [ f
˙ z2
˙ z + f
˙ x2
˙ x + f
˙ y2
˙ y + f
˙
ψ
2
˙
ψ
+f
˙
φ
2
˙
φ [
≤ [ f
˙ z2
˙ z [ + [ f
˙ x2
˙ x [ + [ f
˙ y2
˙ y [ + [ f
˙
ψ
2
˙
ψ [
+ [ f
˙
φ
2
˙
φ [ (48)
Deﬁne
γ
˙
θ
= min¦α
˙ z
+ α
˙ x
, α
˙ z
+ α
˙ y
, α
˙ x
+ α
˙ y
, α
˙
ψ
+α
˙
θ
,
α
˙
ψ
+ α
˙
φ
, α
˙
θ
+ α
˙
φ
, α
˙ z
, α
˙ x
, α
˙ y
, α
˙
ψ
, α
˙
φ
, α
˙
θ
¦ (49)
and ˙ z = C
˙ z
e
−γ
˙ z
, ˙ x = C
˙ x
e
−γ
˙ x
, ˙ y = C
˙ y
e
−γ
˙ y
,
˙
ψ = C
˙
ψ
e
−γ ˙
ψ
, ∀α
˙ z
, α
˙ x
, α
˙ y
, α
˙
ψ
where
15
C
˙ z
, C
˙ x
, C
˙ y
, C
˙
ψ
> 0 [Kreyszig, 1998] for an exponentially stable system. Then
[ d
˙
θ
(t) [ ≤ De
−γ ˙
θ
t
(50)
Solution of the
¨
θ(t) becomes
[
˙
θ(t) [ = [ e
−[f ˙
θ
(
˙
θ,θ,t)+ ˙
θ
]t
˙
θ(0)
+
_
t
0
e
−[f ˙
θ
(
˙
θ,θ,t)+ ˙
θ
]τ
d
˙
θ
(τ)dτ [
≤ e
−P˙
θ
t
[
˙
θ(0) [ + [
_
t
0
e
−P˙
θ
τ
d
˙
θ
(τ)dτ [
≤ e
−P˙
θ
t
[
˙
θ(0) [ +
[ D[e
(P ˙
φ
+γ ˙
φ
)t
−1] [
[ P
˙
φ
+ γ
˙
φ
[
(51)
where P
˙
θ
=
_
t
0
[f
˙
θ
(
˙
θ, θ, τ) +
˙
θ
]dτ. Thus
˙
θ(t) is bounded ∀t ≥ 0. Also
˙
θ(t) → 0 as
1
P˙
θ
+γ ˙
θ

is small. Next, to show that θ(t) is bounded, consider
[ θ(t) [ = [
_
t
0
˙
θ(t)dτ + θ(0) [
≤
_
t
0
[
˙
θ(t) [ dτ+ [ θ(0) [ (52)
Using (51),
[ θ(t) [ ≤
1
[ P
˙
θ
+ γ
˙
θ
[
_
t
0
e
−(P˙
θ
+γ ˙
θ
)τ
dτ
+
_
t
0
e
−P˙
θ
τ
dτ [
˙
θ(0) [
+
_
t
0
1
[ P
˙
θ
+ γ
˙
θ
[
dτ+ [ θ(0) [ + [ θ(0) [
≤ −
e
−P˙
θ
τ
+ 1
P
˙
θ
[
˙
θ(0) [ −
D
[ P
˙
θ
+ γ
˙
θ
[
2
e
−(P˙
θ
+γ ˙
θ
)t
−
D
[ P
˙
θ
+ γ
˙
θ
[
(t −1)+ [ θ(0) [ (53)
Thus, θ(t) is bounded ∀t ≥ 0 as
1
P˙
θ
+γ ˙
θ

is small for RRC ROV II.
16
6 Computer Simulation Results
This section illustrates the performance of the proposed control scheme (in the pres
ence of parameters perturbation) using a ROV Design and Analysis (RDA) package
(see Fig.3) developed by NTU. The platform adopted for the development of RDA
is MATLAB/SIMULINK. The RDA provides the necessary resources for rapid and
systematic implementation of mathematical models of ROV systems with focus on
ROV modeling, control system design and analysis. The package provides examples
ready for simulation. As shown in Fig. 4, the block diagram of the proposed robust
PID control was designed.
Performance of the proposed control scheme has been investigated in computer
simulation using Pentium IV, 2.4 GHz computer. The simulation time was set to 100s.
The RRC ROV II parameters used in simulation can be found in [Koh, et al., 2002b].
As the vehicle is currently ﬁtted only with limited sensors, the desired position com
mands values, v = [0.5 0 1 0 0 0]
T
are chosen with this purpose in mind. The objective
is to regulate the position of the RRC ROV II to x = 0.5m and z = 1m or the error
signal equal to zero.
The PID was selected due to its simplicity in implementation and its widely used
in control application. The control algorithm requires little computing resources, and
is therefore suited for an onboard implementation. The PID control parameters were
obtained from the Response Optimization (using gradient descent method) toolbox
in SIMULINK. The PID parameters are as follows: K
p
=diag ¦11, 11, 9, 1¦, K
i
=diag
¦2, 4, 1, 2¦ and K
d
=diag ¦1, 1, 2, 1¦. For example, the optimization setting for the
response in the x direction was constrained as shown in Fig 7.
As mentioned in the paper, the system is separated into actuated and unactuated
states. In the SIMULINK, it is modelled as a selector that channels the actuated and
unactuated states into two paths. The PID is used only for the actuated states while
the unactuated states are left uncontrolled, since the later are selfstabilizable. The
”robust” PID controller is called due to the perturbation on the mass inertia term as in
(14). The perturbation on the mass inertia term ranges from minimum to maximum
17
scale (−0.001 ≤ M
η
−1
1
≤ 0.01). This can be modelled as a twodimensional (2D)
lookup table as shown in Fig.4 was used. During simulation, only the maximum value
(or worstcase perturbation) is used. Similarly the perturbation for the centripetal
and coriolis and hydrodynamic damping matrix was set at (C
η1
≤ 0.001L +0.001).
Figure 5 shows the simulation results for the robust PID control design with and
without perturbation. It shows the position of the ROV with respect to the inertia
frame. As observed, the ROV control system is able to regulate about the selected
reference position. The roll and pitch motions are selfstabilizable about zero position
(as seen the Sordalen proof). It can be seen from Fig.5 that the actuated states exhibit
asymptotical stable phenomena, as proved by the Lyapunov stability theory. Notice
how, in spite of parameter uncertainty, the ROV converges asymptotically to the
desired position.
In summary, the ﬂow chart shown in Fig. 6 illustrates the robust PID design
ﬂow and methodology using RDA for ROV RRC II. The PID parameters are tuned
oﬄine by the Response Optimization toolbox using gradient descent till the desired
responses are obtained. Besides, tuning of the PID parameters can be performed
iteratively.
7 Conclusion
Besides using the ROV’s bodyﬁxed coordinates in the stability analysis, the earth
ﬁxed coordinate model was analyzed. The stabilizability condition of the underactu
ated ROV was shown. The nonlinear system was separated into actuated and unac
tuated dynamic equations and the asymptotical stability of the closedloop actuated
equations, using the robust PID controller in the earthﬁxed frame was examined.
This is based on the argument that the actuated dynamic equation could converge
exponential to zero by the PID controller. The asymptotic stability proof by Sordalen
in the unactuated dynamic equation provides a validation for the initial argument of
the roll and pitch dynamic convergence to zero. The vehicle in the earthﬁxed frame
with the selfstabilzable unactuated dynamic is globally asymptotically stable with the
18
robust PID controller in the actuated dynamic. This gives a necessary and suﬃcient
condition for GAS of the ROV in earthﬁxed frame.
By simulating the closedloop control system design using the RDA package, the
asymptotical stability of the robust PID controller for ROV RRC II can be observed
as the position responses converge asymptotically to the desired position. The PID is
said to be robust as the ROV’s mass inertia, centripetal and coriolis matrix and di
agonal hydrodynamic damping matrix were allowed to vary within the limit obtained
explicitly from the CAD software, ProE.
In summary, the RDA provides a systematic methodology in control system sim
ulation and stability analysis before implementation. It also provides the necessary
resources for rapid implementation of mathematical models of ROV systems. The
control algorithm proposed is quite simple and requires little computing resources,
and is therefore suited for an onboard implementation. Simulation results show the
eﬀectiveness of the proposed methodology. If a realtime tuning of the PID parameters
is used, the control system could be more robust against larger parameters perturba
tion. As shown in this paper, the PID parameters can be conveniently tuned by the
Response Optimization toolbox in the MATLAB/SIMULINK instead of the trial and
error method of tuning.
Acknowledgment
The authors would like to thank and acknowledge the contributions by all the project
team members from NTU Robotics Research Centre especially Mr. Lim Eng Cheng,
Ms. Agnes S.K. Tan, Ms. Ng Kwai Yee and Mr. You Kim San.
References
T. I. Fossen, ”Guidance and Control of Ocean Vehicles”, New YorkWiley, 1994.
J. Yuh, ”Modeling and control of Underwater Robotic Vehicles”, IEEE Transactions
on Systems, Man and Cybernetics, Vol. 20, No. 6, pp. 14751483, 1990.
19
R. W. Brockett, ”Asymptotic stability and feedback stabilization”,in R. W. Brock
ett, R.S. Millman and H. J. Sussmann, eds,Diﬀerential geometric control theory,
Boston,Birkhauser,pp. 181191, 1983.
C. Byrnes, A. Isidori, ”On the attitude stabilization of rigid spacecraft”,Automatica,
vol. 27, No. 1, pp. 8795, 1991.
K.Y. Wichlund, O.J. Srdalen and O. Egeland ”Control of Vehicles with SecondOrder
Nonholonomic Constraints: Underactuated Vehicles”, Proc. 1995 European Control
Conference, Rome, Italy, pp. 30863091, Sep. 1995.
Ole Jakob Srdalen, M. Dalsmo, Olav Egeland,”An Exponentially Convergent Control
Law for a Nonholonomic Underwater Vehicle”, in Proceeding of 3rd International
Conference on Robotics and Automation,Altanta,Georgia, USA, pp. 790795, 1993.
O. J. Søodalen and O. Egeland,”Exponential Stabilization of Chained Nonholonomic
Systems”,Proceedings of the 2nd European Control Conference, The Netherlands,pp.
14381443,Groningen,1993.
O. J. Sødalen and O. Egeland,”Exponential Stabilization of Nonholonomic Chained
Systems”,IEEE Transactions on Automatic Control,vol 40,no.1,pp. 3549.
T.H. Koh, M.W.S. Lau, E. Low, G.G.L. Seet and P.L. Cheng, ”Preliminary studies of
the modeling and control of a twinbarrel underactuated underwater robotic vehicle”,
in proceedings of the 7th International Conference on Control, Automation, Robotics
Vision, Singapore, pp. 10431047, 2002.
T. H. Koh and M. W. S. Lau and E. Low and G. Seet and S. Swei and P. L. Cheng,
”Development and Improvement of an Underactuated Remotely Operated Vehicle
(ROV)”,MTS/IEEE International Conference of Oceans (OCEANS),Biloxi, Missis
sippi, October,pp. 20392044,2002.
E. Kreyszig,”Advanced Engineering Mathematics”,John Wiley Sons,1998.
M.W.S. Lau, S.S.M. Swei, G.G.L. Seet, E. Low and P.L. Cheng,”Control of an un
deractuated remotely operated underwater vehicle”,Proceeding of IMech Engineering:
Part 1: Journal of System and Control, Vol. 217, No.1, pp. 343358,2003.
20
Appendix
The parameters used in (38) are as follow:
k
a1
= −I
xx
I
yz
2
+ I
xx
I
yy
I
zz
−2 I
yz
I
xy
I
xz
−I
yy
I
xz
2
−I
xy
2
I
zz
f
˙
φ
I1
= I
xy
I
2
yz
−I
2
xz
I
zz
+ I
xy
I
2
zz
+ I
2
xy
I
yy
−I
2
yy
I
xz
−I
xx
I
2
yy
+ I
3
xy
−I
3
xz
−2I
xy
I
xz
I
yy
−I
xy
I
yz
I
yy
+ I
xx
I
yz
I
xz
+ I
xx
I
zz
I
xz
−I
xx
I
yy
I
xy
+I
yy
I
xz
I
yz
−I
2
yz
I
xz
+ 2I
xy
I
zz
I
xz
−I
xy
I
zz
I
yz
+I
xy
I
2
xz
+ I
xx
I
2
zz
−2I
2
xy
I
yz
−I
2
xy
I
xz
+2I
yz
I
2
xz
−I
xx
I
yz
I
xy
+ I
yz
I
xz
I
zz
f
˙
φ
I2
= −I
xx
I
zz
I
xz
−I
xy
I
zz
I
xz
−I
3
xy
−I
xx
I
yz
I
xz
+ I
2
xy
I
xz
+I
xx
I
yy
I
xy
+ I
xx
I
yz
I
xy
+ I
2
xy
I
yz
−I
xy
I
2
xz
−I
yz
I
2
xz
+ I
xy
I
xz
I
yy
+ I
3
xz
f
˙
φ
I3
= −3I
xx
I
2
yz
−I
2
xy
I
zz
−I
yy
I
2
xz
−4I
yz
I
xy
I
xz
+I
xx
I
yy
I
zz
−I
yy
I
xz
I
yz
−I
xy
I
zz
I
yz
−I
xy
I
xz
I
yy
−I
xy
I
yz
I
yy
−2I
xx
I
zz
I
yz
−2I
xx
I
yz
I
yy
−I
2
yz
I
xz
−I
xx
I
2
yy
−I
2
yy
I
xz
+ I
2
xy
I
yz
−I
xx
I
2
zz
−I
xy
I
2
zz
+ I
yz
I
2
xz
+ I
2
xz
I
zz
−I
xy
I
2
yz
+ I
2
xy
I
yy
−I
yz
I
xz
I
zz
−I
xy
I
zz
I
xz
f
˙
φ
I4
= (−I
yz
I
xz
Kp −I
xy
I
zz
Kp −I
yy
I
xz
Kp
−KpI
yy
I
zz
−I
yz
I
xy
Kp + KpI
2
yz
)/k
a1
(54)
21
Figure 1: Thrusters conﬁguration on the ROV platform
Figure 2: Experimental RRC ROV II in swimming pool
22
Figure 3: SIMULINK library browser showing RDA package
23
Figure 4: SIMULINK block diagram of the proposed robust PID control
24
Figure 5: Position response of the proposed robust PID control for x=0.5m and z=1m
25
Figure 6: Flow chart of the robust PID design ﬂow and methodology using RDA on
RRC ROV II
26
Figure 7: Example of response optimization setting on output x
27
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.