You are on page 1of 17

Ocean Engineering 89 (2014) 82–98

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Multiple Impedance Control for object manipulation by a dual arm


underwater vehicle–manipulator system
Hamed Farivarnejad, S. Ali A. Moosavian n
Center of Excellence in Robotics and Control, Advanced Robotics and Automated Systems Lab, Department of Mechanical Engineering, K.N. Toosi
University of Technology, 19 Pardis Street, Vanak Square, Tehran, Islamic Republic of Iran

art ic l e i nf o a b s t r a c t

Article history: In this paper, control of an object manipulation task by a dual arm underwater vehicle–manipulator
Received 14 January 2013 system (UVMS) is studied. The object is a heavy cylinder which should be moved and pegged precisely in
Accepted 22 June 2014 an underwater structure while impacts due to contact are inevitable. The Multiple Impedance Control
(MIC) is customized for this system, and the obtained results are compared with those of the augmented
Keywords: object model (AOM) approach. To this end, first an explicit dynamics model of the UVMS is developed.
Underwater vehicle–manipulator system Hydrodynamic effects, i.e. added mass, drag and buoyancy are considered as external forces/moments in
Dynamics modeling the task space, which are incorporated into the model by corresponding Jacobian matrices of their
Hydrodynamic effects application points. Then, this derived model is used for preparing a MATLAB code by which the
Underwater object manipulation
manipulation task in the 3D space is simulated. Extensive computer simulations show that in the
Multiple Impedance Control
presence of flexibility in the system that is modeled by remote center compliance the AOM does not
Model uncertainties
have a stable performance, while the MIC yields a reliable performance. It is shown that even in case of
appearing an impact due to contact with the environment, also severe model uncertainties, the system
exhibits a smooth and nondestructive reaction.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction which high accuracy and dexterity are necessary. Pioneering


models, ODIN, OTTER and VORTEX, were mainly used as research
Exploitation of mobile robotic systems in various types of test beds and always worked in water tank conditions. AMADEUS,
environment is one of the most interesting topics in recent ANSALDO, SWIMMER, ALIVE and SAUVIM are the next projects in
decades. Although dynamics modeling and control of terrestrial the way of the evolution of UVMSs towards further dexterity and
and space robots have been regarded more frequently than under- autonomy (Sanz et al., 2010a).
water robots (Eslamy and Moosavian, 2011, 2010; Alipour and Today, one of the challenging subjects related to the application
Moosavian, 2011; Moosavian and Papadopoulos, 2007; Moosavian of autonomous UVMSs in underwater missions is accomplishment
and Rastegari, 2006), the utilization of autonomous underwater of the intervention missions which are performed by manned
vehicles (AUV) in marine technologies has been extended in the submersible endowed with robotic arms or by remotely operated
last two decades. The first models of such systems were designed vehicles (ROV). The drawbacks of these systems such as reduced
and used as small submarines which only were able to move freely time for operation, the human presence in dangerous and hostile
in the water and accomplish survey missions such as inspection, environment, very high cost associated with the need for an
subsea observations and oil or gas explorations. As these types of expensive oceanographic vessel with a heavy crane and cognitive
underwater robots were unable to carry out any manipulation fatigue of the operator are why researchers have recently started
tasks and interact with the environment, mounting a dexterous efforts to make UVMSs capable of performing complicated inter-
manipulator on the moving base (vehicle) was focused in the early vention missions autonomously. Besides, some recent challenging
1990s. This idea resulted in the formation of a new type of projects such as removing the oil spill in the Gulf of Mexico
underwater robots called underwater vehicle–manipulator sys- (Bleicher, 2011) highlighted the need for faster and more reliable
tems (UVMS). This type could carry out various tasks such as underwater robotic systems. Similarly, some future research plans
welding, drilling, object manipulation and any other tasks in such as the exploration of extraterrestrial life under the ice-
covered oceans of the Jupiter's frozen moon (Kumagai, 2007;
Kramer, 2013) emphasize the need for highly adaptable UVMSs
n
Corresponding author. Tel.: þ 98 21 8406 3238; fax: þ 98 21 8867 4748. which can execute complicated missions in unknown environ-
E-mail address: moosavian@kntu.ac.ir (S.A.A. Moosavian). ment with high degrees of autonomy.

http://dx.doi.org/10.1016/j.oceaneng.2014.06.032
0029-8018/& 2014 Elsevier Ltd. All rights reserved.
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 83

One of the first efforts to develop Intervention Autonomous compensate various dynamic and hydrodynamic effects during
Underwater Vehicles (I-AUV) was the OTTER I-AUV by which underwater manipulation tasks. Regarding developments in the
studies on automatic objects retrieval were started (Wang et al., structure of underwater manipulators, Fernandez et al. (2013)
1995). Another I-AUV was ALIVE that was developed to be capable have adapted a manipulator, which was initially designed to be
of docking to a subsea structure not specifically modified for AUV teleoperated, for the autonomy needs in an AUV by reduction in its
use (Evans et al., 2003). Sotiropoulos et al. (2012) have presented a dimension and weight, and development in its kinematics model.
two-step method for the determination of the optimum docking Also, Yu et al. (2013) have focused the replacement of conventional
pose for an AUV performing an intervention mission. Angeletti rigidly connected multi-link arm by a small, deployable, and highly
et al. (1998) and Casalino (2001) have studied grasping and maneuverable agent ROV, which eliminates the need for additional
manipulation of objects, an ordinarily necessary task in interven- positioning sensors and allows significant size reduction of
tions, by a robotic work cell composed of two 7 DOF manipulators the agent.
by mainly focusing on its various aspects of modularity, scalability As the key element in underwater intervention performed with
and configurability. Sun and Cheah (2004) have proposed two autonomous systems is autonomous manipulation, and achieve-
simple set point controllers for the coordinated control of multiple ment to such a performance requires the capability of the robotic
cooperative UVMSs holding a common object and presented system in physical contacts with the unstructured environment
Lyapanov like functions for convergence and stability analysis of without continuous human supervision, in this paper, as shown in
the system. Improvement of early claw-like end effectors by Fig. 1, the manipulation and installation of a definite object by a
addition of 6-axis force/torque and position sensors to fingers is dual-arm UVMS has been focused, and the functionality of the
a different research in which increasing the capability of under- Multiple Impedance Control (MIC) (Moosavian and Papadopoulos,
water arms in grasping a wide variety of objects (Meng et al., 1998), in such a mission has been studied.
2006; Wang et al., 2007). Wang et al. (2008) have noticed the
drawbacks of the original impedance control method for control-
ling the contact force between the fingertips and object and have
proposed the position-based neural network impedance control
method to resolve them by compensating the uncertainty of finger
dynamics, object position and stiffness. Shao et al. (2006) have
considered cooperative transportation of a floating object to its
destination by three fish-like robots and have studied the imple-
mentation of a situation based action selection mechanism to
perform this task in the lab environment. Also, Zhang et al. (2007)
have proposed a coordination method for cooperative transport
task in the particular underwater environment by utilizing the
limit cycle approach to control the posture of the robotic fish and
adopting the fuzzy logic method to control the transport orienta-
tion. CManipulator project (Hildebrandt et al., 2008) is the first
deep-see underwater robot capable of detecting previously
defined objects autonomously, grasp them and set them down
Fig. 1. 3D model of the dual arm UVMS performing a manipulation task.
or connect them to other objects. Christensen et al. (2009) have
presented a hardware facility to simulate movements of an ROV. Why the MIC has been chosen for such a mission is its
The described system is capable to let the ROV react in a realistic efficiency in the presence of flexibility in the robotic system and
way to forces emerging while one of the attached manipulators also in case of impacts with an obstacle (Moosavian and
interacts with an object. Development of a multi-limb manipula- Papadopoulos, 1997). As the MIC requires the mathematical model
tion system with tactile force-feedback is the main goal of the of the robotic system, initially the UVMS explicit dynamics model
SeeGrip project (Lemburg et al., 2011) in which providing the including various hydrodynamic effects is developed. Next, the
additional modality of haptic feedback for underwater handling of essential assumptions in the implementation of the MIC, i.e. object
objects in hazy water conditions has been focused. Marani et al. and end-effectors path planning, grasp condition and object
(2008) have described one of the first trials of autonomous dynamics model are dealt with. Following focus on these pre-
intervention performed by SAUVIM. This operation is an under- requisites, the object manipulation on a predefined path and its
water recovery mission which contains a sequence of autonomous collision with an obstacle is simulated in 3D space. To be more
tasks finalized to search for the target and to securely hook a cable realistic, actuators saturation is taken into account. After that, in
to it in order to bring the target to the surface. Hu et al. (2010, order to compare the obtained results with those of the augmen-
2011) have studied underwater cooperative box-pushing by three ted object model (AOM) (Chang et al., 2000), the same procedure
autonomous robotic fish. Lee et al. (2012) have regarded the is followed to apply this control scheme for performing the
vision-based object detection and tracking techniques for under- described task. Finally, the effect of model uncertainty, a highly
water robots performing manipulation tasks. Design and develop- probable disturbance in the underwater environment, on the
ment of an AUV capable to perceive the environment and performance of the MIC is noticed.
autonomously perform simple intervention tasks by means of a
specific hand-arm system are long term objectives of the RAUVI
project (Sanz et al., 2010b). TRIDENT (Prats et al., 2011) is another 2. Dynamics modeling
recent project in which a new methodology for multipurpose
underwater intervention tasks is focused, and a new control In this section, the explicit dynamics model of the considered
architecture, which exploits all redundant degrees of freedom of UVMS which is shown in Fig. 2 is derived. This system is composed
the AUV to satisfy a set of conditions of scalar inequality type, is of two PUMA type robotic arms mounted on a 6 DOF vehicle. Also,
proposed (Simetti et al., 2014). Mohan and Kim (2012) have one of the arms is equipped with a remote center compliance
presented an indirect adaptive control based on an extended (RCC) in order to have the necessary flexibility during the object
Kalman filter to make an autonomous UVMS capable to manipulation and contact with the undersea structure.
84 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

with RC 0 ¼ ðx; y; zÞT and δ0 ¼ ðϕ; θ; ψ ÞT being the base (vehi-


cle) mass center position in a fixed coordinate and Euler angles,
respectively, and qðmÞ
i describes the mth manipulator joint angles
(Fig. 3).

Fig. 3. The generalized coordinates of the UVMS.

By these definitions and applying the general Lagrange for-


Fig. 2. The considered underwater vehicle–manipulator system. mulation, the equations of motion of a space free-flyer in a
microgravity environment are obtained as
2.1. Equations of motion regardless the hydrodynamic effects
Hðδ0 ; qÞq€ þCðδ0 ; δ_ 0 ; q; qÞ
_ ¼ Q ðδ0 ; qÞ ð3Þ
To develop the dynamics model of this system, utilization of the
where H is the N  N mass matrix, C is the N  1 nonlinear
Lagrange formulation via direct path method (Moosavian and
velocity terms vector and Q is the N  1 vector of generalized
Papadopoulos, 2004) is focused. This method has been developed to
forces (N ¼ K þ 6) given by
model space free-flying robots (SFFR) that are different from UVMSs
regarding their surrounding environment. There is no medium ( )
061 if n Nm if T
around a SFFR, and such systems work in the microgravity environ- Q¼ þ ∑ J T0;p F 0;p þ ∑ ∑ ∑ J ðmÞ ðmÞ
i;p F i;p ð4Þ
τK1 p¼1 m¼1i¼1p¼1
ment. However, underwater robots are in the underwater environ-
ment that exerts significant hydrodynamic forces/moments on them.
In which F 0;p is the pth external force/moment applied on the
If these forces/moments can be modeled by the above mentioned
base, F ðmÞ
i;p is the pth external force/moment applied on the ith body
method, the dynamics model of the UVMS will be completed.
of the mth manipulator, if is the number of applied forces/
In this formulation, the vector of the generalized coordinates is
moments on the corresponding body and J ðmÞ i;p is a Jacobian matrix
chosen as
corresponding to the point of force/moment application
q ¼ ððRTC0 ; δ0 Þ; qð1Þ ; …; qðnÞ ÞT
T T T
ð1Þ (Moosavian and Papadopoulos, 2004).
To customize Eq. (3) for a UVMS, the vehicle thrusters, gravity
where and hydrodynamic effects must be included. The effects of the
vehicle thrusters, which are illustrated in Fig. 4, can be incorpo-
qðmÞ ¼ ðqðmÞ ðmÞ ðmÞ T
1 ; q2 ; …; qN m Þ ð2Þ rated into the model by the second term of the right hand side of

Fig. 4. The locations of the thrusters on the vehicle.


H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 85

Eq. (4) as in which


" #( ) " #
I
R0 033 F0 A11 A12
τ0 ¼ ð5Þ MA ¼ ð11Þ
033 ST0 ðδ0 Þ M0 A12 A22

" #
where I R0 is the rotation matrix from the vehicle local coordinate 033  SðA11 ν1 þ A12 ν2 Þ
CA ¼ ð12Þ
to the inertial one and S0 is the matrix which relates the vehicle  SðA11 ν1 þ A12 ν2 Þ  SðA21 ν1 þ A22 ν2 Þ
angular velocities to its Euler angles rates as

ω0 ¼ S0 δ_ 0 ð6Þ ν ¼ ½ νT1 νT2 T ð13aÞ

Also
η ¼ ½ ηT1 ηT2 T ð13bÞ
8 9
> F þF top >
< bottom = where ν1 and ν2 are respectively the linear and angular velocities
F 0 ¼ F rear þ F f ront ð7Þ
>
: F þF >
; expressed in body fixed coordinate, η1 is the mass center position
lef t right
relative to the inertial coordinate, η2 is the vector of Euler angles.
Also, τ is the vector of the external forces/moments acting on the
8 9 body. In addition, in Eq. (12), S is the skew-symmetric cross
> L ðF  F f ront Þ >
< P rear = product matrix.
M0 ¼ LY ðF lef t  F right Þ ð8Þ
>
: L ðF  F > In Eq. (10), it should be noticed that the added mass matrix is
R top Þ;bottom multiplied by time derivative of body fixed frame velocities respect
to the body fixed coordinate, not respect to the inertial one.
To model the gravity and hydrodynamic effects, they are The linear and angular velocities of the ith component
considered as external forces/moments in the task space applied expressed in base coordinate can be written by the Jacobian
on a certain point of the system. By this approach, they are matrix as
assigned to the model by two last terms of Eq. (4) which are
specified to the effect of external forces/moments. However, the
0
V C;i ¼ J L;i q_ ð14Þ
hydrodynamic effects are not computed as simply as the gravity
force is. To incorporate them in the dynamics model, they should 0
ω i ¼ J A;i q_ ð15Þ
be expressed as forces/moments whose magnitude, direction and
application point are definite for a typical rigid body immersed in If local coordinates are assigned to each link of the UVMS on
the water. Therefore, in the next section, these effects which are the basis of the Denavit–Hartenberg method (Craig, 1989) as:
mainly added mass, drag and buoyancy are studied, and their the velocities in Eqs. (14) and (15) can be expressed in the local
equations are developed somehow they can be used in Eq. (4) and coordinate with multiplying by the associated rotation matrix as
added to the UVMS dynamics. (Fig. 5)

Fig. 5. The local coordinates assigned to the links of the UVMS.

2.2. Incorporation of hydrodynamic effects


i
V C;i ¼ i R 0 0 V C;i ¼ 0 R Ti J L;i q_ 9 i J L;i q_ ð16aÞ

2.2.1. Added mass i


ω i ¼ i R 0 0 ω i ¼ 0 R Ti J A;i q_ 9 i J A;i q_ ð16bÞ
This effect originates from the accelerating part of water
surrounding a body. When the body accelerates, this part also
tends to move with it and needs a certain amount of energy in in which
addition to the amount the body needs. If the equation of motion
i
of a free 6 DOF body is written as 0 j1
Ri ¼ ∏ Rj ð17Þ
j¼1
M ν_ þ Cðν; ηÞ ¼ τ ð9Þ
where
the added mass matrix, M A , and its effect on the nonlinear velocity 2 3
0 0 1
terms, C A , can be entered in this equation as (Fossen, 1994) 6
0
R 1 ¼ 4 sin q1 cos q1 0 7
5 ð18aÞ
ðM þM A Þν_ þ Cðν; ηÞ þC A ðνÞν ¼ τ ð10Þ cos q1  sin q1 0
86 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

2 3
cos qj  sin qj 0 (Eq. (9)) as DðνÞν in which
j1 6 0 0 1 7
Rj ¼ 4 5; j ¼ 2; 5 ð18bÞ DðνÞ ¼ diagfX u ; Y v ; Z w ; K p ; M q ; N r g
sin qj cos qj 0
þ diagfX ujuj juj; Y vjvj jvj; Z wjwj jwj; K pjpj jpj; M qjqj jqj; Nrjrj jrjg
2 3
cos q3  sin q3 0 ð25Þ
2 6 7
R 3 ¼ 4 sin q3 cos q3 05 ð18cÞ
The arrays of the first matrix are linear drag coefficients, and
0 0 1
those of the second matrix are quadratic drag ones.
2 3 To add this effect to the UVMS dynamics model, it is written in
cos qj  sin qj 0
6 joint space as
j1
Rj ¼ 4 0 0 17
5; j ¼ 4; 6 ð18dÞ
8 ( ) 9
 sin qj  cos qj 0 >
> J >
< Dðν Þ L;0 q_ > =
0
D0 ¼ J A;0 ð26Þ
So, the time derivative of these velocities respect to the body >
> >
>
:0 ;
fixed coordinate can be obtained as 121

i
V_ C;i ¼ i J L;i q€ þ i _J L;i q_ ð19Þ where index “0” denotes the vehicle.
For cylindrical links of the manipulators, an applicable formu-
i
ω_ C;i ¼ i J A;i q€ þ i _J A;i q_ ð20Þ lation is used. In this method, it is assumed that the drag
component along the main axis of a cylinder is negligible in
Now, the added mass effects can be expressed as external
comparison with the one perpendicular to this axis, which is
forces/moments in the task space as
8 9 8 9 computed as a function of the square of the velocity of the main
< i J L;i q€ þ i _J L;i q_ = < i J L;i q_ = axis points (McMillan et al., 1995).
F A;i ¼ M A;i þ C A;i ð21Þ If it is supposed that the x-axis of the local coordinate coincides
: i q€ þ i _J q_ ; : i J A;i q_ ;
J L;i A;i
with the cylinder axis (Fig. 6), the drag force has two components
As seen, the added mass effect has three separate parts in the along y and z axes which are written as
dynamics equation. The First part is a pure mass effect which can Z l
directly be added to the mass matrix, H. The second part is a f Dy ¼  ρC D;i r i ðV ry ðxÞÞ2 dx ð27Þ
0
velocity effect which does not exist in dynamic equation of a single
body and is due to the articulated nature of the UVMS and its
Z l
configurability. The third part is the centrifugal and Coriolis effects
f Dz ¼  ρC D;i r i ðV rz ðxÞÞ2 dx ð28Þ
of the added mass itself, C A;i . Using Eq. (21) and multiplying by the 0
Jacobian matrix of the application point, these separate parts
notifying with AM;i , AV;i and AC;i respectively, are written in the where V r ðxÞ for ith component is obtained with
joint space as
8 9
i
V ri ðxÞ ¼ i J L;i q_ þ ½i J A;i q_  ½ x 0 0 T ð29Þ
" #T " 0 #
J L;i Ri 033 < i J L;i =
AM;i ¼ M A;i i ð22Þ Now these drags can be considered as external forces which
J A;i 033 0 R : J ; i A;i must be incorporated in the UVMS model. To this end, they must
be transferred from the task space to the joint space the same as
" #T " # ( )
J L;i 0
Ri 033 i
J L;i q
_ what was done for the added mass effects. Therefore, the applica-
AV ;i ¼ M A;i i_
ð23Þ tion point of each force must be definite so that the Jacobian
J A;i 033 0
Ri J A;i q_
matrix of that point can be computed. To determine these points, a
" #T " # ( ) method the same as what is used for computing the mass center of
J L;i 0 033 i
J L;i q
_
AC;i ¼
Ri
C A;i ð24Þ bodies is applied. If the application points of y and z components
J A;i 033 0
Ri i
J A;i q_

As the Jacobian matrices are expressed in the base fixed


coordinate, it is necessary to change the expression of external
forces/moments from the local coordinate of ith component to
that of the base before multiplying by rotation matrices.

2.2.2. Hydrodynamic drag


There are many theories regarding the computation of hydro-
dynamic drag and lift forces/moments. The differences of these
theories are mainly in their initial assumptions about body shape,
velocity range and depth of motion. In this paper, two methods are
used for the vehicle and manipulators separately. The first one
which has been allocated to the drag of underwater vehicles with
low speed (Fossen, 1994) is used to compute the drag of the
vehicle (base). The second theory which has been developed for
cylindrical bodies (McMillan et al., 1995) is used for computing the
drag of the manipulators links. In both formulations, the lift force
can be ignored in comparison with the drag force.
Fossen (1994) has proposed the drag of low speed underwater
vehicles be the combination of two parts, linear and quadratic
drags, and be added to the dynamics equation of a 6 DOF body Fig. 6. Computation of the hydrodynamic drag acting on a cylinder.
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 87

are denoted as xy and xz , they can be obtained with


Z l Z l ,Z
l
f Dy xy ¼ xdf Dy ) xy ¼ ðV ry ðxÞÞ2 xdx ðV ry ðxÞÞ2 dx ð30Þ
0 0 0

Z Z ,Z
l l l
f Dz xz ¼ xdf Dz ) xz ¼ ðV rz ðxÞÞ2 xdx ðV rz ðxÞÞ2 dx ð31Þ
0 0 0

Now, if the Jacobian matrices of these application points are


shown by J Li ;x y and J Li ;x z , the drag effect of the ith link is written
in joint space as
Di ¼ ½J Li ;xy T 0 R i f Dy þ ½J Li ;xz T 0 R i f Dz ð32Þ

Presence of rotation matrices has the same reason as what Fig. 7. The difference between the amounts of q1 obtained with the proposed
explained in the added mass section. model and the one developed with the Newton–Euler approach.

2.2.3. Buoyancy and gravity proposed model by comparing it with another one developed
Gravity and buoyancy can be simply modeled because their by one of the most basic methods of dynamics modeling, i.e.
magnitude, direction and application point are exactly definite. If it Newton–Euler approach.
is supposed that the ith component of the UVMS is a homogenous
body, the sum of gravity and buoyancy forces can be written in the
task space as (Fossen, 1994) 3. MIC implementation
Gi þ Bi ¼ ðmi  ρw Δi ÞI g ð33Þ
The Multiple Impedance Control (MIC) strategy enforces the
in which I g is the gravity acceleration expressed in the task space, same impedance relationship at the end-effectors level and the
ρw is the water density and Δi is the volume of the ith body. manipulated object (Moosavian and Papadopoulos, 1998). In case
Denoted by β , the cumulative effect of the gravity and buoyancy in of mobile robots the same as the one considered in this paper,
joint space is obtained as since the cooperating arms are mounted on a moving base
(vehicle), the MIC also enforces the same impedance law on the
βi ¼  J TLi 0 R I I gðmi  ρw Δi Þ ð34Þ
vehicle so that all participating manipulators, vehicle and manipu-
Now, the complete dynamics equation of the UVMS can be lated object can exhibit the same impedance behavior, as implied
written as by “multiple” in naming the MIC. Among the best-known model-
H UV MS ðqÞq€ þ C UVMS ðq; q_ Þ ¼ Q ð35Þ based control methods, which focus on the impedance control for
object manipulation by cooperative manipulators, the MIC is really
in which integration and improvement of its predecessors that do not
n Nm impose the same impedance law on the whole parts of the system
H UV MS ¼ H þ AM;0 þ ∑ ∑ AM;i ð36Þ and might cause instability and/or damage to the robotic system
m¼1i¼1
and the environment.
n Nm Namely, the Impedance Control (Hogan, 1985), which enforces
C UV MS ¼ C þAV;0 þ AC;0 þ D0 þ β0 þ ∑ ∑ ðAV ;i þ AC;i þ Di þ β i Þ ð37Þ an impedance law only on end-effectors and does not take into
m¼1i¼1
account the dynamics of the object, cannot track the desired path,
In these relations, n is the number of manipulators, and N m is designed for the object, and in case of a collision with the
the number of the links of the mth manipulator. For the presented environment, it does not yield an accurate measure of the contact
UVMS, n is 2, and N m is 6 for both manipulators. force because the force sensors, which are usually installed on the
end-effectors, do not measure the inertial force of the object, and
2.3. Verification of the dynamics model what they feed to the controller as the contact force, significantly
differs from its real value by the amount of the object inertial
One of the efficient methods to verify a dynamics model is to force. This drawback is highlighted when we consider this fact that
obtain it with another method and compare the outputs of these the accelerations engendered in the object by the collision,
models in the same situation i.e., the same initial conditions and normally have high values and produce great inertial forces.
similar input forces/moments. To do this, the Newton–Euler Likewise, the Object Impedance Control (Schneider and
Recursive Algorithm (Craig, 1989) is selected and used for both Cannon, 1992), which imposes an impedance law only on the
kinematic chains which the UVMS is composed of i.e., from the object, does not take into account a predefined path for end-
vehicle to the end of each manipulator. By applying the forward effectors and cannot guarantee a satisfactory guidance of end-
and backward formulations to both kinematic chains, the relation effectors on appropriate paths and may show instability, specifi-
between the applied forces/moments and the UVMS dynamics is cally when the robotic system has some flexible parts (Moosavian
derived, and the model is completed. To compare this model and Papadopoulos, 1998). In addition, although this algorithm
with the proposed one, the same initial conditions are considered proposes a method to compute the contact force based on the
for both models, and the same forces/moments are applied as object dynamics, its plan for computing the object acceleration is
inputs. Then the differences between their outputs (generalized not satisfactory because it suggests the commanded (desired)
coordinates) are studied. As an instant, this comparison is illu- acceleration be used. In fact this value might be very different
strated in Fig. 7 for q1 . from the real value of the object acceleration when it collides with
As it is seen, the difference between each output of the the environment.
proposed model and its equivalence from the Newton–Euler The Augmented Object Model (AOM) (Chang et al., 2000),
model is actually zero. This fact confirms the correspondence of which enforces impedance behavior on the object with taking
the two developed models and verifies the correctness of the the dynamics of the whole system into account and mapping it at
88 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

the object position, might be considered an effective improvement in which F^ c is the estimation of the F c (Moosavian and
in the above mentioned methods. However, it has a serious Papadopoulos, 1998) and is computed as
limitation, i.e. it cannot be applied to mobile robotic systems. That _€
is why the MIC, which is capable of being used for mobile robots F^ c ¼ M X þF ω  F o  GF e ð47Þ
and allows coordinated motion/force control of the robotic system _€
where X is a finite difference approximation of the object accel-
in a manipulation task, has been chosen to be applied to a UVMS, eration and is obtained by
and its performance to be compared to that of the AOM.
In this section, following a brief review on the MIC theory, it is € X_ t  X_ t  Δt
_
X¼ ð48Þ
presented how to formulate and apply this method to the UVMS. Δt
in which Δt is the time step used in the simulation.
3.1. MIC formulation As through the MIC algorithm, the same impedance law is
imposed on the behavior of the vehicle, the manipulators and the
According to Moosavian and Papadopoulos (1998), by consider- manipulated object, the impedance law for the UVMS is written as
ing the dynamics equation of a robotic system in the task space as ~ des e~€ þ k~ d e~_ þ k~ p e~ þ U f F c ¼ 0N1
M c
ð49Þ
~ X€~ þ C~ ðq; q_ Þ ¼ Q~ ð38Þ  T
HðqÞ where U f c ¼ 166 ⋯ 166 is an N  6 matrix, and M des , k~ d
~
the vector of generalized controlling forces in the task space, Q~ , is and k~ p are N  N matrices defined based on M des , kd and kp ,
written as respectively (Moosavian and Papadopoulos, 1997). By Eq. (49) the
motion-concerned part of the applied controlling forces can be
Q~ ¼ Q~ app þ Q~ react ð39Þ written as
where Q~ react is the reaction force on the end-effectors, and Q~ app is ~ _
~  1 ½M € ~ _~ þ k~ p e~ þ U F^ c  þ C~
Q~ m ¼ H~ M des des X des þ k d e fc ð50Þ
the applied controlling force which is divided into two parts,
motion-concerned and force-concerned as The last part of the controlling forces is Q~ react which is obtained as
( )
Q~ app ¼ Q~ m þ Q~ f ð40Þ 061
~
Q react ¼ ð51Þ
 Fe
where Q~ m is the applied control force causing motion of the end-
effectors, while Q~ f is the required force to be applied on the in which
manipulated object by the end-effectors. To obtain proper expres-
sions for these terms, the equations of motion for the manipulated F e ¼ G# ½M X€ þ F ω ðF^ c þ F o Þ ð52Þ
object are considered as
Before applying the above equations to the UVMS, some prere-
M X€ þ F ω ¼ F c þ F o þ GF e ð41Þ quisites and assumptions about the manipulated object, grasp
condition and path planning of the end-effectors must be
where M is the mass matrix, X is the vector of generalized regarded.
coordinates describing the object position and orientation, F ω is
the vector of nonlinear velocity terms, F c describes the contact
3.2. Object dynamics equations
forces/moments, F o is the vector of external forces/moments and
F e is the vector of end-effectors forces/moments applied on the
As mentioned above, computation of F ereq , requires the
object. Also, G is the grasp matrix defined by (Moosavian and
dynamics model of the object. To this end, a method the same as
Papadopoulos, 1997)
2 3 what was done for the UVMS base is followed. The only difference
133 033 133 033 is in the drag formulation. By assuming a cylindrical shape for the
G ¼ 4 ST ½r ð1Þ  ST
⋯ ST ½rðnÞ 
obj e 33 STobj
5 ð42Þ object, the theory which was applied for the cylindrical links of the
obj e 33 obj
66n manipulators is used for computation of the drag force in its
where 133 and 033 denote the identity and zero matrices, dynamics equation. By this approach, the dynamics equation of
respectively. Sobj is a matrix which relates the object angular the object is written as
velocity to its Euler angles rates as ðM þ AM;obj ÞX€ þ C þAV;obj þ AC;obj þ Dobj þ Gobj ¼ τ ð53Þ
ωobj ¼ Sobj δ_ obj ð43Þ in which X is the vector of the generalized coordinate that includes
Also r ðiÞ object center of mass position in the inertial frame and its Euler
e is the position vector of the ith end-effector with
respect to the object center of mass. Now, using the robotic system angles. Undoubtedly, the equivalence of Eqs. (53) and (41) can be
dynamics model and the object dynamics equations, the MIC established easily.
general formulation can be developed. If a desired impedance
relationship for the object motion is chosen as 3.3. Grasp condition
M des e€ þ kd e_ þ kp e þF c ¼ 061 ð44Þ
The cylindrical object is supposed to be manipulated by the
where M des is the object desired mass matrix, e ¼ X des  X is the UVMS manipulators. As seen in Fig. 8, it is assumed that each end-
object position/orientation error vector, and kp and kd are gain effector grasps it rigidly, and their positions and orientations do
matrices, the force-concerned part of the applied controlling forces not change relative to the local coordinate of the object.
is obtained as
( )
061 3.4. Object path planning
Q~ f ¼ ð45Þ
F ereq
The task which the manipulators are supposed to do is the
where adjustment of a bar into a cylindrical slot in an undersea structure
as it is shown in Fig. 9. Based on the positions of the object and the
F ereq ¼ G# fMM des
1
ðM des X€ des þ kd e_ þ kp eþ F^ c Þ þ F ω  ðF^ c þ F o Þg ð46Þ slot, the equations of the object generalized coordinates on this
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 89

Fig. 8. The rigid grasp condition.

vehicle. Therefore, regarding the definition of the impedance


behavior, a predefined path for these parts motion is necessary.
As the manipulation task is assumed to be done in the work space
of the manipulators, it is reasonable to consider the vehicle to
remain stationary during the task. Therefore, the desired path of
the vehicle is considered as
ð0Þ
X~ des ¼ 061 ð55Þ

However, despite the vehicle, the end-effectors path planning


cannot be carried out easily. According to Moosavian and
Papadopoulos (1998), this path needs to be planned based on
the object geometry, grasp condition and the object planned path.
The grasp condition which is assumed in this paper causes
coupling between the object and the end-effectors positions and
Fig. 9. The desired path for the object. orientations. Thus, the end-effectors path planning has to be
carried out in some way so that the rigid grasp condition between
them and the object can be preserved.
path are considered as
Regarding the coordinates assigned to the manipulated object
X des ¼ ðx; y; z; ϕ; θ; ψ ÞTdes and the end-effectors, Fig. 10, the relation between their desired
positions is written as
¼ ð0; 2; 0:1  expð  tÞ; 0; 0; π =6 ðπ =3Þexpð  tÞÞT : ð54Þ
I
T EEdes ¼ I T objdes obj T EE ð56Þ

3.5. Path planning of the end-effectors and the vehicle where


2 3
X objdes
As mentioned in Section 3.1, the goal of the MIC strategy is to I 6 7
T objdes ¼ 4 I R objdes Y objdes 0 0 015 ð57Þ
apply a designated impedance behavior not only on the manipu-
Z objdes
lated object but also on the cooperating manipulators and the

Fig. 10. The coordinates assigned to the end-effectors and the object.
90 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

2 3
0 0 1 0 the manipulated object coordinates is considered as
6 0 1 0 0 7
ð1Þ 6 7 U ðiÞ
obj
T EE ¼6
6 1 0 0  0:5LG 7
7 ð58Þ X~ ¼ Robj _
eðiÞ X ð63Þ
4 0 0 0 1 5
in which
2 3 Robj  I R obj ½r ðiÞ 
0 0 1 0 eðiÞ ¼ ½ 133 e  Sobj  ð64Þ
6 0 1 0 0 7
obj ð2Þ
6 7 The mass and nonlinear parameters of the augmented object
T EE ¼ 6
6 1 0 0 0:5LG 7
7 ð59Þ
4 0 5 model are achieved as
0 0 1
2
~ ðiÞ obj
T
M  ¼ M þ AM;obj þ ∑ Robj
eðiÞ H ReðiÞ ð65Þ
In these transformation matrices, indexes of (1) and (2) denote the i¼1
number of the manipulator, and LG is the distance between two
contact points. The desired positions and orientations of the end- 2
~ ðiÞ
T
F  ¼ C þ AV;obj þAC;obj þ Dobj þ Gobj þ ∑ Robj
eðiÞ C ð66Þ
effectors are obtained by substituting the matrices in Eqs. (57)– i¼1
(59) in Eq. (56) as
In the AOM approach, this new model which represents both the
ð1Þ
X~ des ¼ ð  0:5; 2;  0:1 expð  tÞ; 0; π =6  ðπ =3Þexpð  tÞ; 0ÞT ð60Þ robotic system and the controlled object is used to develop control
laws. Now, by considering the impedance law which was
ð2Þ
X~ des ¼ ð0:5; 2;  0:1  expð  tÞ; 0; π =6  ðπ =3Þexpð tÞ; 0ÞT : ð61Þ described by Eq. (44), the required applied forces/moments from
the end-effectors to move the object are attained as

F ereq ¼ G# F e  ¼ G# fM  M des
1
ðM des X€ des þ K d e_ þ K p e þ F c Þ þF  g: ð67Þ
4. Augmented object model formulation

As Chang et al. (2000) have presented, this method is applic-


able only to fixed base robotic systems. This is another reason for 5. Simulation results and discussions
which the vehicle was supposed to be stationary when its desired
path was planned in the previous section. In this section, the applications of the MIC and AOM strategies
In this method, the dynamics of the whole system is somehow in the manipulation task are simulated, and the performances of
mapped into the object generalized coordinates. In fact, the these control algorithms are compared. The geometric and mass
dynamics of the manipulated object and the robotic system is properties and the hydrodynamic parameters of the UVMS1 and
expressed by the object state variables as (Zarafshan and the object are shown in Tables 1 and 2 respectively (Sarkar et al.,
Moosavian, 2010) 2002; Fossen, 1994; McMillan et al., 1995).
Also, the parameters of the RCC connected to manipulator #2
M  X€ þ F  ¼ GF e  ð62Þ are considered as ke ¼ 1200 ðN=mÞ; be ¼ 500ðN s=mÞ, and the
where M  is the mapped inertia matrix, and F  is the mapped maximum available actuators forces/torques are shown in
vector of velocity terms in the object generalized coordinates, and Tables 3 and 4.
In addition, the initial conditions of the UVMS and object, and
Table 1
the controller parameters are set as
The mass and geometric parameters of the UVMS and the object.
qT ¼ ð016 ; 0:28; 6:13;  2:9; 0:3; 0:96; 4:68;
Mass (kg) Dimensions I xx I yy I zz Δ 6; 6:13;  2:9; 5:98; 0:96; 4:74Þ
(m) (kg m2) (kg m2) (kg m2) (m3)

Vehicle 1073 a; c ¼ 0:5; b ¼ 1 536.5 536.5 214.6 1.047 q_ ¼ 0181


Link #1 32.2 L ¼ 0:5; r ¼ 0:1 0.7513 0.7513 0.16 0.0314 X T ¼ ð0; 1:9; 0:5; 0; 0; 0Þ
Link #2 20.6 L ¼ 1; r ¼ 0:08 0.066 1.75 1.75 0.02 X_ ¼ 061
Link #3 15.7 L ¼ 1; r ¼ 0:07 1.33 0.0385 1.33 0.0153 M des ¼ diagf50; 50; 50; 5; 5; 5g
Link #6 5 L ¼ 0:4; r ¼ 0:1 0.32 0.32 0.14 0.005
Object 50 L ¼ 1; r ¼ 0:1 1.26 1.26 0.24 0.032
kd ¼ diagf900; 900; 900; 200; 200; 200g
kp ¼ diagf180; 180; 180; 80; 80; 80g

Table 2
The hydrodynamic parameters of the UVMS and the object.

Added mass matrix (kg kg m2) Drag coefficients

Vehicle M A ¼ diagf653:8; 653:8; 315; 32:2; 32:2; 0g LD ¼ diagf0:9; 0:9; 0:9; 0:8; 0:8; 0:8g
Q D ¼ diagf1:05; 1:05; 1:05; 1; 1; 1g

Link #1 M A ¼ diagf15:7; 15:7; 3:22; 0:327; 0:327; 0g C D ¼ 1:1


Link #2 M A ¼ diagf2; 20; 20; 0; 1:675; 1:675g C D ¼ 1:1
Link #3 M A ¼ diagf15:4; 1:6; 15:4; 1:28; 0; 1:28g C D ¼ 1:1
Link #6 M A ¼ diagf5:23; 5:23; 0:6; 0:12; 0:12; 0g C D ¼ 1:1
Object M A ¼ diagf15:9; 15:9; 3:5; 0:335; 0:335; 0g C D ¼ 1:1

F e  is the total generalized forces/moments exerted on the object


by the end-effectors, and G is the grasp matrix which was defined 1
Due to their relatively small sizes and masses, the inertial and hydrodynamic
by Eq. (42). If the kinematics relation between the work space and properties of links #4 and #5 are ignored.
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 91

As it is illustrated in Fig. 8, the planned task is the pegging of a As it is seen, in the Z-direction, although the overshoot of the
cylindrical body into a cylindrical slot which belongs to an under- error by the MIC is smaller than that by the AOM, these errors
sea structure. The desired path related to this task was presented decrease with the same rates until the contact occurs at t  1ðsÞ.
in Section 3.4. However, in the middle of this path, the object Then these errors grow by a high rate to the value of the distance
collides with the flat surface beside the slot as it is shown in between the final desired value of the z-direction and the obstacle
Fig. 11. position. In fact, the object motion has been obstructed in this
The flat surface is at Z w ¼ 0, and it is assumed that no torque is direction to prevent the object and the environment from being
developed at the contact surface. The stiffness of this surface is damaged. However, the error in the Y-direction by the MIC
K w ¼ 1e5 N=m. decreases continuously while the one by the AOM shows unstable
In the following, the simulations results are shown, and behavior after 5 s.
the performance of the MIC is compared with that of the In order to more quantitatively compare the performances of
AOM with focusing on the tracking errors, consistent motion of these two methods, Integral of Squared Error (ISE) and Integral of
various parts of the system, the contact force and the actuating Time multiplied by Squared Error (ITSE), which are defined as
torques. (Ogata, 1993)
Z 1
5.1. Position and velocity errors ISE ¼ e2 dt ð68Þ
0

Fig. 12 indicates the position errors of the manipulated object, Z 1


obtained by applying the MIC and the AOM methods. ITSE ¼ te2 dt ð69Þ
0
Table 3
Saturation limits of the vehicle thrusters. and commonly used to measure the quality of system response
over the whole range of time, are taken into account. The values of
Thruster F top F bottom F right F left F rear F front these performance indices for the errors in Y and Z directions are
computed and shown in Table 5.
Saturation limit (N) 300 300 300 300 500 500
Clearly, the ISE which emphasizes transient errors has smaller
values by the MIC than by the AOM. This means that the MIC can
more effectively drive the initial errors toward zero. Also, the ITSE
Table 4 which heavily penalizes errors that occur late in time has smaller
Saturation limits of the actuators of the manipulators.
values by the MIC. This can absolutely be interpreted as the MIC
Actuator τ1 τ2 τ3 τ4 τ5 τ6 high effectiveness to keep errors near zero during the whole task
in comparison to the AOM.
Saturation limit (N m) 120 100 80 30 20 10
Table 5
The values of the ISE and ITSE indices of the object position errors in Y and Z
directions.

ISE ITSE

MIC AOM MIC AOM

eY 0.005 0.0075 0.0144 0.0611


eZ 0.1282 0.1287 1.090 1.091

Table 6
The values of the ISE and ITSE indices of the velocity errors in Y and Z directions.

ISE ITSE

MIC AOM MIC AOM

e_ Y 0.00254 0.00365 0.00057 0.00073


e_ Z 0.0058 0.00626 0.0047 0.005
Fig. 11. The object collision with an undersea structure.

Fig. 12. Object position errors: (a) MIC and (b) AOM.
92 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

Fig. 13. Object velocity errors: (a) MIC and (b) AOM.

Fig. 14. The errors of the pitch angles of the end-effectors: (a) MIC and (b) AOM.

In the same way, the velocity errors of the object and the values Although due to the vehicle large inertial effects, a few small
of the ISE and ITSE for these errors are illustrated in Fig. 13 and differences such as less spiky behavior in its velocity errors are
Table 6, respectively. observed, from an overall view, all parts participating in the
Besides this fact that the MIC provides less spiky behavior, the manipulation task have the same responses not only before
values of the performance indices confirm the MIC superiority the contact but also after it. This means that the main purpose
over the AOM in decreasing the velocity errors. One may assert in the design of the MIC has been achieved. In fact, enforcement
that these controllers have roughly the same performance in the of the same preset impedance on the different parts of the
Z direction, where the collision takes place, and the MIC gives system results in a consistent motion throughout the system like
better results only in the Y direction. However, it should be the motion of a vibration system in one of its natural mode
emphasized that by both controllers, the object motion has been shapes.
obstructed in the Z direction at the wall position, and there is no In contrast, the AOM cannot drive the system to have such
motion in this direction after the contact. That is why the behavior as it is seen in Fig. 16, where the position and velocity
performance indices have near values in the Z direction, speci- errors of the end-effectors and the object are illustrated. Definitely,
fically the ITSE that highlights long time errors, the time that the manipulator #2, which is equipped with a RCC, has errors
object stops at Z ¼ 0. (Fig. 16b), which are noticeably inconsistent with those of manip-
The end-effectors position errors have shown the same beha- ulator #1 and the object. Also, there is a tangible difference
vior as those of the object, so they are not shown and discussed. between the position errors of the manipulators and that of the
However, a very important difference in their orientation errors is object, specifically in the Z direction. This inconsistency is mainly
illustrated in Fig. 14. As seen, the MIC maintains the pitch angle (θ) caused by the fact that position and velocity errors of manipulators
error,2 near zero while it shows unstable behavior by the AOM. are not included in the control effort of the AOM, the important
Undoubtedly, this is a serious deficiency that challenges the characteristic that is incorporated in the architecture of the MIC and
capability of the AOM to carry out this task with reliability. makes it capable to provide the whole system with an accordant
motion.
The above mentioned comparison gives much more credit to
5.2. Consistent motion of various parts of the system the MIC when we consider this fact that this method has been
applied to a robotic system with a moving base while the AOM
The consistent motion of all of the components participating in has controlled a stationary system (this method can only be
the manipulation task is another desired result which is achieved applied to fixed-base robotic systems (Chang et al., 2000)) and
by the MIC. This characteristic is shown in Fig. 15 in which the has not been disturbed by the perturbations caused by the base
position and velocity errors of the vehicle, end-effectors and the motion.
object are presented. It is interesting to know how the MIC provides the vehicle with
The similarity of these graphs shows that the MIC has engen- a motion consistent with other parts of the system. Actually, the
dered an accordant motion in all components of the system. most involved vehicle thrusters are F rear and F f ront whose applied
forces are shown in Fig. 17. The interesting point is the difference
2
As the pitch angles of both manipulators have the same behavior, only the in the steady state values of these two thrusters. In contrast to
graph of one of them has been shown. F rear , F f ront has a non-zero final value of about 100 (N). This value is
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 93

Fig. 15. Position and velocity errors of the (a) vehicle, (b and c) first and second manipulators and (d) the object by the MIC.

caused by presence of the matrix U f c in Eq. (50). This matrix whose To analyze the contact forces in more detail, their power
existence is necessary to make the vehicle motion accordant to spectral density functions are shown in Fig. 19. It is seen that
other parts of the system causes a sense of contact to be induced the one obtained by the MIC has only two discrete frequencies
to the vehicle. To control the vehicle in presence of this virtual which although have different values from the roots of the
contact, the associated actuator exerts the required force. desired impedance behavior (λ1 ¼  0:2; λ2 ¼  17:8) due to the
presence of the RCC, they verify that this controller have success-
fully imposed the desired second order behavior on the robotic
5.3. The contact force and actuating torques system.
However, the contact force obtained by the AOM has a range of
By an overall analysis of the contact forces presented in Fig. 18, frequencies and does not satisfactorily follow the designated
it is seen that although both contact forces go smoothly to a impedance law. Surely, these high frequencies are why the contact
constant value, the MIC has a more satisfactory performance force is spikier and dissipate slowly by the AOM. In addition,
during the transient phase of the contact. In fact, there are three regarding the continuous nature of the robotic system and the
collisions with the underwater structure in this method, while this underwater structure which deserves a wide range of natural
number is six by the AOM. Also, the spiky behavior of the contact frequencies for both of them, the AOM is far more likely to cause
force dissipates more quickly by the MIC than by the AOM. damaging resonance than the MIC.
94 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

Fig. 16. Position and velocity errors of (a and b) the first and second manipulators and (c) the object by the AOM.

Fig. 17. Vehicle thrusters force by the MIC: (a) F rear and (b) F front .

The controlling torque3 applied by the actuators is shown to a steady state value. This value is not zero because the
in Fig. 20. Overall, these torques are almost the same in both actuators prepare the required torques to maintain the system
controllers. It is seen that the major part of the required torque is equilibrium while the wall pushes the object up with the
supplied by τ2 , τ3 and τ5 . Also, the actuators apply great torques constant force, F c .
at the beginning of the task and reduce them quickly to low However, some other observations such as the more consis-
values until the collision occurs. During the transient phase of tent distribution of loads to the arms by the MIC and the
the contact, they have spiky behavior, and then they smoothly go comparison of the consumed energy (this amount is about 26
(J) by the MIC while it is 53 (J) by the AOM) drive us to more
inquisitively compare the actuators performances. To do this, the
3
Since the values of τ1 , τ4 and τ6 are approximately zero during the task, their same procedure as what was done to analyze the contact forces
graphs have not been illustrated. is followed, and the power spectral density functions of the most
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 95

Fig. 18. The time history of contact force: (a) MIC and (b) AOM.

Fig. 19. The power spectral density function of the contact force: (a) MIC and (b) AOM.

Fig. 20. Torques applied by the actuators of the first and second manipulators: (a) MIC and (b) AOM.

involved actuating torques are shown in Fig. 21. Interestingly, it is 6. Effect of model uncertainties
seen that all of the torques provided by the MIC have the same
single frequencies while the ones by the AOM have different ranges In this section, a highly probable possibility in the underwater
of frequencies in addition to a single one which is similar to that by environment, uncertainty in the dynamics model, is considered
the MIC. This means that the Fourier extensions of the torques by and the efficiency of the MIC is examined in a more practical
the AOM have some additional terms with higher frequencies, and condition. The uncertainties in the UVMS model are considered as
although these frequencies have fairly low amplitudes, clearly, they 10% and 20% decrease respectively in the mass and hydrodynamic
are the main causes of more energy consumption and inconsistent properties relative to those of the real system. The simulation
load distribution between the arms. In addition, these frequencies results are shown in Fig. 22.
are other evidence verifying that the AOM has much more potential As it is seen, except a slower response for the vehicle, there are
for excitation of the natural frequencies of the robotic system that no other noticeable differences in the position errors of the object,
might lead to its resonance. end-effectors and the vehicle from those in the case of the
96 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

Fig. 21. The power spectral density functions of the most involved actuators torques: (a) MIC and (b) AOM.

Fig. 22. Position errors in presence of model uncertainties in the MIC: (a) vehicle; (b) object; (c) end-effector #1 and (d) end-effector #2.

complete accordance of the real system and its model. Also, the this vector replaces the τ in the robotic system equation of motion,

consistent motion of the object, end-effectors and vehicle is still H~ X~ þ C~ ¼ τ, as
achieved. This means that the MIC has the capability to be robust
€ 0
~ des X~€ des þ k~ d e~_ þ k~ p e~ þ U f F^ c  þ C~
~  1 ½M
0

in the presence of model uncertainties. This characteristic is due to H~ X~ þ C~ ¼ H~ M des c


ð71Þ
the nature of the impedance behavior induced on all components
If the right hand side of this equation is extended, it is written as
cooperating in the manipulation task. In fact, if the mass0 matrix0
and nonlinear velocity terms of the model are notified as H~ and C~ € 0€ 0 0
H~ X~ þ C~ ¼ H~ X~ des þ k~ d e_~ þ k~ p k^ p e~ þ Fc
0
^ þ C~
0
ð72Þ
respectively, Q~ m is written as 0 0 0 0 0
~  1 k~ d , k~ p ¼ H~ M
by defining k~ d ¼ H~ 0M ~  1 k~ p and F^ c ¼ H~ M
~  1 U f F^ c .
~ des X~€ des þ k~ d e_~ þ k~ p e~ þ U f F^ c  þ C~
0 0
~  1 ½M
Q~ m ¼ H~ M ð70Þ ~ €~
des des des c
des c Now, if a term of H X is added to and subtracted from the right
H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98 97

hand side of Eq. (72) as Chang, K., Holmberg, R., Khatib, O., 2000. The augmented object model: cooperative
manipulation and parallel mechanism dynamics. In: Proceedings of the IEEE
€ 0
€ 0
€ 0
€ 0 0 0
H~ X~ þ C~ ¼ H~ X~ des  H~ X~ þ H~ X~ þ k~ d e~ þ k~ p e~ þ F^ c þ C~ ð73Þ International Conference on Robotics and Automation, ICRA2000. San Fran-
cisco, CA, USA, pp. 470–475.
it can be rewritten as Christensen, L., Kampmann, P., Hildebrandt, M., Albiez, J., Kirchner, F., 2009.
Hardware ROV simulation facility for the evaluation of novel underwater
~ X€~ þ k~d e€~ þ k~p e~ þ F~c þ ðC~  C~ Þ ¼ 0
0 0 0 0 0 0
H~ e€~ þ ðH~  HÞ ð74Þ manipulation techniques. In: Proceedings of the MTS/IEEE Conference on
OCEANS, pp. 1–8.
If d is defined as Craig, J., 1989. Introduction to Robotics, Mechanics and Control. Addison Wesley,
Reading, MA, USA.
~ X€~ þ F~c þ ðC~  C~ Þ
0 0 0
d ¼ ðH~  HÞ ð75Þ Eslamy, M., Moosavian, S. Ali A., 2010. Dynamics and cooperative object manipula-
0 tion control of suspended mobile manipulators. J. Intell. Robot. Syst. 60 (2),
~ des H~  1 , it is rewritten as
by multiplying Eq. (74) by M 181–199.
Eslamy, M., Moosavian, S. Ali A., 2011. Dynamics modeling of suspended mobile
0
manipulators: an explicit approach with verification. Int. J. Model. Simul. 31 (2),
M ~ H~  1 d ¼ 0
~ e€~ þ k~ e_~ þ k~ p e~ þ M ð76Þ
des d des 5300–5323.
Evans, J., Redmond, P., Plakas, C., Hamilton, K., Lane, D., 2003. Autonomous docking
It is seen while the initially designated impedance law is pre- for intervention-AUVs using sonar and video-based real-time 3D pose estima-
served, the effect of the uncertainties is combined with the contact
0 tion. In: Proceedings of the MTS/IEEE Conference on OCEANS, vol. 4, pp. 2201–
force, and a new vector of disturbing forces, M ~ H~  1 d, is 2210.
des
Fernandez, J.J., Prats, M., Sanz, P.J., Garcia, J.C., Marin, R., Robinson, M., Ribas, D.,
generated. In other words, the uncertainties can be interpreted Ridao, P., 2013. Grasping for the eabed: developing a new underwater
as indefinite external forces/moments whose effect are merged robot arm for shallow-water intervention. IEEE Robot. Autom. Mag. 20 (4),
with that of the contact force and can be compensated by the 121–130.
Fossen, T.I., 1994. Guidance and Control of Ocean vehicles. Wiley, Chichester, UK.
impedance law the same as what is done in case of a collision with Hildebrandt, M., Albiez, J., Kirchner, F., 2008. Computer-based control of deep-sea
the environment. manipulators. In: Proceedings of the MTS/IEEE Conference on OCEANS, pp. 1–6.
Hogan, N., 1985. Impedance control: an approach to manipulation. ASME J. Dyn.
Syst. Meas. Control 107 (1), 1–24.
Hu, Y., Wang, L., Liang, J., Wang, T., 2010. Underwater box-pushing with multiple
7. Conclusion
vision-based autonomous robotic fish. In: Proceedings of the IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems, IROS2010. Taipei, Taiwan,
In this paper, moving an object to precisely peg in an under- pp. 4219–4224.
water structure was studied, while impacts due to contact are Hu, Y., Wang, L., Liang, J., Wang, T., 2011. Cooperative box-pushing with multiple
autonomous robotic fish in underwater environment. IET Control Theory Appl.
inevitable. To this end, first an explicit dynamics model of a dual 5 (17), 2015–2022.
arm UVMS was developed. Next, controller design for the manip- Kramer, M., 2013. Incredible Technology: How to Explore Jupiter's Moon Europa.
ulation of a known object by the UVMS arms was elaborated, and 〈http://www.space.com/22146-robotic-exploration-europa-incredible-tech.
html〉 (29.07.13).
the performance of proposed MIC method was compared to that of Kumagai, J., 2007. Swimming to Europa. 〈http://spectrum.ieee.org/computing/hard
the previously proposed AOM method. The obtained results ware/swimming-to-europa〉 (01.09.07).
showed that the MIC has better tracking errors, and in case of a Lee, D., Kim, G., Kim, D., Myung, H., Choi, H.T., 2012. Vision-based object detection
and tracking for autonomous navigation of underwater robots. Ocean Eng. 48
collision with the environment, it dissipates the energy of the (8), 59–68.
impact much effectively than the AOM. As shown by simulation, Lemburg, J., Kampmann, P., Kirchner, F., 2011. A small-scale actuator with passive-
an accordant motion of the manipulated object and the robotic compliance for a fine-manipulation deep-see manipulator. In: Proceedings of
the MTS/IEEE Conference on OCEANS, pp. 1–4.
system was achieved by the MIC while the AOM could not have Marani, G., Choi, S.K., Yuh, J., 2008. Underwater autonomous manipulation for
such a convenient behavior. Also, in the presence of uncertainties, intervention missions AUVs. Ocean Eng. 36 (1), 15–23.
tracking errors and contact force obtained with the proposed MIC McMillan, S., Orin, D.E., McGhee, R.B., 1995. Efficient dynamic simulation of an
underwater vehicle with a robotic manipulator. IEEE Trans. Syst. Man Cybern.
law did not have noticeable differences with respect to what were
25 (8), 1194–1206.
attained in the condition of an exact model without any uncer- Meng, Q., Wang, H., Li, P., Wang, L., He, Z., 2006. Dexterous underwater robot hand:
tainty. This characteristic is one of the main advantages of the HEU hand II. In: Proceedings of the IEEE International Conference on Mecha-
enforcement of the same impedance behaviors on all the parts tronics and Automation, ICMA2006. Luoyang, Henan, China, pp. 1477–1482.
Mohan, S., Kim, J., 2012. Indirect adaptive control of an autonomous underwater
participating in the manipulation task. In fact, any uncertainty in vehicle–manipulator system for underwater manipulation tasks. Ocean Eng. 54
the UVMS model is treated as a disturbing force/moment whose (23), 233–243.
effect can effectively be compensated with suitable tuning of the Moosavian, S. Ali A., Papadopoulos, E., 1997. On the control of space free-flyers
using multiple impedance control. In: Proceedings of the IEEE International
control parameters the same as what was achieved in the collision Conference on Robots and Automation. Albuquerque, NM, USA, pp. 853–858.
with the environment. So, the system under the proposed con- Moosavian, S. Ali A., Papadopoulos, E., 1998. Multiple impedance control for object
troller can efficiently accomplish the considered complex missions manipulation. In: Proceedings of the IEEE/RSJ International Conference on
Intelligent Robots and Systems. Victoria, Canada, pp. 461–466.
even in the presence of unpredictable disturbances such as Moosavian, S. Ali A., Papadopoulos, E., 2004. Explicit dynamics of space free-flyers
uncertainties and sudden collision with the unstructured under- with multiple manipulators via SPACEMAPLE. J. Adv. Robot. 18 (2), 223–244.
water environment. Therefore, the proposed controller can be Moosavian, S. Ali A., Papadopoulos, E., 2007. Free-flying robots in space: an
overview of dynamics modeling, planning and control. Robotica 25 (5),
considered as a promising algorithm for practical implementations 537–547.
in underwater robotic systems. Moosavian, S. Ali A., Rastegari, R., 2006. Multiple-arm space free-flying robots for
manipulating objects with force tracking restrictions. J. Robot. Auton. Syst. 54
(10), 779–788.
References Ogata, K., 1993. Modern Control Engineering. Prentice-Hall, Inc., Englewood Cliffs,
NJ.
Alipour, K., Moosavian, S. Ali A., 2011. How to ensure stable motion of suspended Prats, M., Garcia, J.C., Fernandez, J.J., Marin, R., Sanz, P.J., 2011. Advances in the
wheeled mobile robots. J. Ind. Robot 38 (2), 139–152. specification and execution of underwater autonomous manipulation tasks. In:
Angeletti, D., Bruzzone, G., Caccia, M., Cannata, G., Casalino, G., Reto, S., Veruggio, G., Proceedings of the MTS/IEEE Conference on OCEANS, pp. 1–5.
1998. AMADEUS: dual arm workcell for co-ordinated and dexterous manipula- Sanz, P.J., Ridao, P., Oliver, G., Melchiorri, C., Casalino, G., Silvestre, C., Petillot, Y.,
tion. In: Proceedings of the MTS/IEEE Conference on OCEANS, vol. 2, pp. 947– Turetta, A., 2010a. TRIDENT: a framework for autonomous underwater inter-
952. vention missions with dexterous manipulation capabilities. In: Proceedings
Bleicher, A., 2011. Gulf Spill One Year Later: Lessons for Robotics. 〈http://spectrum. of the IFAC symposium on Intelligent Autonomous Vehicles. Lecce, Italy,
ieee.org/robotics/industrial-robots/gulf-spill-one-year-later-lessons-for-robo pp. 187–192.
tics〉 (20.04.11). Sanz, P.J., Prats, M., Ridao, P., Ribas, D., Oliver, G., Ortiz, A., 2010b. Recent progress
Casalino, G., 2001. Dexterous underwater object manipulation via multirobot in the RAUVI project: a reconfigurable autonomous underwater vehicle
cooperating systems. In: Proceedings of the IEEE International Conference on for intervention. In: Proceedings of the ELMAR Symposium. Zadar, Croatia,
Robotics and Automation, ICRA2001. Seoul, Korea, pp. 3220–3225. pp. 471–474.
98 H. Farivarnejad, S.A.A. Moosavian / Ocean Engineering 89 (2014) 82–98

Sarkar, N., Podder, T.K., Antonelli, G., 2002. Fault-accommodating thruster force Wang, H., Huang, X., Hong, R., Fang, C.H., 2008. Force control strategy and test
allocation of an AUV considering thruster redundancy and saturation. IEEE research of underwater dexterous hand. In: Proceedings of the World Congress
Trans. Robot. Autom. 18 (2), 223–233. on Intelligent Control and Automation. Chongqing, China, pp. 8829–8833.
Schneider, S.A., Cannon, R.H., 1992. Object impedance control for cooperative Wang, H., Huang, X., Qi, X., Meng, Q., 2007. Development of underwater robot hand
manipulation: theory and experimental results. IEEE Trans. Robot. Autom. 8 and its finger tracking control. In: Proceedings of the IEEE Conference on
(3), 383–394. Automation and Logistics, ICAL2007. Jinan, Shandong, China, pp. 2973–2977.
Shao, J., Wang, L., Yu, J., 2006. Underwater transportation of multiple fish like robots Wang, H.H., Rock, S.M., Lees, M.J., 1995. Experiments in automatic retrieval of
using situation based action selection. In: Proceedings of the IEEE International underwater objects with an AUV. In: Proceedings of the MTS/IEEE Conference
Conference on Robotics and Automation, ICRA2006. Orlando, FL, USA, on OCEANS, vol. 1, pp. 366–373.
pp. 3208–3213. Yu, S.C., Yuh, J., Kim, J., 2013. Armless underwater manipulation using a small
Simetti, E., Casalino, G., Torelli, S., Sperinde, A., Turetta, A., 2014. Floating under- deployable agent vehicle connected by a smart cable. Ocean Eng. 70 (17),
water manipulation: developed control methodology and experimental valida- 149–159.
tion within the TRIDENT project. J. Field Robot. 31 (3), 364–385. Zarafshan, P., Moosavian, S. Ali A., 2010. Manipulation control of a space robot with
Sotiropoulos, P., Tosi, N., Andritsos, F., Geffard, F., 2012. Optimal docking pose and flexible solar panels. In: Proceedings of the IEEE/ASME International Confer-
tactile hook-localisation strategy for AUV intervention: the DIFIS deployment ence on Advanced Intelligent Mechatronics, AIM2010. Montréal, Canada,
case. Ocean Eng. 46 (5), 33–45. pp. 1099–1104.
Sun, Y.C., Cheah, C.C., 2004. Coordinated control of multiple cooperative under- Zhang, D., Wang, L., Yu, J., Tan, M., 2007. Coordinated transport by multiple
water vehicle–manipulator system holding a common load. In: Proceedings of biomimetic robotic fish in underwater environment. IEEE Trans. Control Syst.
the MTS/IEEE Conference on OCEANS, vol. 3, pp. 1542–1547. Technol. 15 (4), 658–671.

You might also like