Experimental Study on an Underwater

S.IK. Choi, G.Y. Takashige, and J. Yuh

Autonomous Systems Laboratory
Department of Mechanical Engineering
University of Hawaii at Manoa
Honolulu, Hawaii 96822. U.S.A.

Abstract computer in the autonomous mode or a ground computer in

the remote-tethered mode. The vehicle's mechanical &
As underwater application activities mature, the utilization of electrical architecture were designed to be easily expanded
underwater robotic vehicles becomes more notabile. The for additional components. ODINs specifications are shown
sophistication required by these URVs have significantly in table 1.
increased, and the development of AtJVs are becoming
imminent. The Autonomous Systems Laboratory of the
University of Hawaii has designed the Omni-Directional
Intelligent Navigator (ODIN), its control systems, and its
graphic workstation to develop an integrated, real-time, 3 -
dimensional graphic test platform.

1. Introduction

During the last decade, the development of underwater

robotic vehicles (URV) has been significant and rapid. These
vast strides in the industry are linked to the expansion of General Characteristics
applications - surveying, recovery, mining, maintenance, and
etc. - in underwater, nuclear, and other hazardous
environments. The use of many URVs are no longer limited
to the scientific and military fields; ancl the development
goals have shifted to more efficient and intelligenr vehicles Table 1: ODIN Specifications
for highly complex and sophisticated performance. These
demands have paved the road for enhancement of advanced
URV technologies which will eventually Isad to autonomous
underwater vehicles (AUV).

The design of an efficient and high performance vehicle and

its control system is a major obstacle in the development of
advanced URVs and AUVs. These activities a.re being
researched in the Autonomous Systems Laboratory of the
University of Hawaii. The design and development of
ODIN, its graphic workstation, and its control systems is a
step in the AUV growth process.

2. Vehicle Description

The configuration of ODIN is a closed-framed sphere with

eight thruster assemblies and a manipulator [ l ] as shown in
figure 1. This closed, spherical shaped vehicle coupled with
eight thrusters provide instantaneous, ornni-directional (6
degrees of freedom) prowess. The manipulator is only
capable of 1 dof motion; however, this is sufficient for most
tasks due to the omni-directional capability of the vehicle.
The vehicle can be controlled by ei:her an on-board Figure 1: Photo of ODIN with 4 verticai thrusters.
Funding f o r this research was provided by the National Science Foundation PYI Award Grant No. BCS9l-57896.

0-7803-1808-0194 $4.00 0 1994 lEEE

79

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.
The tethered and autonomous modes of ODIN both where Ui = WiIWil, Cti is a constant, and Wi is the propeller
communicated to the Silicon Graphics Personal Iris 4D/35TG angular velocity. The constant C t is often unknown or
based Integrated Graphic Workstation (IGW) via a 386 DOS- changes for forward and backward motions. The Ct can be
based laptop. Due to the mobility of the laptop, this
derived from results of a thruster experiment. The effect of
configuration allows for high versatility and flexibility. The
motor saturation is introduced by thruster force limits F T ~ ~ ~
on-board computer assembly consists of a Motorola 68030
CPU running at 32 MHz clock speed and an industry- and F T ~ The ~ .desired motion of the vehicle must be well
standard VME bus of size 3U (compact and light weight size) planned considering the thruster saturation limits.
is utilized for the communication architecture to maintain
compatibility with other electronic devices. A detailed 4. SimulationsJExperiments
description of ODIN and IGW can be found in previous
publications [2,3,4]. Both linear and nonlinear controllers were investigated for
O D I " altitude motion. Tests were conducted on ODIN in a
3. Vehicle Dynamics water tank for systems calibration as well as identification of
the dynamics. This section presents some of the initial
In this section, we briefly discuss underwater robot dynamics. experimental results of calibration, system identification, and
More detailed description can be seen in other references [SI. control as well as the simulation results for the proportional-
All external forces and torques can be consolidated into the derivative and adaptive controllers without an in-depth
rigid body equations of motion, then the vehicle dynamic theoretical description for the general audience of this
model can be described by: magazine. The theoretical description and discussion can be
found in detail in [6,7].
MV + C ( V , V , ) + G(X) = F (1)
4.1 Calibration
X = J(X)V
F = RF, Different experiments were performed to determine the
system parameters including the hydrodynamic coefficients
where V = [u,v,w,p,q,rIT is linear and angular velocities in and the Ct of thruster dynamics.
the vehicle coordinates, X = [X,Y,Z,Q,Y,FIT is the position
and orientation in global coordinates, J is a 6x6
transformation matrix, V is a fluid velocity vector, the 6x6
inertia matrix M = M (rigid body inertia) + Ma (added
l5 T
mass), C is a 6x6 matrix including all the nonlinear dynamic
terms with inertia velocity terms, terms associated with the
forces and torques exerted on the vehicle by fluid motion,
drag forces and torques, G is a vector of gravity and
buoyancy and a 6-dimensional vector F represents the forces
and torques generated by the thruster forces FT with a matrix
R that is given by the thrusters and the control surface
configuration. However, 0DI"s symmetric sphere structure Velocity (rads)
simplifies the vehicle equations.

ODIN utilizes a velocity-controlled thruster system. The Figure 2: Thruster Force vs. Motor Velocity and Curve Fitting
velocity-controlled thruster has a servo velocity feedback Results
loop, resulting in equation 4:
A single thruster experiment was set up to examine its
characteristics. Results show the linear relationship between
T,,Qi + Qi = U,i (4) the steady-state values of the input voltage and the motor
speed as expected due to its velocity servo-loop. However,
where the subscript i indicates the i-th thruster. U, is the figure 2 shows the nonlinear relationship between the steady-
servo velocity control input, and T , represents a time state values of the input voltage and the thruster force. It is
constant. The servo velocity loop is usually designed to have proportional to the motor speed with different coefficients for
a much smaller time constant than overall system's time positive and negative speed, as shown in equation 6.
constant, and consequently its dynamics could be ignored.
Thrust force FT of the i-th thruster is proportional to the
absolute square of angular velocity:
Results on figure 2, the two curve fittings to the non-linear
F, = C , U i data. show the values for C t as 6 . 8 3 5 ~ 1 0 -[ ~N / ( r a d / ~ ) for
~]

80

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.
the positive rpm (diving) and 2 . 3 3 9 ~ 1 0 -~~. N / ( r a d / ~for
) ~ ]the
negative rpm (surfacing). The Ct's calculated in this latest Time (sec)
experiment are more accurate than values which had been
presented in previous proceedings [3]. 0 4 8 12 16 20 24
0 1 1 1 11 1 1 1

ODIN's dynamic behavior was also observed by inputting 2

impulse, step, and square wave signals 'or thruster speed. 4
From measured input and output signals, ODIN's h
6
mathematical model was estimated as follows: 5 8
5 10
W = -0.2471 WI w - 0 . 0 5 +
~ 0.101FT + 0.0262 (7)
12
Z=W (8) 14
The results of the various inputs can be seen in figures 3a-d. 16
The dotted line indicates ODIN's motion, and the solid line 18
indicates the desired motion. ODIN's motions follows the
desired motion very closely. This is expected when the
cocfficients of the model are good approximations of the
actual values. Time (sec)
0 4 8 12 16 20 24 28 3 2 36 40
Time (sec)
0 ; ; ; ; ; ; ; I ; ; I
2
0 4 8 1216202428 3235404448 5256 4
h
0 6
2
5 8
h
4 6 10
G
v 6 12
8
9 10 14
4
n
12 16
14
16
18

Figure 3: System responses with different inputs: (a) no input,

(b) impulse input,(c) square-wave input, and (d) constant input.
Time (sec)
0 4 8 12 I 6 20 24 28 32 36 40

2
4
6
8
e 10 h
4
4 12 5 6
n 8
14
10
16
18
a" 12

:;2 f
18

Figure 4: Results with PD-controller

81

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.
This model was also confirmed by observing ODIN's operator. Defining R = C*, + A1 and A0 = Wd where W
performance of a proportional-derivative (PD) controller in a
and a constant d are arbitrary factors of AO, equation 10 will
water environment versus a computer simulation of the model
be satisfied if the predictor equation is
with the same PD controller. The derived dynamics and
controls equations as well as the experimental coefficient
values were used for the simulation programs. Identical C,(q-l)?(k+l) = R(k)V(k)+ W(k)d+Bl(k)U(k)
velocity trajectories were used for both the experimental and (1 1)
simulation cases. As shown in figure 4, the experimental
data (dotted line) of ODIN closely follows the simulation with the following PAA:
data (solid line) and converges on the desired depth.
6 ( k + 1) = 6(k) + F(k + l)@(k)ET(k+1) (12)
4.2 ODIN Control Schemes

The ultimate control objective for ODIN, in the autonomous

mode, is to have redundant control systems to monitor its
missions and provide modifications for its future missions.
The vehicle should have the capabilities of making h, (k)
adjustments to parametric changes during an operation. To where ,-,is used to denote an estimated value, a = -
A2 (k) '
achieve these goals, PD controllers to complex adaptive or
neural network controllers must be developed and tested. 0 < h , ( l ) 5 1 , O < h , ( l ) 5 2 , F(0) > O ,
Various control schemes have been studied [8-121 and most
were investigated with the aid of computer simulation GT(k)= [Bl(k)R(k)W(k)]and
without considering the effects of thruster dynamics. The
adaptive control system considering the effects of thruster
dynamics was designed for ODIN as described below.
QT (k) = [UT(k)VT(k)dT].

The control system is determined using a discrete-time The parameter values estimated by the predictor are then used
approximation of the URV dynamic model, equations I , 3 & at each time step to compute a control signal such that the
5 , which can be expressed by the following vector equation: output of the predictor follows the desired output:

V(k + 1) = A1 * V(k) + A0 + B1* U( k) (9) C,(q-')+(k + 1) = CR(q-')Vd(k + 1) (14)

where k is the k-th sampling time step, and V is the 6 - Substituting equation 14 into equation 11 and solving for
dimensional velocity vector of the URV. If the parameters in U(k):
the discrete-time model were known exactly, a conventional
digital control law could be determined using classical U(k) = Bl~'(k)[CR(q-')Vd(k+l)-R(k)V(k)-W(k)d]
methods. However, since the poorly known hydrodynamic
coefficients are included among the parameters of the (15)
dynamic model, a conventional control scheme cannot
From equations 4 & 5, the servo control input to the i-th
guarantee high performance in URV motion control.
thruster motor, U,i is computed by
Therefore, a Parameter Adaptation Algorithm (PAA) is
introduced to solve this problem. The PAA estimates the
parameters in the discrete-time model at each sampling time
step using input-output measurements from the URV. These
estimates are then used to adjust the controller gains to
provide the required control signals. In this section, a where Ui is the i-th component of U. To summarize, in the
discrete-time adaptive velocity controller for URVs is proposed control algorithm the parameters of the predictor
designed. The basis of this control algorithm is a linear are estimated at each time step using equations 12 and 13,
predictor that is designed with the aim that the prediction then the values of these estimated parameters are used to
error vanishes according to compute the control signal in equations 15 and 16.

lim { E @ + 1) = C,(q-')[V(k+ 1) - $(k+ l)]} = 0 (10) The derived dynamics and controls equations and the
k+m experimental coeffkients values were used for the simulation
programs. The desired output, Vd, was computed by
where $(k+l) is the predicted value of
V ( k + l ) , C,(q-') = l+q-'C; defines the regulation Vd = v, + P(Z, - Z)
dynamics, I is the identity matrix and 9-l is the unit delay

82

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.
 are desired velocity and position,
respectively, generated by the trapezoidal speed profile, and p
is a constant for position error compensation.
0 8 16 24 32 40 48 56 64 72 80
The PD controller and the adaptive controller were tested 0
with respect to vehicle payload changes. The I'D control 2
gains were determined based on values of the system
equation parameters in equations 7 and 8 and implemented
for the same system equations (case 1) and the system
equations with the additional mass of 0.79 kg (case 2). The
same adaptive controller was also implemented for two cases.
Results of case 1 are presented in figure 5 where (a) the
vehicle positions are shown by the dotted line for PD-
controller and the solid line for adaptive controller; and (b)
the vehicle position errors are shown by the dotted line for
PD-controller and the solid line for adaptive controller.
Results of case 2 are presented in figure 6 . It is obsmerved from tive Control f ation ion
figures 5 and 6 that the adaptive controller is needed for
robust position control when the vehicle carries different
payloads during the actual operation.
0.2
- 0
Time (sec) 5 -0.2
0 8 1 6 24 32 40 48 5 6 64 72 80 & -0.4
0
Lc -0.6
0 b
-0.8
2

-1.8
14
16
18
Figure 1
Iv
Figure 10a: PD vs. Adaptive Control § i ~ u ~ a ~ ~ Q n
(vehicle position with no payload)

0.05 5. Discussion

Design and control of an omni-directional underwater robotic

vehicle (QDIN) and its integrated graphic simulation
.I 8 platform, have been described with some initial experimental
-0.05
results. The design of this vehicle was motivated by the
W
following factors with reference to existing state-of-art
E -0.1
0 vehicles: (1) the lack of low-cost vehicle test-beds for
.-
.C
c)
v1
developing advanced vehicle technologies; (2) rigid designs
-0.15 for different tasks and requirements; (3) inefficient and
2 =
--- ineffective operating procedures; and (4) poor performance in
-0.2 terms of position accuracy and response times. The flexible
and redundant design of the vehicle's mechanical and
e (sec) electrical architecture allows easy adaptation to any changes
and modifications required to implement different tasks or to
vs. Adaptive Contrtol ~ ~ ~ ~ ~ test
a different
~ ~ vehiclc
o nconfigurations.
(vehicle position error with no ,payload)

83

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.
We are currently conducting experiments with the adaptive
control system presented in this paper. Experimental results
will be presented in the near future. Some other theoretical
studies conducted at ASL include: adaptive coordinated
motion control system of vehicle and manipulator [8]:
learning control systems [ 131; real-time 3D graphic vehicle
test-bed [2]; underwater object recognition & shape recovery
[14]; and redundant vehicle control systems [4].

References
Choi, S.K., "Underwater Robotic Vehicle Design".
Research Proposal, Mechanical Engineering Department.
University of Hawaii, Honolulu, HI, Dec. 1991.
Yuh, J., Adivi, V., and Choi, S.K., "Development of a 3D
Graphic Test Platform for Underwater Robotic Vehicles".
1992 ISOPE Conference.
Choi, S.K. and Yuh, J., "Design of Advanced Underwater
Robotic Vehicle and Graphic Workstation", 1993 IEEE
Conference on Robotics and Automation.
Yuh, J., Choi, S.K., and Takashige, G., "Qmni-Directional
Intelligent Navigator", 1994 International Symposium on
Robotics and Manufacturing.
Yoerger, D., Cooke, J.G., and Slotine. J.E., "The Influence
of Thruster Dynamics on Underwater Vehicle Behavior
and Their Incxporation Into Control System Design,"
IEEE J. of Oceanic Engr., Vol. OE-15, No. 3, pp. 167-
178, 1990.
Yuh, J., "Modeling and Control of Underwater Robotic
Vehicles," IEEE Trans. Sys., Man and Cyber., Vol. 20.
No. 6, 1990.
Choi, S.K., Takashige, G.Y., and Yuh, J.. "Control of an
Qmni-Directional Intelligent Navigator," to be submitted
to IEEE Control Systems Technology, 1994
Mahesh, H., Yuh, J. and Lakshimi, R.. "A Coordinated
Control of Underwater Vehicle and Robotic Manipulator".
J. of Robotic Systzms, June 1991.
Yoerger, D.N. and Slotine, J.E., "Robust Trajectory
Control of Underwater Vehicles," IEEE J. of Ocean:
Engineering, Vol. QE-10, No. 4, pp. 462-470, 1985.
Goheen, K.R. and Jefferys, E.R., "Multivariable Self-
tuning Autopilots for Autonomous and Remotely Qperated
LJnderwater Vehicles," IEEE J . of Oceanic Engr. Vol. 15.
No. 3, pp. 144-151, 1990.
Cristi, R., Papoulias, F.A., and Healey. A.J., "Adaptive
Sliding Mode Control of Autonomous Underwater
Vehicles i n the Dive Plane," IEEE J . of Oceanic
Engineering, V. 15, No. 3, pp. 462-470, 1991 ,
Yuh, J. and Lakshimi, R., "An Intelligent Control System
for Remotely Operated Vehicles", J. of IEEE Oceanic
Engineering (1992).
Yuh, J., "Learning Control for Underwater Robotic
Vehicles," IEEE Control Systems Magazine, April, 1994.
Kolongani, N. and Fox, J.S., "Photometric Stereo Using
Point Light Sources", 1992 IEEE International Conference
on Robotics and Automation.

84

Authorized licensed use limited to: TUFTS UNIV. Downloaded on October 12,2024 at 14:16:37 UTC from IEEE Xplore. Restrictions apply.