Journal of Electrical Engineering 4 (2016) 133-149 doi: 10.17265/2328-2223/2016.03.004

Depth and Altitude Control of an AUV Using Buoyancy

Control Device

Mahdi CHOYEKH1, Naomi KATO1, Ryan DEWANTARA1, Hidetaka SENGA1 and Hajime CHIBA2

1. Department of Naval Architecture and Ocean Engineering, Osaka University, Suita, Osaka 565-0871, Japan

2. Department of Merchant Marine, Toyama National College of Technology, Imizu, Toyama 933-0293, Japan

Abstract: A new method for depth control was developed for a spilled oil and blow out gas tracking autonomous buoy robot called SOTAB-I by adjusting its buoyancy control device. It is aimed to work for any target depth. The new method relies on buoyancy variation model with a depth that was established based on experimental data. The depth controller was verified at sea experiments in the Toyama Bay in Japan and showed good performance. The method could further be adapted to altitude control by combining the altitude data measured from bottom tracking through a progressive depth control. The method was verified at the sea experiments in Toyama in March 2016 and showed that the algorithm succeeded to bring the robot to the target altitude.

Key words: AUV, depth control, buoyancy device.

1. Introduction

Oil spills produced by accidents from oil tankers and

blowouts of oil and gas from offshore platforms cause

tremendous damage to the environment as well as to

marine and human life [1]. To prevent oil and gas that

are accidentally released from deep water from

spreading and causing further damage to the

environment over time, early detection and monitoring

systems can be deployed to the area where underwater

releases of the oil and gas first occurred. Monitoring

systems can provide a rapid inspection of the area by

detecting chemical substances and collecting

oceanographic data necessary for enhancing the

accuracy of simulation of behavior of oil and gas.

Due to their compactness, the use of AUVs for

full-water surveying is being adopted increasingly [2,

3]. Among the existing types of underwater robots used

to autonomously monitor marine environments in 3-D

space from sea surface to seabed over the long term is

the Argo Float [4] that floats vertically and repeats

descending and ascending in the vertical direction by

using a buoyancy control device. However, it does not

Corresponding author: Mahdi CHOYEKH, Ph.D. student, research fields: naval architecture and ocean engineering.

have a function of active movement in the horizontal

direction. Another method is the underwater glider [5],

which has a streamlined body with fixed wings. It can

descend and ascend also by using a buoyancy control

device, while it moves in the horizontal plane like a

glider for long distance. However, the ratio of vertical

movement distance to horizontal movement distance is

small. An AUV (autonomous underwater vehicle)

called SOTAB-I (the spilled oil and gas tracking

autonomous buoy system), that provides functionality

that lies midway between profiling buoys and gliders,

is being developed. It was designed not only to move in

the vertical direction by a buoyancy control device, but

also in the horizontal direction by two pairs of

rotational fins. SOTAB-I can perform on-site

measurements of oceanographic data as well as

dissolved chemical substances using underwater mass

spectrometry [6]. SOTAB-I has three main surveying

modes. At the first stage, SOTAB-I performs the water

column survey by adjusting its buoyancy. The rough

mode is used to collect rough data on physical and

chemical characteristics of plumes by repeating

descending and ascending on an imaginary circular

cylinder centered at the blowout position of oil and gas

D DAVID PUBLISHING

Depth and Altitude Control of an AUV Using Buoyancy Control Device

134

through the variation of buoyancy and movable wings’

angles. Finally, in case the UMS detects a high

concentration of any particular substance, a precise

guidance mode will be conducted to track and survey

its detailed characteristics by repeating descending and

ascending within the plume. The photograph mode

enables us to have a large visual overview of the area

around the blowout position of oil and gas by taking

pictures of the seabed and making image mosaicking.

SOTAB-I moves laterally using horizontal thrusters

along diagonal lines of a polygon with a radius of 5 m

centered on the blowout position of oil and gas.

Therefore, the depth control is a very important task in

the surveying effort which requires particular attention.

There are several methods used to control the depth

of AUVs. With normal horizontal type AUVs, the

depth control is performed by horizontal wings. For the

vertical type AUVs, a buoyancy control device is

normally used for depth control. There exist a variety

of mechanisms to adjust the buoyancy. In the

submarines for example, the amount of the air/seawater

of trim or ballast tanks is controlled to adjust the

buoyancy. When the submarine is on the surface, air is

filled in the ballast and the submarine becomes

positively buoyant. To start diving, water is introduced

into the ballast tanks while the air is vented outside

until it becomes negatively buoyant leading the

submarine to sink. Compressed air is stored in flasks to

adjust the amount of water inside the ballast tank

during operation. Another widely employed

mechanism in AUVs is to adjust the volume of the

robot through a device that compresses and expands the

air contained in a cylinder. This mechanism is

characterized by its reliability and its relative fast

response. However, the motor pump, used to ensure the

compression and the expansion of the air, generates

noise. Additionally, when the robot is decreasing its

buoyancy, an amount of the ballast water (seawater) is

pulled from one region and then, may be dumped in

another region to increase buoyancy. This represents a

risk of dumping living organisms in an environment

different from their original inhabiting region, which

may harm their new environment. This problem is

referred as the ballast water problem [7, 8]. Among

other existing technologies, there is the metal bellow

mechanism, which imitates the change of state of the

spermaceti oil from liquid to solid and vice versa,

leading to change of density, in the sperm whale [9].

Similarly, AUV using metal bellow mechanism relies

on the change of state of a low melting point liquid,

such as wax [10] or oil [11], by adjusting its

temperature. This mechanism does not make noise and

presents an ecological advantage over other systems

since it does not involve any discharging of ballast

materials, eliminating the ballast water problem.

However, results show that their response time is slow

and is energetically costly since the temperature of the

oil should be maintained. A third mechanism imitates

ray-finned fish, which adjust the volume of their

bladders to adjust their buoyancy [12]. They employ

polymer buoyancy control device [13]. Electrolysis is

used to generate pure hydrogen, which is a clean gas, in

order to expand the volume of an artificial bladder

leading to a displacement of water and an increase of

buoyancy. To reduce the robot buoyancy, extra amount

of gases are simply released outside via a valve. These

systems are characterized by their silent operation.

However, they are more oriented for small devices

operating near the sea surface where the water pressure

is not significant. Due to the reliability and the fast

response as well as their low power consumption, the

buoyancy variation through the adjustment of the air

volume in a cylinder mechanism was employed. The

ballast water problem does not apply for SOTAB-I

since it is designed to operate around the same region

of the blow out gas. The noise caused by the motor

pump may be reduced by choosing a high quality

actuator.

For the same control mechanism, there exist several

control strategies. An implementation of a cascaded

velocity-position PID controller was used in a coastal

profiling float by Ref. [14]. The method consists of

Depth and Altitude Control of an AUV Using Buoyancy Control Device

135

adjusting the velocity set point according to depth error

between the current and target depths. The vertical

velocity is controlled through a PID controller to

achieve the desired depth. The algorithm succeeded to

achieve the desired depth near the sea surface, but at a

high energy cost. Another control strategy is employed

in the underwater gliders where the buoyancy control

device is performed simultaneously with a mechanism

of gravity center movement in horizontal plane. On the

other hand, Argo float uses only buoyancy device to

adjust their depth. To do so, the float relies on the

establishment of a highly accurate ballasting curve [15].

This requires a high precision ballasting experiment to

adjust the robot’s density in a way to become equal to

the density of the seawater, which will be measured by

a highly accurate CTD sensor, at the designated

parking depth. This will lead the robot to reach its

neutral buoyancy point.

There are many challenges and constraints

associated with depth control of underwater vehicle.

For instance, at-sea experiments require enormous

financial and logistic resources limiting the

experiments time. Hence, it is important that the

program should be easy to implement and repeatedly

verified by simulating programs before its real

deployment. On the other hand, environmental

constraints like a considerable variation of the density

of water between the sea surface and the seabed bring

complications in the control because they lead to the

variation of the neutral buoyancy value of the robot.

Even if the neutral buoyancy of the robot is determined

accurately at a certain spatial condition, there is no

guarantee that the robot will keep its vertical position

due to the up-welling and down-welling water currents.

Other constraints are represented by the hardware

limitations. In fact, the buoyancy device employed has

three controlling states: it can be controlled to increase

the buoyancy, decrease it or stay idle. However, it is

not possible to change the rate of variation directly. In

addition, the rate of change of buoyancy is relatively

slow, not symmetric in both directions and varies with

depth. Moreover, the change of the buoyancy variation

orientation is not instantaneous, there is a lag time of 2

s between each change of state. The oil level sensor has

also an inaccuracy within ±0.1%. Previously, a PID

controller was developed for depth control [16]. It gave

good performances and small overshoot, but only for a

depth range up to 100 m. Beyond that limit, significant

overshoot was reported. The previous controller relied

on a very accurate determination of the neutral

buoyancy. In addition, the PID control parameters were

not adaptive. Besides, it does not enable to freeze the

robot at the target depth. For the lacks mentioned

before, it is necessary to develop a new controller that

overcomes the shortening and take in consideration the

environmental and hardware constraints. A new

method for depth control was developed. It is aimed to

work for any target depth and to freeze the robot at the

target depth. The method relies mainly on the

buoyancy variation model with depth established based

on tank and at-sea experimental data. The paper is

organized as follows: Section 2 gives an overview of

SOTAB-I hardware and its buoyancy device. Section 3

studies the buoyancy variation at tank and sea

experiments and establishes its model. Section 4 and 5

deal, respectively, with the depth control and the

altitude control algorithms and the experimental results

obtained of their execution in Toyama Bay

experiments in March 2016. The final section gives the

conclusions of the work.

2. SOTAB-I Overview and Hardware Description

2.1 Outlines of SOTAB-I

The SOTAB-I is 2.5 m long and weighs 325 kg. It

can be submerged in water as deep as 2,000 m. It is able

to descend and ascend by adjusting its buoyancy using

a buoyancy control device while changing its

orientation through two pairs of movable wings. The

SOTAB-I can also move in horizontal and vertical

directions using two pairs of horizontal and vertical

thrusters. A visual overview of SOTAB-I is illustrated

136

in Fig. 1, and

Table 1. Th

installed on

When the

(wireless loc

communicat

transmission

the mothers

through the

The robot

GPS (global

determine th

where the ro

the USBL (

position of t

depth data f

within the bo

DVL (Dopp

robot’s veloc

Fig. 2 Arran

Dep

d its main cha

he arrangem

SOTAB-I is

e robot floats

cal area netw

tion transceiv

n. When the ro

ship and the

acoustic mod

t tracking on t

l positioning s

he robot’s ab

obot is subm

(ultra-short b

the robot in t

from the CT

ottom trackin

pler velocity

cities. The ro

ngement of dev

pth and Altitu

aracteristics a

ment of devi

shown in Fig

on the sea s

work) and an

ver module a

obot is underw

SOTAB-I c

dem.

the sea surfac

system) recei

bsolute posit

merged, tracki

baseline) syst

the water col

D sensor. W

ng altitude fro

logger) is

bot motion an

vices and senso

ude Control o

are summarize

ces and sen

g. 2.

urface, a WL

n iridium sate

are used for

water, the use

can communi

ce is ensured

iver that serve

tion. In the

ing is ensured

tem. The ver

lumn is given

When the robo

om the seabed

able to mea

nd orientation

ors installed on

of an AUV Us

ed in

nsors

LAN

ellite

data

er on

icate

by a

es to

case

d by

rtical

n by

ot is

d, the

asure

n are

give

mea

Fig.

Tab

Tot

Dia

We

We

n SOTAB-I.

ing Buoyancy

en by the

asurement un

. 1 SOTAB-I

ble 1 Principa

tal length [mm]

ameter [mm]

eight in air [kg]

eight in water [k

y Control Dev

compass a

nit).

robot.

al particulars o

] 2,

66

3

kg] ±

vice

and the IM

of SOTAB-I.

,503

67

11.7

3.8

MU (inertiall

Fig. 3 Buoy

An ADCP

and orientat

also fitted w

and physical

obtain a vis

of gas on th

camera.

2.2 Buoyanc

In the bu

pump inject

bladder and

serves to au

brake is use

during the

external pres

the extractio

external pr

In total, six

buoyancy de

power suppl

and another

actuator, one

one specifie

activate/deac

by two digi

and one anal

Fig. 4 dep

in the reser

reservoir. T

Dep

ancy device.

P is employe

tion of water

with a UMS to

l properties o

sual represent

he seabed, th

cy Device

uoyancy cont

s and extract

the internal

utomate open

d to lock the

injection of

ssure of 20 M

on of oil fro

ressure cond

digital input

evice. A digi

ly relay. One

r to close it.

e input is use

s the rotation

ctivate the br

tal outputs th

log output tha

picts the relat

rvoir and th

The “pressur

pth and Altitu

ed to measur

r current laye

o determine th

f the dissolve

tation of blow

he robot is e

trol device,

ts oil between

oil reservoir

ing and closi

pump (Fig.

oil into the

MPa is 243 mL

om the blad

dition, it is

s are employ

tal input serv

input is used

To control

d to run/disab

n direction. On

rake. The feed

hat report th

at provides th

tionship betw

he volume o

re in the re

ude Control o

re the magni

ers. SOTAB

he characteri

ed gas and oil

wouts of plu

equipped wit

an oil hydra

n the externa

r. A motor v

ing cycles, an

3). The flow

e bladder at

L/min, and du

dder at the s

s 349 mL/m

yed to control

ves to control

to open the v

the motor pu

ble it and ano

ne more serve

dback is prov

e valve posit

he oil level.

ween the pres

f the oil in

eservoir” is

of an AUV Us

tude

-I is

stics

l. To

umes

th a

aulic

al oil

valve

nd a

rate

the

uring

same

min.

l the

l the

valve

ump

other

es to

vided

tion,

ssure

the

the

pres

rese

pres

fitte

volt

the

the

vari

Wh

V. W

1.62

and

volt

the

the

mas

volu

typi

Fig.in th

ing Buoyancy

ssure applied

ervoir cylind

ssure of the p

ed with a line

tage is an im

pump drain p

oil volume

iation of rob

hen the oil roo

When the oil

208 V. Becau

d the surface

tage of the po

volume of th

linear volu

ss change ca

ume change w

ically equal to

. 4 Relationshhe oil reservoir

y Control Dev

d by the hy

der. The ma

pump is 0.03

ear potentiom

mage of the p

pressure. The

of the reser

bot buoyancy

om is full, the

l room is em

use the reserv

e of the base

otentiometer

he oil in the

ume/output v

an be obtain

with the dens

o 1,024 kg/m

hip between thr.

vice

ydraulic pum

aximum allo

MPa. The oi

meter whose a

piston displac

e voltage is pr

rvoir, which

y, as shown

e output volta

mpty, the outp

voir has a cylin

e is constant

can be used

reservoir, wh

voltage relati

ned after mu

sity of seawa

m3.

he pressure an

137

mp to the oil

owable drain

il reservoir is

analog output

cement under

roportional to

leads to the

in Table 2.

age is 0.3892

put voltage is

ndrical shape

t, the output

to determine

hich explains

ionship. The

ultiplying the

ater, which is

nd the volume

7

l

n

s

t

r

o

e

.

2

s

e

t

e

s

e

e

s

e

138

Table 2 Vamotor.

Mass change (seawater) (g)7,884.8

7,746.6

7,763.2

3,925.7

171.8

0

3. Establis

The objec

model and

needed for t

estimate the

buoyancy fr

buoyancy m

program. It

buoyancy va

period. We

buoyancy v

water depth

at pressure t

3.1 Experim

3.1.1 Pres

Pressure

calculate the

oil in the

direction is

extracts oil f

to the intern

direction.

Based on

change the b

OUT>IN dir

water depth

modeled as

between the

and 20 MPa

direction, th

from 0 up to

Dep

riation of rob

) Volume change (cm3) 7,700.0

7,565.0

7,581.2

3,833.7

167.8

0

shment of t

ctive in this

a buoyancy

the depth con

e time needed

rom its curren

model is also

enables to e

alue from its

consider es

ariation from

based on the

ank and at-se

ments Results o

ssure Tank Ex

tank exper

e time necess

reservoir in

when the oi

from the exte

nal oil reserv

n Fig. 5 and

buoyancy cor

rection is con

h). From 10

a linear fu

full scale va

a is less than

he buoyancy v

o 10 MPa. B

pth and Altitu

bot buoyancy

Potentiometeoutput voltag1.621

1.599

1.602

1.012

0.443

0.389

the Buoyan

section is to

model. The

ntrol of the ro

d for SOTAB

nt value to a

o needed fo

estimate the

s initial value

stablishing a

m 20 to 85%

e experiments

ea.

of Buoyancy

xperiments

riments were

sary for chan

n both direc

l hydraulic p

ernal oil blad

oir. IN->OUT

Table 3, the

rresponding t

nstant till 10

MPa to 20

unction. The

ariation of buo

n 9 minutes. F

variation tim

Beyond that l

ude Control o

against voltag

er ge (V)

Pressur(MPa)0

0

0.005

0.018

0.030

0.041

ncy Model

establish a t

e time mode

obot. It enable

B-I to chang

target value.

r the simula

variation of

e every samp

a model for

% up to 1,000

s results obta

Variation

e performed

nging 7,500 c

tions. OUT-

pump injects

dder and injec

T is the oppo

e time neede

o 7,500 cc in

MPa (~1,00

Mpa, it can

time differe

oyancy at 0 M

For the IN>O

me is almost s

limit, it beco

of an AUV Us

ge on

re

time

el is

es to

e its

The

ating

f the

pling

the

0 m

ained

d to

cc of

->IN

and

cts it

osite

ed to

n the

00 m

n be

ence

MPa

OUT

same

omes

slig

3

F

tank

0 to

the

fun

of M

vari

m a

vari

deg

3.2

B

vari

line

In

Fig.buo

Tabpres

IN>

OU

ing Buoyancy

ghtly faster.

3.1.2 At-sea E

Fig. 6a confir

k. The buoya

o 100 m. Hen

OUT>IN di

ction. Fig. 6b

March 2015 i

iation in the I

and 700 m w

iation can be

gree polynom

Model of the

Based on th

iation in the O

ear function.

Tc (OU

n the IN>OU

. 5 Relatioyancy variatio

ble 3 Time ssure.

ExternpressurMPa

>OUT

0

10

20

UT>IN

0

10

20

y Control Dev

Experiments

rms the result

ancy variation

nce, the time

irection can

b dates from a

in Toyama Ba

IN>OUT dire

water depths.

e represented

ial function a

e Buoyancy Va

he experimen

OUT>IN dire

UT->IN) = –1

UT direction,

onship betweeon time.

of variation

nal re

Motor rotaspeed rpm

7,865

6,700

5,760

7,865

7,865

8,300

vice

ts obtained in

n rate is almo

e and buoyan

be modeled

an experimen

ay. It shows t

ection from 20

. It shows th

under the fo

as shown in T

Variation with

ntal data, th

ection can be

15.122 * (Bt –

, the main pa

en the press

of robot bu

ation Flow rate

T

cc/min m

331 2

282 2

243 3

331 2

331 2

349 2

n the pressure

st same from

cy models in

d as a linear

nt on the 20th

the buoyancy

0 to 85% at 1

hat buoyancy

form of a 3rd

Table 4.

Depth

he buoyancy

modeled as a

– B) (1)

arameter that

sure and the

uoyancy with

Time

min/7,500 cc

22.6

26.6

30.9

22.6

22.6

21.5

e

m

n

r

h

y

y

d

y

a

)

t

e

h

Fig. 6 Buoyexperiment.

Table 4 Buexperiment.

Direction

OUT>IN

IN>OUT

contributes c

variation spe

and written

function that

Table 5 Com

Experiment 5/25/2015 2:28:43 11/27/2014 9:20:56 11/28/2014 9:48:23

3/20/2015 14:12:33

Dep

(a)

(b) ancy variation

oyancy variat

Depth B

1 to 700 m T

1 m T

700 m T(B

considerably

ed is the depth

under the fo

t depends on

mparison betw

Depth

Air

0~95 m 700 m 0~42 m 0~100 m

95~155 m

155~198 m

198~210 m

pth and Altitu

OUT>IN

IN>OUT

n with depth on

tion model of

Buoyancy variat

Tc = –15.122 * (

Tc = 15.122 * (BTc = 0.0015 * (BBt

2 – B2) + 34.0

in the change

h. The model c

orm of 3rd de

the depth D.

ween the buoya

OrientationOUT>IN IN>OUT OUT>IN IN>OUT OUT>IN IN>OUT

IN>OUT

IN>OUT

IN>OUT

ude Control o

n 20th March

20th March

tion

(Bt – B)

Bt – B) Bt

3 – B3) – 0.26065 * (Bt – B)

e of the buoya

can be general

egree polynom

ancy variation

n Range 95%220%79%220%78%31%

20%

30%4

40%4

of an AUV Us

2015

2015

88 *

ancy

lized

mial

L

coe

buo

esta

foll

Ci (

K

A

A

3.3

and

D

wer

SOT

mod

defi

dev

buo

resu

T

vari

exp

whe

The

exc

sug

dire

the

model and exp

of variation 21% 94% 20% 85% 31% 79%

30%

40%

49%

ing Buoyancy

Tc (IN

C2(D) *

Linear interp

efficients a,

oyancy model

ablished for

lowing equati

(D) for i = {1

Ci (D) = (C

Knowing that

At 700 m C3 =

At 1 m C3 = 0

Comparison

d Experimenta

Data of buoy

re collected

TAB-I and co

del. To estim

fined the rati

viation of the

oyancy variat

ults.

Table 5 sho

iation is cl

periments. Th

en the buoyan

e time estimat

eed 0.5 s per

ggesting a h

ection. On the

maximum va

perimental resu

Tc (experimen1,121 s 1,166 s 859 s 1,274 s 703 s 854 s

178 s

176 s

170 s

y Control Dev

N->OUT) = C

(Bt2 – B2) +

polation and

b and c are

l at a certain d

depths equa

ion shows the

, 2, 3}

Ci (1) – Ci (70

= 0.0015; C2

0; C2 = 0; C1 =

n between Bu

al Data

yancy variatio

from previou

ompared to th

mate the acc

io Time/Ran

e model in s

ion. Table 5

ws the estim

lose to the

he maximum

ncy variation

tion in the OU

1% of buoyan

high accuracy

e other hand, i

alues of devia

ults.

nt) Tc (mod1,119 s 1,164 s 892 s (501,340 s (710s 823 s (50199 s (15184 s (95192 s (15193 s (19169 s (19

vice

3(D) * (Bt3 –

C1(D) * (Bt –

d extrapolat

e used to de

depth based o

al to 1m and

e formula used

00)) * D / (70

= 0.2688; C1

= 15.122.

uoyancy Vari

on time in se

us at-sea exp

he values obt

curacy of the

nge to estim

seconds for e

summarizes

mated time o

values ob

m deviation w

is between 20

UT>IN direct

ncy variation

y of the mo

in the IN->OU

ations were ob

el) Time–0.03–0.03

0 m) +0.56(700 m) +1.01

+0.150 m) –0.6555 m) 5 m)

+2.1 +0.6

55 m) 98 m)

+1.6 +1.5

98 m) –0.11

139

B3) –

– B) (2)

tion of the

etermine the

on the models

d 700m. The

d to calculate

00 – 1) (3)

= 34.065

iation Model

everal ranges

periments of

tained by the

e model, we

ate the time

every 1% of

the obtained

of buoyancy

tained from

was obtained

0% and 30%.

tion does not

n at full range,

odel in that

UT direction,

btained in the

e/range (s/1%)3 3 6 1 5 5

1

9

)

e

e

s

e

e

)

l

s

f

e

e

e

f

d

y

m

d

.

t

,

t

,

e

140

20 to 30%

variation of

model as rel

the model a

model with

depths whic

extrapolation

4. Depth C

4.1 Control A

Dependin

main scenar

robot reache

ascending, l

bring the r

there until

acoustic com

Therefore, w

into two mai

to bring the

target depth

Once the ta

executed to

In the nex

controller an

4.1.1 Pred

As shown

parameters

experiment.

neutral buoy

around it (B

random erro

data (Derr) o

data collecte

Once th

configuratio

executed eve

sensors’ dat

CTD sensor,

measured by

Dep

range, but d

buoyancy. S

liable. It is im

accuracy can

h additional

ch will reduce

n error.

Control

Algorithm

ng on the op

rios to be con

es the target

like in the ro

robot to the

an ascendin

mmunication

we can deco

in steps. As sh

robot from

h (Dt) using a

arget depth is

stabilize it ar

xt part we ex

nd the depth s

dictive Contro

n in Fig. 8, w

of the progr

For instance

yancy (Bn) w

Bm). We also

or of the oil s

of the CTD s

ed in the prev

he program

on is over,

ery second. A

ta, such as th

, and the valu

y a linear po

pth and Altitu

did not excee

o we can con

mportant here

be improved

experiment

e the model

perating mod

nsidered. The

depth and su

ough mode. T

target dept

g order is r

or ascending

ompose the

hown in Fig.

its current de

a predictive

s reached, th

round the targ

xplain in deta

stabilization a

ol

we introduce t

ram at the b

e, the estima

with the marg

o input the v

sensor (Berr)

sensor, based

vious experim

m is exe

the predicti

At first the rob

he depth (D),

ue of the curre

otentiometer

ude Control o

ed 2.1 s per

nsider the ov

e to mention

d by feeding

data at var

interpolation

de, there are

e first is that

ubsequently s

The second i

th and freez

received thro

g timer overf

control prog

7, the first ste

epth (D) to a

depth contro

he second ste

get depth.

ails the predic

algorithm.

the configura

beginning of

ated value of

gin of uncerta

value of sen

and of the d

d on the sens

ments.

ecuted and

ve controlle

bot updates al

provided by

ent buoyancy

image of the

of an AUV Us

1%

erall

that

g the

rious

and

two

t the

tarts

is to

ze it

ough

flow.

gram

ep is

a set

oller.

ep is

ctive

ation

f the

f the

ainty

nsors

depth

sors’

its

er is

ll the

y the

(B),

e oil

leve

such

F

outp

rela

A

of t

Eq.

Fig.

Fig.

Fig.

ing Buoyancy

el. Other data

h as the verti

Fig. 9 show

puts diagram

ated to the pre

At every seco

the time need

(4).

. 7 Depth con

. 8 Predictive

. 9 Predictive

y Control Dev

a can be deriv

cal speed (S)

s the predic

and Table 6

edictive contr

ond, it is poss

ded to reach t

ntrol process.

e depth control

e depth control

vice

ved based on

).

ctive control

defines all th

roller.

ible to have a

the target dep

l flowchart.

l input/output

the raw data,

l inputs and

he parameters

an estimation

pth (Tr) using

diagram.

,

d

s

n

g

Depth and Altitude Control of an AUV Using Buoyancy Control Device

141

Table 6 Definition of the buoyancy control parameters.

Symbol Definition

D Current depth (m). D > 0

Dt Target depth (m)

Dm Margin of tolerance around Dt

S Current speed (m/s). S > 0 Robot descending

Sm Speed margin

B Current buoyancy (%). Range: 20->95%

Bt Output target buoyancy

Bn Neutral buoyancy

Bm Margin of tolerance around the neutral buoyancy Bn

Tc Time needed to change the buoyancy from B to Bt in (s)

Tr Time needed to reach the target depth based on the current speed of the robot

Tm Time margin used for security purpose. It compensates eventual inaccuracy in the buoyancy model

Tr = ((Dt – D) / S) (4)

The buoyancy variation model, established in the

previous section of this paper, enables to estimate the

time (Tc) needed for changing the robot buoyancy from

its current value to the neutral buoyancy. The first step

is based on the continuous estimation of the time to

reach (Tr) and the time to change (Tc) while decreasing

the buoyancy value, till the stop condition is reached.

We introduce Tm, which corresponds to the time error

margin used to compensate eventual inaccuracies in the

buoyancy device model.

If the estimation of Tr > Tc + Tm it means that it is

possible to increase the vertical speed of the robot since

we have enough time to change the buoyancy to its

neutral level. Hence, we decrease the target buoyancy

value.

In the case where Tr <= Tc + Tm, then it means we

have just enough time to change the buoyancy to the

neutral level before the robot reaches its target depth.

Hence, we start to increase the buoyancy of the robot

progressively.

4.1.2 Depth Stabilization

Several algorithms can be used for depth

stabilization. Among the most used are the PID

controllers. However, one of the drawbacks of these

controllers is that they require the actuators to operate

at full time, which increases the power consumption. In

addition, in our robot’s case, the buoyancy variation

speed is not constant and varies with depth.

Furthermore, its variation is not symmetric in both

IN->OUT and OUT->IN directions. For that reason, a

conventional PID controller is not suitable, which

requires the development of an asymmetric PID

controller that adapts its parameters with the robot’s

depth. This will add a lot of complexity to the program

and requires a longer time to implement it and to

validate its performance. For that reason, we chose to

use a heuristic controller for depth stabilization. The

latter provides a simple way to control the depth. It is

based on heuristic rules that enable to adjust the

buoyancy based on the current depth and vertical speed

of the robot. If we take the case where the robot depth

(D) is below the target depth (Dt), we can establish the

following rules, to be executed by priority order:

If the robot is below the maximum tolerated depth

(Dt + Dm), then increase the buoyancy;

If the robot is below Dt and it is descending, then

increase buoyancy;

If the robot is below Dt, but it is ascending fast

above a speed margin (Sm), then decrease buoyancy;

If the robot is below the target depth, and it’s

ascending slowly below (Sm), then the buoyancy

actuator is idle.

Fig. 10 shows the flowchart of all the algorithm. It is

important to mention here that the buoyancy variation

values are limited between Bn + Bm and Bn – Bm.

142

Fig. 10 Heu

Table 7 Par

General param

Target depth

Neutral buoya

Buoyancy ma

Buoyancy con

Buoyancy dev

Predictive con

Time margin

Depth margin

Depth stabiliz

Vertical speed

Depth thresho

Stabilization

4.2 Experim

The dept

Toyama Ba

ordered to re

stay there fo

surface. Hen

3 main step

predictive d

(Step 1). The

heuristic con

for 15 minut

set to its ma

surface (Ste

parameters w

Fig. 11 sh

be observed

depth and fr

Dep

uristic control f

rameter config

meters

(Dt)

ancy (Bn)

argin (Bm)

ntrol threshold

vice accuracy (

ntrol parameter

(Tm)

n (Dm)

zation paramete

d threshold (Sm

old

period

ments Results

th control ex

ay on 17th M

each a set tar

or 15 minute

nce, the contro

ps: In the fi

depth contro

en the depth st

ntrol is execu

tes (Step 2). F

aximum value

ep 3). For th

were configur

hows the resu

d, the robot

freeze there fo

pth and Altitu

flowchart.

uration.

(Bdiff)

(Berr)

rs

ers

m)

xperiment w

March 2016.

rget depth equ

es before asce

ol program ca

irst step, the

l to reach t

tabilization al

uted to freeze

Finally, the ta

e to bring the

at purpose, t

red as shown

ult of the expe

managed to

for 15 minute

ude Control o

Value

500 m

62.5%

2.0%

0.2%

0.05%

Value

20 s

0.5 m

Value

0.02 m/s

5 m

900 s

was conducted

. The robot

ual to 500 m

ending to the

an be divided

e robot uses

the target d

lgorithm using

e the robot d

arget buoyanc

e robot to the

the depth con

n in Table 7.

eriment. As it

reach the ta

es before star

of an AUV Us

d in

was

m and

e sea

into

the

depth

g the

depth

cy is

e sea

ntrol

t can

arget

rting

the

F

spe

Fig.

Fig.

ing Buoyancy

ascent.

Fig. 12a show

ed with dep

. 11 Depth co

. 12 Predictiv

y Control Dev

ws the variati

th under the

ontrol.

(a) Depth V

(b) Depth Vs B

(c) Tc V

ve depth contro

vice

ion of the rob

e effect of th

Vs Speed

Buoyancy

Vs Tr

ol.

bot’s vertical

he buoyancy

l

y

control show

depth (TrDt)

neutral buoy

latter param

be increased

in Fig. 7. It

reach the tar

at a buoyan

algorithm su

speed based

TrDt.

Fig. 13 sh

algorithm.

maintained

tolerance ar

addition, the

robot’s verti

threshold).

In additio

and not 62

program suc

depth.

Fig. 14 s

speed was eq

vertical spee

As a con

succeeded to

overshoot at

a vertical spe

stabilization

the target de

vertical spe

proved its r

buoyancy w

buoyancy, th

depth withou

5. Altitude

There are

SOTAB-I to

photograph

Dep

wn in Fig. 12b

) and the tim

yancy (TcBn)

meters define w

d or decreased

t can be obs

rget depth wi

ncy value ne

ucceeded to

d on the com

hows the res

It shows

the robot s

round the tar

e control pro

ical speed to

n, though the

.5% as set

cceeded in co

shows the ro

qual to 0.4 m/

ed is due to th

nclusion, the

o bring the ro

t a buoyancy

eed close to 0

n algorithm m

epth at a limit

eed. Further

robustness: T

was slightly d

he robot man

ut problem.

e Control

e several op

o get close to t

mode, SOTA

pth and Altitu

b. The time to

me needed t

) are shown

whether the b

d as explained

served that ro

ith a vertical

ear the neut

balance the

mpromise be

sult of the de

that the c

peed within

rget depth eq

ogram succee

o less than 5

e real neutral b

in the progr

ontrolling the

obot’s ascent

/s. The sudde

he change of th

e predictive

bot to the targ

value equal t

0. On the othe

managed to ke

ted overshoot

rmore, the c

Though the e

different from

naged to reach

perating mod

the seabed. F

AB-I needs

ude Control o

o reach the ta

to change to

in Fig. 12c.

buoyancy sho

d in the flowc

obot manage

speed near 0

tral. The con

robot’s ver

etween TcBn

epth stabiliza

control prog

the interva

qual to 5 m

eded to limit

cm/s (check

buoyancy wa

ram, the con

e robot at the

t. The maxim

en variation o

he wings’ ang

control prog

get depth with

to the neutral

er hand, the d

eep the robot

t and a very sm

control prog

estimated neu

m the real neu

h the exact ta

des that req

or example in

to approach

of an AUV Us

arget

o the

The

ould

chart

ed to

m/s

ntrol

rtical

and

ation

gram

al of

m. In

t the

k Sm

as 64%

ntrol

e set

mum

f the

gles.

gram

hout

l and

depth

near

mall

gram

utral

utral

arget

quire

n the

h the

seab

pos

Fig.

Fig.

ing Buoyancy

bed to be abl

ition. Simila

. 13 Depth sta

. 14 Robot’s a

y Control Dev

le to take nea

arly, in the wa

(a) Depth V

(b) Depth Vs B

abilization.

(a) Depth V

(b) Depth Vs B

ascent.

vice

at pictures of

ater column m

Vs Speed

Buoyancy

Vs Speed

Buoyancy

143

the blow out

measurement

3

t

t

144

mode, the ro

to near the se

Hence, it i

controls the

vertical thru

is a risk that

which influ

addition, the

inaccuracies

overcome th

suggest a se

device as an

5.1 Control A

The meth

control algo

altitude cont

robot is onl

seabed when

set target al

range.

Fig. 15 i

control algo

depth contr

control exec

altitude.

Fig. 15 Altit

Dep

obot is require

ea bed to obta

s important

altitude of th

sters to contro

t they mix up

uences the tr

ey will disturb

s in the wat

he previously

econd method

actuator to co

Algorithm

hod consists o

orithm detaile

trol. It is imp

ly able to m

n the bottom

ltitude shoul

illustrates the

orithm. It con

rol followed

cuted to kee

tude control flo

pth and Altitu

ed to dive fro

ain a full wate

to develop

he robot. One

ol the altitude

p the sedimen

ransparency o

b the water flo

ter current m

y mentioned

d that only us

ontrol the alti

of the adapta

ed in the pre

portant to rem

measure the a

tracking is a

d be within

e flow chart

nsists of 3 sta

by an altitu

p the robot

owchart.

ude Control o

m the sea sur

er column pro

a program

way is to use

e. However, t

nts on the sea

of the water

ow, causing s

measurement.

weaknesses,

ses the buoya

itude of the ro

ation of the d

evious sectio

mind here that

altitude from

active. Hence

bottom track

t of the alti

ages of predic

ude stabiliza

at the set ta

of an AUV Us

rface

ofile.

that

e the

there

abed

r. In

ome

. To

, we

ancy

obot.

depth

on to

t the

m the

, the

king

itude

ctive

ation

arget

I

unk

clos

that

be

gua

buo

spe

exp

exp

wat

of a

con

reac

F

S

dep

“Ce

min

on b

the

At f

buo

than

star

to r

set

con

incr

buo

rob

leve

dep

valu

whe

the

S

the

calc

ing Buoyancy

f we conside

known, then t

se to the seab

t the robot can

set near the

arantee that th

oyancy on tim

ed which bec

periment tim

periments, it is

ter depth from

a safe approx

nsiderably to

ch the target a

Following are

Step 1: The r

pth of the robo

ertain depth”

nimum water

board before

depth contro

first, the buoy

oyancy. After

n the neutral

rt diving. The

reduce its buo

as the minim

ntrol device,

rease again it

oyancy level.

ot should hav

el to its neutr

pth. In this ste

ue as shown i

ere L0 = 2.04

DVL sensors

Step 2: When

variable tar

culated as fol

y Control Dev

er the case w

there is a pos

bed at any m

n stop descen

neutral buoy

he robot will b

me. This has a

comes very s

me and its

s possible to g

m the GPS po

ximation of th

the reductio

altitude.

e the altitude c

robot dives w

ot reaches the

”. The “Certa

depth value.

starting the

ol with time e

yancy contro

r the buoyan

buoyancy of

e buoyancy co

oyancy level d

mum buoyanc

with maxim

ts buoyancy

The purpose

ve enough tim

ral buoyancy

ep, the target

in the followi

Dt = Dcert

m is the dista

s, At is the tar

n the robot is n

rget depth c

llows:

Dt = DCert – L

vice

where the wa

sibility that t

moment. Hen

nding the buoy

yancy value

be able to reac

direct impact

slow and then

cost. Howev

get a rough es

osition. The d

he water dept

on of the tim

control steps

with a fast sp

e depth limit

ain depth” is

. Dcert is inpu

descent. In th

estimation sch

ol device will

ncy level bec

f the robot, S

ontrol device

down up to 2

cy level of t

mum speed.

level close t

of this strate

me to change

when reachi

depth (Dt) is

ing equation:

– L0 – At

ance between

rget altitude.

near the certai

control is st

L0 + Amax – A

ater depth is

he robot gets

ce, to ensure

yancy should

in a way to

ch the neutral

t on the robot

n extends the

ver, in real

stimate of the

determination

h contributes

me needed to

:

peed until the

(Dcert) of the

s the certain

ut by the user

his first step,

heme is used.

decrease the

comes lower

OTAB-I will

will continue

0%, which is

the buoyancy

Then it will

o the neutral

egy is that the

its buoyancy

ing the target

set at a fixed

(5)

the CTD and

in depth limit

tarted. Dt is

At (6)

s

s

e

d

o

l

t

e

l

e

n

s

o

e

e

n

r

,

.

e

r

l

e

s

y

l

l

e

y

t

d

)

d

t,

s

)

After pass

that the DV

current altitu

limited up t

depth. In t

estimation is

(Dt) is set eq

range Amax

depth will c

robot decre

variable targ

of SOTAB-I

Therefore, t

buoyancy le

when reachin

result, the ro

Step 3: W

water depth

depth D m

measured by

(L0) between

Once the

to transform

control using

The wate

depth D mea

by DVL. Th

control with

Step 4: W

target depth

depth contr

control with

control prog

around 1

mechanism o

within the t

which has

reaches zero

Dep

sing the certa

VL will dete

ude (A). Ther

o the time (T

this step, th

s still being us

qual to the cu

minus the ta

continuously

ases. Hence,

get depth. At

I is already c

there will n

evel to ensure

ng the target d

obot will dive

When the DV

(Dw) can be

measured by

y DVL taking

n the two sen

Dw

water depth i

m the altitude

g the followin

Dt =

er depth Dw i

asured by CTD

is step is also

h time estimat

When the rob

h plus or min

rol method i

h time estim

gram. The de

m as a co

of the buoyan

target depth

been set on

o, the robot w

pth and Altitu

in zone limit,

ct the seabe

refore, the bu

Tr) needed to

he depth co

sed. However

urrent depth (

arget altitude

change as th

, it is a dep

this point, th

close to the n

not be much

e that the rob

depth, as show

e at a steady s

VL detects the

e calculated a

CTD and t

g in considera

nsors as shown

= D + L0 + A

is known, it b

control to an

ng equation:

= Dw – L0 – A

is defined as

D and the alti

o carried out b

tion scheme.

bot is within

nus the depth

is switched

mation to de

epth margin D

ompensation

ncy device. SO

for a certain

the timer.

will start ascen

ude Control o

, there is a cha

d and outpu

uoyancy chang

o reach the ta

ontrol with t

r, the target d

D) plus the D

(At). The ta

he depth D of

pth control w

he buoyancy l

neutral buoya

h change in

bot is able to

wn in step 2. A

speed.

e seabed, the

as the sum of

the altitude

ation the dista

n next.

A

becomes poss

equivalent d

At

s the sum of

itude A measu

by using the d

the range of

h margin Dm,

from the d

epth stabiliza

Dm is usually

in the con

OTAB-I will

n period of t

When the ti

nding.

of an AUV Us

ance

ut its

ge is

arget

time

depth

DVL

arget

f the

with

level

ancy.

the

stop

As a

e sea

f the

(A)

ance

(7)

sible

depth

(8)

f the

ured

depth

f the

, the

depth

ation

y set

ntrol

stay

ime,

imer

5.2

In

targ

min

wat

was

at le

con

F

con

reac

befo

F

dep

was

exp

the

F

Tab

Cer

Tar

Asc

BT

Dm

Der

Sm

Bn

Bm

Bdif

Berr

Tm

Fig.

ing Buoyancy

Experiments

n this experim

get altitude e

nutes before a

ter depth at th

s unknown, b

east equal to

ntrol were set

Fig. 16 shows

ntrol. It can b

ch near the s

fore ascending

From the DV

pth which was

s composed

plained in Sec

robot ascent.

Fig. 17 illustr

ble 8 Altitude

rtain depth

rget Alt.

cending timer

T min range

m

rr

ff

r

. 16 Altitude

y Control Dev

Results

ment, the rob

equal to 9 m

ascending to t

he place wher

but the water d

724 m. The

as shown in

s the experim

be observed t

seabed and fr

g.

VL data, we

s equal to 766

of 5 steps. T

ction 1. The

.

rates the detai

e control param

72

9

30

24

0.

0.

0.

62

2%

0.

0.

20

control.

vice

bot was order

m then freeze

the sea-surfac

re the robot w

depth was est

parameters o

Table 8.

ment result of

that the robot

reeze there fo

could measu

6 m. The cont

The first fou

fifth step co

ils of the pre

meters configu

24 m

m

00 s

4

.5 m

.007 m

.02 m/s

2.5%

%

.2%

.05%

0 s

145

red to go to a

e there for 5

ce. The exact

was launched

timated to be

of the altitude

f the altitude

t managed to

for 5 minutes

ure the water

trol algorithm

ur steps were

orresponds to

dictive depth

uration.

5

a

5

t

d

e

e

e

o

s

r

m

e

o

h

146

Fig. 17 Altit

control appli

713 m, whic

that the certa

the target al

that the robo

20% at a m

and maintain

the time of

robot’s verti

and the buo

below the m

buoyancy (B

Dep

(a) Dep

(b) Depth

(c)

tude control: S

ied in step 1 w

ch can be calc

ain depth is d

ltitude is equ

ot reached the

maximum vert

ned it for 17

the experime

ical speed was

oyancy was e

maximum va

Bn_Max = Bn +

pth and Altitu

pth Vs Speed

h Vs Buoyancy

Tc Vs Tr Step 1.

with a set targ

ulated using E

defined as equ

al to 9 m. It

e minimum b

tical speed eq

5 s, which e

ent. At the en

s reduced to l

equal to 61%

alue of the e

Bm = 62.5%

ude Control o

get depth equ

Eq. (10) know

ual to 724 m

can be obser

uoyancy equ

qual to 0.49

enabled to red

nd of step 1,

less than 0.15

%, which is 3

estimated neu

+ 2% = 64.5

of an AUV Us

ual to

wing

m and

rved

al to

m/s

duce

, the

m/s

3.5%

utral

%).

Fig.

At t

cha

the

F

with

did

be s

altit

con

con

A

and

ing Buoyancy

. 18 Altitude

that buoyancy

ange to reach

robot detects

Fig. 18 illustra

h variable tar

not detect the

sure that the

tude. In this

nstant. As a

nstant and equ

At 742 m wat

d the 3rd step

y Control Dev

(a) Depth V

(b) Depth Vs B

(c) Tc V

control: Step 2

y value, the b

the neutral b

s the seabed.

ates the result

rget depth. At

e seabed, it de

robot will be

s step, the ro

consequence

ual to 0.14 m/

ter depth, the

was activate

vice

Vs Speed

Buoyancy

Vs Tr

2.

buoyancy dev

buoyancy on

ts of the predi

t this step, sin

escends slowl

e able to stop

obot buoyanc

, the robot s

/s.

robot detecte

ed. The target

vice is able to

time in case

ictive control

nce the robot

ly in a way to

p at the target

cy is almost

speed is also

ed the seabed

t altitude was

o

e

l

t

o

t

t

o

d

s

transformed

As shown

program suc

target depth

a value of bu

To freeze

stabilizer a

implementat

robot succee

set target dep

Fig. 21 ill

The maximu

Fig. 19 Altit

Dep

to an equiva

in Fig. 19,

cceeded to sm

with a vertic

uoyancy very

e the robot a

algorithm is

tion is shown

eded to keep i

pth.

lustrates the r

um speed reac

(a) Dep

(b) Depth

(c)

tude control: S

pth and Altitu

alent target de

the predictiv

moothly reach

al speed almo

close to the n

at the target

used. The

n in Fig. 20.

its depth with

robot ascent t

ched was 0.3

pth Vs Speed

h Vs Buoyancy

Tc Vs Tr

Step 3.

ude Control o

epth Dt = 755

ve depth con

h the robot at

ost equal to 0

neutral buoya

altitude, a d

e result of

It shows that

hin (+/-) 1 m f

o the sea surf

9 m/s.

of an AUV Us

5 m.

ntrol

t the

0 and

ancy.

depth

f its

t the

from

face.

Fig.

Fig.

ing Buoyancy

. 20 Altitude

. 21 Altitude

y Control Dev

(a) Depth V

(b) Depth Vs B

control: Step 4

(a) Depth V

(b) Depth Vs B

control: Step 5

vice

Vs Speed

Buoyancy 4.

Vs Speed

Buoyancy

5.

1477

Depth and Altitude Control of an AUV Using Buoyancy Control Device

148

In summary, the 3 stages of predictive depth control

succeeded to bring the robot to the target altitude

with a buoyancy value close to the neutral and with

vertical speed almost equal to 0. Besides, the altitude

stabilizer succeeded to maintain the robot within 1 m

from the target altitude. The definition of the certain

depth helped to reduce the time to reach the target

altitude.

6. Conclusions

A new method for depth control using the buoyancy

control device was developed. A model of the

buoyancy variation with time was established. It was

built based on the results obtained in high pressure tank

experiment and several at-sea experiments. The depth

control algorithm is based on the comparison between

the time estimated for the robot to change its buoyancy

from its current value to the neutral value, and the time

expected for the robot to reach the target depth. The

method was demonstrated at-sea experiments in

Toyama Bay in Japan in March 2016. It showed the

ability of the control algorithm to smoothly bring

the robot to the target depth without a significant

overshoot. The algorithm is characterized by its

flexibility and does not require a strict determination

neutral buoyancy value. A margin of inaccuracy can be

customized before performing the dive. The method

could be further adapted to perform an altitude control

through a progressive depth control algorithm based on

4 steps. The experiment results showed that it worked

properly.

Acknowledgements

This research project was funded for

2011FY-2015FY by Grant-in-Aid for Scientific

Research(S) of Japan Society for the Promotion of

Science (No. 23226017).

References

[1] Seymour, R. J., and Geyer, R. A. 1992. “Fates and Effects of Oil Spills.” Annual Review of Energy and the Environment: 261-83.

[2] Jakuba, M. V. et al. 2011. “Toward Automatic Classification of Chemical Sensor Data from Autonomous Underwater Vehicles.” Intelligent Robots and Systems, IEEE/RSJ International Conference on Intelligent Robots and Systems: 4722-7.

[3] Harvey, J. et al. 2012. “AUVs for Ecological Studies of Marine Plankton Communities.” Sea Technology 53 (9): 51.

[4] Roemmich, D. et al. 2009. “The Argo Program Observing the Global Ocean with Profiling Floats.” Oceanography 22 (2): 34-43. doi:10.5670/oceanog.2009.36.

[5] Eriksen, C. C. et al. 2001. “Seaglider: Along-Range Autonomous Underwater Vehicle for Oceanographic Research.” IEEE Journal of Oceanic Engineering 26 (4).

[6] Choyekh, M. et al. 2014. “Vertical Water Column Survey in the Gulf of Mexico Using Autonomous Underwater Vehicle SOTAB-I.” Marine Technology Society Journal: 88-101.

[7] Hallegraeff, G. 1998. “Transport of Toxic Dinoflagellates Via Ships’ Ballast Water: Bio-economic Risk Assessment and Efficacy of Possible Ballast Water Management Strategies.” Marine Ecology Progress Series Mar. Ecol. Prog. Ser. 168: 297-309. doi:10.3354/meps168297.

[8] Zhang, F., and Dickman, M. 1999. “Mid-ocean Exchange of Container Vessel Ballast Water. 1: Seasonal Factors Affecting the Transport of Harmful Diatoms and Dinoflagellates.” Marine Ecology Progress Series Mar. Ecol. Prog. Ser. 176: 243-51. doi:10.3354/meps176243.

[9] Clarke, M. R. 1978. “Physical Properties of Spermaceti Oil in the Sperm Whale.” Journal of the Marine Biological Association of the United Kingdom J. Mar. Biol. Ass. 58 (01): 19. doi:10.1017/s0025315400024383.

[10] Mcfarland, D., Gilhespy, I., and Honary, E. 2003. “DIVEBOT: A Diving Robot with a Whale-Like Buoyancy Mechanism.” Robotica 21 (4): 385-98. doi:10.1017/s026357470300496x.

[11] Shibuya, K., Kishimoto, Y., and Yoshi, S. 2013. “Depth Control of Underwater Robot with Metal Bellows Mechanism for Buoyancy Control Device Utilizing Phase Transition.” Journal of Robotics and Mechatronics (JRM) 25 (5): 795-803. doi:10.20965/jrm.2013.p0795.

[12] Bond, C. E. 1996. Swim Bladder, Biology of Fishes (2nd ed.). Orlando: Saunders College Publishing.

[13] Um, T. I., Chen, Z., and Bart-Smith, H. 2011. “A Novel Electroactive Polymer Buoyancy Control Device for Bio-Inspired Underwater Vehicles.” 2011 IEEE International Conference on Robotics and Automation: 172-7. doi:10.1109/icra.2011.5980181.

[14] Barker, L. 2014. Closed-Loop Buoyancy Control for a Coastal Profiling Float. MBARI Intern Report.

[15] Izawa, K., Mizuno, K., Miyazaki, M., Inoue, A., Ando, K., Takatsuki, Y., Kobayashi, T., and Takeuchi, K. 2002. On

Depth and Altitude Control of an AUV Using Buoyancy Control Device

149

the Weight Adjustment of Profiling Floats. ARGO Technical Report FY2001, 18-35.

[16] Kato, N. et al. 2015. “Autonomous Spilled Oil and Gas

Tracking Buoy System and Application to Marine Disaster Prevention System.” Interspill 2015, Amsterdam, Netherland.