Universitat de Girona

Computer Vision and Robotics Group

An Introduction to Applied
Underwater Robotics

Presented by: Dr. Pere Ridao

VICOROB Research Team

Hardware
Anàlisi Percepció Robòtica Visió en temps
d’imatge 3D submarina submarina real
CIRS: Research Center in Underwater Robotics

Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

3
Introduction

OCEANS Exploration
• 71 % earth surface is covered by water
• 37 % of the populations lives at less than 100 km
form the coast
• Oceans are a source of food and resources
• Oceans play an important role in the clima

Technology

Manned ROVs AUVs 4

Submersibles

Introduction

Detph vs Technology

155 m

308 m

600 m

6000 m

10,911 m

5
Introduction: Marine Robots

ASC

Glid
Glider
er
Glider
Hybri
Hybrid
yb
b id
IAUV ROV/
OV//A
AUV
ROV/AUV
ROV
ROV
Survey Hovering
AUV AUV

Introduction: UdG Robot Prototypes

CIRS-UdG Robots
1995 2001 2005 2006 2010

Applications

Industrial Scientific
 Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

Applications: Dam Inspection

[P. Ridao et al., JFR10]

Pasteral dam
Objective: Execute an inspection of a dam wall to search for cracks or other
damages on the concrete.

9


Applications: Dam Inspection

[P. Ridao et al., JFR10]

Pasteral dam

10

Applications: Habitat Mapping

[P. Ridao et al., WPDC10]

Mequinenza dam

Objective: Providing
visual validation for a
sonar-based system
developed to detect
zebra mussel colonies.

11
 

Applications: Habitat Mapping

[P. Ridao et al., WPDC10]

Mequinenza dam

12

Applications: Seafloor Mapping

Dive 1
Dive 2

Dive 4

Dive 3

AZORES Workshop
FREESUBNET
IN COOPERATION
WITH FREESUBNET
RTN NETWORK
 

Applications: Seafloor Mapping

20 m
5m
5m

20 m

Dive
D
Di
ive
ve 4

Dive 4

Di
Divvee 3
Dive

AZORES Workshop
FREESUBNET
IN COOPERATION
WITH FREESUBNET
RTN NETWORK

Applications: Multimodal Mapping

[Escartin et al., GGG08]

Multimodal Maps Very Large Maps

Image Mosaic & Bathymetry registration 20.000 images mosaic (6 days of ROV survey)

Eiffel Tower hydrothermal vent. Lucky Strike Hydrothermal Vent site

Data from IFREMER Data from WHOI 15


Applications: Micro-Bathymetry & 3D Mosaicing

[Nicosivici et al. OCEANS08]

3D Mosaics

16

Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

17
ICTINEUAUV: A bit of history

How did it start ... (2006)

18

ICTINEUAUV: A bit of history

[D. Ribas et al., ICRA07]

How did it continue... (2006)

4 Phases

Pass the Gate Score the Cross

Hit the target Recover

19
Breaking the Surface 2009

ICTINEUAUV: A bit of history

There are other ICTINEUS ...

Narcís Monturiol
1819-1885

ICTINEU II, Model Barcelona ICTINEUAUV, to pay ICTINEU3, Manned

harbour homage to Narcís Submersible under
Monturiol development
20

ICTINEUAUV: The Robot

The Ictineu AUV

Characteristics

Open frame design

Small form factor (74 x
46.5 x 52.4 cm)
Lightweight (52 Kg)
Complete sensor suite
ROV/AUV

21
ICTINEUAUV: The Robot

The Ictineu AUV

Tethered
Buoy
Unthetered 22

ICTINEUAUV: The Robot

The Ictineu AUV

Pressure vessels

Power module (2 sealed

12V 12Ah lead acid
batteries)
Computer module (PC104
and Mini-ITX computers)

23
ICTINEUAUV: The Robot

The Ictineu AUV

Thrusters

2 vertical thrusters
4 horizontal thrusters
Motion controlled in 4
DoF (surge, sway, heave
and yaw)

24

ICTINEUAUV: The Robot

The Ictineu AUV

Cameras

Forward-looking color
camera
Downward-looking b&w
camera

DVL (Doppler Velocity Log)

3D velocities
(bottom/water)
Pressure
Range
ICTINEUAUV: The Robot

The Ictineu AUV

AHRS

Heading, pitch, roll

and heave
acceleration.

Fibre optic gyro

Heading with
low drift rate

ICTINEUAUV: The Robot

The Ictineu AUV

MSIS (Mechanically Scanned

Imaging Sonar)

Generation of acoustic images

of the surroundings
360º scans around the vehicle

Maximum range of 100 m

27
ICTINEUAUV: The Robot

The Ictineu AUV

USBL transponder

Vehicle positioning
Acoustic modem

28

ICTINEUAUV: The Software Architecture

[Palomeras et al. MCMC09]

Robot interface

Software objects that

dialog with the hardware

Two types:
Sensor objects
Actuator objects

29
ICTINEUAUV: The Software Architecture
[Palomeras et al. MCMC09]

Perception module

Navigator object:
Estimate the position
and velocity of the robot
(EKF)

Obstacle Detector:
Determine the position of
obstacles (wall, bottom,
)

30

ICTINEUAUV: The Software Architecture

[Palomeras et al. MCMC09]

Control module

Receives sensor inputs

and sends command
outputs to the actuators.

Behaviours:
GoTo
WallInspection
Distance
Heading
Start/Stop Camera
Check Water
Check Temperature
and Pressure
31
ICTINEUAUV: The Software Architecture
[Palomeras et al. MCMC09]

Mission control

Defining the task

execution flow to fulfill a
mission

32

Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

33
Fundamental Problems in Underwater Robotics...

Where am I?
Navigation

Where are the

amphoras?
Mapping

What path should

What force should I I follow?
f ll ?
Apply to achieve the Path Planning
desired speed?
How should I steer
Control To follow the desired
path

Guidance

Breaking the Surface 2009

Navigation & Mapping

SLAM: Simultaneous Localization And Mapping

Localization Algorithms

Map Environment Robot Pose

Mapping Algorithms
 Navigation & Mapping: 

The Navigation Problem

Navigation: Estimate the position, orientation and velocity of a vehicle

xb
{E} Origin at the centre of the earth.
North Pole
B Earth fixed.
yb
zb

{N} Origin at P=[l,] on the earth

surface. Plane XY tg to earth surface.
Axis pointing North-East-Down

{L} Same origine than N. Rotated wrt to

zN a certain angle to avoid the
singularity in the pole.

From Gade 2008

{B} Vehicle attached frame

Navigation & Mapping: 

Inertial Navigation Systems

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems

Inertial sensors are used for the navigation.

3 Accelerometers are used for the linear motion estimation.

3 Gyroscopes are used for the angular motion estimation.

The sensors are expensive and require an accurate calibration.

The position estimate drifts over time.

Navigation & Mapping: 

Inertial Navigation Systems

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems

Measure acceleration & angular velocity. Computer
linear velocity, position and attitude.

Early INS were based on Strapdown systems avoid

gyro-stabilized gimbaled moving parts using virtual
platforms  gyro-stabilization techniques

Navigation & Mapping: 

Inertial Navigation Systems

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems (Strapdown)


    F
f IB  a IB
g B  a IB
gravitation
m

 +  !122#

IBB Can be measured using a triad acc.

Sagnac effect (1925)

Due to the rotation the

light path is longer cw than
acw.
The phase delay is
proportional to the Ω
 DL


 c
Navigation & Mapping: 

IMU vs. AHRS vs. INS

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems (Strapdown)

Navigation & Mapping: 

Inertial Navigation Systems

[Gade, 2004]

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems (Strapdown)

Attitude,
LB or roll/pitch/yaw

Angular
velocity,
 B
IB
Gyros
Velocity, EB
L

Specific
force,
 IBB Navigation
Navigation
Equations Horizontal E or longitude/
Accelero- Equations
meters position,  latitude

IMU
Depth, z
INS

Navigation & Mapping: 

Inertial Navigation Equations

[Gade, 2004]

 IBB
Gyros

LB 
LB 
IBB   
IEL  EL
L

LB IEL 
LEIEE


dt      

LB
LB

Accelero-  IBB 
EB

L

LB  IBB  BL  IEL 
IEL  EB
L


2IEL  EL   EBL
meters L


dt      

EB
L
EB
L
Assuming:
 spherical earth
1 L
 wander azimuth L EL
L

rEB

EB  EB
L


EL 
EL 
EL
L

Not included:
 vertical direction
 gravity calculation
   

dt
EL

EL
Kenneth Gade, FFI

Navigation & Mapping: 

Doppler Based Navigation

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems 

DVL Based Navigation

A Doppler Velocity Log (DVL) is used for measuring the velocity

An AHRS is used to estimate the vehicle Attitude.

Position is computed through dead reckoning.

The position estimate drifts over time.

time.

More accurate than INS systems 

(only one integral)
 Navigation & Mapping: 

Doppler Based Navigation

Doppler Effect
Change in frequency of a wave for an observer moving relative to
the source of the wave.
The received frequency is higher (compared to the emitted
frequency) during the approach
It is identical at the instant of passing by
It is lower during the recession.

Navigation & Mapping: 

Doppler Based Navigation

Doppler Effect v: Speed of the sound source

c: Sound Speed
T: Periode
1st wave 2nd wave ft : Source (transmitted) frequency
ited
emmited emmited
fr : Received frequency
λr : Received wavelength

ct = vT + c(t − T ) + λr
ct = vT + ct − cT + λr
vT c(t-T) r 0 = vT − cT + λr
λr = (c − v)T ⎫ ⎫
⎪ c−v⎪
1 ⎬ ⇒ λr = ⎪ ⎛ c ⎞
T= ⎪ f ⎬ ⇒ f = f ⎜⎝ ⎟
c − v⎠
t r t
ft ⎭ ⎪
fr λ r = c ⎪
⎭
Navigation & Mapping: 

How the DVL works

[Brokloff, OCEANS94]

1. Moving Sound Source & Static Detector



The acoustic projector emits.
bott
The seafloor acts as a hydrophone

2. Moving Detector & Static Sound Source

The acoustic projector emits.
trans


The seafloor acts as a hydrophone

DVL Doppler Shift

trans
fr = ftbott
bott

Navigation & Mapping: 

How the DVL works

[Brokloff, OCEANS94]

Robot Velocity in the bf-frame (Body)

i axis direction
fa
vfa
vbf i axis velocity
vi vbv
Doppler shift at i axis
bv

LS Solution
Navigation & Mapping: 

Absolute Acoustic Positioning

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems 

DVL based Navigation

Absolute Acoustic Positioning

Navigation & Mapping: 

LBL vs. GIB

Long Base Line vs GPS Intelligent Equipped Buoys

Figures: [Alcocer, 2009]

4 beacons are needed for 3D Navigation (3 if depth is known)
Transponders are fixed at the bottom Transponders are drifting buoys
Transponders position need calibration Transponders position is obtained with
GPS
Navigation is solved on the AUV
Navigation is solved on surface
Proximity to the vehicle allows for They buoys are RF connected
high frequency & high accuracy
The AUV sends sends the depth to the
buoys
 Navigation & Mapping: 

A Simple Trilateration algorithm

[Alcocer, 2009]

Range Only Localization

Problem 2.2.1 (Range-Only Localization). Let p ∈ Rn be the position of a vehicle
and pi ∈ Rn ; i ∈ {1, . . . , m} be the positions of a set of landmarks. Further, let
ri = p − pi  denote the distance between the vehicle and landmark i. Define
r = [r1 , . . . , rm ]T . Compute an estimate p ∈ Rn of p based on a set of range
measurements r̄ = r + w ∈ Rm , where w ∈ Rm is a zero mean Gaussian disturbance
vector with covariance R ∈ Rm×m .

The vehicle is assumed to be

p2
static.
Does not require knowledge
r2
about vehicle dynamics.
r1
p1
p3
r3
p

{I}

Figure: [Alcocer, 2009]

Navigation & Mapping: 

A Simple Trilateration algorithm

[Alcocer, 2009]

Range Only Localization: Unconstrained Least Squares (LS-U)

p2

P = p1 . . . pm ∈ Rn×m Landmark positions

r2

p1
r1

p3
di = ri2 = p − pi 2 Squared distance to landmark p1
r3
p = (p − pi )T (p − pi ) = pT p − 2pTi p + pTi pi
{I}

⎡⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
d1 pT p − 2pT1 p + pT1 p1 1 pT1 pT1 p1
⎢ . ⎥ ⎢ .. ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
d=⎢ .. ⎥ = ⎢ ⎥ = p2 ⎢ ... ⎥ − 2 ⎢ ... ⎥ p + ⎢ ... ⎥
⎣ ⎦ ⎣ . ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
dm pT p − 2pTm p + pTm pm 1 pTm pTm pm

= p2 1m − 2PT p + δ(PT P)

2PT p − p2 1m = δ(PT P) − d̄ + ξ
d̄ = d + ξ
Now we should solve for p
Navigation & Mapping: 

A Simple Trilateration algorithm

[Alcocer, 2009]

Range Only Localization: Unconstrained Least Squares (LS-U)

2PT p − p2 1m = δ(PT P) − d̄ + ξ

We reorganize the equation as Aθ = b + ξ

2PT −1m p = δ(PT P) − d̄ + ξ
p2

A θ b

And solve neglecting the constrain betwen p and  p2

θ ∗ = arg min Aθ − b2 θ ∗ = (AT A)−1 AT b

θ∈Rn+1

Navigation & Mapping: 

SBL vs USBL

Short Base Line vs Ultra Short Base Line 

Figures: [Alcocer, 2009]

Navigation is solved in the boat

Calibration is done at the factory
Shorter baseline means less accuracy

Baseline is reduced to few meters Baseline is reduced to few centimeters

Fixed infrastructure in the ship. Does For very deep water, accuracy is bad
no allows for opportunity boats.
May be mounted easily on ships of
opportunity
Navigation & Mapping: 

Acoustic Wave Transmission

[P. Milne, 1983]

Planar Wave Approximation

Spherical Wave Plannar Wave approximation

Range >> baseline

Navigation & Mapping: 

Transponders vs. Beacons (pingers)

[P. Milne, 1983]

Projector: Converts an electrical signal into an acoustic wave.

Hydrophone: Converts an acoustic wave into an electrical signal.
Beacon: Acoustic device that uses a projector to generate a
periodical acoustic pulse.
Transponder: Acoustic device that uses a projector to generate a
periodical acoustic pulse and then an hydrophone to detect the
corresponding echo.

Beacon Mode Transponder Mode

 Navigation & Mapping: 

How SBL works

[P. Milne, 1983]

bx Beacon-based Short Base Line 

dt3−1 = t3 − t1
by dR3−1 = R3 − R1
dR3−1 = c ⋅ dt3−1
c ⋅ dt3−1 = sin θ x ⋅ bx
⎧ Z a = D cosθ x ⎫ Za Xa
given that ⎨ ⎬⇒ =
⎩ X a = Dsin θ x ⎭ cosθ x sin θ x
x ⇒ X a = Z a ⋅ tgθ x
dR=c· dt the same applies for:
Za
x Ya = Z a ⋅ tgθ y
D
When the vessel is in the vertical ≅ 0
x
Z ⋅ c ⋅ dt3−1
Xa = a
bx
Xa
Z a ⋅ c ⋅ dt2−1
Ya =
by

Navigation & Mapping: 

How SBL works

[P. Milne, 1983]

Transponder-based Short Base Line 

H2 H1
⎧ R12 = ( X a − a )2 + (Ya + b )2 + Z a2
a b a Xa ⎪
b ⎪⎪ R22 = ( X a − a )2 + (Ya − b )2 + Z a2
H4 H3 ⎨ 2
⎪ R3 = ( X a + a ) + (Ya + b ) + Z a
2 2 2
Ya ⎪ 2
⎪⎩ R4 = ( X a + a ) + (Ya − b ) + Z a
2 2 2
y
x R1
( ) ( )
R4
R3 ⎧ R32 − R12 = 4aX a ⎫⎪ R32 − R12 + R42 − R22
R2 ⎪ 2 ⎬ ⇒ X a =
⎪ R4 − R2 = 4aX a ⎪⎭
2
8a
Za ⎨ 2
⎪ R1 − R2 = 4bYa ⎪⎫
2

⇒ =
( ) (
R12 − R22 + R32 − R42)
⎪ R 2 − R 2 = 4bY ⎬ Ya
⎩ 3 4 ⎪
a⎭
8b

Z a = ⎡ R12 − ( X a − a ) − (Ya + b ) + R22 − ( X a − a ) − (Ya − b )

2 2 2 2

⎢⎣

R32 − ( X a + a ) − (Ya + b ) + R42 − ( X a + a ) − (Ya − b ) ⎤ / 4

2 2 2 2

⎥⎦
Navigation & Mapping: 

How USBL works

[P. Milne, 1983]

Transponder-Based Ultra Short Base Line 

b

t

 b· cos

t: wave time delay b: baseline

Δϕ: wave phase delay f: frequency

: incidence angle c: sound speed

b cosθ
Δt =
c ⎛ b⎞ ⎛ Δϕ ⋅ c ⎞
Δϕ Δϕ Δϕ = 2
π f ⎜ ⎟ cosθ θ = cos−1 ⎜
ω= ⇒ Δt = ω
⎝ c⎠ ⎝ 2π f ⋅ b ⎟⎠
Δt ω

Navigation & Mapping: 

How USBL works

[P. Milne, 1983]

Ultra Short Base Line 

Xa

Ya x c⋅t
y R=
2
X a = R cosθ x
R
Ya = R cosθ y
Za R 2 = X a2 + Ya2 + Z a2
Z a = R 1 − cos2 (θ x ) − cos2 θ y ( )

Boat-Transceiver + Vehicle-Transponder
 Navigation & Mapping: 

Map-based Navigation

Navigation: Estimate the position, orientation and velocity of a vehicle

Inertial Navigation Systems 

DVL based Navigation

Absolute acoustic Positioning

Map–based navigation

Ground-fixed, localization given an a priori map of the

environment.
Bathymetric, gravitational anomaly and magnetic field maps. 
An up-to-date map of enough resolution is not always available.

Map may, or may not, be known a priori.

Mapping & Navigation

Environment
Structured
Non Structured
Mapping & Navigation

Sensing The World: Acoustic Imaging

The operation of a MSIS



Mapping & Navigation

Sensing The World: Acoustic Imaging

The operation of a MSIS

 Beam

Bin
Phd Thesis: Underwater SLAM for Structured Environments Using an Imaging Sonar 

Understanding MSIS

Particularities of the MSIS


Indetermination in the vertical position Acoustic beam reflected by a wall:
of the target:

Mapping & Navigation

Sensing The World: Acoustic Imaging

Particularities of the MSIS


Indetermination in the vertical position Reflections of the acoustic beam:
of the target:
Mapping & Navigation

Multibeam vs. Mechanical Scanning

Motion Distortion
Multibeam scanner Mechanical Scanned Imaging Sonar

Full scan in time step k Full scan in m time steps {k → k+m}

Mapping & Navigation

MSIS: Compensating for motion-distortions

Motion Distortion

Distorted Data Corrected Data

Mapping & Navigation

Sensing The World: Acoustic Imaging

The operation of a Side Scan Sonar


Most used underwater imaging method
Uses beams of sound transmitted out a towfish
The beams are narrow (1-2°) in the along-track
direction and wide (40-50°) in the vertical one
Towfish include port and starboard transducers
Sound is reflected back from objects to the
towfish
Hard objects reflect more energy than soft
Projected shadows will be observed behind
objects
The image is built up one line of data at a time


Mapping & Navigation

Sensing The World: Range & Bearing

The operation of a Multibeam Sonar Profiler

 Range & Bearing sensor
Multiple beams simultaneously grabbed
May be operated from a ship or from an AUV.
The higher the altitude, the lower the
resolution.
The higher the frequency the higher the
resolution but the lower the range.

Mapping & Navigation

Terrain Based Navigation

Where am
I?

Mapping & Navigation

Terrain Based Navigation

I Could be
anywere !
Mapping & Navigation

Terrain Based Navigation

What do I
see?

Mapping & Navigation

Terrain Based Navigation

Do I have a
Map?
Mapping & Navigation

Terrain Based Navigation

Let me
check were
I am!

Mapping & Navigation

Terrain Based Navigation

This is what I
expect to see?

This is what I
have seen!
I’M NOT
HERE
Mapping & Navigation

Terrain Based Navigation

This is what I
expect to see?

This is what I have

seen!
BOTH AGREE!

Mapping & Navigation

Terrain Based Navigation

SO I’M
HERE
Mapping & Navigation

Simultaneous Localization And Mapping

> Localization Through SLAM:

k+1
Step:k+1
Step:

Step: k Internal
Sensors
External Sensors

PREDICTION

MEASUREMENT
Y

DATA
UPDATE ASSOCIATION
X
{B}

Experimental Results

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

79
Experimental Results

Abandoned Marina Data-set

Abandoned Marina Data-set

Fluvià Nàutic, St Pere Pescador (Spain)


http://cres.usc.edu/radishrepository/view-one.php?name=abandoned_marina

600 [m] trajectory DVL, MRU, Depth

DGPS Ground Truth MSIS data
80

Experimental Results

Dead Reckoning

Roadmap

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

81
Experimental Results

Dead Reckoning

Dead Reckoning
Prediction
Constant Velocity kinematic model
DVL, MEMS-AHRS, Depth Updates
Drifts over time Correction

DVL MRU

Pressure

EKF (vehicle)

82

Experimental Results

Scan Matching Improved Dead Reckoning

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

83
Experimental Results

Scan Matching Improved Dead Reckoning

[E. Hernandez et. al. IROS09]
Scan Matching Improved Dead Reckoning
Compute the relative displacement of a vehicle between two consecutive configurations
by maximizing the overlap between range & bearing scans.

Minimization: LS over the Mahalanobis distance.

Scan
Matching
^q
k ^q
pIC

do{
• Association
• Minimization
84
}while (!convergence)

Experimental Results

Scan Matching Improved Dead Reckoning

[E. Hernandez et. al. IROS09]

Probabilistic Scan Matching

Dead Reckoning + SM

Mechanical Scanning Imaging Sonar
Polar Range & Bearing scans
Deals with motion induced Distortion
Drifts less, but drifts

85
Experimental Results

USBL Navigation

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

86

Experimental Results

USBL Navigation

DGPS EKF (vehicle)

Prediction

Correction

DVL MRU
MRU

USBL
Pressure
Transceiver

87
Experimental Results

USBL Navigation

EKF (USBL sist.) EKF (vehicle)

Prediction Prediction

Correction Correction

DGPS MRU DVL MRU

USBL Pressure

88

Experimental Results

Map-Based Navigation

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

89
Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

The voting procedure

North

ρ
X

The high intensity bin is likely

to correspond with the tank
boundaries.

The vehicle can only exist
within the boundaries of the
tank.

90

Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

The voting procedure

Number of votes
Distance (m)

X (m)

Distance (m) Y (m)

The vehicle’s X-Y position is determined by the cell with the larges
number of votes. The Z position is obtained from the pressure sensor.
91
Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

Managing compass errors

Compass data may be perturbed when operating close to ferromagnetic structures.
Determining erroneously the angle between the vehicle and the map will produce a
dispersion of the votes and hence, a poor vehicle’s position estimate.

Correct compass measurement Perturbed compass measurement

92

Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

Managing compass errors

Correct compass measurement Perturbed compass measurement

Number of votes

Number of votes
X (m)

X (m)

Y (m) Y (m) 93
Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

Managing compass errors

94

Experimental Results

TBN: Voting Grid-based Localization

[D. Ribas et al., ICRA07]

Voting-based localization: SAUC-E 2006 final run

X (m)
( m)

Y (m)

Uses Range & Bearing Scans

Deals with the magnetic disturbance of the compass
Does not Drift
Requires a priori map
95
 Experimental Results

TBN: Merged Grid Localization and Dead Reckoning

 [De Marina et al. CAMS 2007]

Constant Velocity kinematic model

DVL, MRU, Depth updates
Absolute XY position updates from the Grid Localization
Does not Drift
Requires a priori map

EKF (vehicle)

Prediction

Grid Correction
Localization

DVL MRU

Pressure 96

Experimental Results

TBN: Monte Carlo Localization (PF)

[F. Maurelli et al., MCMC09]

Monte Carlo Localization

100 Particles
Motion model: Dead Reckoning
Measurement Model
ScanGrabbing + Prob. average
Resampling: SIR + random set
of particles
Does not drift

S t r u c t u re d / N o n
structured environment

Requires an a priori map

97
 Experimental Results

SLAM: Simultaneous Localization And Mapping

1 Abandoned Marina Data-Set

2 Dead Reckoning

3 Scan Matching Improved Dead Reckoning

4 USBL Navigation

5 Map based Navigation

6 Underwater SLAM

98

Experimental Results

SLAM: Structured Environment

[Ribas et al. JFR08]

Feature based EKF SLAM

EKF SLAM with line features

Line features from MSIS data

Line Uncertainty from the acoustic

imprint

DVL+MRU dead reckoning

Statistical dependent Local Maps

Does not drift

Does not require an a priori

map

Deals only with structured

environment
99
Experimental Results

SLAM: Non structured Environment

[A. Mallios et al. IROS10]

Pose Based SLAM

ASEKF SLAM with Scan Matching

State = trajectory of scan poses
Non contiguous Scan Matching
Loop Closing
Works as a Network of constrains
Does not drift
Does not require an a priori
map
Structured/nonstructured
environment




100

Experimental Results

Generation of photomosaics
[Garcia et al. IROS01]

Registration
of
consecutive
images

101
Experimental Results

Generation of photomosaics
[Garcia et al. IROS01]

Registration
of SURF method [Bay, 2006]
consecutive
images
A Hessian detector identifies
individual features.
Feature description (gradient
information at particular
orientations and spatial
frequencies).
Matching (Euclidean distance
between descriptors).
Outlier rejection (RANSAC
[Fischler, 1981]).

102

Experimental Results

Generation of photomosaics
[Garcia et al. IROS01]

Registration Registration
of of non-
consecutive consecutive
images images

Detection of nonconsecutive
overlapping images

Motion estimation from

consecutive images.
Vehicle navigation data (when
available).

103
Experimental Results

Generation of photomosaics
[Garcia et al. ICRA02]

Registration Registration
of of non-
consecutive consecutive
images images

Detection of nonconsecutive
overlapping images

Motion estimation from

consecutive images.
Vehicle navigation data (when
available).

104

Experimental Results

Generation of photomosaics
[Ferrer et al. OCEANS 07]

Registration Registration
of of non- Global
consecutive consecutive alignment
images images

Small errors that occur during image

registration cause misalignment.
The image pairings are used as an
input for the global alignment.
Nonlinear optimization (bundle
adjustment) that minimizes a cost
function [Ferrer et al., 2007].

105
Experimental Results

Generation of photomosaics
[Ferrer et al. OCEANS 07]

Registration Registration
of of non- Global
consecutive consecutive alignment
images images

Small errors that occur during image

registration cause misalignment.
The image pairings are used as an
input for the global alignment.
Nonlinear optimization (bundle
adjustment) that minimizes a cost
function [Ferrer et al., 2007].

106

Experimental Results

Generation of photomosaics
[Ferrer et al. OCEANS 07]

Registration Registration
Crossover
of of non- Global
Detection &
consecutive consecutive alignment
Optimization
images images

The mosaic alignment is improved through several iterations of

crossover detection and optimization.
Iterations are repeated until no new crossovers are detected.

107
Experimental Results

Generation of photomosaics
[Ferrer et al. OCEANS 07]

Registration Registration
Crossover
of of non- Global
Detection & Blending
consecutive consecutive alignment
Optimization
images images

108

Experimental Results

Generation of photomosaics
[Ferrer et al. OCEANS 07]

Registration Registration
Crossover
of of non- Global
Detection & Blending
consecutive consecutive alignment
Optimization
images images

109
Experimental Results

Generation of photomosaics

Experimental Setup

110

Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

111
Experimental Results

Conclusions

1 ICTINEUAUV has become a research platform for Navigation & Mapping

Probabilistic Map based Navigation Techniques has been tested (GL, MCL, EKF,
2
Feature Based SLAM, SM Pose based SLAM, Photo-Mosaicing)

3 Aplications: Dam Inspection, Marine Science Surveys, Archeology.

112

Introduction

Applications

ICTINEUAUV, a research testbed

Navigation & Mapping

Conclusion

Future Work

113


The team

114
1
1114

References
[Palomeras et al. MCMC09] Palomeras N., Ridao P., Carreras M., Silvestre C. Mission Control System for an Autonomous Vehicle: Application Study of a Dam Inspection
using an AUV. 8th IFAC International Conference on Manoeuvring and Control of Marine Craft, Guaruja, Brasil, September 2009.

[E. Hernandez et. al. IROS09] E. Hernandez, P. Ridao, D. Ribas and A. Mallios. Probabilistic Sonar Scan Matching for an AUV. International Conference on Intelligent Robots and
Systems (2009)

[D. Ribas et al., ICRA07] D. Ribas, N. Palomeras, P. Ridao, M. Carreras and E. Hernàndez. Ictineu AUV wins the first SAUC-E competition. IEEE International Conference on
Robotics and Automation, Roma, Italy, April 2007.

[De Marina et al. CAMS 2007] G. García de Marina, D. Ribas and P. Ridao. A global localization system for structured environments using an imaging sonar. IFAC Conference
on Control in Marine Systems, Bol, Croatia, September 2007.

[F. Maurelli et al., MCMC09] F. Maurelli, S. Krupinski, A. Mallios,Y. Petillot, P. Ridao. Sonar-based AUV localization using an improved particle filter algorithm. OCEANS 2009
IEEE; Bremen, Germany.

[Ribas et al. JFR08] D. Ribas, P. Ridao, J.D. Tardós and J. Neira. Underwater SLAM in Man Made Structured Environments. Journal of Field Robotics, 25(11-12):898–921,
November - December 2008.

[A. Mallios et al. OCEANS 2009] A. Mallios, P. Ridao, E. Hernàndez, D. Ribas, F. Maurelli, and Y. Petillot. Pose-based slam with probabilistic scan matching algorithm using a
mechanical scanned imaging sonar. Oceans IEEE, Bremen, Germany, May 2009.

[Garcia et al. IROS01] R. Garcia, X. Cufi and M. Carreras. Estimating the motion of an underwater robot from a monocular image sequence. IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS), vol. 3, pp. 1682-1687, Maui, Hawaii, 2001.

[Garcia et al. ICRA02] R. Garcia, J. Puig, P. Ridao and X. Cufi. Augmented State Kalman Filtering for AUV Navigation. IEEE International Conference on Robotics and
Automation (ICRA), pp. 4010-4015, Washington, 2002.

[Ferrer et al. OCEANS 07] J. Ferrer, A. Elibol, O. Delaunoy, N. Gracias, R. Garcia. Large-Area Photo-Mosaics Using Global Alignment and Navigation Data. MTS/IEEE
OCEANS,Vancouver (Canada), Nov. 2007.

[P. Ridao et al., UUVS09] Pere Ridao, Marc Carreras, David Ribas, Rafael Garcia. Visual inspection of hydroelectric dams using an autonomous underwater vehicle. Journal
of Field Robotics. Special Issue: State of the Art in Maritime Autonomous Surface and Underwater Vehicles, Part 1.Volume 27, Issue 6, pages 759–778,
November/December 2010.J.

[Escartin et al., GGG08] Escartin, R. Garcia, O. Delaunoy, N. Gracias, A. Elibol, X. Cufí, L. Neumann, D.J.Fornari, S.E. Humphris, J. Renard. Globally aligned photomosaic of the
Lucky Strike hydrothermal vent field (Mid-Atlantic Ridge, 37º18.5'N): Release of georeferenced data, mosaic construction, and viewing
software. Geochemistry, Geophysics and Geosystems , vol. 9, no. 12, pp. 12(1)-12(17), 2008.

[Nicosevici et al. OCEANS08] T. Nicosevici and R. Garcia. On-line robust 3D Mapping using structure from motion cues. MTS/IEEE Techno-Ocean Conference (Oceans'08), Kobe,
Apr. 2008. ISBN: 978-1-4244-2125-1. DOI: 10.1109/OCEANS KOBE.2008.4531022
[P. H. Milne 1983] Underwater Acoustic Positioning Systems, P.H. Milne 1983, ISBN 0-87201-012-0

[Ridao et al. WPDC 2010] P. Ridao, M. Carreras, R. García, D. Ribas, and J. Batlle. Advancing underwater robotics. Water Power & Dam Construction, 62(5):40–43, May 2010.

[Alcocer, 2009] Alex Alcocer Peñas. Positioning and Navigation Systems for Robotic Underwater Vehicles. Phd. Thesis. 2009.
115