Talk by Dr Siouxsie Wiles (ORCID: 0000-0002-0467-0015) at the Liggins Institute, University of Auckland on the 27 February 2013.
Bioluminescence (literally ‘living light’) has evolved in a wide variety of fascinating organisms with many different purposes. It allows glow worms to glow worms to lure food, fireflies to find a mate and nocturnal squids to camouflage themselves from predators. The light is produced as a by-product of an enzyme (‘luciferase’) reaction, emitted when a substrate (‘lucferin’) is exposed to oxygen. Siouxsie will talk about her research using bioluminescence to better understand infectious diseases, from tracking infections in living animals to discovering new drugs for tuberculosis.
Talk by Dr Siouxsie Wiles (ORCID: 0000-0002-0467-0015) at the Liggins Institute, University of Auckland on the 27 February 2013.
Bioluminescence (literally ‘living light’) has evolved in a wide variety of fascinating organisms with many different purposes. It allows glow worms to glow worms to lure food, fireflies to find a mate and nocturnal squids to camouflage themselves from predators. The light is produced as a by-product of an enzyme (‘luciferase’) reaction, emitted when a substrate (‘lucferin’) is exposed to oxygen. Siouxsie will talk about her research using bioluminescence to better understand infectious diseases, from tracking infections in living animals to discovering new drugs for tuberculosis.
ARGOMARINE ACTIVITIES
1. To develop and to combine marine observig technologies for a more reliable detection and the monitoring of hydrocarbon/oil spills in marine environment, in support of preventive and emergency interventions.
2. To develop an Integrated Communication System (ICS) to ensure reliable and efficient data transmission from different types of sensors to the MIS, providing an accurate geo-positioning of every data item.
3. To develop and to test a Marine Information System (MIS), an operational central unit provided with evaluation adn decision-making power (expert system) where remote sensing data, field experiment results and estimates from simulation models are integrated in order to support the authorities in emergency handling and management.
Q-Learning and Pontryagin's Minimum PrincipleSean Meyn
https://netfiles.uiuc.edu/meyn/www/spm_files/Q2009/Q09.html
Abstract: Q-learning is a technique used to compute an optimal policy for a controlled Markov chain based on observations of the system controlled using a non-optimal policy. It has proven to be effective for models with finite state and action space. This paper establishes connections between Q-learning and nonlinear control of continuous-time models with general state space and general action space. The main contributions are summarized as follows.
* The starting point is the observation that the "Q-function" appearing in Q-learning algorithms is an extension of the Hamiltonian that appears in the Minimum Principle. Based on this observation we introduce the steepest descent Q-learning (SDQ-learning) algorithm to obtain the optimal approximation of the Hamiltonian within a prescribed finite-dimensional function class.
* A transformation of the optimality equations is performed based on the adjoint of a resolvent operator. This is used to construct a consistent algorithm based on stochastic approximation that requires only causal filtering of the time-series data.
* Several examples are presented to illustrate the application of these techniques, including application to distributed control of multi-agent systems.
ARGOMARINE ACTIVITIES
1. To develop and to combine marine observig technologies for a more reliable detection and the monitoring of hydrocarbon/oil spills in marine environment, in support of preventive and emergency interventions.
2. To develop an Integrated Communication System (ICS) to ensure reliable and efficient data transmission from different types of sensors to the MIS, providing an accurate geo-positioning of every data item.
3. To develop and to test a Marine Information System (MIS), an operational central unit provided with evaluation adn decision-making power (expert system) where remote sensing data, field experiment results and estimates from simulation models are integrated in order to support the authorities in emergency handling and management.
Q-Learning and Pontryagin's Minimum PrincipleSean Meyn
https://netfiles.uiuc.edu/meyn/www/spm_files/Q2009/Q09.html
Abstract: Q-learning is a technique used to compute an optimal policy for a controlled Markov chain based on observations of the system controlled using a non-optimal policy. It has proven to be effective for models with finite state and action space. This paper establishes connections between Q-learning and nonlinear control of continuous-time models with general state space and general action space. The main contributions are summarized as follows.
* The starting point is the observation that the "Q-function" appearing in Q-learning algorithms is an extension of the Hamiltonian that appears in the Minimum Principle. Based on this observation we introduce the steepest descent Q-learning (SDQ-learning) algorithm to obtain the optimal approximation of the Hamiltonian within a prescribed finite-dimensional function class.
* A transformation of the optimality equations is performed based on the adjoint of a resolvent operator. This is used to construct a consistent algorithm based on stochastic approximation that requires only causal filtering of the time-series data.
* Several examples are presented to illustrate the application of these techniques, including application to distributed control of multi-agent systems.
National Marine Park of Zakynthos Workshop on “Marine Pollution: Monitoring S...ARGOMARINE
The Management Agency of National Marine Park of Zakynthos (NMPZ) organized a Scientific Environmental Workshop on “Marine Pollution: Monitoring Systems and Treatment” which held on Monday 30th July 2012, at the Cultural Centre of Zakynthos. The Workshop began at 9:45 a.m. and concluded at 15:00 p.m. Afterwards, the guests were guided in the marine protected area of the NMPZ.
The Workshop held in the framework of NMPZ participation (as one of the Scientific Partners) in the FP7 – European Union funded Project: ARGOMARINE “Automatic Oil-Spill Recognition and Geopositioning integrated in a Marine Monitoring Network” which aims to develop an Integrated System for Marine Traffic Monitoring and Marine Pollution Early Warning, particularly for environmental- sensitive sea areas.
After the 1st successful Workshop on “ARGOMARINE: A New Oil Spill Early Warning System” which was organized by NMPZ on 15th December 2011 on Zakynthos island among the Scientific Partners of the ARGOMARINE Project and representatives from the competent Local Services of the island, NMPZ proceeded to the organization of a 2nd Workshop open to the public.
On 07 September 2012, CNR (Laboratory of Signals and Images), in the field of ARGOMARINE project (www.argomarine.eu), launched “Argo Sentinel”: the Mobile App to report oil sliks on the sea.
Reports, geolocated using GPS, are sent to the Maritime Information System (MIS), the “thinking brain” of the ARGOMARINE network able to analyze data and integrate them with mathematical forecasting models.
It is possible to dawnolad the app from Google Play: https://play.google.com/store/search?q=argo+sentinel&c=apps
ARGOMARINE è un progetto teso a sviluppare un sistema di controllo del traffico e dell’inquinamento marino all’interno di aree sensibili dal punto di vista ambientale e di particolare valore naturalistico come quelle dell’Arcipelago Toscano.
Le attività del Progetto prevedono il controllo dell’inquinamento da idrocarburi e degli sversamenti abusivi di petrolio e derivati in mare mediante l’utilizzo simultaneo di tecnologie in grado di pattugliare larghe estensioni di mare.
I dati raccolti vengono integrati con informazioni di geolocalizzazione e geoposizionamento ed inviati attraverso una rete a larga banda al MIS, il cervello pensante di ARGOMARINE ospitato nella sala operativa della Capitaneria di Porto di Portoferraio.
Il MIS (Marine Information System) è una centrale di elaborazione dotata di capacità decisionali e di valutazione, progettata utilizzando tecnologie di supercalcolo ed Intelligenza Artificiale, in grado di elaborare i dati raccolti dalle varie fonti, valutare i modelli matematici e previsionali immagazzinati nella sua memoria, prevedere le evoluzioni dell’evento critico e assistere le autorità preposte nella gestione dell’emergenza.
2. Outline
• CMRE history and mission
• Underwater Monitoring Technologies
• Hardware design and development
• Detection and Localization
• Classification
• Experimental results
• Conclusions
3. • world-class NATO scientific
research and experimentation
facility, La Spezia, Italy
– ocean science, modeling
and simulation, acoustics
and other disciplines
• over 50 years of service
• The Centre disposes of an unique research
structure in the European panorama
– employs scientists from all NATO countries
– two research platforms for experiments at sea
– development systems and laboratories for acoustic
and oceanographic studies
– facilities for various instrument calibration
– electronic and mechanical design laboratories
– autonomous underwater and surface vehicles
4. 1989 End of cold war: many examples of dual-use military technologies
→ Design and implementation of a calibration facility for oceanographic
instrumentation which provides assistance to nearly all the Italian marine
research institutions and to other countries in southern Europe
→ Design and implementation of advanced environmental monitoring systems
→ CMRE starts studies on the effect of anthropogenic noise on marine animals
– main purpose is to draw a mitigation protocol on the influence on animals made by
artificial acoustic emissions used for military or geological applications
– joined several national and international research institution expert on this topic
– from 1999, carried out many big experimental campaigns at sea with the
participation of research institutions from all over the world
. . . . .
→ ARGOMARINE: area access monitoring technologies
6. Acous&c:
Triangula&on
among
distributed
sensors
EM waves
Shore Lab
Cell-phone
GPS
Major differences: Acoustic
• Complex environment waves
• Noise & Sound propagation
• Variety of unknown sound
sources (blind monitoring)
3D VIEW
7. Examples of noise from vessels
Mid-speed, small ship. Spectrogram (dB re. 1 µPa) NURC rubber boat. Spectrogram (dB re. 1 µPa)
30 110 30 110 Slow, mid-size, leisure boat. Spectrogram (dB re. 1 µPa)
30 110
High level @ LF High level in Wide Band High level @ LF & MF
Limited prop. cavitation 100
25
Strong prop. cavitation 100 No cavitation 100
Few spectral lines Several spectral lines Many spectral lines
25
25
90 90
90
20 20
20
Frequency (kHz)
Frequency (kHz)
Frequency (kHz)
80 80
80
15 15
15
70 70
70
10 10 10
60 60 60
5 5 5
50 50 50
40 40 40
0 1 2 3 4 5 6 7 8 5 6 7 8 9 10 11 12 13 0 1 2 3 4 5 6 7 8
Time (sec) Time (sec) Time (sec)
8. Slant Plane
Exploitation of time coherence Hyd 1
of signal received by each
sensor pair of each array Hyd 2
ΔT
Requirements:
• Sparse hydrophones (d>>λ)
• High sampling frequency
X-Correlogram EF2 Tripod 1 - Pair 23
0 τ
α
50 Hyd 2
d Hyd 1
Hydrophone
Pair
Bearing (deg)
100 τ: Time Delay
α : Bearing angle
150 α = acos ( τ cw / d )
0 10 20 30 40 50
Time (sec)
9. System design
• Each Station
– Sparse Tetrahedral Array of
Four low-noise, preamp hydrophones
(100 kHz bandwidth)
– SCU & Digitalizer (192 kHz SF)
– Pan-Tilt-Compass-Depth sensor (serial data
integrated into digital data flow to shore)
• Fiber-optic-cable connection to shore
• Simultaneous acquisition of continuous
data flow from both stations on shore
– Real-time Reception, Acquisition, Display
& Processing of data from both stations
– Integration in the same data files of
acoustic, orientation, and other possible
serial data
10. Deployments
• 2011-2012: La Spezia harbor
– Test of performances and
assessment of performances
degradation
• May 2012: 3 weeks in Elba
Island at sea recording data
with and without ground truth
tracks
– collected several Terabyte of data
to be used for algorithm
assessment and validation and for
classifier training set
11.
12. Localization from one station and from
two stations
P
Water Depth
One Tripod
k
z
y
Eleva.on
Top View
Azimuth
Two Tripods:
x
2D Triangulation
θ2
Tripod1
θ1
Tripod2
13. ARGOMARINE Sea Trials 2012
(NURC & Elba Island)
• Acoustic characterization of the test sites
(ambient noise)
• Oceanographic data on the field (SVP)
• Acoustic data collection under controlled
conditions:
– Simultaneous data acquisition from BOTH
uw stations during the run of an inflatable boat equippedGPS
with GPS antenna
antenna
– Integration of GPS ground-truth position
data into acoustic data files
• Blind acoustic monitoring
14. Enfola (Elba Island) test site
Tripods deployment:
• 40m depth
• About 450-550m off shore
• Sandy-posidonia seabed
• Relatively quiet environment
evidence of
No
thermoclyne
CTD 28/05/'12
5
10
Depth (m)
15
20
25
30
35
Tripod
distance:
120
m
1510151515201525
Sound Speed (m/s)
15. Experimental Results
Fused Track in Geographic Coordinates
GPS ground-truth
42.835 Estimate
42.834 Fused Horizontal Range seen from Tripod 1
Lat N (deg)
700
42.833
600
42.832
500
(m)
42.831 400
42.83 300
10.269 10.27 10.271 10.272 10.273 10.274 10.275 200
10.276
Lon E (deg) 5.446 5.448 5.45 5.452 5.454
4
Time (sec) x 10
16. Experimental Results
Performances:
40 m max error over
700m range (6 %)
Station 2
Station 1
GPS ground-truth
Acoustic Estimate
17. Classification
RVM (Relevance Vector Machine)
classifier
• Supervised statistical method (19 features
selected)
• Binary
• Fully Bayesian model ⇒ provides probabilistic
predictions
• No need of a-priori statistics
• Provides selection of most significant features
• Nice balance between simplicity and power
18. Feature extraction from data
PSD function Spectrogram.
100
22
dB re. 1µPa/√Hz 90
80
Time (sec)
24
70
60 26
50
28
40
0 20 40 60 10 20 30 40
Frequency (kHz) Frequency (kHz)
DEMON Spectrum X-PSD function
0.08 40
Normalized Amplitude
30
0.06
20
dB 10
0.04
0
0.02 -10
-20
0
50 100 150 200 0 20 40 60
Frequency (Hz) Frequency (kHz)
19. Classification results
10
Three
classes
8
Multi-class confusion rate matrix (%)
Feature # 17
6 (Threshold = 0.5)
Slow Fast Ships
Pred
4
small small/
2
boats
mid-
sized
0
15
10 20
40 True boats
5 0
Feature # 16
0 -20
Feature # 2 Slow boats
92
7.0
1.0
Slow, small boats Fast mid-
8.0
89.0
3.0
Fast small/mid-sized boats sized boats
Ships (down to ferry size)
Ships
0.0
0.0
100
21. Complete MIS Integration
AUV positions
&
E-nose data
XML files
To & From CNR
MOOS HTTP
Tracks Database
&
Aco
features MOOS
XML
variables
status
update
XML XML
Style Transfor
sheet
mer
23. Acoustic Detection, Localization and Classification
• Concept successfully validated
• Advanced prototype system
• Possible exploitations:
– Marine mammal survey
– Monitoring of noise sources with important
environmental impacts (wind-farm piling,
regasification ships, etc.)
– Port protection
24. Test at sea
• May 2012: eNose mission integrated into
ARGOMARINE MIS
• Sep. 2012: acquisition of sampling oil
signals
• Nov. 2012: eNose missions with optimal
sampling trajectory and real time MIS
integration
• Cooperation with CNR-IFC, Graaltech and
CNR-ISTI
25.
26. OBJECTIVE: Find the optimum sampling
designs for an AUV-mooring ocean observing
network
• METHODOLOGY: The problem is decoupled into
a) finding the most adequate
Floats Network
• sampling locations for the AUV and b)
-Unevenly distributed
-Same cycling period
-Synoptic measures
to visit these locations in the fastest way.
Glider Network
• Definition of a space filling design. Try to
spread sampling locations throughout the
region, leaving as few holes as possible.
Sampling points are located to minimize a
criterion
• Solution of the Travel-salesman Problem.
Once the sampling locations have been defined,
a trajectory of the AUV is computed to visit all
the locations selected in the fastest way.
27. AUV mission planner
Definition of Operational Constraints
• Area
• Time constraints
• Vehicle speed
• Number of vehicles
• Obstacles
Planning Module
Space-filling Design
Feedback between the space-filling
design generator and the genetic
algorithm until operational conditions
Genetic Algorithm
are satisfied.
AUV Mission
• Waypoints
• Travelled Distance
28. Experimental Design for
ARGOMARINE
Find an optimum
mission
for Folaga AUV to
sample
the selected
marine area,
considering the
existence
of a monitoring
buoy and denied
areas(red).
Mission should
take around 1 hr
( 1m/s)
29. AUV mission -Result for ARGOMARINE
Optimum
trajectory
for the Folaga-
AUV (dash-dot
black line)
compatible with
operational
constraints.
The traveled
distance
is 2962 m.
30. WP4.5 Integration
• Current vehicle capabilities
• Macro Tasks
– Surface navigation
– Gliding mission
– Underwater navigation
– Idle
– Vertical Profiler
• User control mode
– Controlled by external software
– Simple command interface
– Complete control on devices
• Emergency
– Release drop weight
balloon
Valve
pump
Macro tasks and state machine
31. MOOS-IvP
• MOOS: Mission Oriented Operating Suite
• IvP: Interval Programming a mathematical programming model
for multi-objective optimization
• MOOS-IvP is a set of open source C++ modules for providing
autonomy on robotic platforms, in particular autonomous marine
vehicles
– It provides a framework for data exchange/communication
– separation of overall capability into separate and distinct modules
– Front-seat/Back-seat concepts
An Overview of MOOS-IvP and a Users Guide to the IvP Helm
Michael R. Benjamin, Henrik Schmidt, Paul Newman, and John J.
Leonard
32. Moos-IVP integration:
E-Folaga Main
controller
(front-seat driver)
Argomarine MIS
Navigation Navigation commands
data, Heading, speed, depth
vehicle status
Definition of TCP/IP
communication GPRS/3G
protocol
Acoustic modem
MOOS-ivp
Shoreside station
User Control Mode GPRS modem MOOS-iVp
(back-seat driver)
Radio modem
MOOS DB: Log all variables
Behavior examples
n Position
– Wait on position
n Comms
– Search pattern (lawnmower)
n Payloads data
– Goto location
n Mission status
MOOS Set of – E-nose mission on location
database
behaviors
n Etc.
– Go home
33.
34. Many thanks to CMRE team
• Alberto A.
• Alberto G.
• Alessandra T.
• Federico C.
• Lavinio G.
• Piero G.
• Vittorio G.
Alessandro, Marco, Salvatore, Piero