Unmanned Mobile Sensor Net
- Ben Snively
Unmanned Underwater Gliders
Survey and extensions to work from:
COOPERATIVE CONTROL OF COLLECTIVE MOTION FOR
OCEAN SAMPLING WITH AUTONOMOUS VEHICLES;
Derek A. Paley
Problem / Motivation

Manual Ocean Sampling labor, resource, and time intensive.

Objectives/Sensing






Gradient Climbing
Feature Tracking
Boundary Monitoring
Perimeter Surveillance
Mapping
Autonomous Distributed Agents both reduce over costs, improve
accuracy, and reduce latency for critical data.
University of Central Florida
Marine Sensing Equipment

NOAA Manual Ocean Survey

Ships with complex survey equipment
 Performs in both shallow and deep
water collections.

AUV – Autonomous Underwater
Vehicle

Shallow water survey using GPS and
Satellite communication
University of Central Florida
Sensor Network Details

Sink  Satellite Communication

Agents/Sensors  Mobile Gliders


Every Glider in Range of Sink / GPS

Satellite only communicate with satellite when at
surface

Cannot control movement at surface (just floats)
– has control only when submerged.
GCCS – Glider Coordinated Control System

GCCS steers groups/sets of gliders
 Gliders controlled remotely
 No Agenda/Beliefs/Planning
University of Central Florida
GCCS
Overview

“Apply a cooperative control methodology to control a
fleet of autonomous underwater gliders. Underwater
gliders soar through the water on a pair of fixed
wings, collecting valuable oceanographic data for
weeks at a time. We describe the Glider Coordinated
Control System (GCCS), which steers multiple gliders to
a set of coordinated trajectories. The GCCS
automatically controlled up to six gliders continuously for
over three weeks in a 800 km2 region in California’s
Monterey Bay in August 2006. The GCCS enables
oceanographers to specify and adapt glider sampling
patterns with minimal human intervention.”
(Paley)
University of Central Florida
Mobile Sensors and Sink
University of Central Florida
Domain Definition
Domain:
Outer Search Area
University of Central Florida
Domain Example
<domain>
<rectangle>
<x>
<units>deg</units>
<value>-122.3817</value>
</x>
<y>
<units>deg</units>
<value>36.9765</value>
Center Point Longitude
Center Point Longitude
</y>
<a>
<units>met</units>
<value>20000</value>
Width
<units>met</units>
<value>10000</value>
Height
<units>deg</units>
<value>137</value>
Angle
</a>
<b>
</b>
<ori>
</ori>
</rectangle>
</domain>
University of Central Florida
Search Paths
Tracks 
Search Paths
University of Central Florida
Tracks Example
<tracks>
<superellipse>
<name>track1</name>
<x>
<units>deg</units>
<value>-122.2713</value>
</x>
<y>
<units>deg</units>
<value>36.8950</value>
</y>
<a>
<units>met</units>
<value>10000</value>
</a>
<b>
<units>met</units>
<value>6667</value>
</b>
<ori>
<units>deg</units>
<value>47</value>
</ori>
<p>
<value>3</value>
</p>
</superellipse>
</tracks>
University of Central Florida
Shape/Name
Center Point
Width
Height
Glider Groups
Glider Tree:
Entities,
which paths to use,
Comm model.
University of Central Florida
Glider Groups Example
<group>
<group>
<phase>
<value>0</value>
<units>pct</units>
</phase>
<glider>
<mnf>w</mnf>
<sn>7</sn>
<model>e</model>
<track>track1</track>
<direction>1</direction>
<phase>
<value>0</value>
<units>pct</units>
</phase>
<control> sellipse control </control>
<!-- Manufacturer -->
<!-- Serian Number -->
</glider>
<glider> . . .
</glider>
</group>
</group>
University of Central Florida
<!-- Curve Phase -->
YAES Simulations
Simulations Performed
1. 4 Boundary Gliders rotating same direction
2. 4 Boundary and 6 Interior Sensors,
All positioned in ideal/planned position.
3. 4 Boundary and 6 Interior Gliders, all starting from hub location
Actual software from research done in Matlab.
Software allows for both simulated and real tests
University of Central Florida
Background Information on Simulation

XML Configurations drive simulation setup and context.

Much like the Real system, xml configuration files define gliders,
tracks, and other system configuration.
(The XML Schema was slightly modified in order to simplify configuration)

Main Simulation class allows for selection of which
simulation to perform.

JAXB used to parse XML Input / Java 1.6 SDK/Runtime
(JAXB = Java API for XML Binding)
University of Central Florida
Simulation 1: Boundary Test
Sink Communication
at surface
4 Boundary Gliders
University of Central Florida
Simulation 2 : Interior Gliders
6 Interior Gliders covering Area
3 Groups, 2 Gliders in each.
(Gliders communicate w/ Sink at Surface)
University of Central Florida
Simulation 3 : Boundary and Interior Gliders
Sink Communication
at surface
4 Boundary Gliders
6 Interior Gliders
Flow adds error (off track) when
Glider is at surface
University of Central Florida
Extending Network

Current Limitations




Central control/Planning
Planning communication only done at surface
(when gliders are floating and have no control – drift)
Limited Inter-glider communication.
Extending Sensor network principals to the system


Gliders become agents with beliefs, agendas, and planning
Sensor communication models and MDP principals
(Specifically Partially Observable - MDP)

This is critical due to the fluxations and inconsistencies in the robot
control.
University of Central Florida
Glider Agents

Key Difference



Glider communicate with each other
Goal:



Glider has Agenda, Goal, and isn’t pure input/sensing device.
Cover area (and plan) that has
not been covered by other
agents.
Inform others plan/area covered.
Single Sink transmitter could be at surface, to bridge underwater to satellite gap.
University of Central Florida
Planning / Decision Process

PO-MDP (Partially Observable Mark-ov Decision
Process.

Outcome from Glider commands uncertain.
Policy mapping between States / Actions  Perceived new State.

Routing issues isn’t applicable due to every glider having access to
the sink (via satellite).

Routing issues could be introduced in a more complex system where
messages between gliders and to sink are transmitted.
University of Central Florida
Questions?

Questions / Comments?

Thank you.
University of Central Florida