Michael E. West,* Michael Novitzky, Jesse P. Varnell, Andrew Melim, Evan
Seguin, Tedd C. Toler, Tomas R. Collins, John R. Bogle, Matthew P. Brad-
ley, and Andrew M. Henshaw
Georgia Tech Research Institute (GTRI) has developed the Yellowfin, a small
man-portable Unmanned Underwater Vehicle (UUV). The mission for Yellow-
fin is to conduct autonomous collaborative operations. The multi-UUV design
allows for a much wider swath of the ocean to be observed and monitored, while
collaborative operations allow multiple aspects of a mission to be tackled with
distributed systems. Both oceanographic and military missions are aided tre-
mendously by the use of such a UUV network. This paper introduces the mod-
ular and flexible design of the Yellowfin system and describes some of the tech-
nologies integrated within the system construct. The system and software ar-
chitectures of Yellowfin leverage COTS technologies, including software whose
foundation is MOOS-IvP, expanded to include several aspects of autonomy,
communication with the WHOI acoustics modem utilizing the JAUS message
standard, mission planning using MissionLab, mission execution via Falcon-
ViewTM , front seat control with a microcontroller, and visualization with the
Blender open-source, cross-platform suite of tools for 3D graphics.
Unmanned Underwater Vehicles (UUVs) have become indispensable tools for undersea scien-
tific, military, and commercial applications. Using UUVs for Intelligence Surveillance and Re-
connaissance (ISR), Mine countermeasures (MCM) and Anti-Submarine Warfare (ASW) have
shown great potential. 1 UUVs have shown to be invaluable tools for understanding ocean beha-
viors such as Harmful Algal Blooms, Chemical transport (Nitrates and Hydrates) and the impact
of CO2 absorption. The current paradigm for the use of UUVs is a platform centric sensing
system. Our research looks at the use of multiple vehicles which would allow for a net centric
distributed sensing system and interrogate a much wider swath of the ocean. Currently, the US
Navy?s planned deployment of UUVs emphasizes single-vehicle operation through at least 2015
with uncertain capabilities for cooperating vehicles beyond that. 2 3 However, to reduce the over-
all time and cost of acquiring data over large unstructured areas, multiple vehicles must be used.
To address this need, the Georgia Tech Research Institute (GTRI) has developed the Yellowfin
UUV. See Figure 1.

* Senior Research Engineer, Electronic Systems Laboratory, Georgia Tech Research Institute, 400 W. 10th St. Rm. 272
Atlanta, GA 30332-0829.
The Yellowfin is a man-portable UUV that weighs less than 17 lbs and was designed to sup-
port autonomous homogeneous collaborative operations. Advances in multi-agent collaborative
control have been integrated into the system to autonomously control and coordinate multiple
UUVs. The behavior modules were developed and implemented using the open-source Mission
Oriented Operating Suite (MOOS) architecture along with the Interval Programming (IvP) Helm
software module. The MOOS architecture was coupled with MissionLab open-source software
tools developed by Georgia Tech. This suite of software is then used to develop and execute be-
haviors for both single UUV autonomy and for a team of autonomous collaborating UUVs. This
coupling of software allowed for the development of highly capable UUVs using behavior-based
autonomy. By using open source and standards-based development in the design of the Yellow-
fin, we have attained a modular and adaptable suite of platforms that support flexibility in mission
execution. The design also decouples the vehicle autonomy logic from the mechanical control of
the vehicle hardware, so that upgrades and adaptations are readily incorporated. Yellowfin is also
designed to seamlessly incorporate different sensors and payloads, so that collaborative behaviors
enable coordination among the different vehicle sensors and configurations. The result is a stan-
dards-based system capable of supporting intelligent autonomy, from perception to situational
understanding to automated responses. The paper presents the foundations of the Yellowfin de-

A number of design challenges exist for the development of man-portable UUVs. Because
space is at a premium, all of the electronics for power, communications, sensors, computation,
and actuation, along with their packaging, must be carefully considered. A constant tradeoff be-
tween cost and development time must be mitigated, and the use of commercially available off-
the-shelf components must be leveraged when possible. Figure 2 presents the high-level design

Figure 1: Yellowfin CAD Drawing


Figure 2: Yellowfin Design Approach

Design Approach
Yellowfin was designed to rapidly provide a solid foundation of integrated baseline functio-
nality while still being adaptable to a wide range of mission-specific requirements. To keep cost
and development time as low as possible while maintaining the quantity and quality of vehicle
features along with system interoperability, the design of Yellowfin adopted COTS (Commercial,
off-the-shelf) components, open-source software, and industry standards whenever feasible. The
vehicle itself was designed through an iterative process where one prototype was designed, built,
and tested followed by several additional vehicles. The software design also progressed in stages
with the results of the implementation motivating additional cycles of design work.
Design Goals
The initial design goals generated requirements that were categorized into two primary areas:
mission requirements and vehicle requirements 4
Mission requirements
These are specific to the mission that Yellowfin will perform. While Yellowfin is capable of
performing a diverse category of missions, there is a core set of features that is common to the
requirements of most UUV missions. These include
- Control, navigation, and collaborative behaviors
- Validation through real and simulated testing
- Suitability for multiple applications, including oceanographic and military research
Vehicle requirements
These are specific to the Yellowfin vehicle. Particular values for performance specifications
are determined by examining the requirements for various applications and were chosen to meet
or exceed those requirements.
- Operating speed > 2 knots
- Operating duration > 10 hours
- Max weight < 17 lbs
- Cost < $30K
- Ability to accommodate various sensor and payloads
Resulting design
Coupling the mission and vehicle requirements resulted in a vehicle design with the following
features. Figure 3 shows a fully assembled Yellowfin.
- Length: 889mm
- Diameter: 123.8mm
- Weight: 7.7kg
- Front Seat/Back Seat Driver Software Paradigm
- Modular, Behavior-Based Autonomy
- Mission Oriented Operating Suite (MOOS)
- JAUS Interfaces

Figure 3: Fully Assembled Yellowfin



Acoustic communications are enabled through a WHOI micro-modem, which is used
throughout the UUV community and allows the Yellowfin to communicate underwater over fairly
long distances at very low bandwidth. Wi-Fi provides a high bandwidth data link when the Yel-
lowfin is at the surface of the water, but does not work underwater. Radio frequency (RF) com-
munications fall in-between the acoustic and Wi-Fi in terms of bandwidth and works both above
and below water (moderately well). Yellowfin is also equipped with an optional Ethernet tether
that supports in-water drive testing, evaluation of new capabilities, and high-bandwidth connec-
tions when the Yellowfin is docked. Yellowfin uses an open-source implementation of the Joint
Architecture for Unmanned Systems (JAUS) message protocol called OpenJAUS.* OpenJAUS
provides a library of message routines that code and decode a variety of JAUS message types in-
cluding system health, UUV pose, mission directives, and sensor data communication. The Yel-
lowfin software also includes a library of routines that encode and decode the JAUS messages
into and out of the NMEA 0183 standard protocol utilized by the WHOI acoustic modems.

Figure 4: Yellowfin Communications

* http://www.openjaus.com
Yellowfin has many of the most common navigational sensors, including GPS, IMU, com-
pass, and pressure sensor. These are used for low-level motion control and for high-level localiza-
tion. Yellowfin also includes moisture sensors for leak detection and a BlueView forward-
imaging sonar. In contrast with traditional sonars where a single beam is mechanically rotated,
imaging sonars implement multi-beam sensors that form several small acoustical beams at once.
Imaging sonar is effective on moving platforms, whereas movement of traditional sonars during
vehicle operation can cause data errors. Yellowfin uses the sonar in several ways, but this sonar
specifically enables simultaneous localization and mapping (SLAM). See Figure 4. Other sensors
and payloads can be easily incorporated into various-sized nose cones for Yellowfin.

Processing on the Yellowfin is split into two sections: a low-level processor, which man-
ages all hardware and software integral to the sensors and actuators, and a high-level processor,
which manages the autonomy and collaborative behaviors. The so-called ?Backseat-driver? pa-
radigm allows for the decoupling between high-level and low-level control. 5 This dual architec-
ture was created to reduce the difficulty of modifying the more complex high-level software and
to make the software platform-independent. It also reduces the risk of the less stable high-level
software interfering with safety-critical low-level software. The low-level and high-level proces-
sors are able to communicate in a reasonably fast manner (high bandwidth, low latency) to ex-
change data.
The Yellowfin has a complete software package from pre-mission planning to mission execu-
tion. See Figure 5. The pre-mission planning is performed by Mission Lab by organizing availa-
ble behaviors to generate mission behaviors. Mission execution on the vehicle is performed using
the MOOS-IvP suite of applications for high-level autonomy and an XMOS low level controller.
Command and control is performed at a base station, using FalconView. A mission can be ex-

Figure 4: Yellowfin?s Sonar GUI Tracking Buoys
ecuted in the Yellowfin simulator or on the actual vehicles.

Figure 5: Yellowfin's Software For Mission Execution

Mission Planning
Yellowfin uses MissionLab for pre-mission planning. MissionLab was created at the Georgia
Institute of Technology for the purposes of organizing and executing behavior-based architec-
tures.6 In particular, Yellowfin has a database of behaviors that can be organized for a particular
mission based on requirements through the use of a GUI application. See Figure 6. This GUI
translates the organized behaviors into mission files which Yellowfin?s autonomy can execute
during mission deployment.


Figure 6: Mission Lab's Behavior-based GUI

Command and control of multiple UUVs is performed through a base station using the open
source FalconViewTM* Software. FalconViewTM is widely used by the United States Department
of Defense for its aircraft mission planning and mapping capabilities and has over 40,000 users.
FalconViewTM provides for application extensions through a plug-in framework. The UUVs in
this system can communicate to a base station server when or as needed through the Yellowfin
communication system, and the FalconViewTM application plug-in displays the vehicles? posi-
tions and telemetry information in real time using JAUS messages. The base station can also be
used to send JAUS messages to the vehicles, such as waypoint and mission-based commands.

* http://www.falconview.org

Figure 7: FalconViewTM Mission Planning Software

Collaborative Autonomy
The high-level software architecture and autonomy for Yellowfin is provided by MOOS and
MOOS-IvP, respectively. 7 MOOS is C++ cross-platform middleware, created and maintained by
the Oxford Mobile Robotics Group for robotics research. It enables interprocess communication
through a central database using a publish-subscribe architecture. MOOS-IvP, which is main-
tained by MIT's Laboratory for Autonomous Marine Sensing Systems (LAMSS), is a collection
of MOOS-based applications designed for maritime autonomy. MOOS-IvP includes applications
for communication, simulation, data acquisition, pre-mission planning, and post-mission analysis.
IvP Helm is a behavior-based architecture that uses multi-objective optimization to support coor-
dination of multiple competing behaviors. Included with IvP Helm are 17 behaviors and the
ability to create new ones. Behaviors can be clustered according to different mission modes,
making the entire cooperative system flexible to changing mission dynamics.
The software system has been designed to optimize the execution of a course of action to carry
out a specific mission, given the situational awareness derived by the sensors. The mission
represents both the stuational information and operational priorities. It includes a set of rules
which control the execution of a set of behaviors that are not completely known in advance and
occur during the execution of that mission. Thus, this enables
? The ability to react to unforeseen situations (no scripted cases), and
? Autonomous, on-the-fly planning and replanning

Example behaviors needed for collaborative autonomy include Navigation, Avoidance,
Search, Investigate, Attack (independent and assisted), Assist, Rendevous/Loiter, Communicate,
and Negotiate. Each of these operations is implemented as an independent behavior that operates
autonomously within its scope; each conducts real-time planning and analysis of the situation rel-
ative to mission execution, and each responds appropriately to the results of that analysis.
? Search Area Collaboration. When a vehicle completes searching an area, it communi-
cates with its partners informing them that it is available for to assist them. A partner Yel-
lowfin can assist in several ways. One form of assistance is to help another Yellowfin
complete its mission (e.g., re-partition a large search area for one Yellowfin into two
smaller search areas for two Yellowfins). Another form of assistance is if a partner does
not respond (e.g., it may have been removed from the group before completing its mis-
sion), then the assisting Yellowfin will have know the first Yellowfin?s mission and can
assist in completing that mission based on the last communications that occurred between

? Negotiations. Negotiation can occur between Yellowfins in support of a mission for any
situation. Two behaviors (Negotiate and Assistor Negotiate) are designed for coordinat-
ing actions amongst many Yellowfins. A single Yellowfin using Negotiate can initiate
negotiations with multiple vehicles in a single instance. Multiple vehicles use Assistor
Negotiate to indicate they are available to assist in the coordinated operation. These two
behaviors form the foundation for a generic communication between two, where one re-
quests help and determines which partner is best to assist in the task.

? Partitioning / Assigning Search Areas. A capability is being incorporated that allows a
Yellowfin that is designated as the supervisor to receive an entire mission and then dele-
gate individual sub-missions to the group of UUVs under its control, including itself.
Based on the mission, it decides whether to partition one large search area into multiple
smaller areas, assign a single search area to each Yellowfin, or plan specific flight paths
for each vehicle.

? Re-Partitioning Search Areas / Supporting a Partner in Completing a Search Area.
Mission replanning can occur after each Yellowfin completes a given mission. This is in
contrast to when replanning occurs only at rendezvous points. Also being incorporated is
the ability to assist a partner in searching a designated area (e.g., re-partitioning a large
area into two small areas where a second can assist). This includes completing the search
mission of Yellowfins that are out of communication.


Figure 7: Yellowfin Simulator

The Yellowfin simulator was designed for the purpose of developing autonomous behaviors
and functionality in a fully realized 3D environment, as seen in Figure 7. A key aspect of this is
the ability to model the Yellowfin vehicle within the simulator and to allow the Yellowfin soft-
ware package to operate in a manner similar to that in which the vehicle will be deployed. Devel-
oping a simulator also provides the ability to develop multi-vehicle behaviors without requiring
the cost of deploying multiple vehicles.
Visualization Software
The main visualization technology driving Yellowfin?s simulator is the use of Blender, an
open source cross-platform toolset for creating 3D worlds.* Blender provides a programming in-
terface using the Python language as well as several graphical software development tools. A 3D
model of the Yellowfin vehicle was imported into the simulation using Blender's 3D modeling
Bathymetry Data
In order to simulate true deployment conditions, an underwater Bathometry can be imported
into Yellowfin?s simulator using GeoTIFF files of any location of interest. GeoTIFF is a non-
proprietary standard for TIFF image files that contain embedded geographical or georeferencing
data for use in constructing real world geographical images. These GeoTIFF files can be con-
structed from various sources including satellite imagery or elevation models.
Sensor Data Simulation
Yellowfin?s simulator has the ability to simulate sensor data from the 3D environment. Ray
tracing is utilized to create a sonar image of the simulated sonar?s field of view. This allows for
the testing of Yellowfin?s target-tracking and simultaneous localization and mapping (SLAM)
capabilities. These are two active research areas and contribute to Yellowfin?s overall autonomy.

* http://www.blender.org
Simulator to Vehicle Interface
A MOOS database interface has been created for the Blender-based simulation in order to
provide a gateway for the testing of Yellowfin?s autonomy software. MOOS-IvP provides simula-
tion tools, such as iMarineSim, to exercise a vehicle?s autonomy. However, the provided simula-
tor does not include an attractive 3D visualization with perception sensor data such as sonar data
from the simulated sea floor as provided by bathymetry data. Connecting Yellowfin?s simulator
to the MOOS Database allows the Yellowfin autonomy software to receive the same input as the
physical robot would receive during a mission and respond accordingly. This provides the ability
to develop working autonomous behaviors in the simulator before actual deployment.
Yellowfin is a man-portable unmanned underwater vehicle. It has been designed to be dep-
loyed for both scientific and military missions, and its payload is adaptable to mission require-
ments. The design process emphasized low cost, high functionality and interoperability by utiliz-
ing off-the-shelf parts, open-source software, and industry standards. The software of Yellowfin
allows for pre-mission planning through mission execution of a collaborative team of vehicles
with robustness for different missions and real-time situational awareness. The Yellowfin simu-
lator enables testing of the autonomy software in various life-like situations. Yellowfin?s small
size and robust software make it ideal for littoral missions and research into homogeneous colla-
borative operations.
This work was supported by Georgia Tech Research Institute as part of a thrust to develop-
ment of enabling technologies in unmanned underwater vehicles.
1 T. Bean, G. Beidler, J Canning, D. Odell, R. Wall, M. O?Rourke, M. Anderson and D. Edwards, "Language and Logic
to Enable Collaborative Behavior among Multiple Autonomous Underwater Vehicles." International Journal of Intelli-
gent Control and Systems. Vol. 13, No. 1, 2008, pp. 67?80.
2 Navy UUV Masterplan 200, http://www.navy.mil/navydata/technology/uuvmp.pdf.
3 Navy UUV Master Plan Update 2004.
4 D. Furey, et al, "AUVSI/ONR Engineering Primer Document for the Autonomous Underwater Vehicle (AUV) Team
5 M. Benahin, ?White Paper ? Software Architecure and Strategic Plans for Undersea Cooperative Cureing and Inter-
vention?, 2007.
6 R.C. Arkin and T. Balch, ?AuRA: Principles and practice in review.? Journal of Experimental & Theoretical Artifi-
cial Intelligence, Vol. 2, No. 2, 1997, pp. 175-189.
7 M. Benjamin, J. J. Leonard, H. Schmidt, and P.M. Newman, ?An overview of moos-ivp and a brief users guide to the
ivp helm autonomy software.? MIT, Tech. Rep. MIT_CSAIL-TR-2009-028, 2009.

Set Home | Add to Favorites

All Rights Reserved Powered by Free Document Search and Download

Copyright © 2011
This site does not host pdf,doc,ppt,xls,rtf,txt files all document are the property of their respective owners. complaint#downhi.com