NRL'S AUTONOMY RESEARCH - updates on - naval research laboratory
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Spectr A
n a v a l r e s e a r c h l a b o r at o r y
THE MAGAZINE OF THE NAVY’S CORPORATE LABORATORY
WINTER 2014
updates on
NRL’S
AUTONOMY
RESEARCH
NAVAL RESEARCH LABORATORY
LEAD I N G EDGE
THE
Commanding Officer
CAPT Mark C. Bruington, USN
Director of Research
Dr. John A. Montgomery
Welcome to the fourth issue of SPECTRA, a magazine
designed to inform you of exciting science and technology
developments at the U.S. Naval Research Laboratory (NRL).
In the spring of 2012, NRL opened the Laboratory for
Autonomous Systems Research (LASR). Our vision for this
shining, state-of-the-art facility was that it would be a focal
point for basic and applied research related to unmanned and
autonomous systems – a place to develop, integrate, and
test systems and prototypes, to advance the Navy’s scientific
leadership in this domain. Now, a little more than two years
later, the LASR has become the collaborative and energetic An official publication of the Naval Research Laboratory
research space we envisioned, with scientists and engineers
from all across NRL working together on interdisciplinary Publisher
projects in the one-of-a-kind facilities and environments that NRL Office of Public Affairs
LASR offers. Autonomy research for the Navy and Department Richard Thompson, PAO
of Defense benefits greatly from this convergence of diverse
talent and expertise.
Managing Editor
Donna McKinney, Deputy PAO
In its simulated desert, jungle, and littoral environments Editing, Design, and Production
and other well-equipped laboratories, the LASR is hosting NRL Technical Information Services
creative and innovative research in intelligent autonomy, Kathy Parrish, Head
sensor systems, power and energy systems, human–system
interaction, networking and communications, and platforms. Editing
This issue of SPECTRA provides a snapshot of some of the Claire Peachey
LASR projects now moving forward: Design and Production
Jonna Atkinson
• an autonomous underwater vehicle named WANDA that
borrows its design from a coral reef fish, Photography and Archives
• a next-generation hydrogen fuel cell built using 3-D printing, Gayle Fullerton
• a battery that draws its power from the ocean bottom Jamie Hartman
sediment, James Marshall
• a robot that is learning to understand how people think, and
• a vehicle that travels through air and through water – a flying
submarine.
4555 Overlook Ave, SW
We hope you enjoy this issue of SPECTRA and share it with Washington, DC 20375
others. To request additional copies or more information, please (202) 767-2541
email spectra@nrl.navy.mil. www.nrl.navy.mil/spectra
Contents
features
ON THE COVER
2 Bio-Inspired Designs for Near-Shore AUVs
6 Flimmer: A Flying Submarine
10 Cognition for Human–Robot Interaction
14 NRL Autonomy Lab Hosts Shipboard Fire
Robotics Consortium
18 NRL Technology Assists FDNY
20 The Benthic Microbial Fuel Cell
24 3-D Printing in Hydrogen Fuel Cells for
Unmanned Systems Flight-testing the Flimmer UAV.
See page 6.
28 More LASR Projects
news briefs t e c h n o l o g y t ra n s f e r
30 Navy Launches UAV from Submerged Submarine 35 New York City Tracks Firefighters to Scene with
NRL Radio Tags and Automated Display
31 Captain Mark C. Bruington Relieves Captain Anthony J. Ferrari
and Captain Anthony J. Ferrari Receives Legion of Merit
focus on people
outreach
36 Dr. Jeremy Robinson Receives Presidential Early
Career Award
32 NRL and Aerospace Industry Host 10th Annual
CanSat Student Challenge 37 In Memory of Richard J. Foch
making connections
34 Sharing Science ... Starting Conversations
1
NRL FEATURES
Bio-Inspired Designs for
Near-Shore AUVs
Autonomous underwater vehicles inspiration from nature — from fish, The NRL artificial pectoral fin has
(AUVs) have demonstrated many in particular — to design and develop been integrated into a man-portable,
capabilities in inspection, surveil- novel underwater propulsion, control, unmanned vehicle named WANDA,
lance, exploration, and object detec- and sensing solutions. the Wrasse-inspired Agile Near-shore
tion in deep seas, at high speeds, Deformable-fin Automaton. Four
and over long distances. However, For AUV propulsion and control, NRL side-mounted fins, two forward and
the low-speed, high-maneuverability has developed an actively controlled two aft, provide all the propulsion and
operations required for near-shore and curvature robotic fin based on the control necessary for the vehicle. A
littoral zone missions present mobil- pectoral fin of a coral reef fish, the set of custom control algorithms uses
ity and sensing challenges that have bird wrasse (Gomphosus varius). In information about the vehicle motion
not been satisfactorily solved, despite the most recent fin iteration, four rib and surrounding environment to in-
nearly half a century of AUV develop- spars connected by a flexible synthetic form changes to the fin stroke kine-
ment. Very shallow water and littoral membrane define the fin geometry, matics. By weighting and combining
environments are complex operating and control of each rib deflection angle various fin gaits, the magnitude and
zones, often turbid, cluttered with enables the shape deformation needed direction of thrust generated by each
obstacles, and with dynamically to generate 3-D vectored thrust. Using fin can be controlled to produce the
changing currents and wave action. computational fluid dynamics (CFD) desired effects on vehicle motion.
AUVs need to be able to maneuver tools in conjunction with experiments,
around obstacles, operate at low sets of high-thrust-producing fin stroke Computational and in-water experi-
speeds, hover, and counteract surge kinematics, or fin gaits, have been mental results have demonstrated
and currents. To develop AUVs that identified. This kind of artificial fin tech- WANDA’s capabilities. WANDA can
can successfully navigate and oper- nology can adapt to varying flow condi- perform low-speed maneuvers includ-
ate in these dynamic and challenging tions and provide the thrust control ing forward and vertical translation,
environments, Naval Research Labo- necessary for low-speed maneuvering turn-in-place rotation, and station
ratory (NRL) researchers have taken and precise positioning. keeping in the presence of waves.
2 SPECTRA
NRL FEATURES
The bird wrasse (Gomphosus varius) is
a coral reef dwelling fish operating in
depths of 1 to 30 meters. It is known to
use its paired median pectoral fins to
produce thrust for maneuvering in these
complex environments.
The geometry of the bird wrasse was modeled and computational fluid
dynamics simulation was validated against experimental results.
Following design studies for a robotic pectoral fin, CFD simulation
was used to determine effectiveness of this device as a means of
propulsion and control for an AUV.
WANDA can also successfully coordi- water areas, so NRL is also develop- Additionally, these are active sys-
nate maneuvers to achieve waypoint ing new sensor systems for AUVs, tems that emit sound or light, which
navigation. WANDA is designed to and again looking at fish for ideas. A uses energy and exposes an AUV to
operate at speeds in excess of two reliable system for navigating littoral detection.
knots, or hold position in the presence environments requires real-time knowl-
of two-knot currents, giving it the pro- edge of current velocities and object Fish use a system of hair-like flow
pulsion and control authority needed positions. Conventional onboard sen- and pressure sensors, called a lateral
in many harbor and other near-shore sors such as sonar and vision-based line, to detect changes and obstacles
operational zones. systems have shortfalls in a cluttered, in the environment around them. NRL
shallow environment: sonar can suffer has studied these natural systems
Solutions for vehicle propulsion and from multipath propagation issues, and and is developing an artificial lateral
control address only part of the chal- vision-based systems are limited by line, a system of pressure sensors
lenge of operating in shallow under- turbidity. mounted on the AUV hull to provide
WINTER 2014 3
NRL FEATURES
Demonstrating the WANDA AUV in the Littoral High Bay pool in NRL’s Laboratory for
Autonomous Systems Research. The four-fin design enables the propulsion and control
needed for station-keeping and maneuvering in the presence of external disturbances
such as waves and currents. The cameras in the foreground of the top photo are used to
capture the three-dimensional motion of the vehicle in real time.
passive sensing of surrounding water hicle can use to perform maneuvers in NRL’s Laboratory for Autonomous
velocities and near-field objects. This such as orienting into a flow or avoid- Systems Research (LASR). This facil-
system uses quartz crystal–based ing obstacles. NRL’s artificial lateral ity houses vehicle prototyping tools,
pressure sensors with high sensitivity line will help WANDA and other AUVs such as 3-D printers and metalwork-
to measure minute pressure differ- achieve better performance in envi- ing equipment, and large-scale test
entials between various points on ronments where sonar or vision alone environments. The 45 foot by 25 foot
the vehicle hull. The amplitude and cannot be relied on. pool in the LASR Littoral High Bay
phase differences between sensor serves as the primary underwater
signals provide information about the Much of the development and testing test environment for WANDA, where
surrounding environment that the ve- of the WANDA platforms takes place a 16-channel wave generator creates
4 SPECTRA
NRL FEATURES
The WANDA AUV deployed for testing in the Littoral High Bay in NRL’s Laboratory for Autonomous Systems Research.
real-world simulated near-shore con- The WANDA program has spawned into capable propulsion and control
ditions, and a 12-camera underwater other related programs to enable technologies for low-speed operation
tracking system provides ground Navy critical missions using AUVs. in near-shore environments is helping
truth position and orientation mea- NRL’s new Flimmer (Flying Swim- to close a clear gap in AUV technol-
surements for validation of vehicle mer) program seeks to develop an ogy. An unmanned vehicle that can
performance. unmanned platform that will be de- effectively operate in these areas,
ployed from the air, glide to a water where traditional platforms experi-
As WANDA’s fish-inspired technolo- surface landing in an area of interest, ence stability and control problems,
gies are perfected, the AUV is being and transition into a swimming AUV. will improve performance for critical
prepared for payload testing. The ve- NRL is leveraging its expertise in missions including harbor monitor-
hicle’s modular construction enables unmanned air vehicle (UAV) technolo- ing and protection, hull inspection,
easy integration of different mission- gies to design and build the Flimmer covert very shallow water operations,
specific payload packages, and one vehicle. Modifications have been and riverine operations.
such payload that will be developed made to the WANDA fins to enable
and tested on WANDA starting this them to function as aerodynamic
year is a biochemical sensing system control surfaces and to survive the By Jason Geder
NRL Laboratories for Computational Physics
for trace level detection of chemical impact of landing (see page 6 for and Fluid Dynamics
signatures. This sensor system built further details on Flimmer).
onto a capable low-speed platform
such as WANDA will enable missions As the Navy’s focus on autonomy
in plume tracking and target localiza- and unmanned systems intensi-
tion in shallow water environments. fies, NRL’s bio-inspired research
WINTER 2014 5
NRL FEATURES
Flimmer: Flimmer team (left to right): Trent Young, Jason Geder, Dan
Edwards, Marius Pruessner (not shown: Ravi Ramamurti).
A Flying Submarine
The Flimmer (Flying Swimmer)
pathways. The Flimmer program supported by minimal internal struc-
program at the Naval
seeks to investigate the potential of tures with low safety factor margins.
Research Laboratory (NRL) is rapidly flying a submarine over the
merging two research areas ocean’s surface into position, tran- On the other hand, underwater
to provide a novel airborne sitioning from flight to underwater, vehicles are built completely the op-
delivery method for unmanned and then enabling a swimming mode posite way. As an underwater vehicle
underwater vehicles (UUVs). once underwater. descends, the pressure applied by
Underwater vehicles are limited in the water column increases rapidly,
top speed by the drag produced in Major Design Considerations requiring the use of a strong pressure
water. However, air is approximately In general terms, an aircraft is a vessel to protect the electronics from
1000 times less dense than water, so specific outer shape held rigidly in water. These pressure vessels are
the power required to overcome drag place with a series of internal struc- typically thick metal pressure-
is substantially reduced in air. By tural elements. Weight is the enemy resistant shapes with substantial
flying over the surface of the water of an aircraft designer: an increase in seals. Inherent to enclosing air vol-
rather than swimming long distances weight requires additional lift to keep ume is an increase in the vehicle’s
through it, delivery of a UUV to an the vehicle aloft. Drag is increased buoyancy, requiring sufficient ballast
operational area can be accom- due to this extra lift, which requires weight to bring the overall vehicle to
plished much more quickly, and de- more power to overcome that drag. near neutral buoyancy. Instead of us-
livery is possible to areas not directly To keep aircraft lightweight, they are ing ballast weight, many UUV design-
accessible through continuous water typically manufactured as a thin skin ers elect to make the pressure vessel
6 SPECTRA
NRL FEATURES
heavier than required for the desired
depth pressure.
Test Sub maneuvering as a submarine in the Littoral High Bay pool in NRL’s
Laboratory for Autonomous Systems Research.
Combining these two diametrically
opposed vehicles to design a flying
submarine comes down to a balanc-
ing act between buoyancy, weight,
and structural elements. For a sub-
marine to fly, the enclosed air volume,
which is the main driver of weight for
a submarine, needs to be reduced as
much as possible. For an aircraft to
land on the water, its structural ele-
ments need to be more robust to sur-
vive the high impact of splashdown.
The Flimmer program adds a further
complication into the design: flapping
fins are used for underwater propul-
sion. As learned from NRL’s WANDA systems, to include primarily buoy- of descending orbits to a splash-
UUV (see article on page 2), a four- ancy control and kinematic controls. down landing in the water. Test Sub
finned configuration provides high A custom-built buoyancy control flew as any other aircraft, controllable
maneuverability and good stability system uses two inflatable bladders in three axes and exhibiting sufficient
underwater. In air, however, the fins placed at the front and rear inside the stability for man-in-the-loop flight.
add weight and are relatively fragile hull fairing. A control loop pumps air Once nearing the water surface, Test
mechanisms that need to be able to into the two bladders for a combina- Sub was guided along a standard
survive the forces of splashdown. tion of both pitch and heave control approach at an airspeed of approxi-
Bringing all these design elements when the vehicle is stationary (in mo- mately 40 knots before splashdown
together is the central challenge of the tion, the tail surfaces are used). Air is with wings level. Upon touching the
Flimmer program. scavenged from inside the pressure water surface, the aircraft saw a dra-
vessel to inflate the bladders. While matic increase in drag and deceler-
Test Sub using air in bladders is a lightweight ated abruptly. After the splash, Test
The Flimmer program started with just solution, it does have the drawback Sub submerged and started moving
the combination aircraft and subma- of being unstable with depth, requir- underwater.
rine, without the additional complexity ing constant control inputs.
of fins. The “Test Sub” configuration Flying WANDA
was born from combining a traditional The most interesting aspect of run- With the success of Test Sub, the
submarine shape with a traditional ning Test Sub in the LASR pool was Flimmer team applied the lessons
aircraft shape. Test Sub uses a fixed that controlling the vehicle in forward to designing a method for flying
geometry in both air and water, carry- motion underwater was identical to NRL’s WANDA vehicle. The “Flying
ing the drag penalty of large surface “flying” the aircraft. The configuration WANDA” configuration has four fins
area wings in the water, instead of of an aircraft shape and traditional and the addition of a wing, with the
spending weight and complexity on aircraft control surfaces of rudder, el- two aft fins mounted on the tips of
a wing folding mechanism. Test Sub evator, and ailerons functions exactly the wing. This allows keeping the
also carries a weight penalty in the the same whether in air or water. same control techniques developed
aircraft mode so that it can enter the Several test runs showed excellent for the four-finned UUV design, but
water at full flight speed, spending controllability and maneuverability of provides a lifting surface for carrying
weight to make the structures survive Test Sub in the water while moving the weight of all the flapping mecha-
this high impact loading. forward. nisms.
The Littoral High Bay test pool in For flight testing, Test Sub was taken As the fins are a significant surface
NRL’s Laboratory for Autonomous to a local test range. Three free- area relative to the wing area, they
Systems Research (LASR) was home flights started with an air-drop of the have been designed to pull double-
to Test Sub during its initial develop- vehicle from a mother ship at ap- duty. While swimming, the fins act
ment. Underwater testing focused on proximately 1000 feet altitude. Test as flapping propulsors. In flight, the
development of the submarine sub- Sub was guided manually in a series wingtip-mounted fins are turned up
WINTER 2014 7
NRL FEATURES
Test Sub splashdown sequence.
to act as fixed vertical stabilizers moderate to heavy sea states, this incompatible, the Flimmer program is
for lateral stability, and the forward- planing landing is not possible with- showing a feasible path can be found
mounted fins act as close-coupled out large increases in vehicle size or somewhere in the middle. There
canards ahead of the wing. With this autopilot complexity. Therefore, two are important trade-offs between
dual-use of the fin mechanisms, the nose-down landing styles have been enclosed air volume and structural
incremental drag penalty for Flying tested to investigate survivability of weight, with a particular emphasis on
WANDA is only the addition of the the canards in a load condition con- surviving a splashdown in water at
wing. sisting primarily in the drag direction. flight speeds. However, testing has
Since the flapping propulsion mode already shown that Test Sub cruises
Four test flights of this configuration already must be stiff in the drag well above 50 knots in the air, while
have confirmed acceptable stability direction, the fins’ survivability in the top speed in the water is below 10
and control. Half-area forward fins plunge style landing has been better knots, illustrating the ultimate benefit
have been flown and show strong than initially expected. of a flying submarine: assuring quick-
control power for acting as canards. reaction access to underwater areas.
Full-area wingtip fins have been Experimentation with the Flying
flown in a fixed configuration and WANDA configuration continues.
provide sufficient vertical tail surface Future flights will explore the perfor- By Dan Edwards
for stable lateral modes. mance envelope using the fins as ac- NRL Tactical Electronic Warfare Division
tive control surfaces in the air and will
Test flights have also begun explor- continue the landing technique work.
ing the landing mode that will best A hollow, floodable wing is under
protect the fin mechanisms. Flying construction so that the swimming
WANDA is designed for a traditional phase of experimentation can begin
approach and splashdown, with a in the LASR pool.
planing hull surface to stretch the
deceleration time across a longer The Sky’s the Limit for this
landing maneuver. All four fins are Submarine
well protected using this landing While the diametrically opposed re-
technique, as they are on upper quirements for a flying vehicle versus
portions of the vehicle. However, in a swimming submarine vehicle seem
8 SPECTRANRL FEATURES
Flying WANDA takes flight, moments
after departing the pneumatic launcher.
The Flimmer vehicle is an adaptation of
NRL’s WANDA for air-delivery.
Flying WANDA approaches splashdown
in the Potomac River.
WINTER 2014 9NRL FEATURES
for
HUMAN-ROBOT INTERACTION
Scientists from the Naval Research Laboratory’s Navy Center for Applied Research in Artificial Intelligence demonstrate advances in
cognitive science, cognitive robotics, and human–robot interaction with the help of teammate Octavia (center). Left to right: Ed Lawson,
Greg Trafton, Laura Hiatt, Sunny Khemlani, Bill Adams, Priya Narayanan, Frank Tamborello, Tony Harrison, Magda Bugajska.
Our overarching goal is to give tional cognitive architecture ACT-R/E. • Intentional and Imaginal
robots a deep understanding A cognitive architecture is a process- Modules: Provide support for goal-
of how people think. There are level theory about human cognition. oriented cognition and intermediate
three benefits of this to the scien- It provides a rich environment for problem state representations.
tific community. First, by improving modeling and validating different
understanding of how people think hypotheses about how people think. • Visual and Aural Modules: Enable
and interact with the world, we are ACT-R/E provides a set of computa- the architecture to see and hear
pushing the boundaries of the field tional modules, correlated with dif- perceptual elements in the model’s
of cognitive science. Second, we can ferent functional regions of the brain, world.
leverage this knowledge of people to that work together to explain both
help robots be better at tasks they the limitations of human cognition • Configural and Manipulative
typically are unable to perform well, (i.e., we don’t have unlimited work- Modules: Enable the architecture
such as multimodal communication ing memory) and the strengths (i.e., to spatially represent perceptual
and computer vision. Third, this deep we are good at inferring connections elements in the model’s world.
understanding of humans allows between concepts and interacting
robots to better predict and under- with the physical world): • Temporal Module: Allows the
stand a human partner’s behavior architecture to keep track of time;
and, ultimately, be a better and more • Declarative Module: Manages acts as a noisy metronome.
helpful teammate. the creation and storage of factual
knowledge; selects what chunks • Motor and Vocal Modules: Pro-
To accomplish this, we build pro- (facts or memories) will be thought vide functionality for the model to
cess models of fundamental human about at any given time. move and speak with an appropri-
cognitive skills – perception, memory, ate time course.
attention, spatial abilities, and theory • Procedural Module: Manages the
of mind – and then use those models creation and storage of procedural Each module, save the procedural
as reasoning mechanisms on robots knowledge, or chunks; selects module, is associated with a limited-
and autonomous systems. Most of what production (if-then rule) will capacity buffer, representing what the
our work is done using the computa- fire at any given time. model is thinking about / planning to
10 SPECTRANRL FEATURES
say / looking at / etc. Together, the would allow the memory to become see in the real-world environments
contents of these buffers make up familiar enough to be remembered. our warfighters encounter (such as
working memory in ACT-R/E. While jungles, deserts). In the second, ACT-
each of the modules and buffers is ACT-R/E’s fidelity to the way the R/E’s goal structure, together with its
theoretically motivated and validated human mind integrates, stores, and familiarity and context mechanisms,
on its own, ACT-R/E’s strength lies retrieves all this information is what allow an intelligent system to help
in the complex interaction of these provides its true predictive and ex- humans avoid certain classes of er-
components, shown below. planatory power. In this case, model- rors, such as those commonly made
in Navy vehicle maintenance proce-
dures.
Computer Vision
Recent work in computer vision,
performed by Hoiem, Efros, and Her-
bert in 2006, and Oliva and Torralba
in 2007, among others, has shown
that including contextual information
can greatly affect the efficiency of
object recognition in terms of both
the speed and the accuracy of the
processing process. These existing
computer vision approaches, howev-
er, typically rely on static, aggregated
statistics of various features in the
scene to provide their context. While
such approaches are promising, this
limitation renders them unable to
perform competitively with human
perception; a richer, more human-like
representation of contextual infer-
The ACT-R/E architecture. The complex interaction of ACT-R/E’s many ence, learned over time, like that in
theoretically motivated and validated components provides its true predictive and ACT-R/E, may be the key to their
explanatory power. success.
To illustrate, consider an example ing the human’s experiences at the Context in ACT-R/E takes the form
of a conference attendee meeting conference provides both descrip- of associations between related
someone at the conference and then tive information about how a human concepts that are learned over time.
attempting to remember his or her could err in this situation (e.g., people Concepts become associated when
name later in the evening. The at- that look similar, or that were met at they are thought about at roughly the
tendee would need to not only attend similar times or similar parties, may same time; the more they are thought
to the new person’s face, voice, and be easily confused) and predictive about in proximity to each other, the
clothing, but also bind those indi- information (e.g., someone’s name stronger their association becomes.
vidual concepts to the name of the is unlikely to be remembered after a This type of context is rich, in that it
individual. The attendee would then long break without further rehearsal can capture nuanced relationships
need to rehearse the name several or strong contextual cues). between concepts and facts that
times so it would not be forgotten. are not explicitly linked; this context
When the attendee saw the new per- In this article, we present two differ- is also dynamic, in that it is learned
son a bit later, he or she would need ent systems in which ACT-R/E’s fidel- online and adjusted over time based
to take available cues (perhaps only ity to human cognition is leveraged on the model’s experience. Object
visual cues like face and clothing) to develop intelligent systems that recognition algorithms, like LVis, a
and attempt to retrieve from memory can more functionally assist a human biologically plausible object recogni-
the name associated with those teammate. In the first example, ACT- tion system developed by Randall
cues. Priming from contextual cues, R/E’s familiarity and context mecha- O’Reilly at the University of Colorado
like the face, clothing, and the party nisms help an autonomous system Boulder, can utilize this information to
itself, would boost the activation of better perceive ambiguous or ob- improve recognition in cases where
the memory, and the earlier rehearsal fuscated objects that a robot might using visual features alone is difficult.
WINTER 2014 11NRL FEATURES
For example, when looking in the
kitchen, context may suggest related
concepts such as oranges or lemons.
Any ambiguities that might arise from
other similar objects (such as an or-
ange ball) can be quickly resolved by
incorporating contextual information,
resulting in the correct identification.
Our initial experiments using a large
database of objects have shown that
a system that combines context and
LVis’s object recognition algorithms is
able to increase recognition accuracy
as compared to using LVis without
context (see below). While these
results are simple, they shed light on
the powerful tool that context can be,
and demonstrate how it can be used
to build to autonomous systems that
are better able to support and extend
the Navy’s capabilities.
Sample objects from the RGB-D
dataset developed by Lai et al. at
the University of Washington. The
objects in the dataset are relatively
low resolution, similar to what a
robot would see as it moves around
the world. Here, a battery (left) and
dry eraser (right) can potentially
have very similar visual outlines and
contours, but are typically found in
different situations; context can help
differentiate between these two object
classes.
Error Prediction One of our robots, Octavia, correctly recognizes an orange ball when it is in context.
With the rapid rise of communica- Octavia is a humanoid robot we use to investigate how to build intelligent systems
tion technologies that keep people that can serve as better teammates to their human partners.
accessible at all times, issues of
interruptions and multitasking have completion time or errors. Given the have to resume it. As ACT-R/E pro-
become mainstream concerns. For prevalence of interruptions, build- gresses through a task, it maintains a
example, the New York Times in 2005 ing systems that can help remind an representation of where it is currently
and Time magazine in 2006 both individual what they were doing or engaged in the task. For familiar
reported stories about interruptions where they were in a task can have a tasks, this representation always as-
and multitasking, and how they affect large impact on individual and group sociates, via context, an action to the
performance by increasing human er- productivity. next action to be performed. When a
ror. In 2005, the information technol- task is interrupted, the model loses
ogy research firm Basex estimated We built an ACT-R/E model that track of its goal and, upon resump-
the economic impact of interruptions emulates the process people go tion of the task, must remember what
to be around $588 billion a year through as they get interrupted it was working on. Sometimes, con-
due to losses from increased task during a procedural task and then textual cues associate to future steps
12 SPECTRANRL FEATURES
beyond the next correct one, and
the system skips a step. Alternately,
sometimes the familiarity of past ac-
tions surpasses the familiarity of the
current action, and ACT-R/E remem-
bers an action prior to the last one it
completed, and repeats a step.
Using ACT-R/E in this way, we can
explain (and thus can better predict) N av y Ce n t e r fo r
how, when interrupted, people tend A p p l i e d R es e a r c h i n
A rt i f i c i a l I n t e l l i g e n c e
to skip or repeat steps, even in a
familiar task, based on the task’s
context and strengthening. This ac-
complishment will help us develop The Navy Center for Applied Research
intelligent systems that can mitigate in Artificial Intelligence (NCARAI) has
human error risks in dangerous been involved in basic and applied research in
procedures, with both monetary and artificial intelligence, human factors, and human-
functional benefits. centered computing since its inception in 1981. It
is part of the Information Technology Division of
Conclusion the Naval Research Laboratory and is directed by
Our approach to intelligent systems Alan Schultz, who is also director of the Labora-
is multidimensional. First, in the tory for Autonomous Systems Research.
cognitive science tradition, we attain
a deep understanding of how people The Adaptive Systems group conducts research
think: our ACT-R/E models faithfully focused on techniques that allow systems to change
capture people’s behavior as they their level of autonomy on the fly, adapt to changes in
perceive, think about, and act on the their environment and in their own capabilities, learn new
world around them. Second, we use behaviors through interaction with the world, and interact
this understanding to make intelligent more naturally with humans.
systems better by taking advantage
of people’s strengths. Finally, the The Intelligent Systems group performs research in
models help to reveal limitations and cognitive science, cognitive robotics and human-robot
potential failings of human cogni- interaction, predicting and preventing procedural errors,
tion, which our intelligent systems the cognition of complex visualizations, interruptions and
can then take steps to correct. resumptions, and spatial cognition, to enable intelligent
Overall, our efforts to attain a deep systems that can work more effectively with people.
understanding of human cognitive
strengths and limitations allow us The Interactive Systems group develops and enhances
to build more functional intelligent computer interfaces for autonomous and intelligent
systems that are better able to serve systems, spanning human-computer and human-robot
their human teammates. interactions. Research includes linguistic, auditory, and
phonological analyses to link natural language to modes
of human-machine interaction.
By Laura M. Hiatt, Frank P. Tamborello, II,
Wallace E. Lawson, and J. Gregory Trafton
Navy Center for Applied Research in The Perceptual Systems group examines issues related
Artificial Intelligence to sensors required to support autonomous platform navi-
gation, scene interpretation, and teamwork. Techniques
include passive monocular vision, and passive and active
stereo and triocular ranging methods, along with algo-
rithms to rapidly interpret the scene.
http://www.nrl.navy.mil/itd/aic/
WINTER 2014 13NRL FEATURES
NRL Autonomy Lab Hosts
Shipboard Fire Robotics Consortium
The U.S. Naval Research Labo- ties. Add to this scenario a cloistered Thomas McKenna, managing program
ratory (NRL) Laboratory for platform, say many levels down inside officer of ONR’s Computational Neuro-
a seagoing ship, and the challenge is science and Biorobotics programs. “The
Autonomous Systems Research
exponentially increased, resulting in goal of this research is to develop the
(LASR), a partner in the Navy’s extreme risks to human life. Despite mutual interaction between a humanoid
Damage Control for the 21st these risks, a shipboard fire must be robotic firefighter and the rest of the
Century project (DC-21), re- contained and extinguished for the firefighting team.”
cently hosted robotics research safety of the crew and continued mis-
teams from the Virginia Poly- sion readiness of the ship. This highly specialized research, to pro-
technic Institute and State mote advanced firefighting techniques,
University (Virginia Tech) and To mitigate these risks, NRL research- includes development of a novel robotic
the University of Pennsylvania ers at LASR and NRL’s Navy Center platform and fire-hardened materials
(Penn) to demonstrate the for Applied Research in Artificial Intel- (Virginia Tech), algorithms for percep-
ligence (NCARAI), under direction and tion and navigation autonomy (Penn),
most current developments
funding from the Office of Naval Re- human–robot interaction technology,
of advanced autonomous search (ONR), are working with univer- and computational cognitive models
systems to assist in discovery, sity researchers to develop advanced that will allow the robotic firefighter to
control, and damage control firefighting technologies for shipboard work shoulder-to-shoulder and interact
of incipient fires. fires using humanoid robots, an effort naturally with naval firefighters (NCARAI).
led by the NRL Chemistry Division.
Fighting fires can at times prove chal- “These advancements complement
lenging to even the most seasoned “As part of the Navy’s ‘leap ahead’ highly specialized NRL research that
firefighting veteran — a firefighter must initiative, this research focuses on the focuses specifically on the human–robot
deal with extreme unpredictability, high integration of spatial orientation and interaction technology and shipboard-
temperatures, and rapid decline of the shipboard mobility capabilities based spatial interrogation technology,”
environmental and structural integri- of future shipboard robots,” said Dr. said Alan C. Schultz, director of LASR
14 SPECTRANRL FEATURES and the NCARAI. “Developments made from this research will allow a Navy firefighter to interact peer-to-peer, shoulder-to-shoulder with a humanoid robotic firefighter.” The NRL LASR, where the artificial intelligence portion of the research is performed, hosted the consortium of university researchers to demonstrate their most current developments. The LASR facility allows the research- ers from Virginia Tech and Penn to demonstrate, in a controlled environ- ment, progress in the critical steps necessary for shipboard fire suppres- sion using variants of their Shipboard Autonomous Firefighting Robot, or SAFFiR. In 2013, human–robot interaction technology and cognitive models developed by NRL were also demonstrated at the laboratory. “The LASR facility, with its unique simulated multi-environments and state-of-the-art labs allows us to ‘test out’ our ideas before we go to the field,” Schultz said. “In essence, our facility gives us a cost-saving method for testing concepts and ideas before we go to the expense of field trials.” While at LASR, the researchers dem- onstrated the complex motion, agility, and walking algorithms of the robots over natural and manmade terrain and simulated shipboard sea state (pitch and roll) conditions. Also demonstrated were “seek-and-find” algorithms for locating a fire emergency, in this case an open flame, and the use of “artificial muscle” for the lifting and activation of fire suppression equipment, such as opening a water valve, lifting and walking with a fire hose, and activating a nozzle. “SAFFiR is being designed to move autonomously throughout a ship to learn ship layout, interact with people, patrol for structural anomalies, and handle many of the dangerous firefight- ing tasks that are normally performed by humans,” McKenna said. The robot SAFFiR — the Shipboard Autonomous Firefighting Robot. is designed with enhanced multimodal WINTER 2014 15
NRL FEATURES
Researchers demonstrated the complex
motion, agility, and walking algorithms of the
robots over natural and manmade terrain
and simulated shipboard sea state (pitch
and roll) conditions. Also demonstrated were
“seek-and-find” algorithms for locating a fire
emergency, in this case an open flame, and
the use of “artificial muscle” for the lifting and
activation of fire suppression equipment, such
as opening a water valve, lifting and walking
with a fire hose, and activating a nozzle.
16 SPECTRANRL FEATURES
sensor technology for advanced navigation and
a sensor suite that includes a camera, gas sen-
sor, and stereo infrared (IR) and ultraviolet (UV)
cameras to enable it to see through smoke and
detect sources of excess heat. SAFFiR is also
capable of walking in all directions, balancing in
sea state conditions, and traversing obstacles
such as “knee-knocker” bulkhead openings.
SAFFiR
“Today’s display demonstrates the integration of
perception through multiple sensors, and of lo-
comotion through biped walking,” said Dr. Daniel
Lee, director of the General Robotics Automa-
tion, Sensing, Perception Lab and professor at
the University of Pennsylvania. Tasks as humans
we take for granted, such as standing and
remaining upright, become increasingly complex
with the addition of full body mobility required for
walking and lifting. Dr. Brian Lattimer, associate
professor at Virginia Tech’s Department of Me-
chanical Engineering, additionally commented
that what we are now seeing is the result of a
multidisciplinary project combined to perform all
the critical tasks necessary for fire suppression
by a humanoid robot.
“In dark or smoke-occluded and noisy environ-
ments found in shipboard firefighting conditions,
SAFFiR is being designed to move autonomously
tactile feedback — touch — is an important form
throughout a ship to learn ship layout, interact with
of communication between human firefighters,” people, patrol for structural anomalies, and handle
said John Farley, project officer of the fire test many of the dangerous firefighting tasks that are
ship ex-USS Shadwell, NRL Chemistry Division. normally performed by humans.
“Moving forward, the team will integrate NRL’s
human–robot interaction technology with the
SAFFiR platform so that there is a greater focus
on natural interaction with naval firefighters.”
In the short term, however, to protect
robotic mechanisms and electronics from
intense heat, researchers in the Advanced
Materials Section of the NRL Chemistry Division
have developed a class of lightweight, high-
temperature polyetheretherketone (PEEK)-like
phthalonitrile resin that can be molded to any
shape and remain strong at temperatures up to
500 degrees Celsius. The robotic teams are ex-
pecting to soon conduct shipboard trials aboard
Shadwell, the Navy’s only full-scale fire test ship,
moored in Mobile, Alabama.
By Daniel Parry
NRL Public Affairs Office
SAFFiR taking a break after a hard day at work.
WINTER 2014 17NRL Technology Assists FDNY David DeRieux (right) of the Naval Research Laborato RFID tracking system grew out of the realization during the events of 9/11 that first responders did not have FDNY vehicles and helps to answer, Who exactly is on scene? Where are they? Are they safe? NRL received saving work. See details on page 35. (Photo, left to right: George Arthur, Dan Orbach, and David DeRieux.)
ory invented a system for Fire Department New York (FDNY) to automatically track firefighters. NRL’s active- a reliable method to account for all members responding to an incident. The system is now in use on 15 d a 2014 Federal Laboratory Consortium Award for Excellence in Technology Transfer for this potentially life-
NRL FEATURES
Fuel Cell
A benthic microbial fuel cell prospect of using BMFCs to power power generation: to determine the
(BMFC) is an oceanographic these sensors indefinitely, or at least effects of different oceanographic
far longer than possible with batteries, and biogeochemical parameters, to
power supply that can gen- is enticing – we can acquire long-term pair prototype BMFCs with sensors,
erate power indefinitely. The uninterrupted data and significantly and to develop easily deployable
BMFC sits at the sediment/water reduce the cost and logistics burden configurations. We have successfully
(benthic) interface of marine environ- of keeping sensors running. A BMFC powered many useful devices with
ments, where it generates electrical is maintenance free and nondeplet- small-scale BMFCs (less than 0.1
power using organic matter naturally ing: the organic matter and oxygen watt continuous output), and have
residing in marine sediments as its are constantly replenished by natu- just begun development of full-scale
fuel, and oxygen in overlying water rally occurring diffusion and advec- BMFCs (greater than 1 watt continu-
as its oxidant. At the Naval Research tion; the electrode catalysts consist ous output).
Laboratory (NRL), we are developing of self-forming biofilms comprised of
BMFCs to power persistent, in-water microorganisms naturally inhabiting Early Laboratory Experiments
intelligence, surveillance, and recon- the benthic interface; and there are The bottom sediment of many
naissance capabilities presently lim- no moving, degradable, or depletable marine environments is reductant
ited in operational lifetime by battery components. enriched, due to metabolic activity
depletion. of sediment-dwelling microorgan-
NRL has been a leader in developing isms, whereas the overlying water
Thousands of battery-powered sen- the BMFC and optimizing it for Navy is oxidant enriched due to supply of
sors are deployed each year that operational use. We have conducted oxygen from the atmosphere. This
provide valuable scientific informa- laboratory and field research to study transition is referred to as the benthic
tion about marine environments. The the microbial activity underlying redox gradient. As a result of this
20 SPECTRANRL FEATURES
The combined power from six small-scale BMFCs was used to persistently operate a
meteorological buoy (center) in the Potomac River. The buoy measured local weather
conditions and transmitted data every five minutes to a land-based receiver.
gradient, when an inert electrode is replenished at the cathode, and mass reduction of insoluble iron oxide
stepwise inserted into such sedi- transport (assumed to be diffusion) mineral deposits. Such microorgan-
ments from overlying water, its open supplied the anode reactants and isms are fascinating (especially to
circuit potential shifts by as much as removed the anode products. A most an electrochemist like me) for their
−0.8 volts. interesting result was that current ability to transport respired elec-
increased over time to a steady-state trons from inside the cell to insoluble
Taking advantage of this natural volt- level. In most electrochemical experi- oxidants residing outside the cell
age gradient at the sediment/water ments, current decreases over time (referred to as extracellular electron
interface, my colleagues and I set due to depletion of the reactant and/ transport), acquiring a small amount
out to create a battery for powering or diminished catalytic activity of the of energy for themselves in the pro-
oceanographic sensors by embed- electrode. I remember being dumb- cess. It is thought that D. acetoxidans
ding one electrode into reductant- founded watching the current rise. in a BMFC anode biofilm oxidizes
enriched marine sediment, where it organic matter in marine sediment
would act as an anode, and plac- We immediately followed with another and transports the acquired electrons
ing the other in oxidant-enriched benchtop experiment using graphite to the underlying anode as if it were a
overlying water, where it would act electrodes for the anode and cathode, mineral deposit, effectively catalyzing
as a cathode. Our first experiments resulting in a significantly higher cur- the anode reaction.
involved a benchtop shoebox- rent density. After the system gener-
sized aquarium containing marine ated power for some time across a Early Field Deployments
sediment, seawater, and a platinum resistor, we removed the anode from In one of our first field experiments,
anode and platinum cathode, which the sediment and examined it to find at the Rutgers University Marine Field
generated miniscule current across a a biofilm adhering to its surface — a Station in the Great Bay, New Jer-
resistive load. The notion at the time coating comprised of a community of sey in 2002, a BMFC with graphite
was that the anode was oxidizing microorganisms from the sediment. A electrodes generated power continu-
microbial-generated reductants in the genetics-based analysis of the biofilm ously for nine months without indica-
sediment (such as sulfide), and the indicated that it was enriched in a tion of depletion in power before the
cathode was reducing oxygen in the specific class of microorganisms, Del- experiment was terminated. This and
overlying water. Since the net reac- taproteobacteria. The microorganism subsequent longer-term field and
tion is thermodynamically favorable available in pure culture most similar laboratory experiments revealed an
(electron transfer from a reductant to the microorganism enriched on the important attribute of microbial an-
to an oxidant, just like in a battery or BMFC anode was Desulfuromonas ode catalysts: as living entities meta-
fuel cell), it was reasonable to ex- acetoxidans, a dissimilatory metal- bolically benefiting from the reaction
pect that power could be expended reducing bacteria (DMRB) found in they catalyze, they self-maintain
across a resistive load connecting marine sediment that couples oxi- themselves and do not degrade over
the electrodes as long as oxygen was dation of sedimentary acetate with time as long as their environment
WINTER 2014 21NRL FEATURES
is hospitable. In this way, they can powered by the BMFC. This required We are working to optimize the BMFC
maintain catalytic activity indefinitely. a novel solution to convert the low design to maximize power output
Analysis of the anode biofilm of the DC voltage output of the BMFC from while making anode embedment a
New Jersey BMFC indicated enrich- 0.35 volts (dictated by the benthic simple and reliable procedure. Maxi-
ment not only of D. acetoxidans, but redox gradient) to 6 volts to charge a mizing power involves anode designs
also of microorganisms in the Desul- capacitor, which in turn was used to that expose as much mass-transport-
fobulbus or Desulfocapsa genera that power the buoy. The capacitor was accessible surface area to sediment
oxidize and disproportionate sulfur. needed to buffer the low but steady as possible within mass and volume
In sulfide-enriched sediment such as power output of the BMFC with the constraints of the intended deploy-
the New Jersey site, a graphite an- duty cycle of the buoy, whereby ment platform, all while being easy
ode can develop a passivating sulfur nearly 99% of energy consumption by to properly embed into sediment. We
precipitate on its surface that can the buoy occurred during the fraction have been making steady progress,
shut down current. It is thought that of a second during which the data hampered at times by the vagaries of
these microorganisms act to clear the was transmitted. Between trans- field research: in 2012, all our mea-
sulfur from the anode surface, form- missions, the BMFC recharged the surement equipment and 26 BMFCs
ing sulfate and sulfite, both soluble. capacitor. This was the first time that deployed at a New Jersey site were
a BMFC was used as a free-standing lost when directly hit by Hurricane
The next BMFC field experiments power supply. The buoy ran flawlessly Sandy.
were performed in 2003, at 1000 for seven months. The river froze that
meters depth in the Monterey Can- winter, trapping the buoy in ice, but In 2013, we began development of
yon off the coast of California on a it kept running until the ice thawed full-scale BMFCs in NRL’s newly
cold seep, a fissure on the seafloor enough that large pieces began to opened Laboratory for Autonomous
effusing organic-rich water. We had flow down river, dragging the buoy Systems Research (LASR), where we
hypothesized that the high mass and severing the mooring line con- have assembled a six-meter-diameter,
transport rate of organic matter necting it with the BMFC. I got a call 8000-gallon benthic mesocosm,
would benefit BMFC power gen- from the Coast Guard who found the essentially a large indoor aquarium
eration and we were correct. This buoy resting against a pier of the Wil- filled with sediment and seawater to
BMFC, equipped with a spear-like son Bridge. (Lesson: always put your replicate the benthic interface. This fa-
graphite anode inserted vertically phone number on any oceanographic cility enables full-scale BMFC testing
into the seep, generated significantly instrument you want back.) in a controlled environment, greatly
more power per unit footprint area reducing the cost and risk of BMFC
and geometric surface area of the A Practical Power Supply design development compared to field
anode compared to our earlier ex- Since 2004, a number of researchers testing. We have already made seven
periments. This was exciting because have been working to develop BM- deployments of a two-meter-diameter,
it demonstrated that placement of a FCs into practical power supplies that 900-kilogram oceanographic mooring
BMFC in an area with a high rate of are straightforward to deploy. I am equipped with various BMFCs to test
organic matter mass transport, such fortunate to pursue this concept with new designs. The LASR mesocosm is
as a subliming methane hydrate out- my NRL colleagues Dr. Jeff Book, Mr. greatly compressing the BMFC devel-
crop, could result in very high power Andy Quaid, and Mr. Justin Broders- opment horizon and I feel we are on
output, which could be of great use en, oceanographers; Dr. Yoko Furuka- track to transition the BMFC in 2015.
to the Navy. wa, marine geochemist; and Dr. Jeff
Erickson and Mr. Marius Pruessner,
We next deployed a BMFC in 2004 in engineers; with funding from the Of- By Leonard Tender
the Potomac River off the NRL pier. fice of Naval Research and NRL. We NRL Center for Bio/Molecular Science and
Here, the BMFC consisted of graph- have deployed small-scale BMFCs in Engineering
ite electrode slabs attached with zip coastal locations including California,
ties to the top and bottom of milk Gulf of Mexico, Maine, New Jersey,
crate spacers that were positioned North Carolina, and the Adriatic Sea.
on the river bottom with one elec- These BMFCs have operated over
trode in the mud and the other in the durations ranging from less than six
overlying water. An attached buoy months to more than two years with-
floating on the water surface mea- out indication of depletion in power
sured air temperature, pressure, rela- output. They have powered a hydro-
tive humidity, and water temperature, phone with a radio transceiver link, an
and transmitted this data to my office acoustic modem, and a surveillance
every five minutes. The entire buoy camera with a cellular link.
including the radio transmitter was
22 SPECTRANRL FEATURES
BMFC graphite bottlebrush cathodes.
Dr. Leonard Tender (above, and with Marius
Pruessner at right) working in the benthic
mesocosm in the Laboratory for Autonomous
Systems Research. The mesocosm contains 18
inches of seawater and 24 inches of synthetic
marine sediment that has been inoculated with
real marine sediment to provide necessary
microbial activity. On the water surface are BMFC
graphite bottlebrush cathodes; when deployed in
the field, these cathodes are beneath the water The benthic mesocosm in the Laboratory for
surface but above the sediment surface. The Autonomous Systems Research.
vertical tubes indicate the locations of BMFC
anodes embedded in the sediment.
WINTER 2014 233-D Printing in
NRL FEATURES
Hydrogen Fuel Cells
for Unmanned Systems
The NRL crew after the successful flight of a 3-D printed fuel cell aboard the Ion Tiger UAV. Left to right: Dan Edwards,
Drew Rodgers, Ben Gould, Steve Carruthers, Mike Schuette, and Karen Swider-Lyons. Not shown: Chris Bovais.
The Naval Research Labora- In a recent project, an NRL team The only byproducts are water and
leveraged 3-D printing and other low-grade heat, and no combustion
tory (NRL) has been leading
prototyping tools in NRL’s Laboratory occurs. The electrochemical reactions
the development of lightweight for Autonomous Systems Research are efficient, resulting in systems that
hydrogen fuel cell systems for (LASR) for developing and integrat- are over 50% efficient and operate at
ing their next generation of hydrogen low temperatures (near 80 degrees
long-endurance unmanned air
fuel cells into unmanned systems. Celcius). Combined with the high
vehicles (UAVs), starting with Additive manufacturing processes energy of hydrogen, the result is an
the modest 3-hour flight of the such as 3-D printing promise rapid electric power source that is light-
prototyping of new designs so they weight and has high energy per unit
Spider Lion in 2004, followed
can be tested and verified without the weight. For the Ion Tiger system, this
by the 24-hour flight of the Ion expense of traditional manufacturing. meant a sevenfold increase in endur-
Tiger in 2009, the 48-hour flight The team used new tools in the LASR, ance over battery propulsion when
plus commercial tools, to investigate gaseous hydrogen was used for the
of the Ion Tiger in 2013, and the
how 3-D printing can be used toward 24-hour flight. Hydrogen fuel cell sys-
deployment of the XFC tactical improving fuel cells for unmanned tems also respond rapidly to changes
UAV from the torpedo tube of a systems. in throttle, and are easily turned off
and on. The fast response time and
submarine in 2013. NRL also as-
The motivation for NRL’s research into high energy density are two reasons
sisted in a demonstration flight hydrogen fuel cells is clear. Fuel cells major automobile manufacturers such
of Insitu’s ScanEagle vehicle on directly convert the energy in hydro- as General Motors, Toyota, Honda,
gen to electricity by electrochemically and Daimler have invested cumula-
a hydrogen fuel cell in 2010.
combining with the oxygen in air. tively $10 billion to $12 billion in hy-
24 SPECTRAYou can also read
