https://ntrs.nasa.gov/search.jsp?

R=20040010820 2020-07-26T17:16:21+00:00Z

ACTIONS,
OBSERVATIONS,
A~YDDECISION-MAKING:
BIOLOGICALLY
IYSPIRED
STRATEGIES
FOR AUTONOMOUSAERIALVEHICLES

Greg Pisanich ~ a r r A.
y Young'
Corey Ippolito Benton Lau
Laura Plice
euC2emaii.arc.nasa.gov l a m .a.vouneGnasa.gov
QSS Group Inc. ArmyMASA Rotorcraft Division
Compurational Sciences Division
NASA Ames Research Center
Moffett Field, CA

ABSTRACT vehicles as.being highly mobile robots capable of three-

dimensional a c ~ e s s l - ~ .
This paper details the development and demonstration
of an autonomous aerial vehicle embodying search and One new emerging application for aerial flyers is their
find mission planning and execution srrategies inspired poIential use in planetary science missions. In order to
by foraging behaviors found in biology. It begins by find acceptable utility in a planetary science role,
describing key characteristics required by an aeria! though, it wil! be essential to develop robust,
explorer to support science and planetary exploration computationally efficient, and mission-effective means
goals, and illustrates these through a hypothetical of effecting autonomous flight'mission control of these
mission profile. It next outlines a conceptual bio- future planetary aerial vehicles, or "aerial explorers."
inspired search and find autonomy architecture that This is the fundamental goal of the NASA "BEES for
implements observations, decisions, and actions Mars" project4.
through an "ecology" of producer, consumer, and
decomposer agents. Moving from concepts to This paper builds upon preliminary work from the
development activities, it then presents the results of "BEES for Mars" project investigating the feasibility of
mission representative UAV aerial surveys at a Mars defming and implementing bio-inspired flight'mission
analog site. It next describes hardware and software behaviors for autonomous aerial vehicles, acting as
enhancements made to a commercial small fixed-wing surrogate demonstrator platforms for aerial explorers
UAV system, which inc!nde a ncw dpvelopnent far futiire Mars robotic science missiaas5-6. - As an . -
architecture that also provides hardware in the loop offshoot of that initial effort, the "Intelligent Aerial
simulation capability. After presenting the results of Vehicle (IAV)" project has made significant progress
simulated and actual flights of bioinspired flight towards employing bio-inspiration in developing aerial
algorithms, it concludes with a discussion of future expiorer mission scenarios and autonomy architecture
development to include an expansion of system concepts.
capabilities and field science support.
Bio-inspiration can be a powerful tool when applied to
engineering problems', particularly the development of
INTRODUCTION intelligent systems. Animals and Man can be described
as organisms Char operate under a simple be'naviorai
A..tnm---..
ULWLl"llLWUS
.
a r A ' c J
ah;,-I,
J.d,L",bS
nn.oJ
y";oru
LI,r
+- ,-alrnl,,t;
L V lrlwLULlGl'lLU modei. Tiley are moiivaied io do kcitoils, because of
aeronautics. New missions and applications are being the changing state of their emotions and what they
deiiied and demonsn-ared ar an accelerating pace. sense, or Observations they make, in the worid.
Vehicle autonomy will ultimately drive aerial vehicle
design. leading to radically new vehicle configurations In this paper, we describe how the thoughts -- or
and concepts that no longer have to be constrained to Decision-Making -- and actions of animals can be
the limitations of manned flight. Further, many of these described using a behavioral model. We then describe
emerging applications will redefine the classic how a similar model can be applied to autonomous
perception of aerial vehicles into one that sees these uninhabited aerial vehicle (AUAV) operation, and how
inspiration !?om Nature can yield new approaches to
'Senior member, Aerospace Engineer AUAV search and find strategies.

1
American Institute of Aeronautics and Astronautics
 L

PREVIOUS WORK experience with logistics, operational limitations, and

the numerous science opportunities associated with the
Preliminary work on the IAV effort focused on the Mars analog site. This experience proved to be
definition of bio-inspired mission concepts and flight invaluable for follow-on technology and field-science
“behaviors” that would successfully effect Mars aerial demonstrations at Haughton.
explorer demonstrations with terrestrial surrogate
vehicles.

Mission concepts’-6 derive from situations in the

biological world that are oriented toward “search and
find” requirements: a mission using dropped aerial
probes based on dissemination and survivorship curves,
and a terrain-influenced search trajectory derived from
predation strategies. A small compendium of bio-
inspired behaviors was defined for an aerial vehicle.
These behaviors were categorized into general groups:
UAV “primitive” tasks, actions, observations, and
planning/decision-making.The use of stochastic search
strategies wzs a!sc emphzsized in this pre!iminar;.
work.

Figure 1 illustrates an early IAV implementation

demonstrating conventional search and find behaviors.
The demonstration used a 2-meter wingspan UAV
system (Fig. 2a.) that had been enhanced to include a
ground based adaptive decision component. This
decision component allowed the aircraft command
sequence to be altered based on images that were
recognized from video downlinked from the aircraft.
Fig 2. (a) MLB BAT UAV, and (b) Aerial View of
Target (“Orange Tarp”).
e

AERIAL EXPLORERS: M O W T € U X IMAGING FROM

ON-HIGH

In general, prior proposals for Mars flyer missions have

envisioned these vehicles primarily as an imaging
platform. The goal of a Mars flyer, then, has been to
maximize the overflight area covered, taking photos as
it went, until it inevitably crash landed. Two general
Fig i. Eariy UAV B Drop Probe Demonstration. approaches have been proposed for the launch of the
flyer: air-depioyment during entry and descent, or
The mission entailed searching for a target (in this case ballistic launch from a lander or rover. The goal of a
a large orange tarp on the ground, Fig. 2b.), visually flyer that would be shot out from a lander or rover
recognizing the target, and dropping a small aerial might be to peer over the horizon at terrain that might
imaging probe atlonto the position of the target, before be encountered subsequently on a traverse. These
resuming its search. applications of Mars aerial vehicles view the vehicles as
a sensor platform that was an extension of another
We conducted a series of UAV flights at the NASA system (get up high, stay up as long as you can, look
Haughton iVars project on Devon Island, Nunawt, down, and tell me what you see).
Canada. On this initial visit, video and flight data were
gathered from several UAV flights. We also gained
2
American Institute of Aeronautics and Astronautics
 Our vision for an aerial explorer on Mars is an Aerobot' configurations than those for terrestrial UAVs. These
that would consist of an aerial platform, science sensors aircraft design issues are beyond the scope of our work.
(onboard and deployable), an intelligent system (that Nonetheless, despite the physical differences between
can make decisions on its own based on sensor input), planetary (and terrestrial) aerial vehicles, they will have
communications (networked to other subordinate, co- many common autonomy challenges.
equal, or lead systems or aircraft), and a sophisticated
mission profile and goals (that would allow it to make Because of communication time lag (for example,
real time decisions to reconfigure flight goals and several minutes one way toifrom Mars), planetary aerial
reallocate resources based on data measurements). explorers must be fully autonomous. They will not
have the ability to wait a solar day for Earth-based
In this project, we are striving to maintain balance decision-making or an operator in the flight control
between two development goals. The first is that we loop. Planetary UAVs will also have other limitations.
are working to develop an aerial explorer whose They will have limited flight endurance (15 minutes to
primary purpose is to maximize science return. But, 3 hours, by most estimates), no opportunity to refuel,
second, we must also keep in mind that systems such as limited landing opportunities, and little chance of
this will need to operate in a planetary environment, multiple flights. Even with satellite data and advance
with different operational requirements. The next planning, they may have limited prior knowledge of the
paragraphs discuss characteristics of each of these area over which they must fly.
gmls.
These limitations will drive the design of the platform,
Aeriai Explorers for Science -&fission sensors, and the overall mission. With no requirement
to return and land, planetary UAVs may be designed to
UAVs optimized for science return (particularly for fly missions that are broader in scope, than terrestrial
NASA science missions) will likely have different applications. Once an area of interest is located, the
vehicle configurations and mission goals than aerial explorer may need to make decisions on
conventional UAVs. These aerial explorers must resources available, against further exploration
contain imaging (video or photographic) and sensing opportunities. Information returned by drop pods or
capabilities (spectroscopy, atmosphere sampling) that other sensors will provide data to make these decisions.
may be similar to some terrestrial applications (such as Finally, when the aerial explorer does terminate its
military UAVs that may be looking for inorganic flight, it should also have the capacity to safely carry
objects in an organic world). But high altitude sensing instruments to the ground and provide added science
is a necessary but not sufficient fimctionality for return. Where possible, the aerial explorer should also
science capture. 'Our vision of an aerial explorer must be able to choose that landing site in order tc, mziinize
also be able to perform close-in investigation of areas these opportunities.
of interest, to inciude low altitude flight, drop pods, and Gaining operational experience with UAVs in planetary
other deployable sensors. analog sites such as Haughton Crater, or Devil's
Playground in Utah, should continue to provide us with
We are working toward a notion of field science, an a better understanding of what characteristics should be
activity where a scientist is interacting dynamically built into aerial explorers.
with the environment. With access to geologists and
life scientists at NASA Ames, our challenge is to work A Notional Mars Aerial Exulorer Mission Profile
in cocjucctioc with them to Ese UP.Vs to solve t!eir
By working with terresaiai iieid, and pianetary,
science goa!s. Understanding their methcdology will
scientists, we are working to define potentiai mission
help us to better eihance the UAV (sensors,
profiles that continue to build upon the aerial explorer
programming; and mission design). This improved
feature set. The following is an example of a mission
understanding of the field scientist's methods should
description for an aerial explorer that would be tasked
also allow this knowledge to eventually be codified and
to search for water outflow patterns on Mars. These
integrated into vehicle autonomy development.
outflows, that some scientists theorize may still be
active on Martian hillsides (possibly due to seasonal
Aerial Exulorers for Planetary Missions
snow-melt), are prime candidates for investigation as
they may provide important clues in the search for past,
Aerial vehicles developed for planetary missions (for
or still existent, life (Fig. 3).
example, Mars, Titan, Jupiter, and Venus2) must
necessarily embody radically different design
3
American Institute of Aeronautics and Astronautics
 maneuvers. Drop pods containing tethered elements
(Le. “tetherbots”) that would drapeicatch on the hillside,
as they float down, and return data for an extended time
would also be deployed as close as possible to the
outflow area. The result would be a three-dimensional
data map of that site.

If the explorer had sufficient resources remaining, it

could move on to another site. With low reserves, it
would make the decision to terminate its flight in this
area. In addition to the drop pods, the aerial explorer
would also contain contact sensors on the nose, wing
and hselage of the aircraft that could survive impact
with the terrain. By setting up an approach and
Figure 3. UAV photo of hillside snowmelt water-
impacting as close as possible to the outflow area, the
outflow at Devon Island
aerial explorer would provide additional data as long as
In such a hypothetical mission scenario, an aerial power was available. This “survivable crash” would
explorer is deployed from an entry vehicle and also address on of the key problems identified for Mars
descends via parachute to a high search altitude flyers: namely it would provide adequate time for
(-10,OOOft). It separates from the parachute and gathers uplink/transfer data from the aerial vehicle to satellites
a panoramic image and long range sensing data of the and/or directly to mission contio!. Most projected Mars
area around it (terrain, spectroscopy). Upon analysis, flyer flight times are of too short of duration to in-flight
areas of interest are prioritized (flat areas with nearby transmit gigabytes of imagery and other large data sets.
hills, areas showing better spectral results) and the
explorer heads in the direction of the most promising PROJECT
DEVELOPMENT
PHILOSOPHY
sectors. At each decision point in the mission, the
decision, the data supporting it, and subsequent plans There is obviously a large technology gap between the
made by the explorer would be transmitted back to scenario described in the previous section and what is
mission control via satellite relay. possible today. Our work is focused on developing
technologies that will help bridge this gap. Our
As the explorer descends to an intermediate analysis philosophy is grounded in a process of concept
altitude (-2OOOft) it begins a random behavioral search development, simulation, local flight tests and field
of the first area. looking fcr image (image recoglliiloii, demonstrations. This strategy is not unique, but works
rock distribution) and spectral (soil and rock well for our applications.
composition) evidence of riverbeds. The explorer
continually monitors resource constraints, (fuel, power Our development efforts 011this project hcve focused
and time) ana decide whether to prolong a search in a on the creation of biologically inspired flight algorithms
particular sector or direction, based on the state of the and mission concepts. These have been developed
science return. through consultations with domain experts such as
biologists and planetary scientists. The results of which
Upon sensing high evidence of outflows. the explorer
have been algorithms and mission scenarios.
would commit to an area and descend to a low sensing
altitude (-5OOft). It would then use a combination of
We have developed a simulation eiivirmment and
data sources (siope ioiiowing, rock dispersai
architecture (the “Reflection” architecpire, described
distributions, linear feature following) to lead the
!zter ig this pzper) t ! ! ~ al!ows ES to progrcm these
explorer toward the potential source of the outflow.
algorithms and mission scenarios into a simulation of
Drop pods would be dispersed over these areas to
the aerial explorer, and the environment, and test fly the
gather additional information on rock material, contact
resulting mission. As this architecture also allows the
sensing, and visuals of flows at or near ground level.
software to be run on the flight hardware, further
As a potential source of outflows is identified (through checks can be made on the flight readiness of the
spectroscopy, image, and terrain clues) the aerial system.
explorer would extensively map the local area using its
sensors through a series of circling and direct approach As NASA Ames is co-located with the Moffett Federal
Airfield, we have the opportunity to test the resulting
4
American Institute of Aeronautics and Astronautics
 flight code on our UAV. Some of the advantages to BIO-INSPIRATION
& A SEARCH Au~ovokrv
A-W F L ~ D
flying at Moffett Field are that we can fly often and the ARCHITECTCRE FOR A E R U L VEHICLES
airfield provides clearance from other aircraft and RC
frequency control. These are great advantages to our In our previous work’, we described how natural
work. Some disadvantages of working on an operating behavior (bio-inspiration) could be used and combined
airfield are that we have to work in parallel with to achieve mission goals. In the following sections of
manned flights within an airspace that can limit the this paper, we expand on this early work to discuss why
mission scenarios and maneuvers we can attempt. The the use of bio-inspired behaviors may be ideal for
limited topography that the airport presents can be science and planetary missions. We also expand on our
overcome by providing the aircraft with simulated early ”robotic ecology“ work to describe a holiarchy
terrain and sensor data. There are additional limitations architecture based on information.
such as scheduling flight times with other activities, and
the very dense frequency congestion in the Silicon Search and Find for Planetarv Science
Valley, but these limitations are overshadowed by the
ability to fly test fly our work often. There are several reasons why conventional search
algorithms, such as a grid search, or a search around a
The field demonstrations are the ultimate evaluation of last known position, would not work well in a planetary
our vehicle autonomy work. Our goal is to test these science environment and biologically based algorithms
mission scenarios and algorithms in scientifically x s y provide betki pei-fa-zaixe.
interesting analogs of Mars terrain (Mars-analogs). Our
intent is work with scientists that are familiar with the The worid and its extent may not be -known at
sites and have specific science return goals in mind. the onset. Information about the area may be sketchy at
Some challenges to working at field sites include the best (satellite maps, rough spectroscopy). With this
logistics involved in getting to these often remote sites, state of knowledge, aerial exploration would most
flying within terrain constraints, and working with any likely start a search in a promising direction. This
airspace or frequency limitations. All of these factors method points to a directed random search with the
are manageable compared to the valuable lessons- capability of narrowing focus when one or more terrain
learned by flying these aircraft in “actual’? conditions. “features of interest” are identified.
In working on the creation of mission concepts for Goals are hard to define from the outset. A
Mars flyers we have set limitations to the scope of our hypothesis is often put forth, but can be easily amended
effort that have helped maintain satisfactory progress in depending on what information is encountered. What
OF domain. The f k t is that we currently expend was originally searched for -may not be faund, or
limited effort into the development of sensors and something totally unexpected -- but related to another
sensor software and focus primarily on the flight and hypothesis -- may be encountered.
mission aspects of the problem. We use surrogate
sensors (such as the visual recognition of a large orange Limited resources point toward covering larcer
tarp) and simulations as stand-ins for the lightweight regions rather than concentrating onlv on a small
opticalispectral sensors that are still in development. search. Rather than an exhaustive search in a single
By focusing and expanding on the flight and mission location, characterizing an area to some level and then
aspects, we will be ready when these sensors are moving on to additional sites could be preferable. As
ultimately available. the exp!orer wou!d prob251ji cot r e k m to previously
Secor,d!y, a!th=ugh we are a-xare of Lke cons::aints ix,d visited sites, the decision-making employed to
challenges of flight in the atmosphere of iMars, we are determine when to move to a new sitehearch becomes
dortant.
very irL
nor. focusing on tine design of a Xars capabie aircrafc.
We instead concentrate on enhancing and operating a
Scattered areas of interest mav be located far
small commercial fixed-wing UAV to meet our
from each other. separated bv emptv areas. Exhaustive
demonstration goals. When a more advanced design
search in this environment would quickly deplete
becomes available, we should be able to replicate its
resources. The need to look at the world kom varying
resource, sensor, and flight characteristics in our
vantage points (high altitude with low sensor fidelity
simulation and adjust our missions and algorithms to
versus low altitude with high sensor fidelity) points
match that aircraft and goals.
toward the use of “fox and mouse” tactics’.

>

American Institute of Aeronautics and Astronautics

 Genera I Software Arch itecture
“Economy of Nature” Inspired Model
“Decomposer”
Raw Data to Information or Updated Mission Profile &
Knowledge Complex Behauors to simple FCC-compatible
instmc!ions; mission execution 8 operationai overslght
-Science Instrument &
Vision-System Analysis
-0eveloplMaintainWorld

I
Three Semi- Autonomous
Processes on MC; with independent
Watchdog fiinctions

“Consumer”
Information. knowledge &operational data to updated
Mission profile & complex behaviors

.Ongoing assessment of successffailureof mission

.Optimizationor trade-off of energy versus information
Deasion-making or choices in response of assessment,
which results in dynamic updating of mission profile &
Raw Data allowable complex behaviors

FCC Instructions

Fig. 4. Holarchy Conceptual Model of Aerial Vehicle

Natural marriage between bio-inspired search energy and information is assumed needed to
and fmd strategies and statistical random samuline successfully model and sustain vehicle operation and
theorv. Rather than employing an exhaustive search of execution of the planned mission(s). The gathering of
an area, a random sampling (by means of an intelligent Liformation may be regarded as the motivating purpose
random walk flight behavior, for example) of a for sending autonomous vehicles to Mars. The role of
statistically significant portion would yield data that information producer can be filled by instruments and
could be predictive of the actual result. probes that collect information directly !?om the
planetary surface or atmosphere. Information
Robotic Ecolonv consumers take data from producers for processing and
may release information in a different form to other
We are working toward a general architecture for the information consumers.
development of autonomous air vehicles. Although this
will initially involve the use of conventional iMultiple information processing roles may be
programming and representation techniques, we are implemented onboard the same hardware platform. The
striving to move toward a behavioral architecture and steps in the information “food chain” show a similarity
programming model. In our current model, we use a to the layers of abstraction familiar in the design of
producer, consumer and decomposer model borrowed robotic software. Researchers on Earth may be the
fi-om n a t ~ r e ~where
-~ ihe producer provides input ultimate information decomposers about the Maitiaii
(observations) into the system, the consumer evaluates surface.
the data and makes decisions about the input and the
current stite of the machine, md the decoxposer Table 1 is an illustrative ex.mp!e of 2 M x s mhotic
translates those decisions into selections of actions that “ecology” employing aerial explorers. The large
are applied to the world. Figure 4 shows a conceptual number of robotic systems in the ecology is unlikely for
schematic of a holarchical. or “ecology,” architecture as Mars missions in the near future. However, the large
applied to a single aerial vehicle and its sub-systems. A number and variety of robotic systems is entirely
holon is an individual member of the holarchy. possible/feasible for field science testing at Mars analog
sites. Such an ecology has already been mimickedg-’’.
The description of a biological community is based
primarily on the flow of energy. For a bio-inspired
community of autonomous vehicles an analogy of
6
American Institute of Aeronautics and Astronautics
 Table 1. Examples of a Holarchical Taxonomy for if systemsisub-systems acted in isolation from each
Autonomous Aerial Vehicles. other. An illustrative example of an interactive ecology
of aerial explorers and other systems/subsystems
(building off of Table 1) is

These ecology roles and relationships can be initially

prescribed, but more sophisticated modeling may allow
them to evolve with time and information
resourcesidynamics. In this case the ecology
interaction can be moderated by means of techniques
for mzthematical iiioiieling of population dyiiziiics
from biology.

ObservationsProducer In an intelligent air

vehicle, observations may come from several sources.
Some may come from the sensors themselves. These
would include control sensors, video, temperature.
Sensor data could be fused to provide additional
information.

rover-launched Observations could also come from state data,

rocket slider providing a type of memory for the system. This could
Producer & involve trend data (aircraft is descending) or state
#3,a “Mars Consumer change data (we have 2 pods remaining). Observations
rotorcraft”
would also include decisions rnzde and actigns taken by
Producer &
ground deployed Consumer the air vehicle. Observations, as they are updated would
System Producer & be made available to the decision-making system.
IConsumer
BASE Base camp, or Sysrern. man cVirimarej Decision-Maitinw’Consumer Tine decision-
mission conno1 machine /Consumer making component would consume (evaluate) the data
provided and decide whether the data observed was
To fully defme such a robotic ecology is necessary to bringing the aircraft closer towards the flight goals.
define individual systedsub-system roles and
This could be achieved through a conventional
relationships. Table 1 touches upon some of the roles
that might be applicable for Mars exploration, entailing, - -r--- the
ime!!iger,r sysrem 2rchittjctw-e r h t w o d d r-rwent
goals and processing to achieve them as a conventional
in part, the use of aerial explorers. There are four
planner and execution system. The air vehicle would
possible relationships between individual systemshb- be conkolied by pr&FU?ed i;!&?s and ,---2:+;n-q
~“LIUICI”*.U!

systems: independenthsolated (numeric value of 0);

execution of routines based on the data and resources
subordinate (0.5); “self” identification (1); equal (2);
sensed. The output of the decision system would be a
lead (3). For the systems with an “equal” relationship,
single or sequence of behaviors to be performed by the
there can be sub-classes of that relationship:
decomposer. Alternatively, an “emotional holon”
cooperative (numeric value of 2.1); neutral (2.2);
decision-making processimethodology might be
competing (2.3). These relationships between
employed’ (refer to Fig. 6).
systemshbsystems, A, within the ecology hierarchy
can be defined by a “relationship” matrix, R, such that The following set of ~~emotions”
could be proposed for
A-+RA. The matrix R is simply the ‘‘identity” matrix the autonomous aerial vehicle conducting search and
7
American Institute of Aeronautics and Astronautics
 find scientific investigations: “fear” (Le. real-time ActionsDecomuoser The Decomposer would
assessment of level of risk); “desire” (current implement the actions called for by the Consumer
probability of mission success); ’-confidence” (real-time component. These high level commands would be
assessment of system capability, e.g. amount of fuel broken down by the aerial vehicle mission computer
remaining, number of onboard probes, etc.); “altruism” into “primitive” tasks that could be communicated to
(relative assessment of individual versus group success; and executed by the OEM flight computer - as in the
applicable to multiple robotic flyersivehicles). This case of the small fixed-wing UAV used by the IAV
should be considered a minimum list of emotions project. Actions could involve changing the direction
necessary for autonomous UAV operations; there may and altitude of the vehicle (behaviors), deploying sensor
well be others that could be implemented in. UAVs. pods, reconfiguring the sensor system, adding
additional processing steps to the observations, or
Effecting emotional modeling of aerial vehicle goes sending observations back to the base.
hand in hand with definition of motivations and
personalities of the aggregate of producer, consumer.
Fusion of Concepts into an inteerated strateev
and decomposer agents that comprise the
To arrive at an effective autonomous system
“ecology/economy of mind” that allow for the
implementation for the aerial vehicle demonstrator it is
successful autonomous operation and decision-making
necessary to fuse all of the concepts noted so far into an
required for a Mars aerial explorer6. Implementation of
integrated bio-inspired holarchy autonomy architecture.
an aeria! vehic!e persondity CZQ work if at least f o ~ r
Figures 5 and 6 attempt to present a high leve!
attributes are modeled (refer to Fig. 5):
schematic of an integrated system example.
Passive + .Aggressive
(or, rather, Energy Conservanve) +=versus Energy Profligate)
Risk Adverse + Risk Tolerant
(or, rather, Sunple Sensmgmehavlors) + (versus Complex f
Sensinflehaviors)
a
-
Dissemination Conse:-vativc+ ?rof?fgatc
(r-strate9 wrt poddpenetrarors) + (versus K-smteB)
Social + Asocial
(Large exchange of informanon hetween ‘‘equal” system) -+ (No exchange)
A relative numeric scaling (0-10, for example) between
these “personality” attributes allows the pre-flight
tailoring of the vehicles’ ultimate response to
observations and actions during the flight. Finally, it is
imperative that ”search for water’ and ‘hunt for life’
higher-level goals be heuristically expresseddefined in
terms of simple observational rules and subsequent Fig. 5’. Integrating Personality Attributes & Behaviors to the
Emotional Holon Schema.
UAV actions. This also holds true for safety of flight
and health monitoring. Some illustrative conditional
rules are shown below.

8
American Institute of Aeronautics and Astronautics
 DEVELOPMEXT ACTIVITLES

We have so far examined concepts, possibilities and

methodologies for aerial exploration. In the next
sections, we discuss activities of field experiments,
software and hardware development, and demonstration
flights that were accomplished by our team in 2003.

Field Experiments

iMission-Remesentative Aerial Surveys. The

IAV project: in order to refine its bio-inspired mission
scenarios, invested a significant amount of effort
conducting mission-representative aerial surveys at the
Haughton Crater, Devon Island, Mars-analog site'.
Figures 7 and 8 are two representative aerial survey
image mosaics of the Devon Island site. Both sets of
mosaics were taken during a single UAV flight along
the tits ski its ofthe ~ o i 8rai;ii
i ~!aiiitia"." ~ i g m e7
shows a small valley (images taken with a downward
pointing camera); valleys, and how they .form, are
topics of intense interest to Mars planetary scientists.
Figure 8 shows a ridgeline with prominent rock
outcroppings.

Fig.8 Ridge Line Rock Outcroppings

Drop Probe Release and Imaeine We also

made considerable progress towards demonstrating the
potential of drop probes with imaging cameras, and
other sensors, used in conjunction with aerial explorers.
Testing was conducted at the NASA Ames Moffett
Field airstrip as well as the Haughton Crater, Devon
Island site (Fig. 9). The release of the drop probe has
been fully automated to require on!y a rudimentary
vision-system to recognize a target. redirect the UAV
back to the location of the target, and release the drop
probe over the target location.. This has been
accomplished both with ground-based and onboard
software. More sophisticated drop probes'-'' and UAV
behav-iora: responses are plaiiiied for the filkiie.

Fig. 7 Small Valley on the Outskirts of the Von Braun

Planitia.

Fig. 9 Recovery of Drop Probe.

9
American Institute of Aeronautics and .4stronautics
-
 Lessons Learned from HaurJhton We learned
several important lessons from our field work and
subsequent imagery analysis efforts that would not have
been apparent if we hadn’t visited this Mars analog site.
First, in this area largely barren of life, the notion of our
biologically inspired “fox and mouse” mission scenario,
illustrated in our earlier paper’, became more apparent.
In the fox and mouse scenario, the fox avoids the
mouse-barren fields and focuses his efforts on the tree
line, where the mice venture forth from the briar patch.
In viewing the video of our higher altitude flights, we
were struck by the relatively long distances of empty
plains punctuated by interesting riverbeds, valleys, and
rock outcroppings. These were the areas of highest
science interest.

Second; the results from our higher altitude flights

begged for additional flight behaviors. Seeing
interesting features glide by on the video pointed out
the need of circling flight to loiter over an area and look
closer at objects. Also, seeing a hint of an interestins
area along the edge of the video stream, that would not
be over flown by the current flight path plan, pointed
out the need for a gimbaled camera and the ability to
alter the course in that direction based on the
recognition of interesting objects.

Hardware and Software Development Fig 10. - PC-104 “Mission Computer”: (a) outside of
Small UAV and (b) partially installed.
Processor Integration Our initial development
effort was focused on integrating an additional The flight software on the PCI104 performs the high-
processor to the aircraft to support advanced reasoning level mission planning and video processing for each
and behaviors. This “mission computer” angments the experimznt. The current implementation of the
OEM avionics “flight computer” onboard the UAV software is built on components that were tested in
I used for flight operations. simulation, 2nd consists of three main concurrently
operating components. A data monitoring component
The flight system is built on an MLB “BAT” UAV monitors the serial port connections to the flight
outfitted with a PC/104 based CPU stack that interfaces computer and the radio modem, relaying information
through serial communication lines with the MLB flight back and forth between these two hardware components
computer and the onboard radio modem (Fig. 10). The while maintaining the system state for other software
PC/IO4 receives state information by monitoring components in the system by monitoring the telemetry
telemetry information sent from the flight computer to data periodically by the flight computer. A second
the ground station, and uploads new flight plans to the component processes video from the onboard camera
flight computer. The PCI104 also receives video input system and triggers events that are received by the other
from the onboard camera system that is used for image software components. The third component performs
processing experiments. The PC/I 04 stack is powered the high level mission planning, receiving state
from the UAV power supply and consists of a Digital information decoded from the data monitoring module
Logical SmartCore 700MHz Pentium 3 CPU board, a and events from the video processing module,
DCDC power supply board and a video capture board. processing this information, and then uploading new
flight plans to the onboard flight computer accordingly.

With the mission compEter installed and flight tested,

another key project milestone was completed with the
port of the simple trigger (orange tarp) ground-
10
American Institute of Aeronautics and Astronautics
 computer-based image recognition sofhvare to that concurrency, allowing any module to take part in a
onboard computer. The vision-based target recognition multitasking system without the need to consider
and drop probe work completed in 2002529was complex inter-module concurrency issues such as dead
reproduced using the onboard system. locks, race conditions and priority inversion.

"Reflection" Architecture and Simulator In This architecture also provides a rapid development
parallel with this effort, we have teamed with another capability that also supports rigorous testing of
UAV project at Ames to co-develop an architecture and components and systems throughout development. The
hardware-in-the-loop simulation capability. This scripting functionality allows for multiple system
system allows us to quickly develop and test flight configurations to be easily created during development,
algorithms and software both in simulation and on the allowing development of automated regression tests,
actual flight hardware. It also provides post flight simulated scenario tests, stand-alone component tests,
capability to visualize and evaluate the flight etc.
performance of algorithms. Figure 11 shows a display
interface from the simulator. Ultimately, we intend to Traditional methods of testing are usually invasive to
implement the robotic ecology and emotional holon software and can increase code size and complexity,
concepts noted earlier using this architecture. which are two factors that always degrade software
quality. With the Reflection architecture, hundreds of
testing conf;,gur.tions, scripts, scenari~, etc., c;?~?he
created without a single modification to source code,
reducing the amount of resources required to
thoroughly test components. Further, components can
be swapped into and out of the architecture as needed,
allowing for the system to support simulation testing of
varying configurations at various levels of fidelity. For
instance, control modules can be tested with simulated
vehicle information in a UAV application by replacing
sensor interface modules with other reusable modules
implementing a simulated vehicle by changing a few
lines in a script file.

The system also facilitates component reuse, as

developers can selectively credte configurations to test
selected modules or the entire system by utilizing a
Fig 11. - Reflection Simulator Display Interface. growing repository of preexisting nodules that include
generic data visualization modules, vehicle simulations
The Reflection Architecture is a ccmponent-based at various levels of fidelity: data signal generators, and
cross-platform object-oriented application framework interface modules to allow users to manipulate data
for rapid development of high reliability simulations with joysticks or other hardware. For simulation use,
and real-time embedded systems. It operates by the complete large-scale simulations for research or
principle of data reflection, where modular components prototyping can be rapidly assembled by selecting data
operate in a larger system configuration on information nodu!es eon? this repository without writing a single
that is gathered, filtered, and deiivered to the line of code and with functionality customized to the
component without ever interacting directly wirh orher project's requirements.
modules. Data reflection provides a layer of isolation
that surrounds each component, allowing for a reusable The Reflection simulator is composed of several flight
and highly configurable framework. Modules operate simulation components that were assembled and
on data completely within their own private context rewritten to use the architecture. Visible in Figure 1 I
without having to consider where the data comes from are moving map, scene visualization, Mode Control
or where it goes. This greatly reduces the code size and Panel, and Primary Flight Display components. Not
complexity for each module while freeing developers to shown, but critical to the simulation, are aircraft model,
focus entirely on implerr,enting functionality rather that? autopilot, and sensor components.
on larger architectural issues. The principle of data
reflection also great!y simplifies the implementation of
11
American Institute of Xeronautics and Astronautics
 We are also enhancing this software to support real The following sections compare simulation results of
time simulation of sensor data to the aircraft while in these behaviors to actual flight path data. Because we
flight. This capability will allow us to simulate do not currently have access to a complete set of flight
environments and sensor inputs that may be very and autopilot design information on the iMLB UAV,
different than the launch point or location in which the these plots contain several assumptions/approximations.
aircraft is being flown. The first is that the simulations were performed with
using a “scaled” Cessna 172 aircraft simulation model
Work towards an emotional holon svstem The flown at an altitude of 1000 feet and airspeed of 100
purpose of the emotional holon concept is to formulate miles per hour. These simulation results were
and implement a system for controlling the behavior of qualitatively scaled to match the UAV experimental
a UAV that provides adaptive rational decision-making, results. The second is that results show that while the
learning, action selection, and reflexive behavioral autopilot used in the simulation transitions to its next
control in achieving goals*. waypoint prior to reaching it, the autopilot on the UAV
requires it to pass through a waypoint before
The emotional holon theoretical model is largely built transitioning. Finally, the simulated versus actual flight
on the Pandemonium and Global Workspace theories. pathshehaviors compared were chosen to be
Emotional primitives, codelets, are represented as representative of the data collected and are not
control theoretic transformations, with initial correlated in time sequence.
implemeztation c o ~ s i s t i ~ gef rn--c-+nnQk!erivate
Lvr ----I--&

controllers. The activated codelets that are represented Flight Behavior: Hawk Flieht Our first
by the arena analogy in Pandemonium presenr the experiment demonstrated the PC/104 commanding a
system with the active control blocks for the system. circling pattern that drifted randomly over time to
Through combination of emotional modeling theories search an area. As the aircraft drifted the flight system
and control theoretic formulation, we hope to show that computers processed the video input looking for a red
emotional modeling can be an effective tool for target tarp on the ground. Figure 12 plots the ground
designing and implementing adaptive control systems path of the aircraft against the ground plot of a
for aerial explorers. simulated aircraft with the same random algorithm.
The UAV was commanded to fly at an altitude of 400 ft
Terrestrial flight demonstration of behavioral flight AGL with an airspeed of 30 miles per hour.
concepts.
Ciicle Expenrneni RigM Test Veers5 S i r n l l i t i o n
We are in the process of completing a series of UAV Gmund Posnion Cornpanson O w nrne
imn ,
flights at MofferLFieid, with rhe goal of deillonsiidting
and evaluating the utility of several biologically-based
flight behaviors that could be used by an aeria:
explorer. The mission modes and biologically-based
behaviors chosen for these flights include:

I) High altitude preliminary search returning a

panoramic view of the area. For this mode we
have chosen to use a circling, overlapping
flight path ifs exhibited by hawks or eagles
when surveying an area.
2) Mid-level long-range search for an item over
an area with litt!e ~ , T ~ T J ~ Odc?a.
US P. randomly
generated flight path, as used by bacteria,
insects, or other animals searching a new area
was selected for implementation.
3 ) Low-level search of an area where items have
been previously identifed. For this mode, a
“Fox vs. Mouse” terrain terrain-following
behavior was chosen.
Fig. 12. Simulation vs. flight ground path of circling
algorithm.
12.
American Institute of Aeronautics and Astronautics
 Despite the simulation approximations noted, the an under damped heading hold PID autopilot. The
agreement between the simulator and flight test results experiment was initially tested using circular and
are quite good for the "Hawk" circling flight behavior. rectangular ground geometries. Figure 14 shows the
The random permutations to the circling behavior are results of a simulation of the aircraft finding and
clearly captured in Fig. 12, as well. following a ground terrain geometry similar to a
horseshoe gorge. The ground track of the aircraft is
Flicht Behavior: Random-Walk. Our second plotted against a gray scale terrain elevation map,
experiment implemented an ad hoc "intelligent" where dark areas represent low-lying regions. Similar
random-walk algorithm that plotted a random course experiments have been run with ridgelines and crater
over the search area starting from one boundary and rim terrains.
continuing to the other boundary by computing random
length legs and directions while attempting not to return
to points previously searched. When a boundary was
crossed, the algorithm would mirror the search direction
about the boundary to keep the aircraft within the
search area. The result of the algorithm is shown in
Fig. 13, plotted against the actual UAV position and the
simulation results when the aircraft was fed the same
random course. Good agreement was €ound in this
intelligent random-walk flight behavior.

Farda Search Expermen4 Fligtt Ted Verslr SimJatim

Fmnl hrdon tompanson 0 a - T ~

Fig. 14. Simulation ground path of a terrsii;-following . .

flight algorithm.

FUTUREWORK

This past year we completed significant ground work

towards an aerial explorer: a tested flight platform with
on board and data linked processing capabilities, an
enhanced software module development and hardware
Fig. 13. Simu!ation vs. flight ground pazh ofan in the loop test capability, flight test results to include
intelligent random-walk flight algorithm. biologically-based behaviors, and field work in
scientifically interesting areas that has included post
Flight Behavior: Terrain Following. Our third imagery analysis.
experiment implemented a behavior capable of
recognizing and following a low-lying region such as a ExDanding Svstem CaDabilities
ravine or gorge. A terrain following autopilot was
created that accepts input fi-om a ground elevation There are several directions in which we would like to
sensor, computes an estimated ground slope based on expand this aerial exploration work. A primary
the sensor data history, and then computes a desired objective is to expand the sensor and aiiionomy
heading from this information using ad hoc estimation capabilities our system to be more capable of science
algorithms. The heading commands are directed into work.
13
American Institute of Aeronautics and Astronautics
 Inteerate additional low cost vision be necessary. Some near term objectives would
capabilities and modes: Once we have completed our include:
flight testing of the UAV with the secondary processor,
we plan to add a gimbaled camera to the aircraft. This Perform a quantitative assessment of various
will allow random as well as direct observation of search and find strateeies. Evaluate the utility of
ground objects in flight. We would also like to add various algorithms by seeding multiple targets over an
additional vision processing modes, to include image- area or choosing different areas. This should lead to the
processing algorithms that are of the most interest to ability to make quantifiable statements such as 3 of 4
science. targets identified within a rough terrain search area, in a
fifteen minute search, with an Independent Random
Provide additional autonomy software Walk strategy.)
components. The Unix based PC 104 system was Coordinate with a field astrobiologist to
chosen in part because of the breadth of autonomy augment his or her field work. For example, acquire
algorithms (planners, executives, system modules, data detailed mosaic mapping of rock formation faces for
processing) that are available within the NASA and context, based on prior selection of a site.
university robotic community. We would like to
examine how these modules could be integrated and Applv the UAV as an augmentation to a
what affect they might have on our architecture. ground system. For example, accompany a mobile
research vehicie on a mu!tip!e day trek, performing
Explore low cost coordinated image fusion. aerial surveys at each key stopping point. The resulting
Based on our preliminary experience with aerial aerial image mosaics would be deiiverabie to trek
images, we would like to explore the coordination of organizers.
imagery information through a secondary entity (for
example, the location of an area of interest identified by
a radio controlled aircraft or ATC rover surrogate COMCLLBING REMARKS
relayed through a wireless link to the primary aircraft).
Autonomous uninhabited aerial vehicles, acting as
Expmd the utilitv of deploved sensor pods: To “aerial explorers.” can potentially have a significant
include distributed sensor networks (multiple drop pods positive impact on NASA science missions. Whether
and locations); tailored emplacement of sensors for the purpose of acting as robotic field assistants for
(hillsides, canyodvalley walls). and different pod types: scientists conducting research in remote terrestrial
(ground penetrators, gliders, micro-rovers and extreme environments, or in support of planetary
“tetheibots”). science missions, the development and use of aeriai
explorers present a valuable opportunity to maximize
Field Science and Field Robotic Assistants. scientific return on investment.

Equally important is to continue to expand our work Key to the development of such aerial explorers are
with the science community in order to identify new simple robust architectures for vehicle autonomy.
ways in which aerial explorers can be used to enhance Small autonomous aerial vehicles, particularly in the
their work. There is promising follow-on research in context of search and find missions, present unique
that these bio-inspired mission scenario and autonomy challenges as compared to other robotic
coccepts could be applied to rabctic fie!d assistants for platfo~siapplications. The authors presented .m
geologists and biologists performifig terrestrial field outline of a “bio-inspired” architecture for alltonornous
science in extreme environments”. aerial explorers, as well as summarized the
development stztus of their effart. This architecture is
Imaging survey work is only one potential capability heavily dependent upon defining flight “behaviors” -
these aerial robotic field assistants could provide to the actions and observations -- and effecting decision-
field researcher. h4ore than one type of aerial vehicle making via concepts regarding robotic “ecology” (to
configuration might be appropriate for conducting a model information flowiusage through discrete
field campaign. A whole new aerial vehicle design producer, consumer, and decomposer agentsiprocesses)
space might well open up for such robotic field and “emotional holons” (to drive these processes)
assistants. Further, a “system of systems” of robotic
devices (ground and air) and automated tools will likely Finally, representative simulation and flight
demonstration results were presented. Flight
14
American Institute of Aeronautics and Astronautics
 demonstrations entailing three behaviors were Planning and Execution of Missions Using General
conducted at Moffett Field airstrip at Ames Research Classes of Aerial Explorers”, 2003 IEEE
Center. Additionally, an exploratory field Aerospace Conference, Big Sky, MT, March 8-15,
demonstration was conducted at a remote Mars-analog 2003.
site: Haughton Crater, Devon Island, Canada.
6. Plice, L., “Robot Economy,” Robosphere 2002:
Workshop on Self-sustaining Robotic Ecologies,
ACKYOWLEDGMENTS NASA Ames Research Center, Moffett Field, CA,
November 14-15,2002.
The authors gratefully acknowledge the programmatic
support and overall contributions of Dr. Steven 7. Thakoor, S., “Bio-Inspired Engineering of
Zornetzer, Acting Deputy Director for at NASA Ames, Exploration Systems”, Journal of Space Mission
Dr. Butler Hine, Manager of the NASA Intelligent Architecture, Issue 2, Fall 2000, p 49-79.
Systems Program, and Mr. Edwin Aiken, Chief of the
ArmyiP*iASA Rotorcraft Division. The authors also 8. Ippolito, C., Plice, L., and Pisanich, G.,
acknowledge the contributions of Dr. Pascal Lee of the “Holarchical Systems and Emotional Holons:
SETI Institute for making possible the success of the Biologically-Inspired System Designs for Control
autonomous aerial vehicle field demonstrations at of Autonomous Aerial Vehicles,” 7‘h World
Haughtoii Crater, Devoi; Islaid, NT, Canada. Fina!ly, P,.~ca,,..,.p
b”lllrlb.IJbC Systematics an:! Cybernetics,
n..
vI1

the authors also acknowledge the technical Orlando, FL, July 2003.
contributions of Dr. Steve Morris and other members of
the staff of MLB, Inc., for the system integration and 9. Pisanich, G., Young, L.A., Ippolito, C., Lau, B.,
operation of the autonomous aerial vehicle. Plice, L., Sargent, R., and Lee, P., “Initial Efforts
towards Mission-Representative Imaging Surveys
from Aerial Explorers,” International Society of
REFERENCES Optical Engineering (SPIE) Electronic Imaging
Conference, San Jose, CA, January 2004.
1. Aiken, E.W., Ormiston, R.A., and Young, L.A.,
“Future Directions in Rotorcraft Technology at 10. Young, L.A., Aiken, E.W., and Briggs, G.A.,
Ames Research Center,” 56” Annual Forum of the “Smart Rotorcraft Field Assistants for Terrestrial
American Helicopter Society, International, and Planetary Science,” 2004 IEEE Aerospace
Virginia Beach, VA, May 2000. Conference, Big Sky, MT, March 2004.

2. Young, L.A., Chen, R.T.N., Aiken, E.W., and 11. Lee, P., ‘-Mars on Earth: The NASA Haughton-
Briggs. G.A., “Design Opportunities and Mars Project,” Ad Astra: The ilfagazine of the
Challenges in the Development of Vertical Lift NutionaZ Space Society, Volume 14, Number 3,
Planetary Aerial Vehicles,” American Helicopter MayIJune 2002.
Society ( A H S ) Vertical Lift Aircraft Design
Conference, San Francisco, CA, January 2000.

3. Pisanich, G., and Young, L.A., “An Aerobot

Ecology,” Rcbosphere 2002: Workshop on Se!f
Sustaining Robotic Ecologies, KASA Axes
Research Center, Moffett Field, CA. November 14-
15, 2002.

4. Thakoor, S., Chahl, J., Srinivasan, M.V., Werblin,

F., Young, L., Hine, B., Zometzer, S., “Bioinspired
Engineering of Exploration Systems for NASA and
DoD’, Artificial Life Journal, 2002, Vol. 8, Issue
4, pages 357-369.

5. Plice, L., Pisanich, G., Lau, B., and Young, L.,

”Biologically Inspired ’Behavioral’ Strategies for
15
American Institute of Aeronarutics and Astronautics