ASTROBIOLOGY

Volume 20, Number 12, 2020

ª Mary Ann Liebert, Inc.
DOI: 10.1089/ast.2019.2177

Investigating Habitability with an Integrated Rock-Climbing

Robot and Astrobiology Instrument Suite

Kyle Uckert,1 Aaron Parness,1 Nancy Chanover,2 Evan J. Eshelman,1 Neil Abcouwer,1
Jeremy Nash,1 Renaud Detry,1 Christine Fuller,1 David Voelz,2 Robert Hull,2 David Flannery,3
Rohit Bhartia,4 Kenneth S. Manatt,1 William J. Abbey,1 and Penelope Boston5

Abstract
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

A prototype rover carrying an astrobiology payload was developed and deployed at analog field sites to mature
generalized system architectures capable of searching for biosignatures in extreme terrain across the Solar
System. Specifically, the four-legged Limbed Excursion Mechanical Utility Robot (LEMUR) 3 climbing robot
with microspine grippers carried three instruments: a micro-X-ray fluorescence instrument based on the Mars
2020 mission’s Planetary Instrument for X-ray Lithochemistry provided elemental chemistry; a deep-ultraviolet
fluorescence instrument based on Mars 2020’s Scanning Habitable Environments with Raman and Lumines-
cence for Organics and Chemicals mapped organics in bacterial communities on opaque substrates; and a near-
infrared acousto-optic tunable filter-based point spectrometer identified minerals and organics in the 1.6–3.6 mm
range. The rover also carried a light detection and ranging and a color camera for both science and navigation.
Combined, this payload detects astrobiologically important classes of rock components (elements, minerals, and
organics) in extreme terrain, which, as demonstrated in this work, can reveal a correlation between textural
biosignatures and the organics or elements expected to preserve them in a habitable environment. Across >10
field tests, milestones were achieved in instrument operations, autonomous mobility in extreme terrain, and
system integration that can inform future planetary science mission architectures. Contributions include (1)
system-level demonstration of mock missions to the vertical exposures of Mars lava tube caves and Mars
canyon walls, (2) demonstration of multi-instrument integration into a confocal arrangement with surface
scanning capabilities, and (3) demonstration of automated focus stacking algorithms for improved signal-to-
noise ratios and reduced operation time. Key Words: Instrument suite—Robotic platform—Autonomous
navigation—Subsurface—Extreme terrain. Astrobiology 20, 1427–1449.

1. Introduction multiple, independent lines of evidence to support a detec-

tion of astrobiological significance (Shapiro and Schulze-

D etection of extant or extinct life on another planet

is one of the overarching goals of planetary science
exploration and is fundamental to the Mars Exploration
Makuch, 2009).
As we have refined the search for life in our own Solar
System, it has become apparent that many of the best targets
program (MEPAG, 2015), Ocean Worlds exploration for investigation lie in complex terrain that is inaccessible to
(Sherwood, 2016; Lunine, 2017; Hendrix et al., 2019), and current wheeled rovers. For example, the Opportunity rover
even exoplanet research (NRC, 2010). The burden of proof imaged cross-bedding in exposed rock outcrops but could
to confirm detection correlates with the magnitude of the not access the site despite several months of effort (Squyres
claim. Relying on a single piece of evidence or controversial et al., 2009). This has motivated the development of more
interpretations can lead to widespread disinformation that advanced mobility systems such as the Limbed Excursion
can negatively impact the credibility of the mission and the Mechanical Utility Robot (LEMUR) 3 robot used in this
field. This motivates a mission design strategy that uses work (Parness et al., 2017a) (Fig. 1).

1
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California, USA.
2
New Mexico State University, Las Cruces, New Mexico, USA.
3
Queensland University of Technology, Brisbane, Australia.
4
Photon Systems, Inc., Covina, California, USA.
5
NASA Astrobiology Institute, Ames Research Center, Mountain View, California, USA.

1427
 1428 UCKERT ET AL.

FIG. 1. The Limbed Excursion Mechan-

ical Utility Robot (LEMUR) 3 climbing ro-
bot ascends a rocky wall in the Titus Canyon
region of Death Valley National Park, Ca-
lifornia during a field test. A safety line is
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

tethered to the top of LEMUR’s chassis to

prevent damage from a potential fall. Power
and communications are transmitted through
the lower cable. LEMUR, Limbed Excur-
sion Mechanical Utility Robot.

Remote observations of Mars have identified numerous bedded in matrix materials that provide important contextual
geological regions of interest in the search for life, including information. This motivates a science implementation strategy
many that are difficult to access; for example, large canyons that can make multiple measurements of precisely the same site
such as Valles Marineris (Nedell et al., 1987), vast deltas by using complementary analytical techniques and measure the
and floodplains (Moore et al., 2003), skylight entrances to surrounding matrix materials for comparison. Spatial registra-
putative lava tubes in the Tharsis region (Cushing and Titus, tion of datasets is the current state of the art (Paar et al., 2016),
2010; Léveillé and Datta, 2010), seasonal recurring slope but the ability to align multiple instruments to interrogate a
linae (McEwen et al., 2011), and polar ice caps of water and single target without rotating a turret (Doran et al., 2016) or
carbon dioxide ( Jakosky et al., 2003). On Ocean Worlds, moving a deployment mechanism could reduce spatial errors in
the chaos terrain of Europa (Sotin et al., 2002; Kereszturi these data. Further, this approach can increase the operational
and Keszthelyi, 2013) and the active vents on Enceladus tempo of an investigation and allow more science return within
(McKay et al., 2008) are particularly attractive astrobiology a fixed-life mission. Given travel times and bandwidth restric-
targets, both lying in extreme terrain. Accessing targets on tions that may limit human-in-the-loop tactical decisions, the
small bodies, such as volatiles escaping the actively venting automation of these data acquisition activities can dramatically
scarps on Comet 67P-CG (Thomas et al., 2015) or the white improve science return.
faculae within Occator crater on Ceres (Vu et al., 2017), Our prototype system was designed with these factors in
necessitates extreme terrain robotic systems that can ma- mind. It uses a suite of five instruments (three scientific and
neuver despite the low gravity. Figure 2 shows the diversity two engineering/science dual purpose) to generate inde-
of these extreme terrains that future rovers must be able to pendent and complementary lines of evidence for habitable
access and deploy instruments on for in situ investigations. environments and life. The climbing mobility platform
Microbial life and its biosignatures are the highest priority (Parness et al., 2013), which comprises hardware and soft-
targets within these extreme terrains (NRC, 2011). Investiga- ware, can traverse rocky environments independent of
tions may require evaluation of small regions of interest em- gravity and far beyond the reach of wheeled rovers. This
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1429
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 2. Several examples of astrobiologically interesting sites on other planetary surfaces not easily accessible by conventional
wheeled rovers: (a) layered stratigraphy, Victoria Crater, Mars; (b) pit crater, Pavonis Mons volcano, Mars; (c) chaos terrain on
Europa; (d) cometary regolith, Comet 67P/Churyumov–Gerasimenko; (e) recurring slope linea, Newton Crater, Mars; (f) polar ice
terrain (vertical exaggeration), Planum Boreum, Mars; (g) south polar fissures, Enceladus; (h) Occator Crater, Ceres. (Image credit:
a–c, e–h National Aeronautics and Space Administration, d European Space Agency (ESA), public domain).

enables access to many of the high-priority science targets 2011), man-made surfaces such as the exteriors of satellites
of interest. By using alternative grippers (Curtis et al., 2018; ( Jiang et al., 2017), and icy environments (Curtis et al.,
Nash et al., 2020), similar terrains made of ice are also 2018; Nash et al., 2020). The robot uses a PC/104 avionics
accessible with the same robot and autonomous soft- stack to control low-level motor loops and carries an Intel
ware. The instrument suite was implemented in a confocal NUC for high-level autonomy and vision processing. It has
package for two of the three science instruments [deep- been used as a representative robot for extreme terrain ac-
ultraviolet (DUV) fluorescence (Geobiology with ultraviolet cess and as a platform to mature autonomy software and
(UV) Raman Imaging and Laser-induced Autofluorescence develop scientific investigation concepts for habitability and
or GURILA) and X-ray fluorescence (XRF; Advanced PIXL life detection.
{Planetary Instrument for X-ray Lithochemistry} Experi- In this work, LEMUR 3 uses microspine grippers (Parness
ment or APE)] and demonstrated computer vision-based et al., 2013) (Fig. 3) as the end effectors on each limb. These
registration techniques to align all datasets. grippers array hundreds of sharp hooks on compliant mech-
Automated focal stacking algorithms improved both the anisms that allow the gripper to conform to the shape and
quality of data signal to noise ratio (SNR) and the simplicity roughness of the rock’s surface at the centimeter and milli-
of operations by not requiring surface preparation or spectral meter scales. The hierarchical compliance also provides a
re-focus at each point in a hyperspectral map. load-sharing mechanism so that many individually low-force
To mature and evaluate this system, laboratory tests of the engagements of a single hook engaged with a crack or pit or
robot and instrument suite were performed, along with vesicle on the rock can be summed to support large overall
multiple field tests to Mars analog environments in New loads. Demonstrations of rotary percussive drilling in mi-
Mexico and California, as described in Section 2 of this crogravity (Parness et al., 2017b) reacted by microspines as
study. Section 3 presents the Results of these tests, followed well as large-scale gripping with flight-like designs and ma-
by a Discussion of the major lessons learned and their im- terials (Parness et al., 2017c) were demonstrated during de-
plications for future robotic planetary science missions to velopment of the now-cancelled Asteroid Redirect Mission.
extreme terrain. The work is summarized in Conclusions. The platform has been demonstrated in climbing basalt,
granite, and carbonate rocks. Figure 4 shows an autonomous
4.2 m climb on vertical granite performed with the help of a
2. Methods
safety cable and gravity offload system.
2.1. Limbed Excursion Mechanical Utility Robot
2.2. Light detection and ranging/context camera
The LEMUR 3 limbed prototype robot is a quadruped
with seven joints (degrees of freedom) per limb (Parness A Hokuyo UTM-30LX-EW light detection and ranging
et al., 2017a). Swappable grippers allow the robot to climb a (LIDAR) with a 270 field of view is mounted on a per-
variety of terrains, including rocky cliff faces (Parness, ception mast that sweeps the unit 360 around the robot
 1430 UCKERT ET AL.
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 3. (a) A photograph of LEMUR 3 climbing on a rough-textured basalt wall in Four Windows Cave. Two of four
limbs are shown gripping the surface. Each segmented limb provides seven degrees of freedom. (b) Each gripper contains
16 cassettes of microspines. (c) Each microspine cassette contains about a dozen frames with embedded steel hooks to grip
the surface.

to create a three-dimensional (3D) point cloud that rep- by the Portable AOTF (Acousto-Optic Tunable Filter)
resents the rock face (Fig. 5). The sensor is invariant to Spectrometer for Astrobiology (PASA)-Lite instrument,
ambient lighting conditions, functioning well in total providing a scientific context and focusing guidance. An
darkness (inside a cave) and in direct sunlight. Data from additional context camera mounted to the 3-axis instrument
multiple scans are registered with one another and filtered stage images science targets and provides input to the fo-
to continually improve the quality and range of the 3D cusing algorithm used to position the GURILA and APE
map and to better localize the rover within the map. Point instruments, described further in Section 2.6.
cloud data are converted into a voxel map for use by the
robot planning software. More than a byproduct, this 3D
2.3. PASA-lite
surface can also be passed to the science team (with op-
tional color and texture layers) to enhance the science Infrared spectroscopy probes the vibrational transitions of
investigations. a molecule, and the wavelength-dependent reflectivity of an
Four cameras provide additional navigation guidance and irradiated sample can be used to infer its molecular com-
context imagery for science interpretation. A camera on the position and hydration state. The presence of organic com-
perception mast opposite the LIDAR provides contextual pounds and the characterization of minerals indicative of
information on rock color, texture, and fractures. An engi- habitability (including carbonates) can be used to vet sam-
neering camera on the opposite side of the chassis enables ples within a region for their highest astrobiological po-
an operator to monitor instrument proximity to the wall for tential (Korablev et al., 2017; Uckert et al., 2017).
redundant collision avoidance. A camera mounted on the We measured samples in situ as well as those returned
underside of LEMUR 3 images the target surface measured to the laboratory with PASA-Lite (Chanover et al., 2012;
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1431
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 4. A time-lapse photo series of LEMUR 3 climbing a granite outcrop at the Granite Mountain Desert Research
Center. During this test, LEMUR 3 climbed *4.2 m vertically in 7 h.

Uckert et al., 2015)—an active illumination infrared (IR) of PASA-Lite, labeled with major components. The light path
point spectrometer that operates in the 1.6–3.6 mm range through the optical elements is presented in Fig. 6b.
(with a spectral resolving power of l/Dl > 250), over-
lapping with the fundamental vibrational modes of hydro-
2.4. Advanced PIXL Experiment
carbons (C–H), hydroxyls (–OH), and other hydrogen
bonded molecules. Weaker vibrational transitions associated The elemental composition of a sample (including trace
with overtones and combinations of fundamental vibrations elements) is revealed by XRF spectroscopy. Hyperspectral
of other molecules are also present (Pieters and Englert, elemental composition maps reveal subtle changes in chem-
1993; Bokobza, 2002; Uckert et al., 2017). istry across a sampling area, which may indicate the presence
To acquire a near-infrared (NIR) spectrum of a target, of grains, mineral veins, or concretions. These geological
LEMUR 3 positions a sample within the working distance of the boundaries can correspond with elevated organic content or
PASA-Lite auto-focuser, 0–37 mm. A broadband IR light biological activity (identified with Raman or IR spectroscopy)
source is then directed through the AOTF wavelength selection and provide a geological context for the deposition of these
element, resulting in a narrow-band beam of photons that is formations. In XRF, target atoms are bombarded with high-
focused onto a *2 mm spot on a sample and reflected into the energy X-rays, ionizing an atom by ejecting inner-orbital
HgCdTe single-element detector. Figure 6a shows a photograph electrons. Secondary X-rays are emitted as outer-orbital

FIG. 5. (a) A photograph of LEMUR 3 climbing at site 2 in Four Windows Cave (Fig. 9) and (b) a corresponding point
cloud representation of the area generated by the light detection and ranging perception system. The points have a resolution
of 30 mm in the climbing workspace, and the range of the perception system extends from 0.1 to 30 m.
 1432 UCKERT ET AL.
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 6. (a) A photograph of PASA-Lite mounted to LEMUR 3 during a field expedition to Four Windows Cave, showing
the major components of the instrument. (b) An optical path diagram tracing the broadband IR light from the source to the
sample, with reflected light directed back through the off-axis parabolic mirrors to the detector. AOTF, Acousto-Optic
Tunable Filter; IR, infrared; MCT, HgCdTe (mercury cadmium telluride); PASA, Portable AOTF Spectrometer for As-
trobiology; RF, radio frequency.

electrons drop to lower energy states to fill the vacant inner- biosignatures in field and laboratory settings (Flannery
orbital shell. The energy of these characteristic secondary et al., 2016, 2018a; Allwood et al., 2018). The APE gen-
X-rays is measured by a silicon drift detector (SDD) to de- erates X-rays by using a Moxtek X-ray tube (operating at
termine the elemental composition of a sample. 28 kV and 40 mA) and focuses them to a 100 mm spot at a
We developed the APE XRF spectrometer and integrated 2 cm stand-off distance with a custom-built polycapillary
it with the LEMUR 3 rock-climbing robot to determine the optic. A Vitus H50 Ketek SDD detects secondary X-rays
chemical composition of a sample in situ, during climbing (along with scattered source X-rays) with a *140 eV energy
activities. The APE is based on the design of the PIXL resolution (full-width at half maximum [FWHM] at
instrument for Mars 2020 (Allwood et al., 2015). Mapping 5.9 keV). During in situ operations, the Ketek detector was
micro-XRF spectroscopy provides elemental abundance replaced with a more robust but less sensitive Amptek SR-
data that are comparable to the element maps generated by 100SDD, which has a similar energy resolution but requires
energy-dispersive spectroscopy in scanning electron mi- longer integration times due to a smaller collection area.
croscopy, but it is a technique that can now be deployed in We also measured samples returned to the laboratory by
field settings (including martian surface environments) and using the PIXL breadboard, which produces higher resolu-
be used to analyze unprepared rock and soil surfaces (All- tion elemental composition maps with a more precise mo-
wood et al., 2015). torized stage. The PIXL breadboard uses a two-channel
The APE is of a similar design to other PIXL prototype detector system (two Ketex Analytical X-Ray Acquisition
instruments that have been built to effectively map textural System [AXAS] SDDs, energy range: *0.7 to 28 keV,
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1433

energy resolution: *130 eV at FWHM 5.9 keV) and a cooled detector operates from 572 to 4252 cm-1 with a reso-
Moxtek X-ray tube, focusing photons to a < 100 mm spot lution of Dl *4 cm-1. The penetration depth of the 248.58 nm
(under standard operating conditions at 28 kV and 230 mA). photons is dependent on the substrate, but it is expected to be
The X-ray flux attenuation is dependent on the sample >500 mm for most materials (Carrier et al., 2019).
composition and density, primary X-ray energy, and sec-
ondary X-ray energy (Hubbell and Seltzer, 1995). For the 2.6. Confocal integration and scanning
samples measured in this study, the derived energy spectrum stage and operations
represents the composition of the uppermost 100 mm of the
GURILA and APE are a confocal system integrated on
target. All laboratory XRF measurements are conducted in
the same custom 3-axis stage, shown in Fig. 7. These two
an He-rich environment to improve X-ray transmission and
instruments interrogate the same *100 mm spot on a tar-
limit the contribution of atmospheric fluorescence peaks
get, producing a hyperspectral map that combines Raman/
(primarily from Ar) to the energy spectrum, more closely
fluorescence, XRF, and context imagery. A confocal instru-
resembling measurements in a martian atmosphere.
ment suite eliminates uncertainty in instrument targeting and
reduces operation time by allowing for simultaneous mea-
2.5. Geobiology with UV Raman Imaging surements with both instruments. In addition, coincident
and Laser-induced Autofluorescence measurements of a target allow for direct comparisons between
points without the need for interpolation within a hyperspectral
In DUV Raman and fluorescence spectroscopy, an inci-
map or requiring instrument-specific positioning systems.
dent laser pulse interacts with a target molecule in one of
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

The 3-axis stage comprises three linear stages that move

two ways: (1) Stokes Raman scattering: The photon pro-
independently. Each motorized stage has 5 cm of travel,
motes a molecule to an excited virtual state, which then
better than 1 mm precision, and employs custom software to
relaxes to an excited vibration state, transferring energy to
position the instruments in a predefined pattern over the
the lattice and shifting the photon’s wavelength (diagnostic
rock at the APE/GURILA focus point. A visible imaging
of the molecule), or (2) fluorescence: The incident photon is
system with an expanded field of view and a shallow depth
absorbed by the molecule, exciting it to a higher electronic
of field collects a series of images at various distances from
and vibrational state. The molecule returns to the vibrational
the target. A topographic map of the sample terrain is au-
ground state, and it releases a photon as it relaxes to the
tonomously reconstructed by using these images as fol-
electronic ground state. DUV Raman and fluorescence
lows: Image features are identified and images are aligned;
spectroscopy can assess a sample’s habitability potential, its
a sub-group of pixels (16 · 16) within each image is com-
aqueous history, and detect the presence of biosignatures
pressed along a z-stack, where the least compressible image
preserved in rocks by detecting and classifying organics and
represents the focus position within the scene; and a 3D
minerals on a surface (Beegle et al., 2015).
representation of the scene is programmed into the 3-axis
GURILA is a DUV Raman and fluorescence spectrometer
stage to drive the instrument system to the focus point at
and is a predecessor to the Mars 2020 Scanning Habitable
each (x, y) position based on user-defined targeting reso-
Environments with Raman and Luminescence for Organics
lution and area parameters. Each of these intermediate
and Chemicals instrument (Beegle et al., 2015). GURILA
processing steps within this algorithm is represented visu-
measures the native fluorescence and pre-resonance/
ally in Fig. 8. This technique alleviates the need for a rock
resonance DUV Raman scattering of a sample after irradi-
abrasion tool to flatten a surface, which removes potentially
ation by a narrow linewidth, high radiance 248.58 nm NeCu
scientifically interesting surface features or chemically
DUV laser (Photon Systems) focused to a *100 mm spot.
distinct weathering rinds to accommodate instrument po-
The GURILA sensors consist of 32-channel photomultiplier
sitioning requirements.
tube detectors (Photon Systems), which operate from *265
to 430 nm, with a resolution of Dl *5 nm. The optical
2.7. Field deployment and data collection
design is similar to the Wireline Analysis Tool for Sub-
surface Observation of Northern ice sheets DUV Raman Samples were measured in the field with instruments
spectrometer described in the works of Eshelman et al. mounted to LEMUR 3 and in a stand-alone in situ data col-
(2019) and Mellerowicz et al. (2018). lection mode. In some cases, samples were returned to the
GURILA is capable of detecting a range of organics, laboratory for longer-duration measurements with benchtop
microbes, and potential biosignatures, including the classi- instruments. Raw spectral data collected by each instrument
fication of bonds such as C–H, CN, C = O, C = C, NHx, NOx, in situ and by laboratory instruments on returned samples
SOx, POx, ClOx, and OH (Nealson et al., 2002; Bhartia were processed through a standard data reduction pipeline to
et al., 2008, 2012a, 2012b, 2015; Salas et al., 2015; Abbey calibrate the spectra and to normalize to source intensity.
et al., 2017; Sapers et al., 2019). The NIR spectra collected with PASA-Lite were calibrated
We also measured samples returned to the laboratory with against an Infragold reflectance target—a handheld Lamber-
the Jet Propulsion Laboratory ( JPL) Mineral/Organic/ tian reflector that is measured before each in situ measure-
Biological Investigations with UV Spectroscopy (MOBIUS) ment. Normalization to a reflectance target allows for spectral
high-resolution fluorescence/Raman spectrometer to char- comparison between datasets and geochemical reference
acterize the organic and major mineral components within standard libraries and corrects for instrument variability, in-
a sample (Abbey et al., 2017). MOBIUS uses the same cluding detector temperature (Tawalbeh et al., 2013).
248.58 nm laser excitation source as GURILA, which reso- Elemental composition information from XRF spectra
nantly excites nucleobases and amino acids. The MOBIUS collected with APE and the PIXL breadboard prototype in-
instrument uses a 250 mm slit width, and the liquid nitrogen- strument were quantified by using PIQUANT, a calibration
 1434 UCKERT ET AL.
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 7. (a) A CAD rendering of GURILA (blue) and APE (red) components mounted to the 3-axis instrument stage
(gray), showing the major components (labeled) and the co-aligned optical axis. The UV laser path (blue) and X-ray beam
path (red) are outlined. (b) A photograph of the instrument stage integrated with LEMUR 3 during a field test in the Titus
Canyon region of Death Valley. APE, Advanced PIXL Experiment; CAD, Computer Aided Design; GURILA, Geobiology
with UV Raman Imaging and Laser induced Autofluorescence; LPS, Laser Power Supply; PIXL, Planetary Instrument for
X-ray Lithochemistry; PMT, photomultiplier tube; UV, ultraviolet.

and data reduction tool (Elam et al., 2002) derived from the diameter of *10 m (Rogers, 1991; Marinakis, 1997). A map
fundamental parameters approach to peak identification and of Four Windows Cave is presented in Fig. 9 with sampling
fitting (Broll, 1986; Allwood et al., 2015). PIQUANT ac- locations annotated as red points throughout the cave.
counts for sample orientation, source and detector geometry, The El Malpais formation is an extensive series of Plio-
air mass, sample matrix effects, X-ray source spectrum, and cene to Holocene basalt flows that resulted from the eruption
other instrument- and sampling-specific differences to cal- of shield volcanoes, cinder cones, fissures, and spatter cones
culate the absolute intensity of the X-ray spectrum. over *100,000 years (Rogers, 1991; Marinakis, 1997). Four
The Raman and fluorescence spectral processing pipeline Windows Cave formed within the Bandera flow *10,000
includes standard procedures for baseline subtraction, laser years ago (Laughlin et al., 1994). The Bandera flow field
photodiode normalization to account for variable laser pulse contains the largest number of lava tube caves (*30 km of
energy, cosmic ray removal, and wavelength calibration passage) in the El Malpais region, with some of the largest
based on the Raman peak positions of known standards passages in the monument (Marinakis, 1997). The relatively
(Uckert et al., 2019). large tubes in the Bandera flow are partially due to their
recent formation; many large tubes in the area have not yet
2.7.1. Lava tube cave: El Malpais National Monument. undergone collapse due to roof failure after plastic defor-
Four Windows Cave is 1 of out more than 200 documented mation or weathering (Rogers, 1991). In addition, the
lava tube caves located in the El Malpais National Monu- composition of the flows sourced from Bandera crater have
ment, New Mexico. Four Windows Cave is notable for its a more mafic composition with relatively low SiO2, K2O,
relatively long length (1.2 km), having an entrance at each and Al2O3 concentrations and relatively higher Fe2O3, TiO2,
end of the tube, skylights, and a relatively large passage and MgO, leading to low viscosity, tube-fed, pahoehoe
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 8. A demonstration of the focus-merge algorithm applied to an unprepared sample. (a) The 3-axis stage is com-
manded to move perpendicular to the sampling plane, acquiring images every 1 mm. (b) A subset of these images are
shown. (c) A focus-merged composite image representing the best focus for each subpixel region. (d) A heat-map repre-
sentation of the best focus. (e) Mesh representation of the digital elevation model of the sample. The instrument stage linear
motors are commanded to position the instrument focus at these (x, y, z) site frame coordinates.

FIG. 9. A map of Four Windows Cave, annotated with sampling locations. This map has been adapted from the original
survey by B. Rogers (courtesy of El Malpais National Monument).

1435
 1436 UCKERT ET AL.

flows that are conducive for lava tube formation (Hatheway the carbonate rocks that host Girvanella (the Girvanella-
and Herring, 1970; Laughlin et al., 1972; Dunbar and rich stratigraphic layer is highlighted in Fig. 10a).
Phillips, 2004).
2.7.3. Deployment logistics. The robot and instruments
2.7.2. Stratigraphic section: Titus Canyon. Titus Canyon were deployed for multiday field tests in remote areas. To
cuts through the Grapevine Mountains, exposing uplifted, facilitate such tests, the robot was broken down into pieces
Cambrian-aged limestone, dolostone, quartzite, and siltstone that could be transported across rough terrain in hiking
(Niemi, 2002, 2012). The outcrop associated with the field site backpacks: individual limbs (with grippers attached) and the
described in Section 3 is part of an upper Proterozoic/Middle main chassis. Power was provided by portable lithium ion
Cambrian unit (*550 million years old), and the cross- battery packs. Instruments were packed within ruggedized
sections interrogated by the instrument suite are primarily equipment cases for transportation. At the field site, the robot
carbonate rock. These Cambrian carbonates were originally was reassembled, and then instruments were calibrated and
deposited in a warm, shallow sea, and have since been over- assembled onto the robot.
turned and folded, as shown in Fig. 10a (Niemi, 2002). Although prior versions of the LEMUR robot used onboard
Our field site preserves relatively high concentrations of batteries, these required frequent change-out due to their limited
laminated, spheroidal, textural microbial biosignatures, capacity. LEMUR 3 uses off-board power that allows for un-
known as Girvanella (Nicholson and Etheridge, 1878; limited continuous operation with the use of an uninterruptible
Stewart, 1970; Niemi, 2012; NPS, 2015), that were the power supply. Operators controlled the robot and instruments
focus of this field test. Girvanella is believed to reflect the from a work station no >15 m from the robot, determined by
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

metabolic activities of filamentous Cyanobacteria in electrical cable length. Setup times from arrival at field site to
shallow-water environments that were saturated in calcium meaningful field test were reduced to *2.5 h over the course of
carbonate (e.g., Pratt, 2001). It is a form of oncolite that multiple excursions. Repacking time averaged 1.5 h.
may form in environments that are dominated by pro-
karyotic organisms in waters that are regularly agitated. 3. Results
Oncolites have been observed forming in modern settings
3.1. Lava tube cave: El Malpais National Monument
(e.g., Mawson, 1929) and are known from throughout the
early geological record in rocks deposited as early as the We measured basaltic surfaces covered by apparent mi-
Archean (Flannery et al., 2018b). At this field site, Gir- crobial mats, microbial precipitates, and carbonate and
vanella ranges in size from 1 to 20 mm. Figure 10b shows sulfate speleothems throughout Four Windows Cave (see

FIG. 10. (a) A photograph of the testing area in Titus Canyon. The blue highlighted outcrop represents a stratigraphic
layer containing oncolite candidates. The highest concentration of these features is located in the yellow-highlighted band.
(b) An image of the oncolite measured with the instrument suite while mounted to LEMUR 3.
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1437

Fig. 9 for sampling locations) with a suite of in situ and –OH bending modes at *2.4 mm, and broad H2O and HO-
laboratory instruments. During a 5-day expedition to Four stretching vibrations from *2.6 to 3.0 mm (Pieters and
Windows Cave, we conducted 3 days of testing, which in- Englert, 1993; Bishop et al., 2014).
cluded LEMUR 3 robot climbing tests, measurements of Figure 11d shows an annotated XRF spectrum collected
Cyanobacteria-rich basalt walls with APE and PASA-Lite from the same locations while the instrument was mounted
integrated with LEMUR 3, stand-alone in situ measurements to LEMUR 3. Throughout this study, the Ka and Kb lines of
of samples throughout Four Windows Cave, and a collection major and minor elements detected by APE are labeled in
of samples from inconspicuous locations in breakdown talus fluorescence spectra; other lines with lower transition
piles to minimize impact to the cave. probabilities, and peaks related to trace elements are not
We collected samples from regions within the cave near labeled (Deslattes et al., 2005). The APE collected a line
in situ measurement locations, when applicable, for follow- scan measurement across a Cyanobacteria microbial mat/
up measurements with laboratory instruments. Samples basalt boundary; the spectra presented in Fig. 11d represent
were wrapped in aluminum foil (sterilized through a dry the extrema of this scan.
heating technique to 450C for 10 h), and they were sealed Figure 11e shows two spectra collected with the
in plastic containers to limit contamination. MOBIUS laboratory spectrometer of a returned sample
(shown in Fig. 11f); the black trace represents an average
3.1.1. Climbing results. The LEMUR 3 robot success- spectrum of the basalt region, and the gray trace represents
fully climbed on three wall surfaces within Four Windows an average spectrum of the green-colored microbial mat.
Cave, including surfaces coated with Cyanobacteria as well Annotations indicating the broad classification of Raman
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

as basalts without obvious microbial mat coatings. The ro- scattering peaks associated with characteristic frequencies of
bot was operated with a full-gravity offload system, also molecular functional groups are noted, including carbonate
serving as a safety line, which provided an effective weight (*1100 cm-1), organic (*1500–1600 cm-1), and water
of approximately –3 kg. The surfaces were well suited to (*3000–3400 cm-1) spectral features.
microspine grippers with many asperities at the 1–10 mm The returned sample shown in Fig. 11f was measured in the
scale. The robot autonomy software performed well, faulting laboratory with the MOBIUS Raman spectrometer and the
and requiring manual intervention only at large dis- PIXL breadboard spectrometer, with mapping areas outlined
continuities (ledges and corners), cases for which function- in cyan and red, respectively. An intensity distribution map of
ality had not been built into the code operating at the time. the 1630 cm-1 scattering peak is shown in Fig. 11g, alongside
The longest climb measured 2.5 m and occurred on a vertical a context image, showing an association between the green-
face within the cave (climbing site 2 in Fig. 9). Gross posi- colored microbial mat and the strength of this spectral feature.
tioning of the instruments was performed through teleopera- Laser spots were *100 mm in diameter, with 200 mm spacing
tion of the robot by using the context cameras. between points. The XRF false-color maps shown in Fig. 11h
show high concentrations of Fe associated with basalt, as well
3.1.2. Cyanobacteria microbial mat. Cyanobacteria- as higher concentrations of Ca and Si that are associated with
coated basaltic rocks are located in the Twilight Zone of the microbial mat and calcium carbonate crust. The X-ray spot
the cave, where they are exposed to sunlight and greater size is *100 mm at the target, with points spaced every
temperature variability than the microbial communities in *100 mm. In both these hyperspectral maps, the center of
the Dark Zone (Northup, 2004; Northup et al., 2011; each pixel represents a single measurement, with a linear in-
Boston and Kelly, unpublished data). Thin (<1 mm) green- terpolation between individual spectra.
colored coatings are deposited on top of the basalt walls in
this region, forming a nearly homogenous mat. In many 3.1.3. Gold-colored microbial mat: Twilight Zone versus
areas, immediately underlying these green-colored coat- Dark Zone. Actinobacteria-rich microbial mats and crust
ings is a thin (*1 mm) white-colored crust between the are present on the basalt walls near the entrance of the cave
basalt wall and the microbial mat. These coatings are more in the Dark and Twilight Zones, as shown in Fig. 9
sparsely observed within vesicular regions of the basalt and (Northup et al., 2011). A photograph showing the brightly
under overhangs that may block sunlight. At a larger scale, colored biological deposit and the dark basalt host rock
these microbial communities exist in multiple regions acquired by the scientific context camera are presented in
within the Twilight Zone, occupying several square meters Fig. 12a. Figure 12b and c show spectra of basalt walls
of basalt wall. containing these microbial mats measured at each of the
Figure 11a shows LEMUR 3 performing a vertical traverse Actinobacteria-rich measuring sites noted in Fig. 9 by
over the Cyanobacteria-rich microbial mat. Figure 11b shows PASA-Lite (Fig. 12b) and APE (Fig. 12c), where the black
a photograph acquired by the scientific context imager on trace represents a spectrum of the microbial community in
the GURILA/APE instrument stage, which reveals a green- the Dark Zone, and the gray spectrum represents a mea-
colored microbial mat, a white-colored calcified deposit, and surement acquired in the Twilight Zone. Point spectra were
a dark-colored basalt rock. The APE detector and X-ray acquired in both locations of microbial mat-rich regions of
source are visible in this image. the sample to maximize signal. These spectra were ac-
Figure 11c shows PASA-Lite IR spectra of the underlying quired in situ in a stand-alone mode – not while integrated
basalt (black trace) and the calcified deposit and microbial with LEMUR 3.
mat superimposed on the basalt (gray trace) collected
while mounted to LEMUR 3. Major absorption features are 3.1.4. Coralloid deposits. Coralloid deposits, colloqui-
labeled: H2O symmetric bend and asymmetric stretch ally known as ‘‘cave popcorn,’’ are present throughout the
combination band at *1.95 mm, the hydroxyl stretching and main passage of Four Windows. Cave popcorn typically
 1438 UCKERT ET AL.
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 11. (a) A photograph of LEMUR 3 climbing over a Cyanobacteria microbial mat in Four Windows Cave. (b) A
close-up view of the measurement area taken by the instrument context camera, with the APE detector (top) and source
(bottom) within the frame of the image. (c) IR spectra produced by PASA-Lite while mounted to LEMUR 3. (d) XRF
spectra produced by APE while mounted to LEMUR 3. (e) Raman spectra of a returned sample produced by the MOBIUS
laboratory spectrometer. (f) A photograph of a sample returned to the lab with red and blue highlighted areas outlining
mapping locations for the PIXL breadboard spectrometer and the MOBIUS Raman spectrometer, respectively. (g) A context
image (top) and 1650 cm-1 intensity distribution map (bottom) produced by the MOBIUS spectrometer (200 mm/pixel).
(h) A context image (left) and false-color maps representing major elemental composition distribution of the returned sample
(100 mm/pixel). MOBIUS, Mineral/Organic/Biological Investigations with UV Spectroscopy; XRF, X-ray fluorescence.

precipitates from calcium-rich waters that have seeped into APE with elemental peak assignments indicated. Both PASA-
the cave (Hill et al., 1997). A photograph of a coralloid Lite and APE collected measurements of the sample noted in
measured in situ is shown in Fig. 13a. Fig. 13a in situ, unmounted from LEMUR 3.
Figure 13b shows an IR spectrum of the coralloid sample Raman spectra of a coralloid-bearing basalt sample
measured with PASA-Lite, with the gray trace representing the (shown in Fig. 13e) are presented in Fig. 13d. Point spectra
white precipitated structure and the black trace representing at several locations on the target were acquired, denoted by
the underlying basalt. Annotations indicate the most significant the colored circles in Fig. 13e, including white-colored
vibration modes in the reflectance spectrum. Figure 13c shows coralloids, bare basalt, and thin white-colored evaporite
an XRF spectrum of the basalt and coralloid measured with deposits. Two distinct spectra are presented in Fig. 13d: a
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 12. (a) A photograph of the gold-colored microbial mat acquired by the instrument context camera, showing the
APE detector (left) and APE source (right). (b) IR spectra produced by PASA-Lite measured in situ. (c) XRF spectra
produced by APE measured in situ. Measurements of the microbial mat were conducted on similar walls in the Dark and
Twilight Zones of Four Windows Cave (see Fig. 9 for sampling locations).

FIG. 13. (a) A photograph of a coralloid measured in the One-Foot-in-the-Gutters Gallery in Four Windows Cave. (b) IR
spectra produced by PASA-Lite measured in situ. (c) XRF spectra produced by APE measured in situ. (d) Raman spectra of
two distinct regions of a returned sample produced by the MOBIUS laboratory spectrometer. (e) a photograph of the
returned sample measured by MOBIUS highlighting the distinct compositions measured.

1439
 1440 UCKERT ET AL.

carbonate-rich region (black trace, blue-colored spots in salt. Figure 14d shows an XRF spectrum of the basalt with
Fig. 13e), containing strong carbonate spectral features sulfate deposit measured by APE.
(*1100 cm-1) and a sulfate-rich region (gray trace, red-
colored region in Fig. 13e), containing strong sulfate 3.2. Stratigraphic outcrops: Titus Canyon
spectral features (*1000 cm-1). We analyzed a single Girvanella-rich unit within a larger
3.1.5. Sulfate deposits. Several accumulations of a outcrop of carbonate in Titus Canyon. In situ measurements
white, powdery sulfate deposit are located in the upper room acquired by the integrated instrument suite were constrained
at the end of the main passage (labeled as Oe Puna Beach in by the schedule of the 3-day field expedition to Titus Can-
Fig. 9), near the second entrance (Marinakis, 1997). yon, which also included LEMUR 3 climbing tests, stand-
Figure 14a shows a photograph of these deposits, which coat alone measurements, and sample collection.
the floor in the Oe Puna Beach Room (Fig. 9) and appear to We collected samples from a nearby talus pile associated
be emerging from veins in nearby walls, which are en- with the same stratigraphic layer for follow-up measurements
crusted with the same white-colored deposit. We measured with laboratory instruments. Samples were removed from
these deposits and nearby veins with PASA-Lite and APE in inconspicuous locations, packaged in sterile aluminum foil
a stand-alone configuration (Fig. 14b). An IR spectrum of (sterilized through a dry heating technique to 450C for 10 h)
the sulfate deposit measured with PASA-Lite is presented in and sealed in plastic containers to limit contamination.
Fig. 14c, with the gray trace representing the white pre-
3.2.1. Climbing results. The LEMUR 3 robot success-
cipitate and the black trace representing the underlying ba-
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

fully climbed on two stratigraphic surfaces at roughly 70-

and 90-degree (vertical) slopes at the Titus Canyon site,
including surfaces where oncoids were present. The robot
was operated with a full-gravity offload system, which also
served as a safety line that provided an effective weight of
approximately –3 kg. The surfaces were moderately suited
to microspine grippers with a few asperities at the 1–10 mm
scale, but many asperities at the sub-millimeter scale. Long
climbs were not attempted; the limited fieldwork was fo-
cused on scientific integration and data acquisition.

3.2.2. Textural biosignatures. The oncolite shown in

Fig. 15a and b was measured with all three instruments while
integrated aboard the LEMUR 3 rock-climbing robot. The
sample is located within a carbonate matrix and has similar
features found throughout this unit (see Fig. 10a for context).
An IR spectrum of the carbonate matrix (black trace) and the
oncolite (gray trace) is shown in Fig. 15c with prominent
absorption features annotated. A GURILA fluorescence
spectrum is presented in Fig. 15d, showing increased fluo-
rescence signal on the oncolite (gray trace) compared with the
carbonate matrix (black trace). A Raman spectrum of a re-
turned sample is presented in Fig. 15e with an inset showing
the carbonate (*1100 cm-1) and organics (*1500–
1600 cm-1) that are described in greater detail in Section
4.2.1. Each spectrum represents an average of several spectra
acquired from the dark- and light-colored regions associated
with the context image shown in Fig. 15h. The XRF spectra
of the carbonate matrix (black trace) and the oncolite (gray
trace) measured with APE are presented in Fig. 15f. We
present APE and GURILA data that are representative of
spectra acquired from the carbonate matrix and the oncolite
from a line scan that bisected the feature.
Longer integration hyperspectral measurements acquired
with laboratory instruments focused on the returned sample
shown in Fig. 15g were also conducted. Figure 15h shows the
context image (left: blue points indicating laser shot posi-
tion) and intensity distribution map for two prominent
spectral features identified in Fig. 15e. Laser spots were
FIG. 14. (a) A photograph of a powdered sulfate deposit *100 mm in diameter, with 100 mm spacing between points.
measured at Oe Puna Beach in Four Windows Cave. (b) A False-color maps representing the elemental distribution
photograph of APE measuring the sulfate deposit. (c) IR inferred from energy spectra generated by the PIXL bread-
spectra produced by PASA-Lite measured in situ. (d) XRF board instrument are shown in Fig. 15i. The X-ray spot size is
spectra produced by APE measured in situ. *100 mm at the target, with points spaced every *100 mm.
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1441
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

FIG. 15. (a) A photograph of the measurement site at Titus Canyon, Death Valley, with major instrument components
labeled. (b) An expanded view of the measurement area acquired by the instrument context camera, with the APE detector
(top) and source (bottom) within the frame of the image. (c) IR spectra produced by PASA-Lite while mounted to LEMUR
3. (d) Fluorescence spectra produced by GURILA while mounted to LEMUR 3. (e) Raman spectra of a returned sample
produced by the MOBIUS laboratory spectrometer. (f) XRF spectra produced by APE while mounted to LEMUR 3. (g) A
photograph of a sample returned to the lab with red and blue highlighted areas outlining mapping locations for the PIXL
breadboard spectrometer and the MOBIUS Raman spectrometer, respectively. (h) A context image (left) and an intensity
distribution map (right) produced by the MOBIUS spectrometer showing carbonate (*1100 cm-1, blue) and organic
(*1550 cm-1, green) distribution (100 mm/pixel). (i) A context image (left) and false-color maps representing the major
elemental distributions measured in the returned sample (100 mm/pixel).

3.3. Granite: Sweeney Granite Mountains Desert Desert Research Center within the Mojave Desert. The robot
Research Center was operated with a full-gravity offload system, also serving
as a safety line, that provided an effective weight of ap-
The LEMUR 3 robot successfully climbed two vertical proximately –3 kg. The surfaces were well suited to mi-
granite faces at a field site at Sweeney Granite Mountains crospine grippers with many asperities at the 1–10 mm scale.
 1442 UCKERT ET AL.

The robot autonomy software performed well, faulting and 4.1. Lava tube cave: El Malpais National Monument
requiring manual intervention only at large discontinuities
(ledges and corners), cases for which functionality had not 4.1.1. Cyanobacteria microbial mat. Skylights near the
been built into the code operating at the time. The longest cave entrance extend the cave’s twilight zone, allowing for
climb measured 4.2 m on a vertical surface. the growth of an extensive moss garden atop breakdown (as
shown in Fig. 9) and for the support of a Cyanobacteria
3.4. Statistical techniques microbial community inhabiting damp basalt walls.
Comparison of NIR spectra with geologic reference stan-
We applied a principal component analysis (PCA) to each
dards provided by the United States Geological Survey
hyperspectral map to reduce the dimensionality of the hy-
(USGS) (Clark et al., 2007) indicates calcite deposition
perspectral dataset and to determine the dominant organic
overlying basalt host rocks. A definitive classification of the
and mineral vibration modes associated with the microbial
precipitate from the IR spectrum is not possible due to a rel-
community and the elemental composition variability. Be-
atively weak IR reflectance signal, potentially caused by the
fore applying PCA, Raman and fluorescence data were ca-
complex sample surface and difficulty focusing during the test.
librated by using the standard procedures described in
Organic content associated with the Cyanobacteria coating,
Section 2.7. The XRF energy calibration was performed by
evidenced by the increased concentration of C–H stretch
fitting a polynomial to the major, known peak locations of a
modes in the NIR spectra, is not obvious in the IR spectra. The
basalt standard measurement.
low atomic weight hydrocarbon-rich cyanobacterial mat is not
Custom software was written in Python to apply PCA to
detectable by the XRF instrument, which instead reveals the
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

these preprocessed hyperspectral datasets, based on the

calcium associated with the biological carbonate precipitate
implementation described in the work of Wold et al. (1987).
that underlies the microbial mat. XRF measurements of the
Principal components are calculated for all spectra by cal-
basalt wall show strong Ti, Mn, Fe, and Ni peaks relative to
culating the eigenvectors and eigenvalues of the covariance
the calcified deposit, which contains a higher percentage of S
matrix of the spectra intensity values, with the wavelength
and Ca (Fig. 11d). The detection of S in the precipitate could
(Raman) and energy (XRF) dimensions representing the
indicate the presence of calcium sulfates. Strong organic
independent variables. The data are not standardized or
(*1650 cm-1) and bound water Raman signatures associated
normalized. The resulting principal components represent
with the microbial mat texture are observed in Fig. 11e,
the variance in the hyperspectral maps.
demonstrating the ability of this technique to detect organic
Cluster analysis of the resulting PCA score point cloud was
concentration variability that correlates with morphological
achieved by using a simple k-means clustering machine-
features, potentially indicative of microbial mats.
learning algorithm to group data points based on their variance
We applied a PCA to the Raman and fluorescence labo-
relative to principal component loading spectra (Arthur and
ratory hyperspectral data set to determine the dominant
Vassilvitskii, 2007). False-color maps of the target associated
organic vibration modes associated with the microbial
with these cluster analysis results demonstrate spectral vari-
community. An autonomous k-means clustering algorithm
ability correlation with millimeter-scale morphologic features
applied to the PCA score point cloud reveals the presence of
in the context images. This can reveal the boundaries or spatial
three major spectral groups in the dataset associated with the
extent of regions of interest associated with trace elements or the
carbonate deposit (blue), Cyanobacteria microbial mat (red),
relative distribution of organic material throughout a sample
and basalt (green) shown in Fig. 16a. The magenta false-
without a priori knowledge of the spectral features associated
color map in Fig. 16a represents the variance of the first
with this variability. Regions of interest are defined by an opti-
principal component, which accounts for 91% of all vari-
mization of the variability of the region (maximizing the dis-
ance within the dataset; high correlation with the spectral
tinct number of spectral classes, as defined by the k-means
variance loading plot indicates high spectral correlation (i.e.,
cluster results), while minimizing the area of the region (while
high fluorescence and strong organic Raman scattering
still containing some minimum number of points).
peaks coincide with the green microbial mat observed in the
context image). Dark regions in this false-color map indicate
4. Discussion
dissimilar spectra. The depth of focus of the MOBIUS in-
Multiple instruments probing the same target reveal un- strument is large, relative to the magnitude of the surface
ique characteristics of a sample, leading to a more robust roughness for this sample; fluorescence intensity variability
interpretation of its composition or origin. In our case, an IR is likely not caused by the instrument focus at each point in
reflectance spectrometer identifies the bulk mineralogy of a the hyperspectral map.
sample by probing the molecular vibrational modes. Si- The results of PCA applied to the XRF hyperspectral map
milarly, Raman and fluorescence spectroscopy reveal the measured with the PIXL breadboard are presented in
mineralogical and organic composition of a sample. XRF Fig. 16b. In this case, the k-means clustering algorithm
spectroscopy is used to infer the elemental composition. identifies two main groups (basalt—colored in green, and
The interpretation of measurements from a suite of in- the red and blue overlying deposit). The majority of the
struments is enhanced by multivariate statistical techniques variance within the dataset is associated with the difference
along with machine-learning cluster analyses to classify between basalt-rich spectral signals and all other spectra. An
the most significant sources of variance within a hyper- additional 15% of variance is determined to exist within the
spectral dataset. Spectral maps, such as those produced in calcified deposit (blue and red coloring), where the red re-
the lab or by APE and GURILA when not under strict time gions poorly match the placement of the Cyanobacteria
constraints, allow for better correlation between visible microbial mat shown in the context image. The red-colored
context images and spectral features. region is defined by higher metal content than the blue-
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1443

basalt, compared with approximately <1% for the calcified

region), Ca (*17% for basalt, compared with *40% for
the calcified region), and Ti (*1% for basalt, compared
with *0.1% for the calcified region).
The application of multivariate statistics and unsuper-
vised cluster analyses demonstrate an autonomous method
to classifying regions of interest. Regions are ranked based
on their variability, distribution of distinct principal com-
ponents, and presence of boundaries associated with spectral
variability, which in this case may provide ideal measure-
ment locations due to a potential overlap between organic
material and underlying host rock.

4.1.2. Gold-colored microbial mat: Twilight Zone versus

Dark Zone. Unlike the Cyanobacteria microbial mat dis-
cussed in Section 4.1.1, these gold-colored microbial mats
are not photosynthetic and extract energy instead from or-
ganic detritus. Differences in the microbial communities
inhabiting the twilight and dark zones of the cave may be
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

revealed through distinct spectral signatures. The IR PASA-

Lite spectrum of the sample located in the Twilight Zone
shows enhanced absorption of the H2O stretch at *1.9 mm
and –OH stretch at *2.4 mm. These features may indicate
increased adsorbed water content on the target surface,
likely caused by the higher relative humidity at this site due
to its proximity to the entrance of the cave and the skylights
(shown in Fig. 9). The XRF spectrum in Fig. 12c appears to
show an enhanced spectral signal for all peaks in the Twi-
light Zone, likely caused by improved focusing at this lo-
cation. Normalizing these spectra to the count rate reveals
similar peak areas, indicating a nearly identical composition
at each of these sites.
The relatively small size of the microbial mat patches and
the tendency for these biomats to conform to concave features
within the pitted basalt wall contribute to the difficulty in
focusing on these targets. Spectra presented here are, there-
fore, more likely to represent a mixture of the microbial mat
and basalt wall, which fluoresces efficiently in the X-ray re-
gion, potentially overwhelming any signal contribution from
the microbial mat. A high-density hyperspectral map could
reveal spectral differences associated with the microbial mat
by applying a spatial de-convolution to the dataset or by using
the multivariate technique described in Section 3.4. We were
FIG. 16. (a) Analysis of the MOBIUS Raman and fluo- unable to locate a sample from this region to return to the lab
rescence laboratory measurements of the Cyanobacteria mi- for longer integration measurements, and generating high-
crobial mat returned sample. The results of the cluster resolution maps with APE in the field was not possible due to
analysis applied to the PCA of the fluorescence spectra are the prohibitively long time required for acquisition.
shown in the RGB map, with average fluorescence spectra
associated with each of these regions presented below. The 4.1.3. Coralloid deposits. Coralloid features are typically
magenta intensity distribution map represents the variance in precipitated out of solution in caves from water enriched in
the first principal component of the PCA applied to the Ra-
man spectra, with the corresponding loading plot presented. In Ca ions. Higher overall reflectivity of the coralloid is noted in
both false-color maps, the pixel scale is 200 mm/pixel. (b) the IR spectrum in Fig. 13b as well as the presence of car-
Analysis of the PIXL breadboard XRF measurements show- bonate absorption features associated with calcite (Gaffey,
ing the cluster analysis of the PCA applied to the XRF spectra 1987; Clark et al., 2007), which are absent in the basalt host
(RGB map, 100 mm/pixel), with corresponding average spec- rock. The XRF spectra produced by APE are similar in both
tra. RGB, red/green/blue; PCA, principal component analysis. sample regions, with differences in the overall signal strength
associated with poorer focusing of the X-ray beam on the
colored region, which may indicate an association with a small coralloids. Similar to the instrument focusing difficul-
thinner calcified coating, allowing the XRF to probe the ties described in Section 4.1.2, the small size of the coralloid
underlying basalt more easily. An application of PIQUANT and high fluorescence yield of the basalt results in the spectral
to quantify these spectra reveals the most significant com- signature of the basalt overwhelming the signal associated
positional differences to be associated with Fe (*28% for with the coralloid. The low IR reflectivity of basalt limits the
 1444 UCKERT ET AL.

potential signal degradation in the IR spectra presented in modern and ancient ooids (e.g., Davies et al., 1978; Summons
Fig. 13b, revealing the spectral differences associated with the et al., 2013; Flannery et al., 2019). The spatially resolved
carbonate precipitate more easily. association of carbonate minerals, elemental enrichments, and
Raman spectroscopy measurements of a basalt sample re- organics, and the putative textural biosignatures visible in
turned to the laboratory containing coralloids also show context images provide several lines of circumstantial evi-
carbonate-rich features with some sulfate-rich deposits in an dence that help to inform a biological/abiological interpreta-
isolated region on the left side of the sample. Our laboratory tion of the features. In this case, the textural biosignatures are
instrument is unable to create a high-resolution hyperspectral closely associated with organic molecules, as well as the ele-
map of this sample due to its complex topographic profile, but ments and minerals that would be expected to preserve these
several point scans of coralloids, bare basalt, and evaporite molecules had they formed via the metabolic activities of
deposits reveal a distribution of mineral precipitates. Direct Cyanobacteria living in a habitable environment.
measurements of coralloid (white circular protrusions) are Results of the in situ measurements presented in
carbonate rich, consistent with results from PASA-Lite. The Fig. 15c, d, and f reveal the presence of carbonate-rich
overall Raman scattering signal is stronger on more highly minerals, as well as increased fluorescence associated with
concentrated precipitate deposits (inferred by a whiter color), the oncolite, compared with the carbonate matrix, possibly
which appear thicker with distance from the bare basalt vis- due to the increased concentration of organics, described in
ible in the lower portion of the image. Higher concentrations more detail later. No significant spectral differences are
of carbonates on the right side of Fig. 13e transition to sulfate- observed in the NIR spectra associated with the concretion
rich deposits on the left side of the sample. In addition, higher and the mineral matrix; higher reflectivity in the oncolite
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

concentrations of carbonates are observed with closer prox- may be attributed to better instrument focusing at this lo-
imity to the bare basalt with increasing sulfate composition cation. The presence of organic Raman scattering features
with distance from this region. This reveals the aqueous al- associated with both regions measured suggests the pres-
teration of this precipitate layer—carbonates likely precipi- ence of organic matter, possibly with an increased contri-
tate out of solution first, followed by sulfates, possibly as bution coincident with the oncolite, as shown in Fig. 15h.
liquid flows down gradients toward the bare basalt. This The carbonate-rich band cutting through the right side of
sample was collected from a breakdown pile in the main the oncolite in Fig. 15h does not appear in the context
passage of the cave, and therefore its original orientation image and is hypothesized to be a small surficial carbonate
during the formation of these features is not known. scratch, possibly caused during deposition in the talus pile
from which the sample was retrieved.
4.1.4. Sulfate deposits. Sulfate deposits lie within veins Results from a PCA applied to the fluorescence and Ra-
along the walls, and large deposits lie on the ground in heaps man MOBIUS hyperspectral maps are shown in Figs. 17a
near Oe Puna Beach shown in Fig. 9. The largest deposit has and b, respectively. The k-means clustering algorithm ap-
been steadily shrinking since it was first identified in the early plied to the principal component score point cloud identified
1990s. Both wall and floor deposits were measured with PASA- two primary clusters (cyan and magenta in Fig. 17a) cor-
Lite and APE. The IR spectra reveal similar spectral features in responding to a distinct double-fluorescence feature at 300
the wall and floor deposits, possibly indicating that the white and 340 nm, which appears to be absent in spectra more
sulfates, which have a powdery consistency, are precipitated out closely correlated with the carbonate matrix. The fluores-
of solution in large veins along the walls of the cave and de- cence peaks in the oncolite region of the hyperspectral map
posited onto the nearby floor by air flow from the nearby en- could be attributed to large-structure organic compounds.
trance. The reduction in size of the large sulfate deposit may Cluster analysis results of the PCA of the Raman spectra
indicate that the precipitation rate has been variable over the show a more intense Raman organic feature associated with
decades-long timescale since they were documented. Alter- this same region in the sample, indicating that these fluo-
natively, their reduction may simply be caused by increased rescence features are correlated with organics. The PCA was
human impact as access to the cave has improved over this time. also applied to the Raman MOBIUS spectrum, revealing
The XRF spectrum of the sodium sulfate powder shows an variance associated with the carbonate-organic Raman
enhanced concentration of sulfur (by *40%) and a diminished scatter feature peak ratios. Spectra associated with the on-
concentration of Ca (by *10%), Fe (by *35%), and Ti colite (red-colored points) are enhanced in organics relative
(by *3%) relative to the basalt background. to the matrix spectra (green-colored points). The blue-
colored points are associated with a high carbonate signal.
The XRF spectrum of the carbonate matrix appears similar
4.2. Stratigraphic section: Titus Canyon to the oncolite target (shown in Fig. 15f), however a PCA
4.2.1. Oncolites. The concentric features presented in applied to the dataset presented in Fig. 15i, shown in Fig. 17c,
Fig. 10b likely formed via incremental microbially induced reveals subtle variability highly correlated with the oncolite
precipitation of carbonate around a small grain that served as a feature visible in the context image. The principal component
nucleation site. Biological mediation of these features is pos- variability plot shows positive and negative correlation be-
sible, especially given the setting—a warm, shallow sea tween spectral features associated with each element. Spectra
(Niemi, 2012)—and evidence for Cyanobacteria reported from more closely associated with the carbonate rock (green
similar features in this and other units [e.g., Nicholson and points) have relatively higher peak areas associated with Ca
Etheridge (1878) and microfossils reported in the work of Liu and lower concentrations of Fe, Si, Ti. Oncolite spectra are
et al. (2016)]. However, concentrically laminated features elevated in Fe, Si, Ti, and depleted in Ca, possibly due to a
accreted around a nucleus are known to form abiotically, and biofilm adhering to detrital grains or selective diagenetic re-
there is an extensive literature debating the biogenicity of placement of carbonate due to the presence of organics. A
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1445

quantification of these spectra is consistent with this result, In addition, because measurements are collected near-
with the carbonate containing *40% Ca and *19% Si, simultaneously, environmental degradation of the
compared with *38% Ca and 24% Si in the oncolite. sample between measurements is unlikely.
 Engineering instruments (LIDAR and navigational
cameras) were implemented to identify science targets
5. Conclusions
in the field, and additional work in autonomously se-
Field tests conducted in a basaltic lava tube and along lecting these targets by using pattern recognition soft-
stratigraphic sections exposed in cliffs simulated two types ware is in progress. Autonomous vetting of scientific
of semi-autonomous planetary science missions investigat- targets based on these surveys could enable a fast op-
ing astrobiology targets. Autonomous climbing with the
LEMUR 3 robotic platform provided access to science targets
in extreme terrain. A confocal arrangement of nondestructive
instruments allowed interrogation of a specific target simulta-
neously and without the need for complex image co-registration
and precision pointing. An autofocusing algorithm that con-
trolled the 3-axis instrument stage maximized the signal from
the contact science instruments.
The instrument suite used in this study (Raman spec-
trometer, XRF spectrometer, NIR spectrometer) was se-
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

lected to identify distinct astrobiologically relevant classes

of rock components. The combined measurements of the
elemental, mineralogic, and organic rock components
provided greater confidence for a biogenic interpretation
of textural biosignatures encountered in the field as the
results showed organic and elemental compositional var-
iability correlated with textural features, demonstrating
the importance of interrogating samples with multiple
instruments to evaluate the biogenicity of a sample.
Although the exact instrument suite and the design of
the robotic platform are likely to be different for future in
situ astrobiology missions to extreme terrains, and chosen
specifically for the science goals and terrain of the given
target, the lessons learned by the work presented here can
be applied to the architecture of future missions. Key ta-
keaways included:
 Physical co-registration of the instruments provided the
ability to examine more deeply a target with simpler
post-processing of the data. Both instruments interro-
gated the same point on a target; therefore, interpola-
tion of spectra in a hyperspectral map is not required
for direct comparison of points at the *100 mm scale.

‰
FIG. 17. (a) Analysis of the MOBIUS fluorescence lab-
oratory measurements of the carbonate and oncolite returned
sample. The results of the cluster analysis applied to the
PCA of the fluorescence spectra are shown the false-color
map (100 mm/pixels), with average fluorescence spectra as-
sociated with each of these regions presented. (b) Results of
the cluster analysis applied to the PCA of the Raman spectra
are shown in the false-color map (100 mm/pixel), with av-
erage Raman spectra associated with each of these regions
presented below. (c) Analysis of the PIXL breadboard XRF
measurements showing the cluster analysis of the PCA ap-
plied to the XRF spectra (RGB map, 100 mm/pixel). The
principal component variance spectrum shows anti-
correlation between red and green spectrum contributions
(i.e., increased Ca is generally correlated with decreased Fe,
associated with green-colored regions in the map). Negative
variance is represented by the red colored spectral segments;
the absolute value of the variance plot is presented here to
enable plotting on a logarithmic scale.
 1446 UCKERT ET AL.

erational tempo that is beneficial to cost-effective or cycles of development provide the foundation on which fu-
time-sensitive mission operations, and it would be es- ture extreme terrain rovers can be built.
sential on a future mission to a subsurface environment
where a bandwidth-restricted mission architecture Acknowledgments
could limit ground-in-the-loop operations. The authors thank National Park Service and the El Malpais
 Automated focal stacking techniques reliably demon-
National Monument for their support during their field expedi-
strated autonomous focusing of the GURILA and APE tions. They specifically thank Eric Weaver (ELMA) and Kevin
instruments in the field without requiring manual focus Wilson (DEVA) for logistical support and guidance at these
adjustments for each point in the hyperspectral map or field sites. They also thank Tasha La Doux and Jim Andre for
rock abrasion to generate a flat surface. their support of LEMUR 3 climbing operations at the Sweeney
 The ability of the robot to both access targets in ex-
Granite Mountains Desert Research Center. The authors thank
treme terrain and grossly position the instruments Aaron Curtis, Christopher Heirwegh, Joseph Razzell Hollis,
within *1 cm of a target proved essential at the Abigail Allwood, Lawrence Wade, and Sabrina Feldman for
various field sites. Investigations solely of the flat intellectual discussions that have improved this project. Thanks
terrains in these areas would not have uncovered the are due to Rina Onishi for assistance with the mechanical con-
biosignatures readily observable on the walls and figuration of GURILA. The authors thank William Timothy
ceilings and other extreme terrain. Elam for development of PIQUANT, used in this study to
 A macro-mini approach to pointing far outperformed an
quantify XRF measurements. They also thank Marco Tempest
all-in-one solution. The robot positioned the instrument
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

for providing video documentation of their Death Valley field

suite within *1 cm of a target while the instrument expedition. The authors are grateful to the reviewers of this
scan stages and focus mechanisms performed the fine article, including Chris McKay, for providing helpful feedback
micrometer-level pointing and focusing required by the on the presentation and discussion of these results.
instruments. Requiring a larger range of motion from
the instrument stages or a finer resolution of movement Author Disclosure Statement
from the 28 degree of freedom robot each would have
incurred large costs and mass increases to approach the No competing financial interests exist.
performance achieved by the macro-mini system.
Finding the correct balance between rover requirements Funding Information
and instrument stage requirements is a key trade de- This work was carried out at the JPL, California Institute
serving early evaluation in any new extreme terrain of Technology, under contract with the NASA. Government
mission architecture. sponsorship is acknowledged. The authors thank the NASA
Overall, the LEMUR 3 robot was successfully deployed Planetary Science and Technology through Analog Re-
in the field more than 10 times at four different geographic search for funding under NNH14ZDA001N.
sites with varied terrain. It climbed rough vertical rock fa-
cies largely autonomously and carried a suite of five in- References
struments, three scientific and two for engineering. The Abbey WJ, Bhartia R, Beegle LW, et al. (2017) Deep UV
system’s instrument suite successfully identified bio- Raman spectroscopy for planetary exploration: the search for
signatures, inferred from compositional variability, trace in situ organics. Icarus 290:201–214.
elements, and the presence of organics within a sample. Allwood A, Clark B, Flannery D, et al. (2015) Texture-specific
Comparative laboratory analysis showed similar results, but elemental analysis of rocks and soils with PIXL: the Planetary
these larger datasets and the multivariate statistics applied to Instrument for X-ray Lithochemistry on Mars 2020. In 2015
them allowed for improved correlation between spectral IEEE Aerospace Conference, IEEE, Big Sky, MT, pp 1–13.
signatures and morphology present in context images. Al- Allwood AC, Rosing MT, Flannery DT, et al. (2018) Re-
though many of the samples measured in this study have assessing evidence of life in 3,700-million-year-old rocks of
higher organic content than what is expected to be found on Greenland. Nature 563:241.
the surface of Mars, the instrument development and inte- Arthur D and Vassilvitskii S (2007) k-Means++: the advantages of
gration, operational techniques, and analytical methods careful seeding. In Proceedings of the eighteenth annual ACM-
SIAM symposium on Discrete algorithms, Society for Industrial
discussed by this study are applicable to astrobiology ex-
and Applied Mathematics, Philadelphia, PA, pp 1027–1035.
plorations of other planetary surfaces.
Beegle L, Bhartia R, White M, et al. (2015) SHERLOC:
Future work should include many more mock missions in a scanning habitable environments with Raman & lumines-
wide variety of extreme terrains of Earth, such as those cence for organics & chemicals. In 2015 IEEE Aerospace
supported by National Aeronautics and Space Administration Conference, Big Sky, Montana, pp 1–11.
(NASA)’s Planetary Science and Technology from Analog Bhartia R, Hug WF, Salas EC, et al. (2008) Classification of
Research program, to prepare for successful exploration of organic and biological materials with deep ultraviolet exci-
these terrains across our solar system. Some of these will take tation. Appl Spectrosc 62:1070–1077.
the form of climbing robots such as LEMUR 3 and its ice- Bhartia R, Hug WF, and Reid RD (2012a) Improved sensing
climbing sibling, IceWorm (Curtis et al., 2018; Nash et al., using simultaneous deep UV Raman and fluorescence de-
2020), whereas others will use alternative architectures such tection. In Chemical, Biological, Radiological, Nuclear,
as rappelling (Nesnas et al., 2019), hopping, perching, and and Explosives (CBRNE) Sensing XIII 8358:83581A,
more. These field campaigns and the iterative design/test/fix Baltimore, MD.
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1447

Bhartia R, Hug WF, Reid R, et al. (2012b) Noncontact, re- Flannery DT, Summons RE, and Walter MR (2018b) Archean
agentless, nondestructive, detection of organics, bio- Lakes as analogues for habitable Martian paleoenvironments.
signatures, and water. In Sensors and Systems for Space In From Habitability to Life on Mars, Elsevier, pp 127–152.
Applications V 8385:83850E, Baltimore, MD. Flannery DT, Allwood AC, Hodyss R, et al. (2019) Microbially
Bhartia R, Hug WF, Reid RD, et al. (2015) Explosives detection influenced formation of Neoarchean ooids. Geobiology 17:
and analysis by fusing deep ultraviolet native fluorescence 151–160.
and resonance Raman spectroscopy. Laser Based Opt Detect Gaffey SJ (1987) Spectral reflectance of carbonate minerals in the
Explos 4:67–98. visible and near infrared (0.35–2.55 um): anhydrous carbonate
Bishop JL, Lane MD, Byar MD, et al. (2014) Spectral prop- minerals. J Geophys Res Solid Earth 92(B2):1429–1440.
erties of Ca-sulfates; gypsum, basanite, and anhydrite. Am Hatheway AW and Herring AK (1970) Bandera lava tubes of
Mineralog 99:2105–2155. New Mexico, and lunar implications. Commun Lunar Planet
Bokobza L (2002) Origin of near-infrared absorption bands. In Lab 9:299–327.
Near-Infrared Spectroscopy, edited by Siesler HW, Ozaki Y, Hendrix AR, Hurford TA, Barge LM, et al. (2019) The NASA
Kawata S, Heise HM, Wiley-VCH, Weinheim, pp 12–16. roadmap to ocean worlds. Astrobiology 19:1–27.
Broll N (1986) Quantitative X-ray fluorescence analysis. The- Hill CA, Forti P, and Shaw TR (1997) Cave Minerals of the
ory and practice of the fundamental coefficient method. World (Vol. 238). National Speleological Society, Huntsville.
X-Ray Spectrom 15:271–285. Hubbell JH and Seltzer SM (1995) Tables of X-ray mass at-
Carrier BL, Abbey WJ, Beegle LW, et al. (2019) Attenuation of tenuation coefficients and mass energy-absorption coeffi-
ultraviolet radiation in rocks and minerals: implications for cients 1 keV to 20 MeV for elements Z = 1 to 92 and 48
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

Mars Science. J Geophys Res Planets 124:2599–2612. additional substances of dosimetric interest (No. PB-95-
Chanover N, Tawalbeh R, Glenar D, et al. (2012) Rapid as- 220539/XAB; NISTIR-5632). National Institute of Standards
sessment of high value samples: an AOTF-LDTOF spec- and Technology-PL, Gaithersburg, MD (United States). Io-
trometer suite for planetary surfaces. In 2012 IEEE nizing Radiation Division. Available online at [Link]
Aerospace Conference, Big Sky, Montana, pp 1–10. .[Link]/pml/x-ray-mass-attenuation-coefficients
Clark RN, Swayze GA, Wise RA, et al. (2007) USGS Digital Jakosky BM, Nealson KH, Bakermans C, et al. (2003) Sub-
Spectral Library splib06a (No. 231). US Geological Survey, freezing activity of microorganisms and the potential habit-
Reston, VA. ability of Mars’ polar regions. Astrobiology 3:343–350.
Curtis A, Martone M, and Parness A (2018) Roving on ice: field Jiang H, Hawkes EW, Fuller C, et al. (2017) A robotic device
testing an ice screw end effector and sample collection tool. In using gecko-inspired adhesives can grasp and manipulate
2018 IEEE Aerospace Conference, Big Sky, Montana, pp 1–17. large objects in microgravity. Sci Robot 2:eaan4545.
Cushing GE and Titus TN (2010) Caves on Mars: candidate Kereszturi A and Keszthelyi Z (2013) Astrobiological impli-
sites for astrobiological exploration. In Astrobiology Science cations of chaos terrains on Europa to help targeting future
Conference 2010: Evolution and Life: Surviving Cata- missions. Planet Space Sci 77:74–90.
strophes and Extremes on Earth and Beyond, League City, Korablev OI, Dobrolensky Y, Evdokimova N, et al. (2017)
TX, 1538, 5414. Infrared spectrometer for ExoMars: a mast-mounted instru-
Davies PJ, Bubela B, and Ferguson J (1978) The formation of ment for the rover. Astrobiology 17:542–564.
ooids. Sedimentology 25:703–730. Laughlin AW, Brookins DG, and Causey JD (1972) Late Ce-
Deslattes R, Kessler EG, Indelicato P, et al. (2005) X-ray nozoic basalts from the Bandera lava field, Valencia County,
Transition Energies. NIST Standard Reference Database. New Mexico. Geol Soc Am Bull 83:1543–1552.
Available online at [Link] Laughlin AW, Poths J, Healey HA, et al. (1994) Dating of
transition-energies-version-10 Quaternary basalts using the cosmogenic 3He and 14C
Doran G, Thompson DR, and Estlin T (2016) Precision in- methods with implications for excess 40Ar. Geology 22:
strument targeting via image registration for the Mars 2020 135–138.
Rover. In International Joint Conference on Artificial In- Léveillé RJ and Datta S (2010) Lava tubes and basaltic caves as
telligence, New York City, NY. astrobiological targets on Earth and Mars: a review. Planet
Dunbar NW and Phillips FM (2004) Cosmogenic 36Cl ages of Space Sci 58:592–598.
lava flows in the Zuni-Bandera volcanic field, north-central Liu L, Wu Y, Yang H, et al. (2016) Ordovician calcified cya-
New Mexico, USA. Tectonics, geochronology, and volca- nobacteria and associated microfossils from the Tarim Basin,
nism in the Southern Rocky Mountains and Rio Grande rift. Northwest China: systematics and significance. J Syst Pa-
New Mexico Bureau Mines Miner Resources Bull 160:51–59. laeontol 14:183–210.
Elam WT, Ravel BD, and Sieber JR (2002) A new atomic Lunine JI (2017) Ocean worlds exploration. Acta Astronautica
database for X-ray spectroscopic calculations. Radiat Phys 131:123–130.
Chem 63:121–128. Marinakis HA (1997) The lava tube cave systems of New
Eshelman EJ, Malaska MJ, Manatt KS, et al. (2019) WATSON: Mexico’s El Malpais. National Speleological Society News,
in situ organic detection in subsurface ice using deep-UV pp 161–167.
fluorescence spectroscopy. Astrobiology 19:771–784. Mawson D (1929) Some south Australian algal limestones in
Flannery DT, Allwood AC, and Van Kranendonk MJ (2016) process of formation. Quart J Geolog Soc 85:613–620.
Lacustrine facies dependence of highly 13C-depleted organic McEwen AS, Ojha L, Dundas CM, et al. (2011) Seasonal flows
matter during the global age of methanotrophy. Precambr Res on warm Martian slopes. Science 333:740–743.
285:216–241. McKay CP, Porco CC, Altheide T, et al. (2008) The possible
Flannery DT, Allwood AC, Summons RE, et al. (2018a) origin and persistence of life on Enceladus and detection of
Spatially-resolved isotopic study of carbon trapped in *3.43 biomarkers in the plume. Astrobiology 8:909–919.
Ga strelley pool formation stromatolites. Geochim Cosmoch Mellerowicz BL, Zacny K, Eshelman E, et al. (2018) Devel-
Acta 223:21–35. opment of a deep drill system with integrated deep UV/
 1448 UCKERT ET AL.

Raman spectrometer for Mars and Europa. In 2018 AIAA In Robotics and Automation (ICRA), 2017 IEEE Interna-
SPACE and Astronautics Forum and Exposition, p 5290. tional Conference, Marina Bay Sands, Singapore, pp 5467–
MEPAG (2015) Mars scientific goals, objectives, investiga- 5473.
tions, and priorities: 2015. Hamilton V, ed., 74. White paper Parness A, Willig A, Berg A, et al. (2017b) A microspine tool:
posted June 2015 by the Mars Exploration Program Analysis grabbing and anchoring to boulders on the asteroid redirect
Group (MEPAG). Available online at [Link] mission. In IEEE Aerospace Conference, Big Sky, Montana,
[Link]. pp 1–10.
Moore JM, Howard AD, Dietrich WE, et al. (2003) Martian Parness A, Evans T, Raff W, et al. (2017c) Maturing micro-
layered fluvial deposits: implications for Noachian climate spine grippers for space applications through test cam-
scenarios. Geophys Res Lett 30:1–4. paigns. In AIAA SPACE and Astronautics Forum and
National Park Service (NPS) (2015) Geologic Formations, Exposition, Orlando, FL, p 5311.
National Park Service. [Link] Pieters CM and Englert PAJ (1993) Remote Geochemical
nature/[Link]. Analysis: Elemental and Geological Composition. Cambridge
National Research Council (NRC) (2011) Visions and Voyages University Press, New York, pp 43–77.
for Planetary Science in Decade 2013–2022. National Aca- Pratt BR (2001) Calcification of cyanobacterial filaments: gir-
demies Press, Washington, DC. vanella and the origin of lower Paleozoic lime mud. Geology
National Research Council (NRC) and Space Studies Board 29:763–766.
(2010) Panel Reports—New Worlds, New Horizons in As- Rogers BW (1991) General geology and development of lava
tronomy and Astrophysics. National Academies Press, tubes in New Mexico’s El Malpais National Monument. In
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.

Washington, DC. Proceedings of 6th International Symposium on Vulcanos-

Nealson K, Tsapin A, and Storrie-Lombardi M (2002) Search- peleology, Hilo, Hawaii, pp 58–62.
ing for life in the universe: unconventional methods for an Salas EC, Bhartia R, Anderson L, et al. (2015) In situ detection
unconventional problem. Int Microbiol 5:223–230. of microbial life in the deep biosphere in igneous ocean crust.
Nedell SS, Squyres SW, and Andersen DW (1987) Origin and Front Microbiol 6:1260.
evolution of the layered deposits in the Valles Marineris, Sapers HM, Razzell Hollis J, Bhartia R, et al. (2019) The cell
Mars. Icarus 70:409–441. and the sum of its parts: patterns of complexity in bio-
Nesnas IA, Kerber L, Parness A, et al. (2019) Moon diver: a signatures as revealed by deep UV Raman spectroscopy.
discovery mission concept for understanding the history of Front Microbiol 10:679.
secondary crusts through the exploration of a lunar mare pit. Shapiro R and Schulze-Makuch D (2009) The search for alien
In 2019 IEEE Aerospace Conference, Big Sky, Montana, life in our solar system: strategies and priorities. Astrobiology
pp 1–23. 9:335–343.
Nicholson HA and Etheridge R (1878) A Monograph of the Sherwood B (2016) Strategic map for exploring the ocean-
Silurian Fossils of the Girvan District in Ayrshire with Spe- world Enceladus. Acta Astronautica 126:52–58.
cial Reference to Those Contained in the ‘‘Gray Collection,’’ Sotin C, Head III, JW, and Tobie G (2002) Europa: tidal heating
I, Volume 1. Blackwood, Edinburgh, p 341. of upwelling thermal plumes and the origin of lenticulae and
Niemi NA (2002) Extensional tectonics in the Basin and Range chaos melting. Geophys Res Lett 29:74-1.
province and the geology of the Grapevine Mountains, Death Squyres SW, Knoll AH, Arvidson RE, et al. (2009) Exploration
Valley region, California and Nevada. Doctoral dissertation, of Victoria crater by the Mars rover opportunity. Science 324:
California Institute of Technology. 1058–1061.
Niemi NA (2012) Geologic map of the central Grapevine Stewart JH. (1970) Upper Precambrian and Lower Cambrian
Mountains. Inyo County, California and Esmeralda and Nye strata in the southern Great Basin California and Nevada.
Counties, Nevada, Digital Maps Charts Ser., DMC12, scale, United States Geological Survey Professional Papers,
1(48,000):28. United States Department of the Interior, 620:206.
Northup DE (2004) Studies of Bacterial Communities from Summons RE, Bird LR, Gillespie AL, et al. (2013) Lipid bio-
Four Windows Cave, El Malpais National Monument, New markers in ooids from different locations and ages: evidence
Mexico. In 2004 National Speleological Society Convention, for a common bacterial flora. Geobiology 11:420–436.
Marquette, MI, p 111. Tawalbeh R, Voelz DG, Glenar DA, et al. (2013) Infrared
Northup DE, Melim LA, et al. (2011) Lava cave microbial acousto-optic tunable filter point spectrometer for detection
communities within mats and secondary mineral deposits: of organics on mineral surfaces. Opt Eng 52:063604.
implications for life detection on other planets. Astrobiology Thomas N, Sierks H, Barbieri C, et al. (2015) The morpho-
11:601–618. logical diversity of comet 67P/Churyumov-Gerasimenko.
Paar G, Deen RG, Muller JP, et al. (2016) Vision and image Science 347:aaa0440.
processing. In Contemporary Planetary Robotics: An Ap- Uckert K, Chanover NJ, Voelz DG, et al. (2015) Demonstration of a
proach toward Autonomous Systems, edited by Yang Gao, portable AOTF IR spectrometer for in situ exploration of plane-
Wiley-VCH Verlag GmbH & Co., Weinheim, Germany, tary surfaces. In 2015 IEEE Aerospace Conference, Big Sky,
Chapter: 3, pp 105–181. Montana, pp 1–9.
Parness A (2011) Anchoring foot mechanisms for sampling and Uckert K, Chanover NJ, Getty S, et al. (2017) The character-
mobility in microgravity. In Robotics and Automation ization of biosignatures in caves using an instrument suite.
(ICRA), Shanghai, China, pp 6596–6599. Astrobiology 17:1203–1218.
Parness A, Frost M, Thatte N, et al. (2013) Gravity-independent Uckert K, Bhartia R, and Michel J (2019) A semi-autonomous
rock-climbing robot and a sample acquisition tool with mi- method to detect cosmic rays in Raman hyperspectral data
crospine grippers. J Field Robot 30:897–915. sets. Appl Spectrosc 73:1019–1027.
Parness A, Abcouwer N, Fuller C, et al. (2017a) LEMUR 3: a Vu TH, Hodyss R, Johnson PV, et al. (2017) Preferential for-
limbed climbing robot for extreme terrain mobility in space. mation of sodium salts from frozen sodium-ammonium-
 INSTRUMENT SUITE ON A ROCK-CLIMBING ROBOT 1449

chloride-carbonate brines–Implications for Ceres’ bright

spots. Planet Space Sci 141:73–77. Abbreviations Used
Wold S, Esbensen K, and Geladi P (1987) Principal component 3D ¼ three-dimensional
analysis. Chemometr Intell Lab Syst 2:37–52. AOTF ¼ Acousto-Optic Tunable Filter
APE ¼ Advanced PIXL Experiment
DUV ¼ deep-UV
Address correspondence to: GURILA ¼ Geobiology with UV Raman Imaging and Laser-
Kyle Uckert induced Autofluorescence
Jet Propulsion Laboratory JPL ¼ Jet Propulsion Laboratory
California Institute of Technology LIDAR ¼ light detection and ranging
4800 Oak Grove Drive MOBIUS ¼ Mineral/Organic/Biological Investigations with
UV Spectroscopy
Pasadena, CA 91109
NASA ¼ National Aeronautics and Space Administration
USA PASA ¼ Portable AOTF Spectrometer for Astrobiology
PCA ¼ principal component analysis
E-mail: [Link]@[Link] PIXL ¼ Planetary Instrument for X-ray Lithochemistry
PMT ¼ photomultiplier tube
Submitted 3 October 2019 SDD ¼ silicon drift detector
Accepted 8 July 2020 XRF ¼ X-ray fluorescence
Associate Editor: Christopher McKay
Downloaded by Stanford University from [Link] at 12/12/21. For personal use only.