Technology Development and Testing for

Enhanced Mars Rover Sample Return Operations

Richard Volpe Eric Baumgartner Paul Schenker Samad Hayati
Mailstop 198-219 Mailstop 82-105 Mailstop 125-224 Mailstop 180-603
818-354-6328 818-354-4831 818-354-2681 818-354-8273

Jet Propulsion Laboratory

California Institute of Technology
Pasadena, California 91109
Email: [email protected]

Abstract—This paper describes several Jet Propulsion Labo- raise cameras and take panoramic images of the surrounding
ratory research efforts being conducted to support Mars sam- terrain, a shorter arm for sample acquisition and instrument
ple return in the coming decade. After describing the 2003/05 placement, and a sun sensor to accurately determine heading
mission scenario, we provide an overview of new technolo- while driving. These features were demonstrated during field
gies emerging from three complementary research efforts: tests in the Mojave Desert in 1997, which led directly to the
Long Range Science Rover, Sample Return Rover, and FIDO acceptance of the 03/05 missions [15].
Rover. The results show improvements in planning, naviga-
tion, estimation, sensing, and operations for small rovers op- The selected 03/05 mission concept, however, requires an en-
erating in Mars-like environments. larged rover that has the added functionality of carrying a drill
for rock sampling, larger wheels for enhanced mobility, and
a significantly upgraded science instrument suite (as opposed
TABLE OF C ONTENTS the Sojourner rover). Therefore, to support continued field
tests with the selected science team for the 03/05 mission,
1. I NTRODUCTION a new Field Integrated Design and Operations (FIDO) rover
2. M ARS S AMPLE R ETURN M ISSION S CENARIO was conceived, designed, integrated during a 12-month pe-
3. L ONG R ANGE S CIENCE ROVER T ECHNOLOGIES riod, and demonstrated in desert tests in April 1999 [14]. The
4. S AMPLE R ETURN ROVER T ECHNOLOGIES FIDO rover reflects the current engineering sensors and sci-
5. FIDO ROVER T ECHNOLOGIES ence instrument suite that are planned for the 03/05 mission.
6. S UMMARY While this rover will continue to act as an operations testbed
7. ACKNOWLEDGMENTS for mission scientists, it has a second function as an integra-
tion testbed for new technologies that continue to be devel-
1. I NTRODUCTION oped by ongoing research efforts.

Even before Sojourner made its first wheel tracks on Mars in The JPL core robotics technology program has been sup-
1997, it was anticipated that this rover would be only the first porting these research efforts, which include the Rocky 7
in a series of surface exploration spacecraft targeted for the rover as well as another platform, the Sample Return Rover
planet. While it will be Sojourner’s flight spare that drives (SRR) [13]. Each rover has been dedicated to increasing
on Mars in 2002, the next leap in technical capability exhib- autonomy in two respective halves of the exploration prob-
ited by rovers will be in the 2003/05 mission set, where much lem: autonomous motion away from and back to the lander.
larger rovers will perform rock and soil sample collection for New techniques used include estimation and visual localiza-
return to Earth. These rovers will have greater innate capa- tion, on-board path and sequence re-planning, and natural and
bilities, opening the door for the insertion of new robotics man-made target recognition and tracking.
technologies that have been in development since the incep-
tion of the Pathfinder mission five years ago. Among these This paper describes these techniques, as well as the details
are on-board stereo vision processing, autonomous lander- of the mission scenario in which they will be used. Sec-
less operations, manipulation and instrument positioning by tion 2 describes the Mars Sample Return’s Athena rover mis-
arms, precision navigation for rover/lander rendezvous, and sion. The technology developments associated with the Long
distributed ground operations. Range Science rover task are discussed in Section 3, while
the Sample Return Rover and the development of rendezvous
Of fundamental importance to the incorporation of the new techniques are described in Section 4. The Athena terrestrial
capabilities on next-generation rovers is the use of a more prototype rover, FIDO, is discussed in Section 5, and sum-
capable electronics, sensing, and instrumentation infrastruc- mary remarks are provided in Section 6.
ture located on-board the rover. For instance, as Sojourner
was being prepared for flight, JPL was constructing a new
prototype, Rocky 7 [16]. Several key features were added to
support long trips away from the lander: a deployable mast to

0-7803-5846-5/00/$10.00 c 2000 IEEE

 During the mission, communication with the rover will nom-
inally take place twice per day, relayed by the lander. Each
communication window will allow a limited set of images
and data to be transmitted to operators on earth, while new
instructions are provided to the rover based on the previous
communication cycle. Typically the rover will receive in-
structions for the day in the morning, and transmit the results
and status at the end of the day.

After obtaining its first set of rock samples, the rover will re-
turn to the lander and deposit them in the MAV. This must be
accomplished by successfully aligning with the base of the
ramp, driving up its two narrow rails, and accurately detect-
ing the proper position for sample transfer. At this location,
the MAV payload door will be opened and the sample transfer
will be completed robustly and autonomously; thermal con-
siderations require that the payload door be open less than
that of the typical communications cycle.

Since mission constraints limit surface operations to less than

90 days, the sample acquisition and return to MAV cycle can
Figure 1. Preliminary drawing of the Mars Sample Return
be performed only three times at most. It is likely, however,
Athena-Rover.
that each cycle will see the rover venturing farther from the
lander.
2. M ARS S AMPLE R ETURN M ISSION
S CENARIO After the last sample return operation, the rover will move off
the lander deck and far enough away from the lander to pre-
Launching in 2003 and again in 2005, NASA’s Mars Sample vent its being damaged during the MAV lift-off. This launch
Return (MSR) spacecraft will place two science rovers on the is expected to damage the lander communication system and
surface of the planet. These rovers will be carried to the sur- prevent it from acting as a relay for the rover. Therefore,
face on the top deck of a three-legged lander that is roughly the rover will use auxiliary communication to an orbiter, to
1 m tall and 3 m in diameter. This size, substantially larger enable it to perform the extended mission of exploring the
than previous landers of the 1996, 1998, and 2001 missions, surface.
makes it possible to carry a sample-return Mars Ascent Ve-
hicle (MAV) as part of the lander payload. The lander size Obviously, the complexity and accuracy of the autonomous
also enables the use of a rover that is approximately twice the operations described above directly influence the amount of
size of the Pathfinder mission rover, Sojourner, in every di- science operations that will be performed. For this reason,
mension. (Sojourner was 60 x 40 x 35 cm.) A preliminary JPL research projects aim to introduce new functionality and
drawing of this “Athena-class” rover is shown in Figure 1. features into Mars rovers to enable greater science return
from all upcoming Mars rover missions. Techniques for more
The larger size of this MSR rover is needed to support rock autonomous and robust exploration and return to the lander
sampling operations and to carry the seven science instru- have been developed and implemented in field tests. Each of
ments that make up its payload. Rock sampling is accom- these efforts is described next.
plished using a coring drill, which relies on the rover mass to
provide the force behind it in its vertical operating configu- 3. L ONG R ANGE S CIENCE ROVER
ration. The science instruments are used to select the target T ECHNOLOGIES
rocks for sampling, to determine the composition of the rock
samples obtained, and to study rocks in the surrounding area. To improve rover navigation, exploration, and autonomy, the
Four of the instruments, located on a five degrees-of-freedom Long Range Science Rover (LRSR) research task has been
arm, require close proximity to the sample to be measured. improving the rover’s ability to navigate through the environ-
Two others, located on a 1 m mast, require only line of sight ments, while maintaining an accurate sense of its position.
to the target. The drill, itself, is the seventh instrument. This section describes advances in four pertinent areas: dy-
namic sequence generation, autonomous path planning, vi-
The rover will begin its mission on the lander top deck by sual localization, and state estimation. All research was con-
obtaining a panorama of the surrounding terrain for science, ducted with the prototype rover, Rocky 7, shown in Figure 2.
engineering, and public outreach purposes. From these im-
ages, a ramp deployment direction will be selected, as well as
initial travel routes and goals for the first rover traverse. Af- Dynamic Sequence Planning
ter ramp deployment, the rover will drive to the surface and
begin navigating the terrain. The maximum distance driven On-board planning with dynamic sequence generation allows
each day will be 100 m, and often much less, especially when ground controllers to provide much higher level commands,
the rover is positioning itself for science operations. while increasing the optimality and robustness of rover opera-
 Original planned path
Re-planned path

Figure 3. After encountering a previously unknown obstacle

shown in light orange, CASPER replans the sequence
of targets. Blue crosses are the goal locations, and dark
orange shapes are obstacles known a priori.
Figure 2. The Rocky 7 research prototype.
cameras, select a path through the terrain to the edge of the
effective stereo range, and repeat the process until the goal
tions on the surface. For instance, during the Pathfinder Mis- is achieved. A representative example of a partial panorama
sion, the Sojourner rover [10] was provided with extremely and the resulting elevation map of the terrain are shown in
detailed sequences daily, fatiguing operators while also dis- Figure 4.
allowing contingency operations when the flow of execution
was non-nominal. Contrary to this, we have been experiment- Rocky 7 uses a local sensor-based path planner called Rover-
ing with on-board replanning that can change the execution Bug [8], [9]. This algorithm was developed for vehicles that
of daily activities based on unanticipated variations in quanti- have limits on sensor range, field of view, and processing.
ties such as position, terrain, power, and time. To accomplish The two main modes of operation are motion-to-goal and
this, we have used a dynamic on-board planing system called boundary-following, which are used to provide global con-
CASPER (Continuous Activity Scheduling, Planning, Execu- vergence.
tion, and Replanning) [5], [6].
Rover-Bug works by using the local elevation map to con-
Figure 3 shows an example scenario in map form, where dark struct a map of convex hulls around all obstacles in the sens-
orange shapes represent obstacles known a priori (e.g. from ing range. These hulls are then merged and grown to provide
lander descent imagery). In this case, the initial plan for the a configuration space representation of the sensed terrain. A
traverse will bring the rover to an unexpected obstacle near tangent graph is constructed to determine if there is an un-
the first goal, represented as a light orange shaded shape. Cir- obstructed path to the envelope of the sensed region in the
cumnavigation around this obstacle will move the rover closer direction of the goal. If one exists, it is followed, and the pro-
to other goals, triggering CASPER to recognize the situation cess is repeated at the end of the path segment to the sensory
and re-plan to visit the closest goal first. We are currently envelope.
evaluating this and similar scenarios experimentally.
If a free path does not exist, stereo images are obtained to
Autonomous Path Planning the most promising side of the current view, and the process
is repeated. In some cases, the free path places the rover at
For the longer traverses required of upcoming missions, au- the edge of the sensory envelope but still obstructed from the
tonomous path planning is desirable since operators will not goal by an obstacle. In this case, the rover will begin to use
be able to see three-dimensional terrain features out to the its body-mounted cameras to reactively boundary follow until
more distant goal locations. Whereas Sojourner drove a total there is a clear path to the next goal.
of 84 m during its entire mission, the MSR Athena rover will
be capable of driving this distance in a single day. However, Figure 5 shows experimental data obtained from Rocky 7
stereo imagery of the terrain provided to operators will only while using this algorithm to traverse our MarsYard test area.
have an envelope of 20 m at best resolution. Therefore, the The start position is in the lower left corner, 21 m from the
path planning advances described here will allow the rover to goal in the upper right. Four sets of sensory data are shown as
be its own operator. It can image the terrain from periscopic
Figure 4. Steps of on-board terrain sensing: panoramic mosaic view from rover mast stereo imager, composite range map
extracted from stereo views, and elevation map created from range data.

Goal position

Localization target

Figure 5. Experimental results from a multi-step run using

Rover-Bug in the JPL MarsYard.

orange wedges along the path. Projected on to these data sets

are the pink convex hull representations of the sensed obsta-
cles. The green and blue lines passing through the obstacle
field are the planned and executed paths, respectively. The
gap between data sets is due to a lack of merging of data from Figure 6. Example of automatic target selection for local-
both the mast and body-mounted cameras, and is currently ization.
being corrected. The sudden changes in the path direction at
the far side of each wedge does not indicated actual rover mo- The technique relies on obtaining a stereo elevation map from
tion; rather it is an artifact of the estimated rover position that the initial rover position, as shown in Figure 4. This imagery
is updated by localization at this point in the traverse. This is automatically analyzed based on the range data quality, and
localization is discussed next. the quality expected to be seen at the goal point specified by
path planning [11]. Typically, there will be a prominent rock
Visual Localization between the two locations. From the analysis, it will be se-
lected as the localization target, and viewed by the rover after
Visual localization uses the same terrain imagery as path reaching the end of the local path segment. Figure 6 shows
planning, but for the purpose of monitoring the apparent mo- an example of the automatic target selection, with four images
tion of three-dimensional ground features after the rover has concatenated to provide a reasonable number of potential tar-
completed a move. In this way, the on-board position esti- gets.
mate of the rover can be updated to compensate for errors
caused by wheel slippage or rock bumping. On Pathfinder, To match the two stereo views of the terrain, a multi-
this localization functionality was performed manually by op- resolution search technique is used to provide the best esti-
erators viewing Sojourner from the fixed position lander cam- mate of the displacement of the original elevation map from
eras, restricting the update to once a day and permitting oper- the final one [12]. Typical results provide position error es-
ations only within the stereo envelope of the lander. In con- timates that are 1% of the distance traveled. However, the
trast, the terrain-based localization described here has appli- search uses the on-board estimate of the rover position as its
cation to many forms of landerless operations: incremental starting point, so more reliable results are obtained if the ini-
long traverses, local operations within range of a prior stereo tial estimate is more accurate. Therefore, to obtain better con-
panorama, localization in descent imagery, and closed chain tinuous position estimates, we have been developing a new
rover moves with estimate smoothing. estimation technique, discussed next.
 20 10 5 10

ξ1 (deg)

ξ4 (deg)
EST EST
ξ1 (deg)

ξ4 (deg)
10 SIM 0 SIM 0 0

0 −10 −5 −10

−10 −20 −10 −20

0 2 4 6 8 10 0 2 4 6 8 10 0 5 10 15 20 25 0 5 10 15 20 25

10 5 1 20

ξ2 (deg)

ξ5 (deg)
EST EST
ξ2 (deg)

ξ5 (deg)
5
SIM 0 SIM 0 10
0
−5 −1 0
−5

−10 −10 −2 −10

0 2 4 6 8 10 0 2 4 6 8 10 0 5 10 15 20 25 0 5 10 15 20 25

10 10 2 10

ξ3 (deg)

ξ6 (deg)
EST EST
ξ3 (deg)

ξ6 (deg)
5
SIM 5 SIM 1 0
0
0 0 −10
−5

−10 −5 −1 −20
0 2 4 6 8 10 0 2 4 6 8 10 0 5 10 15 20 25 0 5 10 15 20 25
t (s) t (s) t (s) t (s)

Figure 7. Simulation results showing true values and esti- Figure 8. Experimental results from Rocky 7 showing the
mates for wheel contact positions, where the left/right estimates for wheel contact angles.
three wheels angles are shown by the left/right side
graphs.

Kinematic State Estimation

In addition to periodic visual localization of the rover, we

have developed real-time position and heading estimation us-
ing all other sensors on the vehicle: angular rate, accelerom-
eter, sun sensor, wheel angle rates, and mobility system link-
age (rocker-bogey) configuration. This technique moves far
beyond the simple dead reckoning of Sojourner and improves Figure 9. The Sample Return Rover.
upon our previous advances in position estimation with sun
sensing [15]. The results aid navigation during path execu-
tion, provide better input to localization, and replace both in wheel contact points change only slightly, the much larger
visually featureless terrain (e.g., sand dunes) or in the case of changes in the right side are correctly determined. Note, in
visual sensing failure. the final configuration, the bogey wheels were still on the ob-
stacle, while the steering wheel had completely traversed the
The estimator uses a Kalman filter framework, with the pro- obstacle. The estimated wheel contact angles were within
cess model chosen so that the inertial sensor data are used as 5 degrees of the true values measured independently. Further
an input to drive the process equation. The procedure avoids experiments are in progress, as well as integration of visual
the difficulty of modeling the detailed process dynamics [1] position estimation, as described in the next section.
by exploiting the ability of the Kalman filter to perform the
appropriate least-squares averaging of the action of each kine- 4. S AMPLE R ETURN ROVER T ECHNOLOGIES
matic chain in the rover. These forward kinematic chains have
velocity components defined by the sequence of links joining In 1997, JPL began development of the Sample Return Rover
the rover frame to each estimated wheel contact point, and (SRR), a small, lightweight rover that investigated focused
a component given by the slip between the wheel and the technology advances in the areas of rover-to-rover and rover-
ground. The deterministic component of the slip is used to to-lander rendezvous. SRR, shown in Figure 9, is a 7 kg
capture the effects of a known steering action or a known av- rover with a 4-wheel rocker mobility system that is capa-
erage slip rate over different kinds of terrain. ble of traversing over obstacles up to 15 cm in height. The
rover includes 4-wheel steering and carries a three degrees-
Both simulation and experimental results with this estimation of-freedom manipulator arm with a 1 degree-of-freedom grip-
technique have been conducted. Figures 7 shows the ability of per. This robot arm is used for panoramic imaging along
the estimator to correctly track the wheel contact angle over with sample pick-up and transfer. The rover also includes
an undulating terrain. The simulation is particularly valuable a posable rocker joint for variable ground clearance and rover
since ground truth can be known exactly. For instance, using reconfiguration in difficult terrain. SRR carries a PC104+
simple integration of the wheel velocity results in a 2% error derived electronics system, including a 300 MHz AMD K6
in measured position, while the estimator reduces the error by processor, motion control I/O boards (D/A and encoder read-
an order of magnitude. ers with closed-loop control realized in software), A/D board,
PCI color framegrabbers, and a wireless Ethernet.
Experimental results further support these conclusions. Fig-
ure 8 shows the results from Rocky 7 driving over an obsta- The original concept for SRR derived from the previous ver-
cle on the right side of the vehicle only. While the left side sion of the MSR mission where an Athena-class rover tra-
versed long distances, stopped at interesting science sites, and of the rover from 25 m from the lander to 5 m in front of
acquired and stored rock and soil samples. During a second
 the ramps. Finally, visual acquisition of the lander ramps is
mission, a small, lightweight rover would land on the Martian accomplished using cooperative markings from which rover-
surface and rendezvous with the science rover to retrieve the to-ramp position and heading information is obtained. The
sample collected during the primary science mission. Such successive visual acquisition of the lander ramps brings the
a rover-to-rover rendezvous was demonstrated by the SRR rover from 5 meters to within 5 cm of the bottom of the lan-
technology team during the summer of 1998 [13]. This ren- der ramps. Many successful experiments have been accom-
dezvous was accomplished using an RF beacon transmitter plished in both laboratory and outdoor settings. These results
and receiver pair for non-line-of-site navigation to the sci- indicate that the rover can reliable and robustly navigate to the
ence rover, visual tracking of the static science rover during bottom of the ramps with an absolute precision of 1 cm in lat-
line-of-sight navigation, determination of rover-to-rover pose eral and longitudinal offset and less than 1 degree orientation
using man-made features located on the science rover, and, error with respect to the ramps. As such, the combination of
terminal guidance to the sample cache container and pickup these rover navigation techniques leads to a single-command
of this container using the on-board manipulator arm. autonomous sequence associated with the return to the lander
and the regress of the rover up the lander ramps to deposit the
In 1999, the SRR task turned its attention to the rover-to- sample cache in the MAV.
lander rendezvous problem in support of the MSR mission
and the requirement that the Athena rover return its cached Finally, the SRR task has, over the past two years, developed
samples to the lander and the awaiting MAV, as described an alternative form of rover state estimation for the accurate
in Section 2. Some of the technology developments estab- and reliable determination of rover position and orientation
lished during the rover-to-rover rendezvous application were relative to a fixed reference frame. This technique uses an ex-
transferred to the rover-to-lander rendezvous problem. The tended Kalman filter framework based on the work described
requirements of the Athena rover mission, however, necessi- in [3]. Within the SRR development, the registration of suc-
tated the development of new techniques for the robust and cessive range maps generated by the rover’s forward-looking
accurate navigation of SRR with respect to the lander such hazard avoidance cameras are utilized to determine the frame-
that autonomous lander acquisition, lander rendezvous, and to-frame translation and rotation of the rover. This informa-
ramp climbing are possible with minimal, if any, ground sup- tion is combined with dead reckoned estimates of the rover’s
port. Figure 10 depicts the scenario associated with the rover- translation and rotation to produce an optimized determina-
to-lander rendezvous problem in terms of the multi-phase op- tion of the rover pose. This work is described in [7] and [2]
erations and technologies used during the return to the lander. and illustrated in Figure 12 for a 6+ m traverse within a soft-
The rover-to-lander rendezvous problem is divided into the soil, rock-filled indoor sand pit.
following phases:
5. FIDO ROVER T ECHNOLOGIES


Long distance visual tracking using lander texture features

derived in the wavelet space As described in Section 2, the 2003/05 MSR’s Athena rover


Multi-point tracking of lander features for heading and mission represents a significant increase in complexity over
range estimation of the rover relative to the lander the recent Sojourner rover mission. The Athena rover carries


Ramp location determination using lander features seven science payload elements, including a multi-spectral


Ramp recognition using cooperative ramp features for imaging system, a microimager, a sample acquisition system,
heading and range estimation of the rover relative to the and four different spectrometers as compared to Sojourner’s
bottom of the ramps single science instrument, the Alpha Proton X-ray spectrom-
eter. In addition, the Athena rover has over eight times the
The long-distance navigation of SRR relative to the lander is volume and six times the mass of the Sojourner rover. As
accomplished with a novel wavelet-based detection algorithm such, the Athena rover represents a true sciencecraft and ful-
for the long-range visual acquisition of a lander from greater fills the need for a robotic field geologist on Mars.
than 100 m using a single, black-and-white rover imaging
system (20 degree field of view) [4]. This information en- To facilitate the successful operation of the Athena rover dur-
ables autonomous correction the rover heading with respect ing the 2003/05 Mars Sample Return missions, a terrestrial
to the lander, and to guidance of SRR to within 25 m of the prototype of the Athena rover is being used by the Athena sci-
lander. Such a navigation sequence is shown in Figure 11. ence team for science and engineering testing associated with
Within 25 meters of the lander, a visual technique that takes flight mission operations. The development of this robotic
advantage of the known geometry of the lander structure is vehicle, known as FIDO (for Field Integrated, Design, and
used to track multiple features (e.g., lander leg struts, lander Operations) rover began in early 1998 with the conceptual
deck, etc.) to determine the pose of the rover relative to the design of the vehicle and culminated nine months later with
lander. Preliminary results indicate that precision on the order the full-scale integration of the rover and its science payload
of 50 cm in range and 1 to 2 degrees in orientation is possible [14]. In April 1999 (less than 14 months since the initial paper
using this approach. designs), the FIDO rover was successfully operated at the Sil-
ver Lake field test site outside Baker in California’s Mojave
Likewise, the known ramp geometry allows for the visual ac- desert. Figure 13 shows FIDO operating during this desert
quisition of the lander ramps and the relative positioning of field test.
the rover relative to the ramps. These techniques combine to
produce a navigation strategy for the autonomous guidance FIDO’s mobility subsystem consists of a 6-wheel rocker-
 Figure 10. The operations scenario associated with the return to the lander.

Figure 11. Wavelet-based lander detection and navigation.

Position Est. Error wrt. Ground Truth, K.F. vs. D.R.

3

Kalman Filtered
Dead Reckoned
2.5

2
Position Error (m)

1.5

0.5

0
0 10 20 30 40 50 60 70 80
Time (Navigation Iterations)

Figure 12. State estimation for rover pose determination.

 Figure 14. The miniature core drill acquiring a core sample
from a carbonate rock at Silver Lake

The science payload on FIDO is analogous to the Athena pay-

load. In particular, the remote sensing suite located on the
FIDO mast includes a multi-spectral, narrow field-of-view
Pancam stereo imaging system; a monochromatic, wider
field-of-view Navcam stereo imaging system; and the optics
for a near-IR point spectrometer that operates in the 1200 to
2500 nm wavelength region. The multi-spectral capability
associated with the Pancam system is realized using a Liq-
uid Crystal Tunable Filter (LCTF) that is tuned to the three
near-IR wavelengths of 650, 750 and 850 nm. The in situ
instrument suite attached to the end-effector of the FIDO in-
strument arm consists of a color microimager and a Moess-
bauer spectrometer that are used to determine the iron content
of target rocks. A miniature core drill system, body-mounted
Figure 13. The FIDO rover during the Silver Lake field trial. to the rover, provides the capability to acquire and cache rock
and soil samples. All of these instruments, with the exception
of the near-IR point spectrometer (the flight mission uses a
bogie suspension system that has been scaled up by a factor miniature thermal emission spectrometer in the mid-IR wave-
of 20/13 from the Sojourner design. This suspension sys- length region), are breadboards of the Athena flight instru-
tem allows for the safe traverse over obstacles up to 30 cm in ments.
height. Each wheel is independently driven and steered us-
ing a Sojourner-derived actuation and encoder system. The In total, the science instrument and engineering sensing suites
top ground speed of the vehicle is 9 cm/sec. Approximate and the resulting FIDO rover system represent a terrestrial
rover dimensions are 1 m in length, 0.8 m in width, 0.5 m in analog of the Athena rover that can be used to test and vali-
height, and 0.23 m ground clearance. The rover carries a four date the Athena mission scenario and associated engineering
degrees-of-freedom deployable mast that stands 1.94 m off functions. As such, the Athena science team led by Professor
the ground surface at full extent. This mast provides the nec- Steven Squyres, Athena PI, and Professor Raymond Arvid-
essary pan-and-tilt control for panoramic imaging and point son, Athena Co-I, has worked with the FIDO engineering
spectroscopy. FIDO also carries a four degrees-of-freedom team since March 1999 to perform rover operations testing in
instrument arm that is used to place the in situ suite of instru- support of the Athena rover mission. In particular, the desert
ments on rock and soil targets. field trial at the Silver Lake test site represented the first ever
demonstration and validation of the sample acquisition phase
The FIDO electronics are similar in nature to the SRR elec- of the Athena rover mission through the identification of tar-
tronics with the CPU being a 80586 AMD processor running get rocks, approach to the target rock, placement of the minia-
at a 133 MHz clock speed. The rover uses a PC104-based ture core drill over the target rock, successful acquisition of
platform for all I/O functions, including a motion control a core sample using the miniature core drill, and return of
system (D/A and encoder readers with closed-loop control these samples to a simulated landing site. Figure 14 shows
in software) that can control up to 30 actuators simultane- the FIDO rover and associated core drill during the success-
ously, two monochromatic and one color framegrabbers, dig- ful acquisition of a core sample from a carbonate rock. Future
ital I/O boards, A/D boards, low-pass filter and analog mul- field trials in 2000 and 2001 will focus on flight-like rover op-
tiplexer boards, and a wireless Ethernet. Engineering sensors erations, with the rover and science teams being sequestered
include front and rear stereo hazard avoidance camera sys- at JPL while the rover is located at a remote test site. During
tems, an inertial navigation system, and a sun sensor for abso- such “blind” field trials, the full Athena mission scenario will
lute heading determination. A differential GPS unit is also in- be the further validated, including sample acquisition and re-
tegrated within the rover electronics for ground-truthing pur- turn to the lander as well as long-range science exploration
poses only. and discovery.
 6. S UMMARY craft. In IEEE Aerospace Conference, Aspen CO, March
1999.
This paper has reviewed several recent research efforts that [6] T. Estlin, G. Rabideau, D. Mutz, and S. Chien. Using
support upcoming MSR mission scenarios. Three research Continuous Planning Techniques to Coordinate Multiple
tasks, LRSR, SRR, and FIDO, have been developing and test- Rovers. In IJCAI Workshop on Scheduling and Planning,
ing new capabilities in prototype rover platforms. LRSR re- Stockholm, Sweden, August 1999.
search with Rocky 7 has developed four new techniques to en- [7] B. D. Hoffman, E. T. Baumgartner, T. L. Huntsberger,
hance the functionality of autonomous long-range Mars rover and P. S. Schenker. Improved Rover State Estimation in
navigation: intelligent sequencing, sensor constrained path Challenging Terrain. Autonomous Robots, 6(2), 1999.
planning, natural terrain visual localization, and Kalman Fil- [8] S. Laubach. Theory and Experiments in Autonomous
ter state estimation. SRR has developed a return-to-lander Sensor-Based Motion Planning with Applications for
capability with visual tracking at various ranges, using tech- Flight Planetary Microrovers. PhD thesis, California In-
niques suitable to those ranges. It has also displayed visual stitute of Technology, May 1999.
odometry techniques. Finally, FIDO represents the culmina- [9] S. Laubach and J. Burdick. An Autonomous Sensor-
tion of the two core robotic technology programs since its de- Based Path-Planner for Planetary Microrovers. In IEEE
velopment is based on the experiences gained within LRSR International Conference on Robotics and Automation,
and SRR. As a result, it has become a standard integration and Detroit MI, 1999.
test system for the validation of rover navigation and control [10] J. Matijevic et al. Characterization of the Martian Sur-
strategies. face Deposits by the Mars Pathfinder Rover, Sojourner.
Science, 278:1765–1768, December 5 1997.
7. ACKNOWLEDGMENTS [11] C. Olson. Subpixel Localization and Uncertainty Esti-
mation Using Occupancy Grids. In IEEE International
Large systems like these require the support of many people Conference on Robotics and Automation, pages 1987–
beyond the authors. We would like to recognize the follow- 1992, Detroit, Michigan, May 1999.
ing individuals for their invaluable contribution to the work [12] C. Olson and L. Matthies. Maximum-likelihood Rover
described in this paper: the LRSR team includes J. (Bob) Localization by Matching Range Maps. In IEEE Inter-
Balaram, Clark Olson, Sharon Laubach, Tara Estlin, Richard national Conference on Robotics and Automation, pages
Petras, Darren Mutz, Greg Rabideau, and Mark Maimone; 272–277, Leuven, Belgium, May 1998.
the SRR team includes Hrand Aghazarian, Terry Huntsberger, [13] P. S. Schenker et al. New Planetary Rovers for Long
Yang Cheng, Mike Garrett, and Lee Magnone; the FIDO team Range Mars Science and Sample Return. In Intelligent
includes Hrand Aghazarian, Terry Huntsberger, Anthony Lai, Robots and Computer Vision XVII, SPIE Proceedings
Randy Lindemann, Paul Backes, Brett Kennedy, Lisa Reid, 3522, Boston, November, 1998.
Mike Garrett, and Lee Magnone. [14] P. S. Schenker et al. FIDO Rover and Long-Range
Autonomous Mars Science. In Intelligent Robots and
The research described in this paper was carried out by the Computer Vision XVIII, SPIE Proceedings 3837, Boston,
Jet Propulsion Laboratory, California Institute of Technology, September, 1999.
under a contract with the National Aeronautics and Space Ad- [15] R. Volpe. Navigation Results from Desert Field Tests of
ministration. Reference herein to any specific commercial the Rocky 7 Mars Rover Prototype. International Jour-
product, process, or service by trade name, trademark, man- nal of Robotics Research, 18(7), 1999.
ufacturer, or otherwise, does not constitute or imply its en- [16] R. Volpe et al. Rocky 7: A Next Generation Mars Rover
dorsement by the United States Government or the Jet Propul- Prototype. Journal of Advanced Robotics, 11(4):341–
sion Laboratory, California Institute of Technology. 358, 1997.

R EFERENCES
[1] J. Balaram. Kinematic State Estimation for a Mars
Rover. Robotica, Special Issue on Intelligent Au-
tonomous Vehicles, Accepted for publication, 1999.
[2] E. T. Baumgartner, P. C. Leger, P. S. Schenker, and T. L.
Huntsberger. Sensor Fused Navigation and Manipulation
from a Planetary Rover. In Sensor Fusion and Decen-
tralized Control in Robotic Systems, SPIE Proceedings
3523, Boston, November, 1998.
[3] E. T. Baumgartner and S. B. Skaar. An Autonomous
Vision-Based Mobile Robot. IEEE Transactions on Au-
tomatic Control, 39(3):493–502, March 1994.
[4] J-K. Chang and T. L. Huntsberger. Dynamic Motion
Analysis using Wavelet Flow Surface Images. Pattern
Recognition Letters, 20(4):383–393, 1999.
[5] S. Chien, R. Knight, R. Sherwood, and G. Rabideau. In-
tegrated Planning and Execution for Autonomous Space-
Richard Volpe, Ph.D., is the Principal Investigator for the Eric T. Baumgartner, Ph.D., is a group leader in the Me-
Long Range Science Rover Research Project, and the System chanical and Robotics Technology Group and a senior mem-
Technologist for the Athena-Rover sample collection robot ber of engineering staff in the Science and Technology De-
for the 2003 Mars Sample Return Project. His research in- velopment Section at NASA’s Jet Propulsion Laboratory in
terests include real-time sensor-based control, robot design, Pasadena, CA. At JPL, he serves in a systems engineering ca-
software architectures, path planning, and computer vision. pacity for the development of advanced planetary rovers and
also contributes to technology developments in the areas of
Richard received his M.S. (1986) and Ph.D. (1990) in Ap- robotic sensing and control. Prior to his tenure at JPL, he
plied Physics from Carnegie Mellon University, where he was an Assistant Professor in the Mechanical Engineering -
was a US Air Force Laboratory Graduate Fellow. His the- Engineering Mechanics Department at Michigan Technolog-
sis research concentrated on real-time force and impact con- ical University in Houghton, MI. He has published over 30
trol of robotic manipulators. Since December 1990, he has articles in the area of robotic, controls, and state estimation
been at the Jet Propulsion Laboratory, California Institute of and is active in the SPIE and ASME. He received his B.S. de-
Technology, where he is a Senior Member of the Technical gree in Aerospace Engineering from the University of Notre
Staff. Until 1994, he was a member of the Remote Surface In- Dame, the M.S. degree in Aerospace Engineering from the
spection Project, investigating sensor-based control technol- University of Cincinnati in 1990, and the Ph.D. in Mechani-
ogy for telerobotic inspection of the International Space Sta- cal Engineering from the University of Notre Dame in 1993.
tion. Starting in 1994, he led the development of Rocky 7, a
next generation mobile robot prototype for extended-traverse
sampling missions on Mars. In 1997, he received a NASA
Exceptional Achievement Award for this work, which has led
to the design concepts for the 2003 Mars rover mission.
Paul S. Schenker, Ph.D., is Supervisor of the Mechani- Samad Hayati received his M.S. and Ph.D. in Mechanical
cal and Robotics Technologies Group, Mechanical Systems Engineering with a specialty in controls from the University
Engineering and Research Division, Jet Propulsion Labora- of California at Berkeley in 1972, and 1976, respectively. He
tory. His current work emphasizes planetary rover devel- joined the Jet Propulsion Laboratory in 1979 and worked in
opment and robotic sampling technologies; he is task man- the guidance and control of the Jupiter orbiter Galileo space-
ager for NASA/JPL’s Sample Return Rover (SRR) and Ex- craft, currently orbiting Jupiter. From 1983-1999 he per-
ploration Technology Rover Design, Integration, and Field formed research in robotics at JPL. He was also a lecturer of
Test R&D efforts (”ET Rover/FIDO” – a Field Integrated robotics at the California Institute of Technology from 1986
Design & Operations rover prototype supporting the NASA to 1988. He has published numerous conference and journal
Mars ’03-’05 sample return missions, and related terres- papers and holds two US patents related to robotics control.
trial mission simulations), as well as new NASA tasks on His pioneering work in robot calibration was used to develop
use of cooperating robotic assets/rovers for Mars exploration techniques to utilize manipulators as an aid in brain neuro-
and future human habitation. Schenker also recently led surgery at the Long Beach Memorial Hospital in California
JPL’s Planetary Dexterous Manipulators R&D under NASA in 1986. His most recent research efforts were in the develop-
funding, work that proto- typed a robotic sampling con- ment of long range science rover and rendezvous and sample
cept which flies on the NASA Mars Polar Lander mission retrieval technologies for NASA’s planned missions to Mars.
now in route to the red planet. Schenker’s other recent
robotics R&D activities include a role as founding co-PI for Currently, Samad is the manager of Robotics and Mars Ex-
NASA/MicroDexterity Systems Inc. development of a Robot ploration Technology Programs at JPL.
Assisted Microsurgery high-dexterity tele-operative worksta-
tion and a longer standing involvement in various tele-robotic
technology and system developments for orbital servicing
and autonomous robotic exploration. Schenker is a member
of AAAI, IEEE, and SPIE; he is a Fellow and 1999 Presi-
dent of the last. Schenker is widely active in external techni-
cal meetings, publications, and university collaborations in
the areas of robotics and machine perception, having con-
tributed about 100 archival articles to same. Dr. Schenker
received his B.S. in Engineering Physics from Cornell Uni-
versity, and completed his M.S., Ph. D. and postdoctoral stud-
ies in Electrical Engineering at Purdue University. Prior to
joining JPL/Caltech in 1984, Schenker was with the Electri-
cal Sciences faculty, Brown University, and later the Research
Section Chief for Signal & Image Processing, Honeywell Inc.