traverse performance characterization for the mars science laboratory rover

Traverse Performance Characterization for the MarsScience Laboratory Rover

• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • •

Matt Heverly, Jaret Matthews∗, Justin Lin, Dan Fuller, Mark Maimone, Jeffrey Biesiadecki, and John LeichtyJet Propulsion Laboratory, California Institute of Technology, Pasadena, California, 91109

Received 14 December 2012; accepted 19 July 2013

It is anticipated that the Mars Science Laboratory rover, named Curiosity, will traverse 10–20 km on the surface ofMars during its primary mission. In preparation for this traverse, Earth-based tests were performed using Marsweight vehicles. These vehicles were driven over Mars analog bedrock, cohesive soil, and cohesionless sand atvarious slopes. Vehicle slip was characterized on each of these terrains versus slope for direct upslope driving.Results show that slopes up to 22 degrees are traversable on smooth bedrock and that slopes up to 28 degreesare traversable on some cohesive soils. In cohesionless sand, results show a sharp transition between moderateslip on 10 degree slopes and vehicle embedding at 17 degrees. For cohesionless sand, data are also presentedshowing the relationship between vehicle slip and wheel sinkage. Side by side testing of the Mars ExplorationRover test vehicle and the Mars Science Laboratory test vehicle show how increased wheel diameter leads tobetter slope climbing ability in sand for vehicles with nearly identical ground pressure. Lastly, preliminary datafrom Curiosity’s initial driving on Mars are presented and compared to the Earth-based testing, showing goodagreement for the driving done during the first 250 Martian days. C© 2013 Wiley Periodicals, Inc.

1. INTRODUCTION

When driving the rover on Mars, a sequence of motion com-mands is sent to the vehicle roughly once per Martian day,otherwise known as a sol. Various levels of rover autonomycan be invoked to drive the rover anywhere from a few me-ters to 100 meters or more per sol. The final location of thevehicle is typically dictated by the science objectives, butthe traverse path to reach the final location is constructed tomaximize vehicle safety. This process of selecting a sciencetarget, evaluating the terrain, and constructing a sequenceof motion commands must be done in only a few hoursin order to allow new commands to be sent to the vehicleeach sol. It is therefore important to have accurate and con-cise metrics to be able to evaluate vehicle safety and predictvehicle performance on a given terrain. One such metric isvehicle slip. While slip is not inherently unsafe, it does resultin deviation from the planned path, which could result inthe vehicle encountering unsafe terrain such as large rocks.In soft soil, excessive slip can lead to vehicle embedding.The Mars Exploration Rover (MER), named Opportunity,encountered a soft sand ripple on its 446th day of surfaceoperations. High wheel slip and sinkage resulted in embed-ding, and it took more than five weeks to extricate the rover.

∗Formerly of the Jet Propulsion Laboratory, currently at SpaceX,Hawthorne, CA.Direct correspondence to: Matt Heverly, e-mail: [email protected]

Due to these safety concerns, it is important to characterizethe Curiosity rover’s slip potential in a variety of terraintypes.

The ability to predict terrain traversability as a functionof slope was not only helpful in tactical operations, it alsoaided in selecting a safe landing site for the rover. Sinceterrain slopes can be measured from orbital data and terraintype can be generally classified by orbital images, the abilityof a vehicle to traverse can be mapped to first order priorto its arrival on Mars. This allowed for the selection of alanding site that was tolerant to uncertainty in the preciselanding position based upon the probability of landing ontraversable terrain.

Sections 2 and 3 of this paper describe the Curiosityrover as well as the Earth test vehicle known as Scare-crow, shown in Figure 1. Section 4 presents a GlobalPositioning System (GPS) -based method as well as animage-based motion-tracking method used to measurerover motion during terrestrial tests. Section 5 describesthe vehicle test venue at the Jet Propulsion Laboratory (JPL)known as the Marsyard, as well as naturally occurring Marsanalog terrains used for testing in the Mojave Desert of Cal-ifornia. Section 6 presents slip versus slope data for variousterrains, as well as slip versus sinkage data in cohesionlesssand. Section 7 compares terrestrial test results for the MarsScience Laboratory test rover with results from identicaltesting done with the smaller Mars Exploration Rover testvehicle, showing the impact of wheel diameter in soft sand.The paper concludes in Sections 8 and 9 with a discussion

Journal of Field Robotics 30(6), 835–846 (2013) C© 2013 Wiley Periodicals, Inc.View this article online at wileyonlinelibrary.com • DOI: 10.1002/rob.21481

836 • Journal of Field Robotics—2013

Figure 1. The Mars weight Scarecrow rover during traversetesting at Dumont Dunes in the Mojave Desert of California.

on how these data were used in selecting a landing site forCuriosity, and slip versus slope data are presented from thefirst 250 sols of operations on Mars.

2. BACKGROUND

The Mars Science Laboratory (MSL) rover, named Curios-ity, is a mobile geochemistry lab that successfully landed atGale Crater near the equator of Mars on August 5th, 2012.The mission’s objective is to assess whether the planet everhad an environment that was able to support life. To dothis, the rover has a suite of seven science instruments, 17cameras, and a robotic arm for sample inspection and acqui-sition. The 899 kg vehicle measures 2.8 m from the outsideof the front wheels to the outside of the rear wheels, is 2.8 mwide, and 2.2 m from the ground to the top of the imagingmast. It uses a six-wheel-drive rocker bogie system similarto the Mars Exploration Rovers (MER) Spirit and Oppor-tunity launched to Mars in 2003, as well as the Sojournerrover launched to Mars in 1996. The rocker bogie systemis a passive suspension with a differential connecting theleft and right arms of the rocker. On the forward end of therocker arms are the left and right front wheels. The aft endof the rockers connects to a freely pivoting bogie with theforward end of the bogie connecting to the mid wheels andthe aft end connecting to the rear wheels. This suspensionsystem allows for nearly equal weight distribution acrossall six wheels independent of pose when articulated overobstacles or undulating terrain (Bickler, 1998; Harrington,2004). It also allows for articulation over obstacles in excessof one and a half wheel diameters (75 cm) in height.

The four corner wheels of the rover are steerable, butthe mid wheels are fixed. This dictates that with Ackermansteering, the rover always drives in a constant curvature arc(for a given set of wheel steering angles) pivoting about apoint along the line connecting the middle wheels. The cur-vature of the arc can range from zero, resulting in straight-

line motion, to infinite, where the rover pivots around itsorigin (otherwise known as a point turn). The top speed ofthe vehicle is 4.2 cm/s, so all motions are quasistatic. Addi-tionally, each drive actuator has sufficient torque such thatthe rim thrust is never the limiting factor for driving overobstacles or up slopes.

For the MER rovers, testing was done prior to the ve-hicles’ arrival on Mars to characterize traverse performanceusing a Mars weight vehicle on a tilt table (Lindemannet al., 2005). This provided valuable data on vehicle slipas a function of slope. These data are still in use by thosecommanding the Opportunity rover to predict the vehicleperformance and to plan the safest and most efficient routespossible. This vehicle slip testing is also complementary tothe computer modeling of traverse routes for predicted per-formance (Iagnemma et al., 2011; Trease et al., 2011).

Other terrestrial work has been performed characteriz-ing the slope climbing performance of potential planetaryvehicles (Sutoh et al., 2012). Experimental testing has beenperformed at the single wheel level to determine the influ-ence of several physical wheel parameters such as diameter,width, grouser size, and spacing (Ding et al., 2011). Simi-lar experiments have been performed to characterize soilflow under a wheel in granular materials (Moreland et al.,2012; Senatore et al., 2012). A single wheel test apparatushas also been used in an attempt to characterize the soilproperties of planetary bodies based upon wheel soil inter-action (Sullivan et al., 2011). The effect of gravity has beencharacterized at the single wheel level using aircraft duringvariable gravity maneuvers (Kobayashi et al., 2010).

Due to the size of the Curiosity rover, and a desire todo vehicle level performance characterization, laboratory-based experiments similar to those mentioned above werenot feasible. This necessitated the use of the outdoor testfacility at JPL known as the Marsyard. There was also adesire to test in naturally occurring terrain analogous towhat was expected on Mars. Testing in the wind-blownsand dunes and desiccated lake beds of the Mojave Desert inCalifornia ensured that the terrain properties were as closeas possible to those expected on Mars. This paper presentsthe methods and results of this testing.

3. TEST PLATFORMS

The Curiosity rover, shown in Figure 2, was assembled in aclass 10,000 cleanroom to ensure that the mechanisms andinstruments were free of debris, and to ensure that Earthcontaminants were not inadvertently taken to Mars. Insidethis cleanroom, the vehicle only drove a few tens of meterson flat ground and over gentle ramps to do the most basicof mechanism checkouts.

A second identical copy of the vehicle was built to fa-cilitate software and hardware testing. Initially, this vehicleonly had the rover chassis, wheels, and suspension system.At this point, the vehicle had a mass of 436.8 kg, which on

Journal of Field Robotics DOI 10.1002/rob

Heverly et al.: Traverse Performance Characterization - Mars Science Laboratory Rover • 837

Figure 2. Traverse testing of the flight rover in the assembly cleanroom prior to launch.

Earth gave it a weight of 128% of what the flight vehicleweighs on Mars. This was the lowest mass configurationpossible with the flight-identical components, and in thisconfiguration the vehicle was known as the Dynamic TestModel (DTM). The DTM rover was used to do a subset oftraverse testing to quantify performance while in this lowmass configuration. The rover was then populated with theremainder of the mechanical components, science instru-ments, cameras, and computing elements to make it a trulyidentical copy of the flight vehicle. The test rover in thisfinal configuration is known as the Vehicle System Test Bed(VSTB). Since Mars gravity is approximately 3/8 of Earthgravity, the VSTB weighs 260% of what Curiosity weighson Mars. This increased weight and ground pressure meansthat the vehicle is not a good analog for traverse character-ization and the VSTB is used primarily for software andautonomous navigation testing.

The Scarecrow rover is a 340 kg vehicle that weighs onEarth what Curiosity weighs on Mars. The center of gravity(CG) of the vehicle is within 5 cm of the CG location on theflight rover. It has identical wheels and suspension system,but it has a minimal chassis with commercial electronics.The vehicle has onboard batteries and is commanded viaa wireless network, so it does not require a tether. Teleme-try, similar to the flight vehicle, including motor currentand roll/pitch/yaw from an onboard inertial measurementunit (IMU), and suspension angles are all recorded onboardthe vehicle at 2 Hz. Additionally, the vehicle has ultrasonicrange finders mounted to the axle of each wheel to mea-sure wheel sinkage. Table I summarizes the vehicles of theMars Science Laboratory program as well as their mass andweight in their operating environments.

The Mars Exploration Rover project had a similar strat-egy for vehicle testbeds. A development version of the Spiritand Opportunity rovers known as the Surface System TestBed (SSTB)-Lite was often configured to weigh on Earthwhat the rovers weigh on Mars. The 66 kg version of the ve-hicle was used extensively for traverse performance charac-terization both during the early MER mission developmentand again in the years following the landing of the MERvehicles in 2004. It was used in 2012 to perform head-to-head vehicle comparisons in identical terrain quantifyingthe impact of the wheel diameter for vehicles with similarground pressure. Table II summarizes the vehicles of theMars Exploration Rover program as well as their mass andweight in their operating environments.

All of the flight Mars rovers to date have had rigidwheels. Wheels for the Sojourner Rover, Mars ExplorationRovers, and Mars Science Laboratory Rover are shown inFigure 3. The MSL wheels are 50 cm in diameter and 40 cmin width and are machined from a single piece of aluminum(Haggart et al., 2008). The six wheels have chevron-shapedgrousers with a 15 degree spacing. The grousers protrude7.5 mm radially from the 0.75-mm-thick wheel skin. In themajority of the terrains that the vehicle encounters, theground deforms to allow for a contact patch that distributesthe weight of the vehicle over a nominal area. Since thecontact area is highly terrain-dependent, a metric has beendeveloped at JPL to define the effective ground pressure(EGP), shown in Eq. (1). The total vehicle weight is dividedby the number of wheels to provide the average forceF of the wheel on the ground. An assumption of wheelsinkage gives a contact patch area equal to the width of thewheel, w, multiplied by the wheel radius r. In the case of



Table I. Mars Science Laboratory rover mass/weight comparison.

Vehicle Curiosity DTM Scarecrow

Mass 899 kg 437 kg 340 kgWeight on Earth – 4,287 N 3,329 NWeight on Mars 3,336 N – –

Table II. Mars Exploration Rover mass/weight comparison.

Vehicle MER SSTB-Lite

Mass 177 kg 66 kgWeight on Earth – 643 N + 10 N of unsupported tetherWeight on Mars 655 N –

noncylindrical wheels where the outer diameter of thewheel is convex, the largest radius, usually at the midplane,is used. When the wheel’s periphery has cleats, lugs, orgrouser bars, a determination is made as to the amountof their height added to the wheel diameter, based on theprojected area ratio of these tractive elements. This simplemodel gives an easy metric to compare a nominal groundpressure for vehicles with rigid wheels independent ofterrain,

EGPF

r · w(1)

Table III shows the effective ground pressure for theMSL and MER rovers. This table makes the simplifyingassumption of a cylindrical wheel, but it does not include theheight of the grousers in the wheel radius. Since the Earthtest rovers (Scarecrow and SSTB-Lite) have identical wheels

to the flight vehicles and ground pressure scales directlywith gravity, the ground pressures of the Earth-based testvehicles are the same as their siblings on Mars.

4. MEASUREMENT TECHNIQUES

Vehicle slip is the metric with which the rover operationsteam is most concerned. To characterize this, both the ide-alized motion that the vehicle was commanded to take, aswell as the actual motion achieved, must be known. The for-mer, for straight drives, is simply a matter of commandedwheel rotation multiplied by the wheel radius. The latter ismeasured with a ground truth system attached to the rover.

Initial testing used a differential GPS to track both roverposition and orientation with reference to a fixed local co-ordinate system. This GPS system consisted of three GPSantennas mounted to the rover as well as one stationary



Figure 3. Mars rover wheels. Left to right: Mars ExplorationRover, Sojourner, and Mars Science Laboratory.

Table III. Rover ground pressure comparison.

EffectiveWheel Wheel GroundRadius Width Pressure

Curiosity 25.0 cm with grousers 40 cm 5.75 kPa24.3 cm without grousers

MER 13.1 cm with grousers 16 cm 5.72 kPa12.4 cm without grousers

antenna to cancel out GPS noise and to provide an ori-gin for the local coordinate system. The primary antennamounted on the rover measured the rover position withrespect to the stationary base antenna. The other two anten-nas on the rover were mounted along the vehicle X and Yaxes to give vehicle roll, pitch, and yaw. This system pro-vided positional accuracy of the vehicle of approximately+/− 5 cm and +/− 0.5 degree at 1 Hz.

To increase accuracy, the test team switched to animage-based motion capture system. A series of infraredemitters and cameras is used to simultaneously track a se-ries of reflective elements mounted to the vehicle. Figure 4shows the tracking cameras mounted on tripods as well asthe reflective markers mounted to the rover. This systemgives both the position and orientation of the rover originwith a positional accuracy of +/− 1 mm and 0.1 degree ata rate of 100 Hz. This increased accuracy allows not onlyfor ground truth in traverse performance characterization,but also for the accuracy characterization of visual odom-etry, which is how the vehicle tracks its actual motion inoperations on Mars (Johnson et al., 2008).

5. TEST VENUES

A majority of the traverse testing is done at a facility knownas the Marsyard, shown in Figure 5, which is located within

Figure 4. A motion capture system used to track the actualmotion of the Scarecrow rover.

the Jet Propulsion Laboratory campus. This is an approxi-mately 2,500 m2 outdoor facility with various terrains thatare representative of what Curiosity is expected to en-counter on Mars. There are three main terrain types thatare constructed in various configurations: bedrock, cohesivesoil, and cohesionless sand. These materials where chosenbased upon the terrain encountered with the Spirit and Op-portunity rovers, as well as the terrain anticipated at thepotential Curiosity landing sites (Biesiadecki, 2005; Leger,2005).

Bedrock terrain in the Marsyard is reconstructed us-ing Arizona Rosa flagstones. These approximately 50 cm ×50 cm sandstone tiles are used to build tracks that are ap-proximately 5 m wide and 9 m long. One track is constructedon level ground, and four other tracks are constructed atconstant 5, 10, 15, and 20 degree slopes. These tracks areslightly wider than the 2.8-m-wide vehicle, and they haveenough length to allow for three vehicle lengths of drivingat constant slope.

The native material in the Marsyard is cohesive soilwith a mix of grain sizes in a 20 cm layer over a formerasphalt parking lot. Soil cohesion changes with weatherconditions since this is an outdoor facility. In addition tovariable soil moisture from weather, the soil is typicallycompacted from foot and vehicle traffic. The majority of thearea in the Marsyard is made of this material, and trackssimilar in size to the bedrock tracks are constructed withconstant slopes of 5, 10, 15, and 20 degrees as shown in anaerial view of the Marsyard in Figure 6.

Cohesionless sand was also used for testing in theMarsyard, as shown in Figure 7. Retaining walls were as-sembled in a “U” configuration reaching heights of 2.4 mallowing cohesionless sand to be configured in slopes rang-ing from flat to a 25 degree incline. Similar to the bedrockand cohesive soil, the sand track is 5 m wide × 9 m long, al-lowing for approximately three vehicle lengths of traverse ata constant slope. The cohesionless sand used is well sorted,



Figure 5. 5 degree, 10 degree, 15 degree, and 20 degree slopes of alternating bedrock and cohesive soil in the Marsyard.

Figure 6. Aerial view of the Marsyard.

Figure 7. The Dynamic Test Model (DTM) rover attempting toascend a 20 degree slope of cohesionless sand in the Marsyard.

with grain sizes ranging from 250 to 425 μm. The sand wascovered while not in use to minimize the moisture content,but some moisture was in fact detected in the subsurface.This led to some behavior at high sinkage, such as packingbetween the wheel grousers, which would not be evident intrue cohesionless material.

A soil compaction meter (cone penetrometer) was usedprior to each test in the cohesive soil and the cohesionlesssand to determine the bearing capacity (resistance to pene-tration) of the soil. The meter is used by fitting a 12.7-mm-diameter cone to an extension that is attached to a load cell.An ultrasonic range finder on the unit measures the depthof penetration. This unit records the bearing capacity every2.5 cm of penetration depth. Representative cone penetrom-eter data for the sand as well as the cohesive soil are shownin Figure 8.

In addition to onsite testing in the Marsyard, testingwas performed in the Mojave Desert of California. Mem-bers of the MSL project science team selected two areas as



Figure 8. Cone penetrometer data for cohesive soil and cohe-sionless sand in the Marsyard.

appropriate analogs for the terrains that might be encoun-tered at Gale Crater on Mars. Specifically, an area knownas Dumont Dunes was selected in the Mojave Desert as anAeolian dune field analogous to wind-blown sand dunesseen from orbit covering a portion of the rover’s landingtarget ellipse on Mars. Tecopa, California was chosen for itsdesiccated lakebeds having slopes and stair step topogra-phy similar to what is expected at the base of Mt. Sharp,a 5-km-tall mound of stratified rock in the center of GaleCrater. The Scarecrow rover, the SSTB-Lite rover, and themotion capture tracking system were taken to these sitesfor field-testing in May of 2012.

6. CHARACTERIZTION DATA

The prime metric of use for the rover operations team is slipvs slope for various terrains. Data are collected at 1 Hz forGPS-based data and 2 Hz for image-motion-capture-baseddata. A slip measurement is computed at each instance,comparing the commanded distance, dc, and the traversedistance, dt to those from 10 seconds prior. Commandeddistance is calculated in Eq. (2) and is based solely upon thecommanded wheel rotation angle θ , assuming rigid terrainand wheel radius r. The wheel radius used is the distancefrom the axis of wheel rotation to the tip of the grousers.Commanded distance is treated as an absolute value so thatit is always positive, independent of drive direction. Sinceall commanded motions in these data are straight drives,the rotation angles for all six wheels are the same. Data areonly included for the time when the rover is completely onthe constant slope. Traverse distance is measured using theGPS or image motion capture data and is calculated as theEuclidean distance, in Eq. (3), between the measured posi-tions at the respective times. Since the Euclidean distanceis used, the calculated traverse distance is nonzero during

embedding as the rover sinks into the terrain, even thoughthe rover is no longer making upslope progress. This is aknown limitation of this technique that results in slip neverreally reaching 100%, but the effect is small when vehicleprogress is being made. Vehicle slip is calculated in Eq. (4)and is expressed as a percentage of the commanded motion,

dc = |(θt2 − θt1)r| (2)

dt =√

(xt2 − xt1)2 + (yt2 − yt1)2 + (zt2 − zt1)2 (3)

slip % =(

1 − dt

dc

)· 100 (4)

For each trail run, the rover was commanded to driveforward (rocker first) up the steepest gradient of the slope atthe vehicle’s top speed (4.2 cm/s). Rover tilt, measured bythe GPS or motion capture system, is reported as the slope toshow local variations in the terrain during the test. At leastthree-vehicle lengths worth of motion were commanded foreach trial run up the approximately constant slope. Theseindividual upslope tests were then compiled for a terraintype and are presented in the subsequent sections.

6.1. Bedrock

Figure 9 shows vehicle slip vs slope for the DTM roverand the Scarecrow rover collected over three trial runs withthe DTM rover and four trial runs with Scarecrow. Slopesfor the trials range from flat to 30 degrees on simulatedbedrock in the Marsyard. A special test was conducted ata 30 degree slope to find the limit of the vehicle’s capa-bility on flat flagstone simulated bedrock. In this test, the

Figure 9. Slip vs slope data for bedrock terrain in theMarsyard.



vehicle showed a limit cycle behavior in which the roverwould climb until all six wheels were on smooth surfaces,at which point the rover would dynamically slide downthe slope. At this incline, the rover was not able to climbmore than a fraction of a meter up the slope after approx-imately 50 m of commanded motion. The low-slip outliersin the 30 degree slope case can be attributed to instancesin which the wheel grousers were able to grab the edges ofthe bedrock flagstones and make upslope progress. Afterslight progress, however, the grousers would lose contactwith the disparity in the rock and the vehicle would slidedownhill. The rover would limit cycle, climbing some smalldistance and then sliding backwards, without making anysignificant upslope progress. Additionally, there are someslip numbers that are shown to be negative. Traditionallythis represents skid, where the vehicle makes more progressthan commanded. For the upslope driving represented inthis data set, however, the negative slip numbers are likelydue to noise in the GPS measurements.

It is believed that smooth flagstones represent thebounding case for bedrock performance. Qualitative test-ing shows that as there is more texture in the bedrock, thewheel grousers are able to grab the terrain and the rover isable to climb significant slopes, up to 30 degrees. The dataset is sparse between 20 and 30 degrees, but it is expectedthat the location of the inflection point (where a small in-crease in slope results in a large increase in slip) is highlysurface-texture-dependent and will be between 20 and 35degrees depending upon the strength and exposed textureof the bedrock.

6.2. Cohesive Soil

Figure 10 shows vehicle slip vs slope data for the DTM roverand the Scarecrow rover collected in both the manmadeslopes in the Marsyard and at the desiccated lake bed terrainat the Tecopa field test site. Four trial runs with the DTMand two with Scarecrow were performed in the Marsyard,and four runs were performed with Scarecrow at the Tecopasite. Slopes for these tests ranged from flat to 28 degrees.

The results show that unlike bedrock, the transitionfrom low to high slip is more continuous. In the Marsyardon 20 degree slopes, the vehicles saw as much as 75% slip,but they continued to make steady uphill progress withoutembedding themselves into the terrain. At the Tecopa site,the Scarecrow rover performed extremely well, climbingslopes as high as 28 degrees with less than 50% slip. The soilhad low enough compaction that the grousers were able topenetrate, but it had high enough shear strength that it didnot fail as the vehicle ascended the slope. It should be notedthat cohesive soil has the most variability in physical prop-erties of all the terrains tested. Minor changes in moisturecontent, compaction, and cohesion can have a significantimpact on the vehicle’s ability to traverse up slopes.

Figure 10. Slip vs slope data for cohesive soil in the Marsyardand in the desiccated lake beds of Tecopa, California.

Figure 11. Slip vs slope data for cohesionless sand in theMarsyard.

6.3. Cohesionless Sand

Figure 11 shows vehicle slip vs slope for the DTM roverand the Scarecrow rover collected over five trial runs withthe DTM and three trial runs with Scarecrow. Slopes forthe test range from flat to 20 degrees on cohesionless sand.The data, collected on the artificially created sand slopes inthe Marsyard, show that the Scarecrow and DTM vehiclesare able to traverse up 10 degree slopes with marginal slip,but that both vehicles become embedded when on a slopeof approximately 17 degrees.



Figure 12. Slip vs slope data for the Scarecrow rover in cohe-sionless sand at Dumont Dunes, California.

Figure 12 shows similar slip vs slope data from eighttrial runs with the Scarecrow rover in the Aeolian sanddunes at Dumont Dunes in the California Mojave Desert.This data set is more densely populated due to the num-ber of tests conducted at the field site and the variety ofnaturally occurring slopes. The colors in the plot representthe different tests run at approximately constant slope, butall tests were conducted with the Scarecrow rover in thesame type of cohesionless sand. Similar to the data fromthe Marsyard, the Dumont Dunes data also show that theinflection point for the transition between upslope progressand embedding is at a slope between 10 and 17 degrees. Thevehicle showed less than 50% slip at slopes less than 12.5degrees, but quickly transitioned to slip of approximately75% at 16 degrees. At 20 degrees the vehicle experiencedslip of approximately 90% and was no longer able to makeupslope progress. Due to the density of the data points rang-ing from slopes of 4 to 21 degrees, a third-order polynomialcurve was fit to the data with coefficients of c0 = 26.197,c1 = −8.849, c2 = 1.281, and c3 = −0.034. This curve is validbetween the slopes of approximately 4 and 21 degrees. Atslopes below 4 degrees, we expect nearly constant slip ofapproximately 10%. At slopes above 21 degrees, we expectslip in excess of 90% and vehicle embedding.

The Scarecrow rover was fitted with ultrasonic rangefinders on each of the six wheels. These devices measurethe distance from the wheel strut to the undisturbed groundadjacent to the wheel with an accuracy of +/− 2 mm at afrequency of 2 Hz. This directly measures wheel sinkagefor the eight trail runs of slip vs slope data shown previ-ously. The average sinkage of all six wheels is plotted ver-sus slip in Figure 13. Wheel sinkage is measured to the topof the grousers such that there would be zero sinkage on

Figure 13. Slip vs sinkage for the Scarecrow rover in cohesion-less sand at Dumont Dunes, California.

rigid terrain. In sand, even on flat terrain, the wheels sinkapproximately 15 mm due to the grousers and deformationof the sand.

What these data show is that as slip increases, the sandis excavated from under the wheel, causing it to sink. Thissinkage creates more opposition to forward motion due tothe bulldozing resistance in front of the wheels (Wong, 2008,p. 187). This further increases slip, which results in the rapidtransition from moderate to high slip due to a small increasein slope in soft terrain. This is in contrast to cohesive soil,where sinkage is minimal. In these firmer terrains, forwardprogress can still be made even with high slip since theterrain deformation and consequently the bulldozing resis-tance remains small.

7. MARS EXPLORATION ROVERCHARACTERIZATION

During field testing in the Dumont Dunes, data were alsocollected with the Mars weight version of the Mars Explo-ration Rovers, known as the Surface System Test Bed (SSTB)-Lite, shown in Figure 14. Both the Scarecrow rover and theSSTB-Lite vehicles were tested on the same cohesionlesssand during the field test at Dumont Dunes in the Mo-jave Desert of California. Since the SSTB-Lite is driven withbrush motors and commanded with a simple DC powersupply, commanded distance was recorded as the numberof wheel rotations based upon a repeating unique patternin the wheels. Traverse distance was measured simply witha tape measure adjacent to the vehicle. While these meth-ods are not as accurate as those used for Scarecrow, theresults at high slip are still very pronounced. The Scare-crow and SSTB-Lite rovers have ground pressures of 5.75



Figure 14. Mars weight version of the Mars ExplorationRovers known as the Surface System Test Bed (SSTB)-Lite.

and 5.72 kPa, respectively, a difference of less than 1%. TheScarecrow rover, however, is able to traverse up a 15 degreeslope, while the SSTB-Lite vehicle becomes embedded in a12 degree slope, as shown in Figure 15.

Any differences in the terrain are removed as a side-by-side test was run between the two vehicles on the same12 degree slope. The Scarecrow vehicle successfully as-cended to the top while the SSTB-Lite became embedded.The key difference between the vehicles is wheel diameter,which is 50 cm for Scarecrow and 26.2 cm for SSTB-Lite.As vehicle slip and consequently wheel sinkage increase,the motion resistance is increased due to the undisturbedterrain in front of the wheels (Wong, 2008). For the same

Figure 15. Direct performance comparison of the Scarecrowrover and the SSTB Lite rover in the cohesionless sand in Du-mont Dunes, California.

sinkage, the high curvature of the smaller diameter wheelresults in higher motion resistance than the larger diame-ter wheel of the same ground pressure (Skonieczny et al.,2012). This extra motion resistance at a given sinkage can bethought of as a steeper slope in front of the smaller wheeldue to its higher curvature.

8. LANDING SITE SELECTION

One of the critical uses of the data described above wasin the selection of the landing site for the Curiosity rover.The final Gale Crater location was not selected until July of2011. From November of 2008 until then, four landing siteswere considered, each with a unique scientific appeal, andeach with unique terrain. Orbital data of these sites allowedfor terrain-type classification and also provided elevationdata at 1 m horizontal separation. Slope maps were derivedfrom these topographic data to quantify the average slopeover a 2 m × 2 m area. These 2 m2 cells were then tiledtogether to make a global discretized slope map. The entirelanding ellipse was evaluated for traversability based uponthe type of terrain and the slope. An algorithm was devel-oped to find potential paths from points distributed over theentire ellipse to the potential science targets. The traverseperformance characterization of the MSL rover was used asinput to this algorithm constraining the paths. This analy-sis showed that, from an engineering perspective, each ofthe potential sites was traversable and that driving perfor-mance was not a necessary discriminator in selecting a site(Golombek et al., 2012).

Once the MSL mission was launched, it was shownthat the in-flight performance of the attitude determinationsystem, a combination of star scanner and sun sensors, wasbetter than expected. This reduced the error with whichthe vehicle entered the Martian atmosphere, and in turn al-lowed the Entry Decent and Landing (EDL) team to reducethe probabilistic landing ellipse from 20 km wide × 25 kmlong to 7 km × 21 km. This allowed for the ellipse to beplaced closer to the base of Mt. Sharp where the ultimatescience objectives lie, minimizing the traverse distance forthe rover once it landed on the surface. The moving of thelanding site, however, placed an Aeolian sand dune field ina significant portion of the ellipse. A detailed analysis of thedune field was performed using orbital-based slope maps.Our characterization of the vehicle’s performance in sand,showing that slopes less than 12.5 degrees were traversable,revealed that the dune field did not contain any areas thatwould trap the rover. This ultimately allowed for the land-ing ellipse to be moved to the optimal location, whichcontained the dunes.

9. MARS DATA

On Mars, the rover has the ability to measure its slip us-ing visual odometry. Visual odometry is a technology thatprovides a full 6-degrees-of-freedom motion estimate given



Figure 16. Panorama taken by the Curiosity rover between sols 3 and 13 showing the hummocky terrain around the landing sitein Gale Crater on Mars.

two sets of stereo images of nearby terrain, taken beforeand after a small drive step (typically a 1 m drive or a turnin place of less than 20 degrees). This technology was pre-viously employed on the Mars Exploration Rover mission(Maimone et al., 2007), and the algorithm has been improvedfor MSL by employing multiresolution processing and moredata-driven search window sizes (Johnson et al., 2008). MERresults have shown visual odometry to be capable of mea-suring motions to a resolution of 2 mm (Maimone et al.,2007), given substantial overlap between images.

As of May 1, 2013, Curiosity has traversed a total of723 m, primarily in hummocky terrain at Gale Crater. Thisterrain is shown in a panorama taken by the rover at itslanding site in Figure 16. This is well-consolidated soil withgentle rolling topography most similar to the cohesive soilin our Earth-based testing. Slip results can be calculatedusing visual odometry for the measurement of actual rovermotion, dt, in Eq. (3). The slip measurements computed thusfar are shown in Figure 17 alongside the Scarecrow rovercohesive soil data set.

Figure 17. Slip vs slope for the DTM and Scarecrow rovers oncohesive soil in the Marsyard and Tecopa, California presentedwith data for the Curiosity rover in hummocky cohesive soilon Mars as of sol 250.

Slip vs slope data are not only used by the human op-erators on Earth, but also used by the autonomous softwareonboard the rover. As the rover drives, it tracks its positionon the surface based on commanded wheel odometry, as-suming zero slip. This position estimate can be refined byestimating the vehicle’s slip based upon measurements ofthe rover’s roll and pitch. A third-order polynomial modelof slip vs slope is used to estimate the downslope slip com-ponent of the rover’s open-loop motion. The model’s co-efficients are adjusted based upon terrain type and wereinitially set based upon the terrestrial testing described inthis paper. As more Mars data are gathered, the model willbe adjusted as needed.

10. CONCLUSIONS

The Mars weight Scarecrow rover, as well as other terrestrialrovers, were tested on various terrain types, and vehicle slipwas characterized as a function of slope along the rover’spitch axis. The sensitivity of this relationship to the com-paction, cohesion, and shear strength of the terrain is evi-dent. The rover is able to traverse up slopes of 28 degrees instrong cohesive soil with only 30% slip, but it experiences75% slip in cohesionless sand at 16 degrees. The sensitiv-ity to wheel diameter is also evident, showing that groundpressure is not the only worthwhile metric when designingrover wheels; geometry must also be considered. For futuremissions, improving wheel designs can lead to increasedaccess to previously unreachable science targets as well asan increased margin against a vehicle embedding in softterrains.

While only a limited variety of terrains and slopes havebeen encountered to date with the Curiosity rover on Mars,the initial data show good agreement with the terrestrialtesting. The a priori characterization of the vehicle’s traversecapabilities, based upon terrain type, will give the Curiosityoperations team the ability to predict traverse performanceand provide for safer operations as the vehicle continues toexplore the surface of Mars.

ACKNOWLEDGMENTS

The research described in this paper was carried outat the Jet Propulsion Laboratory, California Institute of



Technology, under a contract with the National Aeronau-tics and Space Administration. Copyright 2012. All rightsreserved. Special thanks go to Rob Sullivan of Cornell Uni-versity for his help with the design of the Marsyard terrain.Thanks to Scott Moreland of Carnegie Mellon University forthe installation of the wheel sinkage sensors on the Scare-crow rover. Thanks to Ray Arvidson and the MSL SurfaceMaterials and Mobility Working Group for their identifica-tion of Earth analog terrains suitable for testing and for geo-logical characterization of the terrains during the field tests.Thanks to Jack Dunkle for his logistical support at the fieldtest. Thanks to Brian Trease, Nathan Stein, and ChristinaKreisch for measuring rover pitch and wheel slippage forthe SSTB-Lite traverses on the Dumont Dunes.

REFERENCES

Bickler, D. (1998). Roving over Mars. Mechanical Engineering,74–77.

Biesiadecki, J., Baumgartner, E., Bonitz, R., Cooper, B., Hart-man, F., Leger, C., Maimone, M., Maxwell, S., Trebi-Ollennu, A., Tunstel, E., & Wright, J. (2005, October). MarsExploration Rover surface operations: Driving opportu-nity at Meridiani Planum. In Proceedings of the IEEE Con-ference on Systems, Man, and Cybernetics, Kona, HI.

Ding, L., Gao, H., Deng, Z., Nagatani, K., & Yoshida, K. (2011).Experimental study and analysis on driving wheels’ per-formance for planetary exploration rovers moving in de-formable soil. Journal of Terramechanics, 48, 27–45.

Golombek, M., Grant, J., Kipp, D., Vasavada, A., Kirk, R., Fer-gason, R., Bellutta, P., Calef, F., Larsen, K., Katayama, Y.,Huertas, A., Beyer, R., Chen, A., Parker, T., Pollard, B., Lee,S, Sun, Y., Hoover, Y., Sladek, H., Grotzinger, J., Welch, R.,Noe Dobrea E., Michalski J., & Watkins, M., (2012). Selec-tion of the Mars Science Laboratory landing site. SpaceScience Reviews, 170, 641–737.

Haggart, S., & Waydo, J. (2008). The Mobility System Wheeldesign for NASA’s Mars Science Laboratory mission. InProceedings of the 11th European Conference of the In-ternational Society for Terrain-Vehicle Systems, Torino,Italy.

Harrington, B. D., & Voorhees, C. J. (2004). The Challengesof designing the Rocker-Bogie suspension for the MarsExploration Rover. In Proceedings of the 37th AerospaceMechanism Symposium, Galveston, TX.

Iagnemma, K., Senatore, C., Trease, B., Arvidson, R., Shaw, A.,Zhou, F., et al. (2011). Terramechanics modeling of Marssurface exploration rovers for simulation and parameter

estimation. In Proceedings of the ASME Design Engineer-ing Technical Conference, Washington, DC.

Johnson, A. E., Goldberg, S. B., Cheng, Y., & Matthies, L. H.(2008). Robust and efficient stereo feature tracking for vi-sual odometry. In Proceedings of IEEE International Con-ference on Robotics and Automation, Pasadena, CA.

Kobayashi, T., Fujiwara, Y., Yamakawa, J., Yasufuku, N., &Omine, K. (2010). Mobility performance of a rigid wheelin low gravity environments. Journal of Terramechanics,47, Issue 4, 261–274.

Leger, C., Trebi-Ollennu, A., Wright, J., Maxwell, S., Bonitz, R.,Biesiadecki, J., Hartman, F., Cooper, B., Baumgartner, E.,& Maimone, M. (2005, October). Mars Exploration Roversurface operations: Driving spirit at Gusev Crater. In Pro-ceedings of the IEEE Conference on Systems, Man, andCybernetics, Kona, HI.

Lindemann, R. A., & Voorhees, C. J. (2005). Mars ExplorationRover mobility assembly design, test and performance. InProceedings of the IEE International Conference on Sys-tems, Man, and Cybernetics (Vol. 1, pp. 450–455).

Maimone, M., Cheng, Y., & Matthies, L. (2007). Two years ofvisual odometry on the Mars Exploration Rovers. Journalof Field Robotics, special issue on Space Robotics. 24, 169–186.

Moreland, S. J., Skonieczny, K., & Weggergreen, D. S. (2012).Motion analysis system for robot traction device evalua-tion and design. In Proceedings of the International Con-ference on Field and Service Robotics, Matsushima, Japan.

Senatore, C., Wulfmeier, M., Jayakumar, P., Maclennan, J., &Iagnemma, K. (2012). Investigation of stress and failure ingranular soils for lightweight robotic vehicle applications.In Proceedings of the Ground Vehicle Systems Engineeringand Technology Symposium, Dearborn, MI.

Sullivan, R., Anderson, R., Biesiadecki, J., Bond, T., & Stew-art, H. (2011). Cohesions, friction angles, and otherphysical properties of Martian regolith from Mars Explo-ration Rover wheel trenches and wheel scuffs. Journal ofGeophysical Research, 116.

Sutoh, M., Yusa, J., Ito, T., Nagatani, K., & Yoshida, K. (2012).Traveling performance evaluation of planetary rovers onloose soil. Journal of Field Robotics, Special Issue on SpaceRobotics, Part II, 29(4), 648–662.

Trease, B., Arvidson, R., Lindemann, R., Bennett, K., Zhou, F.,Iagnemma, K., et al. (2011). Dynamic modeling and soilmechanics for path planning of Mars Exploration Rovers.In Proceedings of the ASME Design Engineering TechnicalConference, Washington, DC.

Wong, J. Y. (2008). Theory of ground vehicles. Hoboken, NJ:John Wiley & Sons.


traverse performance characterization for the mars science laboratory rover

Documents