NASA JPL MarsYard and Planetary Robotics Laboratory

NASA/JPL-Caltech. Public domain (NASA / US government work).

The JPL Robotics section runs two connected terrain venues in Pasadena: the MarsYard, a 66 x 36 m outdoor field where full-scale rover twins drive, climb, sample and get stuck under Earth atmosphere and Earth gravity on seven slopes variable in both angle and cohesion [1], and the indoor Planetary Robotics Laboratory, 3000 sq ft of prototyping and test floor half of which holds two large sand pits [2]. The MarsYard is the third yard of that name, roughly six times the area of its predecessor.

The same flight programs use these venues in sequence with the JPL space simulator chambers, which hold Mars pressure, Mars temperature and simulated solar flux but cannot drive. Neither venue reproduces the other’s environment, and the gap between them is where several documented surprises have lived.

A rover test venue has to supply three things a flight vehicle will meet on another body: terrain with representative slope, cohesion and rock distribution, a ground truth independent of the rover’s own sensors, and a vehicle whose mass and mechanics stand in for the flight article. The outdoor yard is the one choice that gives tens of meters of continuous drive at the cost of Earth gravity, daylight and ambient weather. The flown programs converged on pairing it with a Mars-weight surrogate (Scarecrow) and a full avionics twin (the vehicle system testbed) rather than trying to make the yard itself more Mars-like [5][6]. The sections below cover the yard, the indoor laboratory, what each does not reproduce, and the research and flight-software work that has used the yard as its field stage.

Laboratory data

Parameter	Value
Operator	NASA JPL, Section 347, with flight projects [1]
Location	Pasadena, California, United States
Commissioned	Present outdoor yard is the third iteration, MarsYard III
Type	Outdoor analogue terrain field and indoor prototyping laboratory [1][2]
Floor area	Yard 66 x 36 m, over 2000 m2 of terrain; PRL 3000 sq ft
Capabilities	MarsYard III, PRL sand pits
Simulant or terrain	Beach sand, decomposed granite, brick dust, volcanic cinders; basalt rocks [1]
Instrumentation	Total stations and 3D tracking [4]; no permanent motion capture published [1]
Ground truth	Total station and 3D tracking position, independent of rover odometry [4]
Fidelity limits	Sand results not repeatable enough to report quantitatively [3]. See below
Access	Not published
Cited by	Curiosity [3][4], Perseverance [5]

Capabilities

MarsYard III

NASA/JPL-Caltech. Public domain (NASA / US government work).

Parameter	Value
Working volume	66 x 36 m site, over 2000 m2 of test terrain [1]
Test article limits	Articles to date up to the 900 kg-class rover testbeds [1][5]
Vacuum	Not applicable. Outdoors at ambient pressure
Temperature	Not controlled. Testing stops above about 32 C ambient [4]
Illumination	Not published. Terrestrial daylight, no solar simulator
Simulant or terrain	Beach sand, decomposed granite, brick dust, volcanic cinders [1]
Slope	Seven slopes, variable angle and cohesion; 15, 18 and 30 degrees cited [1][3][4]
Gravity offload	No offload rig; Scarecrow used instead, about 3/8 mass [5]
Instrumentation	Total stations and 3D tracking [4]

The yard was built with geologists and engineers to mimic terrains on Mars, the Moon and beyond. A robot garage houses the Curiosity and Perseverance vehicle system testbeds side by side, and a visitor platform overlooks the area [1]. The soil is not a compositional simulant: beach sand, decomposed granite, brick dust and volcanic cinders, chosen for mechanical behavior and color. Rocks are basalt in several textures and colors, fine grained and vesicular, in red and black, placed to a size distribution matched to Mars, with the explicit caveat that the large rocks are not Mars-like in composition, being less dense but easier to move for testing. Bricks and trenches are added for specialized work, and the yard is sculpted into a representative terrain per task rather than held in a fixed configuration. Named features in the published campaigns include a north slope with gravel and flagstone surfaces, a flat sand box, a sand slope and a slip-differential hill [3][4]. No tonnage, relative density target or preparation procedure is published, and the individual slope angles and cohesion settings are not published either.

Campaigns cite 15 degree, 18 degree and 30 degree slopes, the last being the maximum grade traversable by the MSL rover [3][4]. Mars weight is approximated not by an offload rig but by the Scarecrow test rover, whose lighter chassis puts its mass at approximately 3/8 that of the flight rover so that its Earth weight matches the flight rover’s Mars weight [5].

The temperature limit is operational rather than nominal. Curiosity was designed for Gale Crater at -94 to 0 C; the yard routinely exceeds 32 C in summer, past the point where the vehicle system testbed’s Earth-only cooling system can hold the vehicle in a safe operating range, at which point outdoor testing stops and the testbed moves into the garage [5]. The Mars 2020 testbed carries a gaseous nitrogen cooling system added specifically to permit operation through Pasadena summers [5].

Planetary Robotics Laboratory sand pits

Parameter	Value
Working volume	Two large sand pits within 1500 sq ft; pit dimensions not published
Temperature	Not controlled
Simulant or terrain	Loose rocky soil and steep cliffs, simulated indoors; products not published
Slope	Steep-terrain descent supported; angles not published

Source: [2].

The laboratory is described by its operator as a facility for rapid prototyping of advanced robotic systems and for developing the algorithms that control them, covering design, fabrication and testing of wheeled and limbed mobility and of manipulation. It holds 1500 sq ft of assembly, integration and test floor for mechanical and electronic subsystems, and a further 1500 sq ft containing the two sand pits, with workstations in both areas [2]. Two lander-mounted manipulators kinematically equivalent to the Mars Polar Lander four degree of freedom arm live there, as do 3 and 4 degree of freedom arms fitted to the rovers. A self-contained field trailer associated with the laboratory supports remote field operations with satellite communications back to the lab, used extensively for Mars Exploration Rover operations rehearsals [2].

This section is thin because the source is thin. The operator’s page is the only description found, and it gives no pit dimensions, no soil product, no depth, no instrumentation and no slope range.

Instrumentation

In the yard the measurement problem is position rather than temperature. Mobility commands used in regression testing have been verified against total stations and 3D tracking hardware to confirm they meet accuracy requirements, so subsequent campaigns can compare telemetered rover position against planned values as sufficient analysis [4]. Slip is computed as distance traversed against distance commanded over a 10 m drive step, the longest single straight drive step MSL permits [3]. No permanent motion capture array is published for either venue.

Site support for the yard is the garage and the visitor platform [1]; the indoor laboratory carries UNIX, Linux and Windows workstations for software development and testing in both of its halves [2].

What it does not reproduce

Repeatable sand. The clearest published limit. In the 2018 rimmed wheel campaign, quantitative sand results were excluded from the paper because of concerns about consistency, with, in the operators’ words, “significant variability in all test metrics between individual trials of all configurations” [3]. Earlier slip testing for MSL and for Mars 2020 had already shown that the soil preparation needed for consistent, repeatable sand trials is intensive and that performance and consistency depend strongly on temperature and moisture content; that preparation was out of scope for the campaign [3]. Moisture content is a Pasadena weather variable with no Martian counterpart.

Gravity. Neither venue offloads. The Scarecrow rover exists to close part of that gap: it carries none of the flight instruments or key avionics, but its lighter chassis brings its weight to approximately 3/8 of the flight rover’s, matching what the flight rover weighs on Mars, which is why certain terrain performance verification activities can be at least partially run on it [5]. The same document states that without sufficient fidelity in avionics, instrumentation and weight distribution, Scarecrow falls short as a venue for full verification and validation of mission requirements [5].

Sinkage has no flight comparison. Rimmed wheels increased Scarecrow’s sinkage by 1 to 2 cm against full wheels, but there is no flight metric to compare it against because Curiosity has no way to measure wheel sinkage [3].

Mars pressure and temperature. Neither the yard nor the sand pits hold either. That work goes to the space simulator chambers, which in turn cannot drive a rover.

The testbed is not the rover. Beyond cooling and heaters, the vehicle system testbed uses umbilical cabling in place of the radioisotope generator and the flight radio, does not carry all the flight platinum resistance thermometers, and adds fuses to protect the batteries from ground power supply surges [5]. The umbilical is itself an operational hazard: a test buddy is assigned to keep the testbed from dragging its power and telemetry cable across the yard and to press the panic button [4].

Campaigns run there

MSL rimmed wheel campaign, first two months of 2018. Scarecrow was run in baseline, two-rimmed and three-rimmed wheel configurations to establish whether Curiosity could continue operating after deliberately shedding the inner two thirds of a damaged wheel [3]. Straight 10 m drives up and down 18 degree gravel and flagstone slopes on the north slope showed the only statistically resolvable slip increase was 2.95 percent for the three-rim configuration driving uphill on stone, against a mobility fault protection threshold that does not respond below that level [3]. Average drive motor current differences across all six motors in all cases were under 0.1 A and unbiased. Rimmed wheels caused a consistent drift off course, more pronounced on gravel, with all three-rim gravel trials ending outside the positive one-sigma bound of the baseline. Arm activities were simulated on 18 and 30 degree slopes at four cardinal headings by moving a 40 kg weight 1.5 m forward of the center of mass by hand, applying a 300 N upward force with a fish scale to represent drill preload, and running a 1.1 g, 60 Hz vibration motor on the deck for 30 seconds to represent sample ingestion shake [3].

MSL R13 flight software regression, July and August 2022. Mobility regression for a software update six years in the making, run on the vehicle system testbed [4]. Test shifts ran 8 to 9 hours and averaged 62.41 m of driving, over twice Curiosity’s typical sol [4]. The campaign comprised 94 test cases after redlines: 10 redlined for redundancy, 2 redlined as impossible in the testbed configuration, and 3 retested. The traction control algorithm was exercised on a 15 degree hill with half the wheels on loose sand and half on firm gravel, where slip without traction control rose from 8.2 percent at the first drive step to 38.4 percent by the fourth, and the rover yawed noticeably off target; with traction control enabled slip was reduced and heading held [4]. Test procedure was transcribed by script from the procedure document into activity reports and into Rover Markup Language command files, which removed commanding transcription error across four primary test conductors [4].

Mars 2020 vehicle system testbed operations, 2020 onward. The Perseverance twin is used to train the operations team on anomaly handling before landing, to dry-run first-time activities with flight-like kinematics, and to shadow the rover during milestone events such as the first drive and the first coring activity [5]. The Ingenuity helicopter watchdog timer fix, which had interrupted the transition from preflight to flight mode during the first rotor spin, was tested on the testbed before being sequenced on Mars. Sampling subsystem work on the testbed established that the rover could apply the 50 N preload needed to drill while on a slope, which had not been validated before because drilling operations had previously been confined to a fixed-tilt hexapod [5].

Earlier laboratory programs. The FIDO rover for Mars technology development and mission rehearsals, the SRR rovers for sample return, robot work crews and steep-terrain descent, and the LEMUR limbed robots for walking and manipulation were all developed in the Planetary Robotics Laboratory [2].

Use beyond the flight twins

Mobility characterization. Before landing, the MSL project ran Mars-weight slip-versus-slope testing with the Scarecrow rover, a 340 kg vehicle that weighs on Earth what Curiosity weighs on Mars [6]. On smooth flagstone bedrock the maximum traversable slope was 22 degrees; on artificial cohesionless sand slopes slip was moderate at 10 degrees and the rover embedded at 17 degrees. The soil is native JPL parking-lot soil and flagstone, not a simulant, and its properties varied with weather without tight control, so the values are specific to those materials [6]. An earlier yard of 20 x 20 m, designed as an analog to the Viking lander sites, hosted slip compensation work on the Rocky 8 research rover, which held visual odometry error under 1.5 percent of distance traveled on a 29 m field traverse with a total station as ground truth [7].

Autonomy and perception. The yard is the standard field stage for navigation research before flight. An active mast planner improved visual odometry position error by 20.8 to 82.7 percent over a fixed mast on eight of ten Mars Yard runs, with the fixed mast failing on two more; the authors warn the gains are environment dependent [8]. A closed-form clearance checker for articulated rovers was exercised on Scarecrow over a 42 day qualitative campaign in which the yard could not be held constant and each run took 20 to 30 minutes at 0.04 m/s [9]. Perseverance’s flown autonomous navigation was verified against Monte Carlo simulation and Mars Yard scenarios scored pass or fail, not as calibrated traction coefficients [12]. Earlier, the FIDO rover in the yard supplied navigation imagery for an onboard dust devil detector in a multi-asset science-alert demonstration [15], and a 7-sol walkabout run by MSL scientists evaluated the Pathogen onboard planner [16]. The PUFFER foldable rover program ran autonomous mapping trials in the yard [17], and CADRE multi-agent autonomy work used monocular imagery collected there [19]. A digital twin study describes the yard as built to physically simulate Mars terrain [18], and a mission-autonomy review describes it as a 66 x 36 m outdoor site [20].

Lunar-analog configurations. For the CADRE three-rover lunar demonstration, the yard was reconfigured with rocks over 0.05 m every 1 to 2 m and over 10 craters up to 1 m; 15 experiments totaling about 25 hours, indoors and in the yard, explored 2724 m2, with night tests lit by a flood light rather than low-sun lunar illumination [10]. A half-scale Endurance rover concept drove 4.1 km in 5.2 hours (0.8 km per hour average) on a 200-waypoint route on 2 October 2025 under the MEXEC autonomy framework, with the battery swapped by hand instead of recharged [11]. CADRE flew to the Moon later; Endurance remains a mission concept [10][11].

Software test venues. Mars 2020 flight software also runs in the Surface System Development Environment on commercial avionics, built because the flight-like testbeds ran three shifts a day, seven days a week and were still short of verification demand. The Mars Yard stays the highest-fidelity mobility venue in that hierarchy [14][5].

Comparable venues

The ESA Automation and Planetary Robotics Lab at ESTEC is an indoor bed of 8 x 8 m [13] of sand, gravel and rocks with a small crater, boulder field, dune and gravel slope, instrumented with infrared motion tracking cameras. No simulant, grain size, bearing strength or slope angle is published [13]. It trades the yard’s 66 m drive length [1] for indoor repeatability and a motion tracking array the JPL yard does not publish. Neither is a reduced-gravity or vacuum venue; those are covered under space simulator chambers.

What is not established

No grain size distribution, bulk density, cohesion, friction angle, or preparation procedure is published for the yard soil, for the individual slopes, or for the laboratory sand pits [1][2]. Operator figures conflict: 66 x 36 m is 2376 m2 against “over 2000 m2” [1]. Soil moisture and temperature changed results in sand testing, and one campaign excluded its quantitative sand data [3]. Surrogate results (Scarecrow, the testbed) transfer as differences between configurations, not as absolute Curiosity or Perseverance values [3][6]. The Mars rock size-frequency distribution the yard matches is not cited by the operator [1]. This page cites 20 works against a family minimum of 30: the corpus does not hold more on the yard, because the held, curated sources are research papers that mention it as a field site and the yard’s own characterization is not among them.