Carnegie Mellon University Field Robotics Center

Carnegie Mellon University Robotics Institute.

The Field Robotics Center at the Carnegie Mellon Robotics Institute runs a set of small, purpose-built planetary rigs rather than one large chamber: a glass-walled single wheel soil imaging testbed, a 0.76 m thermal vacuum chamber, a raised terrain bed lit by cinema HMI heads for perception work, an indoor sandbox for prototype mobility, and outdoor slag terrain used for full-vehicle mobility characterization [5].

The rigs were built by and for the programs that used them: Scarab in 2007 [5], the single wheel imaging work from 2011 [4], and the MoonRanger microrover from 2020 onward [12]. Two of them have no name of their own, CMU writing only “CMU’s TVAC system” and “MoonRanger’s perception testbed”, which is why they appear below under descriptive headings.

Laboratory data

Parameter	Value
Operator	Carnegie Mellon University, Field Robotics Center, Robotics Institute [3][5]
Location	Pittsburgh, Pennsylvania, United States
Commissioned	Rig by rig: 2007, Scarab [5]; 2011, single wheel imaging [4]; 2020, MoonRanger [12]
Type	Bench-scale planetary rigs: soil imaging, thermal vacuum, perception lighting, outdoor terrain
Floor area	Not published
Capabilities	Wheel rig, TVAC, Perception, Outdoors
Simulant or terrain	GRC-1 [3]; lunar-like perception bed material [6]; outdoor mixed grain slag [5]
Instrumentation	ATI Delta six-axis force and torque sensor; vertical-axis optical encoder; DSLR imaging [3][4]
Ground truth	Optical flow on imaged grains; torsional shear tester after every bed preparation
Fidelity limits	No gravity offload anywhere; half-width wheel against glass [3][5]. See below
Access	Not published. No external user route, lead time or fee schedule was located
Cited by	scarab, moonranger, cuberover, iris

Capabilities

Licensed photographs exist only for the perception bed, both above and in its own section below. No image of the wheel imaging rig, the thermal vacuum chamber or the outdoor terrain was found.

Wheel imaging rig

Parameter	Value
Working volume	Soil cross-section 31 cm wide by 22 cm high at the sidewall [3]. Bin length not published
Test article limits	Half-width wheels, 23 to 50 cm diameter, dead weight payload on a free vertical axis
Illumination	External halogen floods, normal to the glass, for imaging only
Simulant or terrain	GRC-1, loosened then drop-tamped to a repeatable state before every run
Gravity offload	Not applicable
Instrumentation	ATI Delta six-axis force and torque; vertical-axis encoder; DSLR at 8 fps [3][4]

A half-width wheel is run against tempered glass at a commanded slip ratio while a traveling camera photographs the soil through the sidewall. Optical flow on the imaged grains recovers the sub-surface displacement field and the shear interface beneath the wheel, which is what vehicle telemetry cannot show [3][4].

The rig is a glass-walled soil bin, a wheel module on a belt-driven linear travel axis and a camera that translates with the carriage. The wheel module is position or velocity controlled against the carriage so that a commanded constant slip is held while the wheel advances, and a linear rail leaves the wheel free vertically so that sinkage develops naturally and the dead weight payload is transmitted. Slip curves were generated from 5 to 60 percent in 5 percent increments, with the wheel tangential rim speed held at an equivalent no-slip ground speed of 2 cm/s and the carriage speed varied to set the ratio [3][4]. The reported configuration used a 23 cm diameter by 5.72 cm wide rim under a 10 kg payload, giving 22 kPa average ground pressure measured from the contact patch at 20 percent slip with no grousers [3][4]. Soil strength is verified after each preparation with a torsional shear tester modified for constant normal pressure control [3][4].

TVAC chamber

Parameter	Value
Working volume	30 in diameter by 30 in deep, about 0.76 m [1]
Vacuum	5 x 10^-4 torr or better
Temperature	Article to about 160 C radiative; cold plate at -190 C [1][2]
Illumination	Internal xenon lamp, 40 W to the article, used as a heat source [1]

A stainless steel chamber bought as a bare body and completed in house with pumps, feedthroughs and data acquisition. Roughing is an Alcatel 2004A rotary pump of 3.2 cfm reaching 5 x 10^-2 torr in about 20 minutes, followed by an Alcatel 5081 turbopump of 80 l/s reaching 5 x 10^-4 torr in about 10 minutes more, with the turbopump mounted directly into the chamber wall to shorten the molecular flow path [1]. Sealing is primarily ConFlat all-metal knife-edge joints crushing annealed copper gaskets, and electrical and fluid penetrations use ConFlat flanges with wires sealed by brazed alumina or melted glass, which both seals and electrically isolates them.

Cooling is an aluminum and copper cold plate fed with liquid nitrogen at -190 C under PID-modulated solenoid control behind a shroud; heating is the internal xenon lamp and resistive spot heaters emulating electronics dissipation, and all three can be imposed concurrently [1][2]. Test configuration is varied by how the article is mounted: for wheel module thermal isolation work the outer rim is placed in direct conductive contact with the cold plate to stand in for contact with cold regolith, and for radiative studies the article is suspended above the plate so that conduction is negligible.

Perception bed

Parameter	Value
Working volume	Raised terrain table. Dimensions not published
Illumination	ARRI Arrimax 18/12 HMI head, 18 kW electrical, near grazing incidence
Simulant or terrain	Lunar-like material and color, with a line of rectangular blocks at surveyed locations
Instrumentation	Surveyed blocks as independent terrain geometry

Source: [6].

A raised terrain table of lunar-like material with embedded rocks and craters, lit by a cinema HMI head positioned at bed height and set near grazing incidence because sun elevation at the lunar pole is a few degrees. Blackout curtains on the walls and floor prevent diffuse reflection back onto the terrain, so shadows stay close to black [6]. The stated target is the 1350 W/m2 and full infrared content reaching the lunar surface, against 1120 W/m2 at the Earth surface.

Outdoor terrain

Parameter	Value
Working volume	Indoor sandbox, dimensions not published [8]. Outdoor slag terrain, extent not published [5]
Slope	Outdoor slag slopes of 20 to 25 degrees, a property of the site rather than a setting [5]
Simulant or terrain	Loose mixed grain size slag; a twist course of two sawtooth ramps; a railroad tie trench

Full vehicle mobility work is done outdoors on slag slopes and craters, on a twist course of two sawtooth ramps set half a period out of phase so that the two sides of the suspension rock in opposition, and over a trench of two railroad ties [5]. Prototype driving for egress risk work is done in the indoor sandbox [8].

Instrumentation

Force measurement in the single wheel rig is by a six axis ATI Delta force and torque sensor, which resolves net thrust in the travel direction as well as actuator torques [3]. Sinkage comes from an encoder on the free vertical axis rather than from the drive train, so it is independent of wheel slip. The imaging chain uses a digital SLR with a 50 mm macro lens at 8 frames per second, sufficient for travel speeds of order 1.5 inches per second, framing a 12 inch wide by 8 inch deep patch of soil for wheels of 23 cm diameter [4]. Processing clusters each image into soil and non-soil regions, computes flow magnitude per pixel within the soil, and uses adaptive clustering to separate moving soil from static soil, giving the shear interface, the flow direction field, and the boundary between forward and rear flow.

Structural characterization uses roving hammer modal testing, with accelerometers reading frequency and magnitude while the article is tapped, in the mounting configuration it will fly in [9].

What it does not reproduce

Gravity is the largest omission. Scarab drawbar pull was measured at full weight, giving 2,000 N in mixed grain sized sand, approximately 0.7 times vehicle weight, and 2,700 N on concrete pavement, approximately equal to vehicle weight, with the authors recording that similar tests under a lunar gravity offload system remained to be run [5]. The same document notes that the loads induced during 1 m coring in simulant were relatively moderate and that a gravity offload system would be needed to make the drilling loading conditions flight-like.

The shear imaging technique substitutes a symmetry-plane boundary condition for the real mid-width condition of a full wheel. The authors state the requirement explicitly: the zero shear stress plane of symmetry is equivalent to a half-width wheel against the glass only if the soil particle to glass interface is sufficiently low friction, ideally zero [3]. Nothing in the published description measures the residual wall friction.

The vacuum specification is a thermal criterion. The stated reason for 5 x 10^-4 torr is that it reduces convective heat transfer to a negligible level, being about 1.5 million times less air than at atmospheric pressure [1]. Nothing in the description addresses outgassing of a dusty article, dust adhesion in vacuum, or holding pressure with regolith in the chamber.

Illumination fidelity is bounded by what a cinema lamp can do. The operators describe lunar light as more intense, less diffuse and lower in the sky than terrestrial sunlight, and note that the Earth atmosphere both reduces the constant from 1350 W/m2 to 1120 W/m2 and scatters the beam so that shadows are neither sharp-edged nor truly black [6]. The consequence they identify for camera design is that lunar terrain in mixed lighting is nearly binary: sunlit regions saturate at exposures long enough to see into shadow, and shadowed regions lit only by the rover at about 1 W/m2 read as almost entirely black [7].

Campaigns run there

Scarab mobility characterization, 2007. Laboratory drawbar pull tests established 2,000 N in mixed grain sized sand and 2,700 N on concrete for the rubber skid loader tires, and outdoor testing on 20 to 25 degree slag slopes measured the effect of body leaning on cross-slope traction: holding a straight path over 11 m, slip was 15 percent downhill without leaning and 7 percent with the body leaned into the slope, attributed to weight transfer onto the uphill wheels and to the wheels cutting terraces rather than skiing across the slope face [5]. Rock obstacles of 40 to 50 cm were climbed and a 50 cm trench between two railroad ties was bridged. A single continuous 1 km traverse was run, chosen to represent the distance between drill sites [5]. NORCAT’s coring drill breadboard was integrated on the vehicle and drilled to its full 1 m depth in lunar soil simulant [5]. See scarab.

Grouser geometry study, 2011 to 2012. The single wheel imaging rig was used to photograph the shear interface below rigid wheels with varying grouser height and spacing in GRC-1, at 20 percent slip for all combinations and over full slip curves for a subset [3][4]. The result was a design relation between grouser spacing and the extent of the shear zone, obtained from optical flow on the imaged grains rather than inferred from drawbar pull alone [3][4].

Push-pull locomotion, 2012. Carnegie Mellon contributed the soil imaging rig, filled with GRC-1 prepared by loosening, levelling and compacting, to a joint study with NASA Glenn on inching locomotion; the vehicle-scale drawbar pull comparisons in the same study were run in the Glenn SLOPE soil bin of approximately 8 m by 3 m by 0.3 m, also filled with GRC-1 [10]. The division of labor is the pattern the CMU rigs follow: sub-surface soil mechanics at Carnegie Mellon, vehicle-scale traction at a larger NASA bin.

MoonRanger thermal vacuum campaign, 2021. Components and assemblies were tested to verify the conductance of thermal interfaces, with the stated purpose of letting tested performance take precedence over prior analysis and simulation, and of catching interface problems before full system test [2]. The wheel module was run in the chamber with its rim on the cold plate and subsequently passed launch vibration testing [2]. See moonranger.

MoonRanger perception and lighting tests, 2021. The terrain bed with surveyed blocks was used to check that the terrain geometry the rover reconstructs matches the scene, under 18 kW HMI illumination at near-grazing incidence with blackout curtains [6]. A parallel exposure study established that images taken while the camera swept through a turn at 50 ms exposure were blurred beyond use for stereo, while the same exposure from a stationary camera was sharp, which bore directly on whether the rover could image while in continuous motion rather than stopping for every frame [7].

MoonRanger egress risk tests. Prototype driving in the CMU sandbox, combined with Monte Carlo rock distribution modeling, set the rock protrusion height thresholds used to classify drop-deploy egress risk: success at 6 cm, negotiable at 8 cm, high risk above 10 cm despite single 10 cm rocks having been driven over, and high risk of belly contact and internal damage at 14 cm [8].

MoonRanger modal survey. Roving hammer tests on the inverted chassis, bolted in its lander attachment configuration, returned response frequencies of 107, 119 and 170 Hz against a requirement that the lowest frequency exceed 100 Hz [9]. Mode shapes agreed with Ansys predictions while the frequencies differed from earlier vibration test results. A wheel module tapped separately gave a first mode of 127 Hz for the older module against 135 Hz from vibration testing, and 142 Hz for the 60 mm wide wheel module [9].

CubeRover. Carnegie Mellon co-developed the initial CubeRover prototypes, and mobility verification moved to NASA KSC Granular Mechanics and Regolith Operations laboratory. That campaign ran more than 150 mobility tests with 11 wheel sets, covering drawbar pull, slope and point turn performance, with some wheel sets climbing 30 degree slopes [11]. See cuberover and gmro-lab.

References