DLR Institute of Robotics and Mechatronics Planetary Exploration Laboratory

DLR (CC BY-NC-ND 3.0) [2].

A rover wheel-soil model needs two things no orbiter or lander telemetry can supply: a controlled patch of ground whose mechanical state is known, and a way to compare a simulated trajectory against a measured one on that same ground. Earth gravity, ambient air and a soil that has never seen the Moon or Mars all stand between a lab bin and the body a rover will drive on. The founding critique of the field’s governing equations argues that a pressure-sinkage or shear-displacement curve measured on one plate, in one soil, at one scale, is a property of that test setup and not a transferable soil constant [15]. Facilities answer that problem in one of three ways: they measure the ground they have rather than assume a textbook value, they hold gravity, vacuum or temperature fixed and accept the loss, or they move outdoors and substitute terrain variety for environmental fidelity. The Planetary Exploration Lab at DLR Oberpfaffenhofen, run by the Institute of Robotics and Mechatronics, took the first path indoors and, since 2024, the third outdoors [1][2].

The lab is an indoor soil bin of 55 m2 including a 3 m by 5.5 m variable slope partition, a purpose-built portable bevameter that characterizes the soil in the bin, and a 1500 m2 outdoor Moon-Mars test site beside it [1][2]. The original bin was specified from what a multibody wheel-soil simulation needs as input and output: large enough to avoid side wall and bottom effects for an ExoMars breadboard-sized rover, a surface elevation measurement to build the digital elevation model the contact model consumes, a pose tracking system to compare simulated against measured rover position, and a soil measurement device to produce a Bekker parameter map of the bin rather than one averaged value [1]. That last choice answers Wiendieck’s objection: the bevameter measures the bin’s own pressure-sinkage and shear response plate by plate, at the scale the wheel sees [15].

All of these facilities are ground test at 1 g. No hardware from them has flown, and the wheel models and rover results validated in them are laboratory or field-analog evidence, not flight data [1][2][16].

Two other DLR facilities are separate pages. TROLL, the single wheel rig, is described on the same institute’s site as a force-controlled industrial robot with automated soil preparation, a 3 m traverse, wheel loads to 1 kN and speeds above 1 m/s [3]. Where the Planetary Exploration Lab validates a whole-vehicle multibody model against a tracked rover, TROLL isolates one wheel and folds mechanical loosening and soil fluidization into the same unattended sequence as the traction run [3][4]. It grew out of a 2015 feasibility study of whether an industrial robot arm could carry the force control a single-wheel rig needs [5]. A companion study on the same rig found that soil preparation, not measurement, limits how fast a testbed can turn over soil states, and that semi-automated compaction beats a human board walk on repeatability even though a skilled operator can still reach a higher peak density by hand [7]. The Landing and Mobility Test Facility in Bremen is a robot-on-rail facility that DLR used for lander footpad drop and drag tests in sand [17].

Laboratory data

Parameter	Value
Operator	DLR Institute of Robotics and Mechatronics [2]
Location	Muenchner Strasse 20, Wessling, Oberpfaffenhofen, Germany [1]
Commissioned	Indoor bin described 2009; outdoor test site opened 2024 [2]
Type	Indoor rover soil bin with pose tracking, plus an outdoor analog terrain
Floor area	Indoor 55 m2; outdoor 1500 m2 [2]
Capabilities	Soil bin, Moon-Mars site, bevameter
Simulant or terrain	Dry quartz sand indoors; basalt, lava and granite outdoors [2]
Instrumentation	Eight-camera infrared optical tracking, better than 3 mm, 60 fps [1]
Ground truth	Tracked pose against simulation; bevameter map of the bin
Fidelity limits	1 g, Earth ambient, no vacuum, no thermal, no solar simulator [2]
Access	Not published. Institute-run
Cited by	rosalind-franklin [1]

Design space

Terramechanics testbeds trade three things against each other: how faithfully they reproduce gravity and vacuum, how large or varied a terrain they hold, and how directly a measured run ties back to a soil-contact model. The families below differ in which they give up.

Indoor soil bins with model validation. The DLR bin belongs here: a bounded, instrumented bed, a tracked vehicle and a contact model to check [1]. Its closest analog is the Canadian RCAST effort, which coupled a commercial soft-soil tire model to a multibody chassis simulation and checked it against single-wheel tests on Viking-derived lunar soil types; the model reproduced sinkage well but overpredicted drawbar pull once slip passed about 0.5, where the measured curve levels off [9]. ESA’s ExoMars-era trade study took a different route: EADS Astrium built three full-scale breadboards and tested them in three facilities in the UK, Canada and Switzerland, using a flexible-wheel model developed at DLR-Bremen to optimize the wheel [8].

Single-wheel rigs. TROLL measures one wheel with the soil preparation automated [3][4]. On it, over 500 forced-slip runs at 100 N wheel load showed drawbar pull and tractive efficiency falling sharply above about 0.2 m/s: for a rigid ExoMars-style grousered wheel in olivine sand, drawbar pull dropped 8 percent from 0.01 to 0.2 m/s and 63.7 percent from 0.2 to 1 m/s at 20 percent slip, and a flexible wheel gave over twice the drawbar pull at 1 m/s [16]. A Tohoku University testbed reached the same direction of result with Tohoku Silica Sand No. 5 at speeds to 1.0 m/s, while its companion DEM simulation predicted traction rising with speed, the opposite of the measurement [27]. Both are 1 g results; the 100 N load at DLR is argued, not demonstrated, to correspond to a 350 kg rover under lunar gravity [16]. At NASA Glenn, imaging of the shear interface under a grousered wheel in GRC-1 attributed the drawbar gain from grousers to reduced compaction resistance and not to added thrust, a theory resting on assumptions about grouser packing [30].

Vehicle-scale rovers as testbeds. Rather than a fixed bin built around one model, ESA’s Automation and Robotics Section iterated whole-rover platforms, ExoTeR and MaRTA, across five years of campaigns [10]. A MaRTA field campaign in the Bardenas Reales semi-desert produced about 1.7 km of teleoperated traverses with force-torque sensing at all six wheel legs, on a dried river bed rather than regolith [24]. NASA’s Lunar Rover Optimization Platform is a 20.4 kg four-wheeled rig for sweeping wheel geometry, load and center of gravity, scored by observed pass or fail rather than measured drawbar pull [32]. Scarab was characterized on GRC-1 and seven soil types, with the authors concluding that wheel design dominates traction and that Earth-gravity results still need a validated model to extrapolate [29].

Outdoor analog yards. The DLR Moon-Mars site belongs with JPL’s MarsYard III, 66 m by 36 m with seven variable angle and cohesion slopes on beach sand, decomposed granite, brick dust and volcanic cinders [19]; the Canadian Space Agency yard at 7200 m2, CNES SEROM at 4000 m2 and the Altec yard with a tilting platform, which DLR’s own survey compares [2]; and two indoor yards in Luxembourg built for vision-based navigation, which deliberately used graded basalt sand instead of simulant to buy a larger area at some cost in fidelity [22]. The yard family trades soil fidelity for terrain variety. The JPL page gives no grain size distribution, density or strength for its soil, so its terramechanic fidelity is not established [19]. Rover field testing at Earth analog sites is older: the August 2002 MER-FIDO test near Flagstaff ran a prototype rover through a 20-sol baseline under blind remote commanding [23].

Reduced gravity. None of the DLR indoor or outdoor sites can reduce gravity, and the evidence that this matters is direct. Parabolic-flight tests on a single-wheel testbed found that reduced-weight testing on Earth overstates mobility, with drawbar pull to weight falling 20 percent or more and sinkage rising up to 40 percent in lunar gravity at equal wheel load; preparing GRC-1 to equal cone index gradient also overstated drawbar pull, by 44 percent mean squared percentage error; granular scaling laws came within 8.5 percent on drawbar pull [20]. A NASA study of gravity offset testing reached the same direction: slip on a slope measured at reduced weight on Earth understates slip at true lunar gravity [14].

Soil and simulant characterization. DLR measures the bin soil with the bevameter [1]. Other laboratories characterize the material itself: NASA Glenn engineered GRC-1 from four commercial silica sands to match the cone-index gradient Apollo astronauts measured, with a full geotechnical characterization including bevameter parameters [11]. ESA’s LUNA facility spans 1000 m2 with a 750 m2 mare area holding about 20 tons of EAC-1, whose abrasivity and density were characterized on 5 kg samples, not on the beds in use [18]. Standardized ASTM testing of six lunar simulants showed that matching the lunar average particle size distribution within one standard deviation still leaves substantial differences in shear strength and compressibility [21]. The Exolith Lab simulants carry modal mineralogy, oxides and particle size distributions but no geotechnical properties [28]. NASA Kennedy’s GMRO laboratory lists a 120 ton BP-1 bin and a full soil mechanics apparatus, as a capability list without conditions [26].

Contact modeling. A JPL study showed a meshfree particle formulation reproducing the classical Bekker pressure-sinkage curve for a rigid wheel where a Lagrangian finite-element mesh penetrates the wheel surface once soil flows [12]. DLR’s own work compared homogeneous and heterogeneous multi-tier Bekker-based Soil Contact Models against a six-wheeled rover multibody model [6]. NASA’s survey of contact models ran five wheel-soil models on identical hardware and found a span of roughly five orders of magnitude in speed, from a Bekker-based solver at 0.024 times real time to a discrete-element model at 9600, with only the slowest models reproducing the rise in drawbar pull as slip approaches 1 [13]. A data-driven alternative fits slip and sinkage regressions to field data from a 21.63 kg rover on Tohoku Silica No.5 sand and runs them inside a 30 Hz photorealistic simulator; its reported errors compare the simulation to the fitted data, not to independent validation [25].

Outside the field. Automotive vehicle dynamics research reached a similar design question with the Illinois Roadway Simulator, a scaled treadmill-based hardware-in-the-loop testbed, which trades full-scale fidelity for repeatable, instrumented runs [33].

The DLR lab does not simulate reduced gravity and does not claim to. It concentrates on making the one thing it can measure, the bin’s own soil state, traceable enough that the multibody model built from it is explicit about the ground it was validated against.

Capabilities

Indoor soil bin

Parameter	Value
Working volume	55 m2 of floor; original bin 10 x 5.5 x 0.5 m deep [1][2]
Vacuum	Not applicable. Ambient pressure
Temperature	Not applicable. No thermal capability
Illumination	Not published. No solar simulator described
Simulant or terrain	Dry quartz sand; tonnage and density control not published [1]
Slope	3 x 5.5 m variable slope partition; angle range not published [2]
Gravity offload	Not applicable. None in this bin
Instrumentation	Eight-camera infrared tracking, better than 3 mm, 60 fps [2]

The bin was specified from what a multibody simulation needs as input and output: large enough to avoid side wall and bottom effects for an ExoMars breadboard-sized rover, divisible into two parts so that two different soft soils can be held at once, with a surface elevation measurement to build the digital elevation model the contact model consumes, a pose tracking system to compare simulated against measured rover position at matched samples, and a soil measurement device to produce a Bekker parameter map of the bin [1]. The frame is built from standard aluminum tooling profile so the bin can be extended or moved.

The 55 m2 figure and the 3 m by 5.5 m variable slope partition come from DLR’s own 2024 survey of test sites, which is the most recent published description of the lab [2]. The original 2009 specification gives the bin as 10 m long by 5.5 m wide and 0.5 m deep [1]. No open document gives the partition’s slope range, its setting mechanism, or whether it moves under load. No article mass limit is published; the sizing argument is geometric rather than structural.

Bevameter characterization in this bin used dry quartz sand, Martian Soil Simulant D and milled Eifel lava, fitting all three plate sizes to the pressure-sinkage curve at once rather than by Bekker’s original two-plate averaging; the quartz fit was good and the two simulant fits only acceptable [1]. Neither a tonnage, a relative density target nor a preparation procedure between runs is published. The soil state is instead measured rather than controlled, by mapping Bekker parameters over the bin with the bevameter. DLR’s laboratory and infrastructure index carries images of the outdoor test field and of TROLL, but none of this hall [1].

Moon-Mars test site

Parameter	Value
Working volume	1500 m2, 4 m of elevation difference, about two thirds Moon
Test article limits	Crater traversable by wheeled and legged robots to 150 kg
Simulant or terrain	Sand, breccia and gravel of basalt, lava and granite
Slope	Crater slopes 15 to 35 degrees; 15 degrees loose, 35 degrees solid
Instrumentation	Optical tracking, 2 mm in four volumes, cm site-wide

Source: [2].

The outdoor site opened in 2024 and complements the indoor labs and the Scout rover mobility test bed [2]. It packs a crater, a hill, a tunnel, a canyon, bumps, cracks and rough walls into 1500 m2, with boulders of basalt, lava, granite, suevite and conglomerate, 230 V and 400 V three phase power, network access and Wifi, and a dedicated fiber link to the German Space Operations Center for telerobotic experiments from the International Space Station.

The terrain was derived from use cases rather than from a scenic brief. The crater requirements are stated as eleven numbered items: 10 m wide and at least 2 m deep, a bottom at least 2 m wide of loose material, traversable by wheeled and legged robots up to 150 kg, placed next to a hill giving 4 m of height difference top to bottom, with a 15 degree loose slope and a 35 degree solid slope with distributed rocks, mostly of volcanic material, draining rain easily, built to reduce weeding, and visible from a visitors area [2]. The 35 degree slope is drainage concrete, chosen to simulate a rocky descent and to stop the slope sliding off. The tunnel is 4 m long at 1.5 by 1.5 m with 1 m and 0.5 m branches, the smallest exiting directly into the crater [2].

Ground truth is an optical tracking system giving up to 2 mm accuracy in four dedicated tracking volumes and centimeter accuracy across the whole site, used for control as well as measurement [2].

Portable bevameter

Parameter	Value
Working volume	Tripod, used inside the bin; plate radii 0.025, 0.05, 0.075 m
Instrumentation	Electromechanical actuators, 0.05 mm displacement precision
Ground truth	Bekker parameter map of the bin itself, not of a sample

The bevameter was designed without reference to existing bevameter layouts, arriving at a tripod with electromechanical actuators and sensors in a real-time computing environment, delivering 500 N normal force and 20 Nm shear torque at 400 Hz [1]. Portability is the design decision that matters: it characterizes the soil actually in the bin rather than a sample removed from it, which is what makes a Bekker parameter map across the bin possible. Three plate radii, 0.025, 0.05 and 0.075 m, are used, and displacement precision is 0.05 mm [1]. NASA Glenn’s more recent DRAGON M rig automates the same measure-in-place idea a step further, loosening a bin through 96 auger holes and then running a cone penetrometer and bevameter across it, checked for repeatability across nine bin locations with no significant edge effects in the top 100 mm [13]; DLR’s bevameter solves the measurement half of that problem without automating the loosening half.

Instrumentation

The three measurement systems in the indoor lab were each specified against a simulation need. Pose comes from eight-camera infrared optical tracking at better than 3 mm position and 1 degree orientation accuracy at 60 frames per second, which is what allows a tracked rover trajectory to be compared against a simulated one at matched time samples [1]. Terrain geometry comes from a five-camera digital elevation mapping beam on a linear rail, giving about 1.5 mm in-plane resolution and 2 to 4 mm altitude error, which produces the digital elevation model the contact model consumes. Soil state comes from the bevameter.

Outdoors the equivalent function is the optical tracking system, at up to 2 mm in four dedicated volumes and centimeter accuracy over the whole 1500 m2, which serves both as ground truth and as the control input for multiple robots at once [2].

Instrumentation

Pose comes from eight-camera infrared optical tracking at better than 3 mm position and 1 degree orientation at 60 frames per second, which lets a tracked rover trajectory be compared with a simulated one at matched time samples [1]. Terrain geometry comes from a five-camera digital elevation mapping beam on a linear rail, giving about 1.5 mm in-plane resolution and 2 to 4 mm altitude error, which feeds the contact model. Soil state comes from the bevameter. These are stated design figures; no validation run is reported [1].

Outdoors the equivalent function is the optical tracking system, up to 2 mm in four dedicated volumes and centimeter accuracy over the 1500 m2, serving as ground truth and as control input for multiple robots [2].

What it does not reproduce

Gravity. Neither the indoor bin nor the outdoor site reduces gravity, and neither carries an offload rig. A rover’s slip response measured at Earth weight, even on the right soil, does not extrapolate cleanly to lunar or Martian gravity [14][20].

Vacuum, temperature and illumination. None is provided indoors, and the outdoor site is under natural weather and lighting [1][2].

A mineralogical simulant. The bevameter work used dry quartz sand, Martian Soil Simulant D and milled Eifel lava [1]. The outdoor substrates are terrestrial basalt, lava and granite gravels and sands chosen for mechanical and visual relevance, and fine sand and dust were deliberately excluded [2]. Neither is characterized the way GRC-1 was against Apollo cone-index data [11].

A controlled soil state. No preparation procedure, relative density target or tonnage is published for the indoor bin. The lab measures the soil state with the bevameter rather than resetting it [1]. The fits themselves are not clean: the identified cohesive modulus was negative for all three soils and the deformation exponent reached 2.53 for MSS-D, which the authors attribute to a two-layer bed and choose not to model, so the parameters are fitting constants for those beds [1].

Campaigns run there

ExoMars locomotion model validation, 2009. The testbed and bevameter were built to close the loop between hardware and the three-dimensional multibody wheel-soil model DLR contributed to ExoMars: bevameter-derived Bekker parameters become simulation inputs, and tracked rover pose becomes the comparison data [1]. See rosalind-franklin. The wheel design built on the flexible-wheel model EADS Astrium and DLR-Bremen had optimized in the locomotion trade study [8].

Moon-Mars test site inauguration, 2024. In 2024 three experiments with the Lightweight Rover Units LRU1 and LRU2 and the Scout rover demonstrated the site: autonomous navigation and mapping, autonomous manipulation and sampling, and advanced mobility tests [2]. The demonstrations are qualitative, with no success rates or measured localization error [2]. The site’s fiber link to the German Space Operations Center is meant to support telerobotic experiments of the kind run in ESA’s METERON Analog-1, in which an astronaut on the ISS drove a rover and arm in 2019 [31]. The link is described as enabled, not exercised [2].

What is not established

No open document gives the indoor bin’s slope range, its setting mechanism or its behavior under load, and no tonnage, relative density target or preparation procedure fixes the soil state between runs [1][2]. Access terms are not published. The outdoor tracking accuracy is quoted as a planned capability, not a measured one [2]. Neither site reduces gravity or provides vacuum, thermal control or solar simulation, and the outdoor substrates are not characterized against a lunar or martian reference simulant [1][2][11]. How the bin’s measured Bekker parameters compare against the newer contact models, which differ by orders of magnitude in speed and in high-slip behavior, is not addressed in either primary source [13]. No published comparison exists against the automated single-wheel cycle at TROLL [3][4] or against the ExoTeR and MaRTA campaign histories [10], so the relative throughput and repeatability of these facilities is a gap in the open literature.