ESA ESTEC Automation and Robotics Laboratories

ESA / G. Porter. ESA, CC BY-SA 3.0 IGO.

The Automation and Robotics Laboratories are the robotics laboratories of the European Space Research and Technology Centre at Noordwijk, run by the Automation and Robotics Section of ESA’s Directorate of Technology, Engineering and Quality. Four laboratories sit under that heading: the Orbital Robotics Laboratory, the Planetary Robotics Laboratory, the Human Robot Interaction Laboratory and the Regolith Interaction Facility. The group holds ISO 9001:2008 certification [1].

What a robotics campaign gets here is a 9 by 9 m indoor terrain bed for locomotion, navigation and arm placement; a 45 m2 air bearing flat floor that relieves two translational and one rotational degree of freedom for rendezvous, docking and capture work; and a small dusty thermal vacuum chamber in which a wheel, a drill or a joint meets regolith at vacuum and at lunar temperatures [1][2][3].

ESTEC houses two distinct organizations. The robotics laboratories belong to the Directorate of Technology, Engineering and Quality and are entered through ESA’s Third Party Activities route [1]. The Test Centre is a separate 6000 square meter cleanroom complex operated under contract by European Test Services on ESA’s behalf, holding Europe’s largest vacuum chamber, its loudest acoustic facility and its most powerful hydraulic shaker [14]. Its Large Space Simulator has its own page, a different operator and a different access route.

Laboratory data

Parameter	Value
Operator	ESA Automation and Robotics Section, Directorate of Technology, Engineering and Quality, ESTEC [1]
Location	Keplerlaan 1, 2201 AZ Noordwijk, the Netherlands [1][3]
Commissioned	Terrain bed in service by 2013 [2]; ORBIT floor built and characterized 2015 to 2016 [3]
Type	Indoor robotics laboratories: terrain bed, air bearing flat floor, dusty thermal vacuum chamber [1]
Floor area	Terrain bed 81 m2; ORBIT floor 45 m2 [1][2][3]
Capabilities	Mars Yard, PUT, ORBIT, DTVAC
Simulant or terrain	Sand, gravel and rock in the yard; simulant bed in DTVAC. Tonnages not published [1][2]
Instrumentation	Vicon on ORBIT [3]; infrared motion tracking in the yard [2]; RIEGL and Nikon metrology [1]
Ground truth	Motion capture and laser scanning, independent of the article under test
Fidelity limits	No vacuum, thermal or offload capability in the yard; DTVAC bed 0.5 x 0.5 x 0.2 m [1]
Access	ESA Third Party Activities; ORBIT also through the ESA Academy Experiments Programme [1][11]
Cited by	rosalind-franklin

The laboratories maintain their own rover testbeds rather than relying on visiting hardware. HDPR is a six-wheeled 6x6x4 platform on two rocker-bogie passive suspensions, rated to IP4 and 3.6 km/h, for outdoor high speed work; ExoTeR is a half-scale ExoMars reproduction with a five degree of freedom arm and a locomotion and navigation subsystem mimicking the flight rover, limited to below 10 cm/s; MaRTA is the half-scale successor, which adds six-wheel steering [1]. No mass or footprint limit is published for the terrain bed, and the rover fleet is in practice what defines the article envelope.

Shared metrology across the laboratories is listed below [1].

Instrument	Capability
COMAU six degree of freedom arm	130 kg payload; flies entry, descent and landing or rendezvous trajectories
RIEGL LMS-Z210i laser scanner	300 m range; 1 mm resolution at short range
Nikon K600 optical calibration	0.1 mm position accuracy in a 4 x 2 x 2 m volume
Trimble differential GNSS base station	1 to 3 cm at 1 Hz
SenseFly eBee mapping drone	2 cm per pixel, 1 hour endurance

All Orbital Robotics Laboratory facilities can be driven through a common API that synchronizes data between systems, and are remotely operable through a dedicated interface [1][11].

Access is open to ESA projects and to research institutions and companies of ESA member states through the Third Party Activities management system, which covers testing requests, laboratory access, training and consultancy. ORBIT is additionally opened to Bachelor, Master and PhD students through the ESA Academy Experiments Programme. No lead time and no fee schedule is published for any of the laboratories [1][11].

Capabilities

Mars Yard

The Mars Analogue Terrain, called the Mars Yard as a nickname rather than as an official name, is the terrain bed of the Planetary Robotics Laboratory: sand, gravel and rock of several sizes laid out as a small crater, a boulder field, a sandy dune and a gravel slope area, in which small to mid-size rovers demonstrate locomotion and navigation [1][2].

Parameter	Value
Working volume	9 x 9 m bed; 8 x 8 m in ESA’s older description [1][2]. Bed depth not published
Temperature	Not controlled, ambient [1]
Illumination	Not published. No solar simulator described
Simulant or terrain	Sand, gravel and rock. No tonnage or grain size distribution published [1][2]
Slope	A gravel slope area in the bed. No adjustable rig, no angle range published [2]
Instrumentation	Infrared motion tracking cameras [2]; RIEGL LMS-Z210i terrain scans [1]

ESA describes the bed as a sandbox filled with sand, gravel and rock of different sizes. Nothing is published about tonnage, grain size distribution, relative density target or how the bed is prepared between runs, so two campaigns in it are not comparable on soil state. The material is graded sand and rock rather than a mineralogically matched simulant, which is the deliberate consequence of what the bed is for [1].

Planetary Utilization Testbed is ESA’s older name for the same bed, carried on a legacy page that gives it as 8 by 8 m against the 9 by 9 m of the current description, with a flagstone slope section and infrared motion tracking for position ground truth [1][2]. It is the bed in which ExoTeR ran sample tube detection and autonomous navigation work.

ORBIT

ORBIT, the Orbital Robotics Bench for Integrated Technology, is a 45 m2 epoxy floor leveled to gravity to within 0.7 mm across its whole surface, on which platforms ride on air bearings [3]. Two translational and one rotational degree of freedom are relieved, which is the whole trade the operators state: planar microgravity is bought “at the cost of reducing the system to three DoF”. It is the free-floating half of the Orbital Robotics and GNC Laboratory, whose other half is the 33 m GRALS rail arm [4].

ESA. ESA, CC BY-SA 3.0 IGO.

Parameter	Value
Working volume	45 m2 epoxy floor: 5 x 9 m in the operators’ papers, 9 x 4.8 m on ESA’s page [1][3]
Test article limits	250 kg, the laboratory crane limit; floor loading 500 kg/m2 [3][4]
Slope	None; the floor is leveled to gravity, maximum slope below 0.3 mm/m [1][3]
Gravity offload	Air bearings, 3 DoF; flatness 0.7 mm, residual acceleration 1.5 mm/s2
Instrumentation	Vicon Bonita B10 motion capture, sub-millimeter, to 250 Hz [3]

The floor was built to a flatness specification rather than laid and then measured. Ground concrete was covered by rows of leveled concrete screed set to gravity with a Leica AT402 absolute laser tracker, the gaps were filled, and low-viscosity epoxy was applied in layers with sanding between them, with laser tracker verification at every step [3]. The result is 0.7 mm of maximum deviation over 45 m2, with no general trend to tilt in any direction, and a surface whose roughness, RA below 0.9 um, RQ below 0.6 um and RZ below 1.8 um, is one to two orders of magnitude below the air gap of a few tens of microns the bearings ride on. Because the floor is not on the ground floor of the building, building movement is a standing concern, and the flatness was re-measured six months after commissioning specifically to check for it; no apparent change was found. ESA restates the flatness as 0.7 plus or minus 0.1 mm with respect to gravity, a maximum height variation of 0.67 mm, a maximum slope below 0.3 mm/m and a residual natural acceleration of 1.5 mm/s2 [1].

Access to the floor is by a fold-away ramp from the laboratory’s raised computer floor. Three rows of 45 mm aluminum profile run along the walls at 55, 140 and 225 cm above the floor, each to plus or minus 1 cm, for mounting targets and mock-ups [3].

The platforms are the variable part. MANTIS, the Maneuverable Testbed for In-orbit Simulation, rides on three carbon air bearing pucks at 25.82 plus or minus 0.05 kg dry with a moment of inertia of 1.667 plus or minus 0.09 kg m2 and about 50 kg of payload capacity [3]. ACROBAT is the large platform at 128.85 plus or minus 0.05 kg and 8.154 plus or minus 0.09 kg m2, which with the SATSIM propulsion and RECAP stack reaches 221.67 kg, 12.223 kg m2, 102.5 cm tall and 35 cm in radius, and carries 4 kg of 300 bar breathing air [3][8]. ROOTLESS is the actively leveled platform, about 15 kg of payload, rebuilt on a Neobotix MPO-500 mobile base by 2018 [3][4]. ROMA carries a UR10e arm on a 35 kg platform with a 12.5 kg payload [1]. Platform masses across the set run 35 to 250 kg with payloads to 200 kg by default and more “with minimal engineering effort”. Platform actuation is eight 10 N air thrusters, eight air propellers to 6 N and a 4 kg reaction wheel, with about 30 minutes of autonomy on integrated tanks, plus an ejection mechanism giving representative delta-v and tumbling conditions [3].

ROOTLESS inverts the facility’s own design premise. Instead of demanding a perfectly level floor, it levels itself on three supports using a high-accuracy inclinometer and three precision linear stepper motors, and can then deliberately tilt its payload plate to introduce a controlled lateral force that emulates a chosen low gravity level [4][5][7]. The operators note that the approach “allows to have less stringent requirements on the floor flatness which makes testing cheaper”, which is a direct statement that extreme floor flatness is expensive to buy and to keep.

By 2018 the laboratory had been rebuilt and extended: the compressed air installation was changed for a new robotic gantry, 300 bar refilling was moved inside the flat floor area, the 33 m GRALS rail arm was added, and ROOTLESS was rebuilt on its mobile base [4]. GRALS carries satellite and terrain models and the illumination for vision-based navigation work, and is used together with the floor: GRALS flies the approach while ORBIT supplies the free-floating target and the contact dynamics [4][5][7].

Dusty Thermal Vacuum Chamber

The Dusty Thermal Vacuum Chamber is the principal instrument of the Regolith Interaction Facility, a small thermal vacuum chamber with a simulant bed inside it, held between -150 and +125 C. Its published parameter list contains one number that most vacuum facilities do not publish at all: the base pressure with dust in the chamber, an order of magnitude worse than the base pressure clean [1].

Parameter	Value
Working volume	Shroud 0.85 m diameter by 1 m high; regolith bed 0.5 x 0.5 x 0.2 m
Vacuum	10^-5 mbar clean, about 7.5 x 10^-6 torr; 10^-4 to 10^-5 mbar with regolith present
Temperature	-150 to +125 C, liquid nitrogen; shroud and base plate controlled separately
Simulant or terrain	Bed fillable with different simulants; product, tonnage and density not published

Source: [1].

The chamber is small by design and its geometry states what it is for. The shroud and the base plate sit on separately controlled liquid nitrogen circuits, so the soil and the surrounding radiative environment can be set to different temperatures. That separation is the feature: a lunar mechanism’s problem is usually a cold tool working into cold soil while some other part of it is warm, and a single-zone chamber cannot pose that [1].

The facility exists because regolith properties are difficult to model, and, in ESA’s words, “any equipment designed to operate in contact with regolith, requires early testing to confirm that the assumptions taken in the design phases hold against the difficulty to model regolith properties” [1]. ESA lists six application areas, and they define the article envelope more usefully than a mass limit would: wheel and soil interaction for mobility systems; regolith sampling, drilling, excavation and conveying for scientific sampling and in-situ resource utilization support; regolith beneficiation by size separation or mineral enrichment; dust and temperature effects on motors, drive mechanisms, cameras, lidars and force-torque sensors; sealing technologies for robotic joints; and deployment and pointing mechanism development for the lunar surface. Pumped volume, pump type, pumpdown time and cleanup burden between campaigns are not published.

Human Robot Interaction Laboratory

The fourth laboratory of the group shares its equipment with ORBIT: a human-rated floating chair carrying a VR headset, mounted on the same standard platform base as the attitude actuators and the manipulator stack [1]. No room dimensions or instrumentation are published.

Instrumentation

Position ground truth in the Mars Yard comes from infrared motion tracking cameras [2]. On the ORBIT floor the equivalent is a Vicon system: the 2016 facility paper’s abstract says fourteen cameras and its body text says twelve Vicon Bonita B10 cameras at up to 250 Hz, giving sub- millimeter real-time localization from at least four reflective markers per tracked body, while ESA’s current page states sub-millimeter accuracy without a camera count [1][3]. The value of tracking both a manipulator and the object it touches on one system is that the contact is measured independently of either body’s own state estimate. Terrain ground truth for rover experiments is generated by the RIEGL LMS-Z210i scanner, and calibration of sensor frames by the Nikon K600, at 0.1 mm position accuracy over a 4 by 2 by 2 m volume [1].

Mass properties on ORBIT are measured rather than modeled. A center of gravity rig places the platform on a 900 mm honeycomb plate over three load cells, good to 150 kg with 10 g weight accuracy and sub-millimeter center of gravity resolution; a moment of inertia rig suspends the platform on springs and measures the oscillation with a gyroscope, at 1 to 2 percent accuracy [3]. Both matter because a floating platform’s dynamics are set by properties that change every time a payload is bolted on, which is why the published platform figures carry their measurement uncertainties [1][3].

The air plant supports two distinct pressures. A Bauer Oceanus E dive compressor fills the platforms’ onboard bottles to 200 or 300 bar at 140 l/min; a separate Atlas Copco SF11 oil-free scroll compressor supplies 10 bar at 900 l/min across four outlets, two pressure-regulated to 8 bar and two flow-monitored, for tethered operation, alongside a nitrogen outlet [3].

In the Dusty Thermal Vacuum Chamber nothing is published. ESA’s facility panel lists the shroud dimensions, the bed dimensions, the two vacuum levels and the temperature range, and gives no instrument. A campaign there brings its own measurement, or the article under test is itself the instrument, which is consistent with the application list where cameras, lidars and force-torque sensors appear as objects of study rather than as facility equipment [1][3].

What it does not reproduce

Vacuum, thermal or reduced gravity in the terrain bed. The Mars Yard is an ambient-pressure, ambient-temperature, one-g bed whose published parameters cover terrain geometry, rover testbeds and metrology and nothing else [1]. Its material is sand, gravel and rock rather than a mineralogically matched simulant [2], so it supports locomotion, perception, navigation and arm placement work and not tribology, dust adhesion or electrostatics.

Vehicle scale, where dust and vacuum are available. The Dusty Thermal Vacuum Chamber’s shroud is 0.85 m in diameter by 1 m high and its bed is 0.5 by 0.5 by 0.2 m [1], which admits one wheel, one drill string or one end effector. Nothing traverses it.

Its own clean base pressure, once regolith is inside. The published figures are 10^-5 mbar clean and 10^-4 to 10^-5 mbar dusty [1]. Publishing both is unusual and useful: it states, as a facility parameter, that the environment a dusty test actually runs in is up to an order of magnitude coarser than the specification a clean test meets.

A documented soil state in the chamber. The bed “can be filled with different simulants” [1] and nothing is published about which, how much, how deep, how prepared or to what density. Two campaigns in that chamber are not comparable on the published record. There is also no published cleanup burden, simulant handling procedure or contamination policy. Illumination is absent as well, so the thermal environment is radiative from the shroud rather than solar, and the extreme surface gradients of a lunar terminator or a permanently shadowed region rim are not produced by a beam. Soil behavior in the bed is at 1 g overburden [1].

Three of the six degrees of freedom on ORBIT. The operators state the trade in the first paragraph of the facility paper: planar free floating is achieved “at the cost of reducing the system to three DoF” [3], and ESA’s own description repeats the qualification, that the facility is an analogue to microgravity “albeit constrained to a plane, simulating microgravity in two translational and one rotational dimension” [11].

Free drift for longer than a few seconds. This is the facility’s most useful published number. After a defined contact, the platform’s velocity vector deviates 10 percent from nominal after 4.9 s at a 30 mm/s approach, 6.9 s at 50 mm/s and 11.4 s at 100 mm/s, which the operators attribute to “local imperfections of the flatness of the floor on the sub-millimeter level”, noting that “while at higher velocities this effect is barely visible, it becomes more significant in lower velocity regimes or for longer durations” [3]. Slow approaches, which are the operationally realistic ones for rendezvous and capture, are the ones the floor corrupts fastest. Users concur that “the small slopes that exist on the floor are enough to introduce significant disturbances into the system” [8].

Closed-loop tracking to better than tens of centimeters. A university group flying an overactuated platform with eight thrusters and a reaction wheel reports 100 percent success in simulated trajectory finding and following but real-floor straight-line tracking only within tens of centimeters [9], and attributes the gap to floor unevenness: “given the large weight of the system and the comparably low force capabilities by the thrusters, high actuation is required to compensate even small disturbances induced by the unevenness of the flat-floor” [8].

One agreed number for the floor’s flatness. Three figures circulate. The operators’ laser tracker measurement is 0.7 mm maximum deviation over the whole surface [3]; ESA’s page gives a maximum slope below 0.3 mm/m and a residual acceleration of 1.5 mm/s2 [1]; and the simulation heightmap a user built from the operators’ data is characterized as “a maximal deviation of 1 mm over one meter” [9]. The first is a global bound and the others are local gradients, so they are not the same quantity and should not be substituted for each other.

Unlimited test duration on the floor. A run ends when the platform’s cold gas does, with about 30 minutes of autonomy on integrated tanks and less when thrusters are working hard [1][9].

Anything above 250 kg on the floor. Not because the floor cannot carry it, at 500 kg/m2, but because the crane that puts a platform on the floor lifts 250 kg [3][4].

Campaigns run there

ExoMars autonomous navigation, 2018 to 2019. ExoTeR, the half-scale ExoMars testing rover, was driven in the 9 by 9 m terrain bed running CNES-developed autonomous navigation software, with the yard laid out in red sand, gravel and rock and instrumented for position ground truth [1]. The laboratory maintained ExoTeR as its primary testbed for about a decade before MaRTA was built to replace it [1]. See rosalind-franklin.

MaRTA force-torque sensor assessment, 2023 to 2024. MaRTA, carrying six force-torque sensors mounted above the wheels alongside an IMU, was driven over varying terrain, slopes and speeds in a field campaign at Bardenas Reales, Spain, producing the BASEPROD dataset [12]. The finding was negative and specific: drawbar pull could not be read from the above-wheel force-torque sensors in a meaningful way, the vibration superimposed on the signals made them harder to interpret than expected, and the authors recommend cheaper conventional sensors, or mounting in the drive hub and correlating with motor current, over the above-wheel configuration. They also record that confirming the candidate drawbar pull extraction needs experiments with external ground truth on defined slopes with known ground parameters, which a field campaign cannot supply [12]. For terrain classification the force-torque sensors did outperform the IMU [12].

BASEPROD dataset, July 2023. The underlying dataset covers 24 traverses of 6.85 m to 202 m, about 1.7 km and roughly 450 GB of logs, recorded between 20 and 23 July 2023 [13]. MaRTA carried an Xsens MTi-680 IMU with RTK GNSS receiver, a Realsense D435i tilted down 20 degrees, a Bumblebee XB3 on a mast pan and tilt unit with a 24 cm baseline, an Optris PI 640i thermal camera, a KVH DSP-1760 fiber optic gyro for yaw and six ATI mini45 force-torque sensors, one per wheel leg. A DJI Mavic 2 Pro produced a georeferenced photogrammetric map and a portable LIBS instrument sampled terrain composition along the routes [13].

ROSPA cross-validation against GMV platform-art, 2016. A Mitsubishi PA10 arm mounted on the ORBIT floor pushed a spherical probe with one degree of freedom of spring compliance into a cylinder carried by the MANTIS platform, at approach speeds of 30, 50 and 100 mm/s, with Vicon tracking both bodies [3]. What was measured was the post-contact free-drift velocity vector against a one degree of freedom contact model; what was found is the drift budget quoted above, which bounds how long after a contact the floor still represents free space [3].

Uncooperative rendezvous and capture, 2018. GRALS carried a vision-based navigation camera and the COSMIC gripper toward a Launch Adapter Ring target while ORBIT supplied the free-floating target and the contact dynamics at the moment of capture [4].

Landing with surface contact, 2018. GRALS flew the descent with camera and altimeter and forwarded the landing state vector and attitude to an ORBIT platform, which performed the landing force test [4].

ROOTLESS active leveling, 2018. Validation that a slight commanded tilt of the payload plate introduces a controlled lateral force, so that microgravity or a chosen lower gravity level can be emulated on a floor that is not perfectly level [5][6][7].

Overactuated platform trajectory control, 2022. The ACROBAT, SATSIM and RECAP stack with eight solenoid thrusters of about 10 N each and one reaction wheel, modeled in Gazebo against a digital elevation model of the floor [8][9]. Monte Carlo trajectory finding and following succeeded in every simulated case; on the real floor, straight line tracking held only to tens of centimeters, with the disturbance dominated by floor unevenness [8][9].

Olympus quadruped in-flight attitude control, 2026. The Norwegian University of Science and Technology’s Olympus, with motors limited to 12 Nm, mounted on a rotating rod over a floating platform, with motion capture supplying the state estimate [10]. Single-axis reorientation times were compared between simulation and hardware, and a 90 degree reorientation was achieved in 2.6 s on hardware [10]. This is the campaign shown in the ORBIT photograph above.

Docking and capture generally. The flat floor is used for close-range rendezvous, docking, berthing of free-floating objects and landing on low-gravity bodies, with an ejection mechanism providing representative delta-v and tumbling conditions, and the ROMA platform carrying a UR10e for docking interface and in-space servicing autonomy tests [1].