Tohoku University Space Robotics Laboratory

Source: [1]. CC BY 4.0.

The Space Robotics Lab in the Department of Aerospace Engineering at Tohoku University, Sendai, works on the mechanics, dynamics and control of spacecraft and of planetary exploration robots, and is closely involved in JAXA missions [6]. Its ground facilities are bench-scale terramechanics rigs rather than a large yard: a slip-controlled single wheel testbed used for wheel and grouser characterization, a modular sinkage bench for legged robot end effectors with adjustable load and ground angle [2], and rover platforms taken to external analogue sites for full-vehicle work [3]. The lab’s soils are Toyoura sand, Tohoku Silica No. 5 and FJS-1 lunar simulant [1][3].

Laboratory data

Parameter	Value
Operator	Tohoku University, Space Robotics Laboratory, Department of Aerospace Engineering [1][6]
Location	Aoba Aramaki, Sendai, Japan [6]
Commissioned	Not published. The single wheel testbed appears in campaigns from 2019
Type	Bench-scale terramechanics rigs: a slip-controlled wheel bed and a tilting sinkage bench [1][2]
Floor area	Not published
Capabilities	Wheel testbed, sinkage bench, rovers
Simulant or terrain	Toyoura sand, FJS-1 lunar simulant, Tohoku Silica No. 5 [1][3]. Tonnages not published
Instrumentation	Force sensors, encoders, hand vane, wheel trace imaging [1][2]
Ground truth	Hand vane bulk density before every run; angle of repose by fixed funnel [1]
Fidelity limits	1 g; minimum wheel load 5 to 10 N against a 4.1 N lunar target. See below
Access	Not published. Laboratory research equipment, no external user route in the literature
Cited by	rashid

Capabilities

Single wheel testbed

Parameter	Value
Working volume	Bed near 1.6 x 0.30 x 0.20 m; stable experimental region 0.4 m long
Test article limits	Minimum vertical load 5 to 10 N. Wheels run to date 200 mm diameter, 80 mm wide
Vacuum	Not applicable
Temperature	Not applicable
Illumination	Not published. Ordinary laboratory lighting
Simulant or terrain	Toyoura sand at 1440 kg/m3; FJS-1 at 1740 kg/m3, both 24.5 percent relative density
Gravity offload	Not applicable. Reduced gravity is reached by load scaling and simulation
Instrumentation	Force sensors, encoders, wheel trace imaging, hand vane

Source: [1].

The testbed sets the rotational and translational velocity of the wheel independently, so that a fixed slip ratio characterizes the traveling state through the run [1]. The procedure is fixed: the soil surface is tilled to loosen it, levelled with a flat plate to a consistent horizontal height, checked for bulk density with a hand vane, then the rotational and translational velocities are set for the desired slip ratio and constant motion begins. Force sensors and encoders record force, horizontal and vertical displacement, and wheel trace images are taken. The bed is then reset and each condition repeated three times, with the experimental error taken as the standard deviation of the three run averages.

The bed dimensions are inferred rather than stated. The group publishes that its discrete element replication of 0.65 x 0.14 x 0.12 m is 40 percent of the length, 47 percent of the width and 60 percent of the depth of the real bed, which puts the physical bed near 1.6 m long, 0.30 m wide and 0.20 m deep [1]. Vertical load is commanded with the wheel’s own mass accounted for.

Modular sinkage bench

Parameter	Value
Working volume	Sand box 446 x 332 x 218 mm, filled to 100 mm depth
Test article limits	Manually actuated linear turret, force meter resolution 0.01 N; 66 N in the published campaign
Simulant or terrain	Silica sand
Slope	Box -45 to +45 degrees; load angle 0 to 90 degrees, set to 0.1 degree
Instrumentation	Force meter at 0.1 s; turret displacement to 0.01 mm; universal joint wrist

Source: [2].

A related climbing mechanism under development at the lab replaces the sinkage bench’s gripper with an 8-fingered wheel-gripper that retracts into a cylindrical wheel for driving and extends its fingers as a micro-spine gripper for steep or uneven ground, driven by a single actuator through a tether and pulley; on a 3D-printed rough-surface analog it held 40 N in the normal direction but, unlike a dedicated comparison spine gripper, produced no holding force at all when the surface was flat under the gripper’s tested orientation [14].

The sinkage bench is built for end effectors rather than wheels. A manually actuated linear turret carries a force meter connected rigidly to the gripper; the turret sits on a curved rail so the load angle can be dialled between 0 and 90 degrees, and the sand box below is mounted on an inclination-adjustable frame covering -45 to +45 degrees, so that load direction and ground slope are set independently [2]. The wrist is a universal joint with roll and yaw freedom so that load acts through the gripper center at any inclination. Load is set to the case of a hexapod standing on three legs.

Rover platforms

The lab’s rover platforms are exercised outside these benches, at external analogue sites. EX1, a four-wheeled 21.63 kg rover of 0.82 by 0.52 by 0.67 m with a passive spring-damper suspension developed at the lab, was run at the JAXA Space Exploration Field, over a 20 m by 20 m area of Tohoku Silica No. 5 about 0.3 m deep, at wheel velocities of 0.23 to 1.17 m/s on flat ground and 0 to 18 degree slopes at up to 0.47 m/s, plus rapid acceleration and deceleration runs for transient slip [3]. The suspension concept behind EX1 was first evaluated in simulation under a 1.625 m/s2 lunar gravity field against 1 to 12 cm steps, a 10 cm hemispherical rock, a 1.5 m outcrop and 1.5 m slopes from 5 to 30 degrees, at 0.05 to 1 m/s [5].

An earlier generation of rover platforms ran on the same pattern of bench plus external sand field. The four-wheel-drive, four-wheel-steer El Dorado rover, 2 to 30 cm/s, was driven on a sandy beach for a vision-based slip angle estimator that reads the wheel’s own trace with a rear-facing camera, reaching under 2 degrees RMS error against laser ground truth on a 7 m traverse [9]. A related four-wheel-steer test vehicle, 13.5 kg, was run in an indoor sandbox of 2.0 by 1.0 m filled to 8 cm with Toyoura sand tilted to 5 and 10 degrees, comparing a model-based feed-forward steering law against sensor-based slip feedback, the feedback cutting lateral path error from 0.051 m to 0.011 m [8]. The El-Dorado II rover climbed a 12 degree Toyoura sand slope under an energy-optimal, equal-slip-ratio control law developed from the lab’s single wheel testbed data, though the reported minimum effective motor work differed by a factor of 2.6 between that experiment and the accompanying simulation [13]. A still earlier six-wheeled testbed, 5.8 kg on 90 mm wheels, established the lab’s basic method of pairing a Bekker-Wong wheel-soil model with multibody dynamics: it drove on dry sand up an increasing slope until slip went to 1 at about 12 degrees and the wheels dug in, a failure mode used since as a benchmark case rather than a general limit, since it is a property of that vehicle and that sand [7].

High-speed single wheel testbed

A second, larger single wheel rig, purpose-built and separate from the bench above, drives wheels at translational speeds up to 1.0 m/s over a 6.85 m bed of Tohoku Silica Sand No. 5, roughly a hundred times the speed of a flown planetary rover [15]. At that speed a slick wheel needs a slip ratio of 0.25 before drawbar pull turns positive, against about 0.1 at 0.2 m/s, and its traction coefficient falls to about 0.02, a tenth of what grousers reach at low speed; a companion discrete element simulation reproduces the ordering between wheel designs but not this speed dependence, predicting traction rising with speed where the bench shows it falling [15]. The same rig, run at slower, quasi-static speeds of 0.02 to 0.04 m/s, has also compared a 200 mm Rashid-rover-pattern grouser wheel in Toyoura sand against FJS-1 lunar simulant across 0 to 50 percent slip and 5 to 70 N vertical load [16], and been scanned with a LiDAR profiler to fit ruts left at fixed 20 percent slip and three vertical loads with an inclined sine function, though several of the resulting height differences sit within the profiler’s own 3.2 mm measurement noise [17]. A parallel resistive-force-theory model of grouser geometry, validated against single-wheel tests in the same Tohoku silica sand and against discrete element simulation, chose the tilted octagonal grouser flown on JAXA’s LEV-1 lunar hopping rover; the wheel then operated on the Moon after the January 2024 SLIM landing, though the ground and simulation numbers have no flight telemetry to check against [18].

Instrumentation

Soil characterization is treated as a measurement in its own right. Particle density is measured to JIS A 1109:2020 and minimum and maximum bulk density to JIS A 1224:2020, and the static angle of repose is measured by the fixed funnel method over repeated runs, giving 34.3 degrees averaged over Toyoura runs and 38.0 degrees over seven FJS-1 runs, with the error taken as the standard deviation between the left and right side angles of each run [1]. The angle of repose is then used as the calibration target for discrete element parameters, so that a soil property measured independently of the wheel constrains the simulation the wheel results are compared against. Terrain deformation left by the wheel is used as a measurement rather than a by-product: grouser pitch is read from top-down images of the trace by measuring across several pitches and dividing, and the lab’s terrain deformation model takes vertical load and wheel pose to update a height map, with the load to depth relation obtained from discrete element single wheel runs [4].

What it does not reproduce

Gravity is not reduced anywhere in the published setups. The lab’s route to lunar performance is a discrete element model calibrated against Earth-gravity bench data and then run at 1.62 m/s2, and the comparison it produces is a prediction, not a measurement: 4.1 N under lunar gravity matched the 24.5 N Earth gravity results for traction coefficient and tractive efficiency, with a 1 to 2 mm difference in total sinkage, and tractive force and resistive torque both reduced by a factor of six [1]. The load floor blocks the direct experiment that would test that prediction: the bench could not go below about 5 to 10 N, so the 4.1 N lunar-equivalent load could not be applied on Earth.

Soil reproduction is bounded on both sides. FJS-1’s measured bulk density range could not be covered in the matched simulation without raising particle density above the measured value, so the higher packing densities were not studied [1]. In the field, the sand used is not a mineralogical simulant at all: Tohoku Silica No. 5 is dry loose silica sand described as a low-fidelity lunar regolith simulant whose sparse grain distribution presents harsher sinkage and slip conditions than the real material [3].

Scale bounds the wheel runs. The stable region of the bed is 0.4 m long, which sets how much steady-state travel is available after transients, and at the higher speeds the group is interested in, above about 0.1 m/s, transient effects occupy a growing fraction of that distance [1]. The discrete element replication is smaller still, 0.65 m by 0.14 m by 0.12 m, sized so that boundary interaction does not impede soil flow rather than to match the bed, and its particles are represented at 1 mm radius against the sub-millimeter real grain size, a scaling of more than ten times that the authors identify as common practice and a known source of error [4].

Manipulator and free-flying testbeds

Before the lab’s terramechanics work, its ground facilities centered on manipulator control and free-flying capture, and those testbeds still frame how it verifies control laws before trusting simulation. TREP, a two-joint arm mounted on a deliberately flexible double beam, was used to show that a reaction null-space control law lets the arm track a path without exciting the base while a separate loop damps whatever vibration the base already has, tested at two very different traverse speeds against a point-to-point move that vibrated the base strongly by comparison; the beam was designed so that reaction torque and one reaction force component could be neglected, which simplifies the problem relative to a real spacecraft [10]. An air-floating table gives the lab’s free-flying capture work its planar microgravity: a compliant-wrist impedance controller aimed at a zero coefficient of restitution on contact with a non-cooperative target reached 0.016 +/- 0.011 in practice, with the gap attributed to wrist friction and sensor noise rather than the contact model [11], and a later dual-arm study on the same class of rig showed repeated deliberate impacts, rather than a single grapple, damping a spinning target’s angular velocity to a steady low-velocity state within about six seconds [12]. Both are two-dimensional rigs and neither reproduces the three-dimensional tumbling or out-of-plane torques a real capture would see.

Campaigns run there

Rashid-1 wheel traction evaluation. A wheel modeled on the Emirates Lunar Mission Rashid-1 rover, 100 mm radius, 80 mm wide, 14 grousers of 20 mm and 1.18 kg, was run in Toyoura sand and FJS-1 at the nominal per-wheel load of 24.5 N for a 10 kg rover, at 0.02 m/s nominal horizontal speed and 0.2 rad/s at zero slip, over slip ratios from 10 to 50 percent [1]. Traction coefficient and dynamic sinkage agreed between experiment and discrete element simulation over the 0 to 50 percent slip range in both soils. Grouser pitch in the wheel trace was found to be unchanged by gravity variation, which makes it usable as a visual slip indicator [1]. The wheel used on the bench differs from the flight wheel in grouser length, mass and material, with vertical load adjustment used to hold the correct load despite the mass difference. See rashid.

Legged robot gripper sinkage, 2024. An underactuated tendon-driven microspine gripper for the SCAR-E hexapod was pressed into silica sand at slope angles from 0 to 35 degrees under a load representing the robot standing on three legs [2]. Maximum sinkage rose from 12.98 mm at 0 degrees to 36.19 mm at 35 degrees, with the increase accelerating beyond 25 degrees; the operators attribute the steep rise at 35 degrees to granular collapse at the slope surface, and report agreement with the coupled multibody and discrete element model to within about 10 percent at 66 N [2].

Wheel trace deformation model, 2024. Discrete element single wheel runs were used to derive a load to depth relation for a 200 mm diameter, 80 mm wide wheel with 14 grousers of 20 mm, in a virtual bed of 0.65 m by 0.14 m by 0.12 m of Toyoura sand represented at 1 mm particle radius, and the resulting trace amplitude and profile were fed into a real-time height-map deformation model driven by vertical load and wheel pose [4].

Data-driven simulator calibration, 2026. Slip and sinkage regressions derived from the single wheel testbed, from discrete element simulation and from the EX1 field campaign were combined and integrated into the OmniLRS lunar surface simulator, targeting flat terrain and slopes to 20 degrees [3]. The group notes that including grouser length in the effective wheel radius is necessary for the slip ratio to come out positive during acceleration and steady-state driving, and negative during deceleration, as it should [3].

References