Asagumo

The problem

A lunar lava tube offers shielding from radiation and temperature swing, and a possible cache of ancient volatiles. Reaching the interior means descending a skylight or crossing a rockfall into terrain a wheeled vehicle cannot cross, with no line of sight to a lander for communication or sunlight. A robot there has to place discrete footholds on rubble, hold a grip or a stance on slopes of loose material, and keep working through link dropouts and darkness. Mass is limited by the delivery vehicle: Asagumo was sized to a single-unit CubeSat frame, 10 cm on a side, at 1.3 kg [1].

Legs are chosen for this because a leg treats the ground as a set of footholds, not a rolling contact. Analysis of quadruped creeping gaits showed that three of the six nonsingular creeping gaits can keep a rigid point-foot machine statically stable at all times, and that a quadruped needs four legs for a positive stability margin when leg transfer takes nonzero time [4]. Running and hopping machines kept the footfall-placement premise and added dynamic balance, first demonstrated on a one-legged hopper on a flat laboratory floor [5]. Legs and grippers cost more mechanism and control than wheels, so they are chosen only where the destination requires a foothold or grip a wheel cannot make.

What has flown

No legged rover has operated on the Moon, and Asagumo did not. Spacebit developed it as a four-legged walking micro-rover for a program aimed at a swarm of walking rovers deployed from a wheeled carrier to map lava tube interiors, and it was manifested on Astrobotic’s Peregrine lander as a technology demonstration [1]. The planned demonstration was to deploy, walk at least 10 m from the lander under teleoperation over the lander’s wireless network, and exercise the legs, wide-field cameras and three-dimensional lidar for up to eight days [1].

Peregrine Mission 1 launched on 8 January 2024 and lost its landing capability when a helium pressure control valve failed, about 92 minutes after separation [2]. The lander operated for 10 days 14 hours in cislunar space and returned data from all nine active payloads before a controlled re-entry on 18 January 2024; its legs and terrain relative navigation were among the subsystems not exercised [2]. Asagumo’s status is canceled, and no surface operation of any kind took place.

The nearest flown analogue is Colmena-1, five reversible 12 cm micro-rovers and a deployment module on the same lander [17]. After the landing was lost, no lunar surface operation occurred, so its mobility, deployment and traction objectives were untested; the authors report that 75 percent of technical objectives were met through improvised cruise operations [17]. Flight heritage for small rovers on Peregrine is therefore cruise-phase only. Wheeled flight heritage elsewhere is the JPL lineage from the 11.5 kg Sojourner to the 1025 kg Perseverance [3], none of which carried legs.

What is being tried now

None of the families below has flown on the Moon. Each is labeled by how far it has been tested.

Lava tube legged and limbed rovers. Ground tested: LIMBERO is a 9.7 kg quadruped with single-motor spine grippers that climbed pumice and lava rock at 0.017 m/s under 1 G, teleoperated, with spine wear after 100 cycles [19]. A hexapod with tendon-driven micro-spine feet, SCAR-E, was characterized on a bench: a 66 N static load sank a foot 12.98 mm into silica sand on flat ground and 36.19 mm at 35 degrees, near the angle of repose [8]. Microspine anchors themselves have held basalt above 150 N in any orientation and survived over 100 engage and release cycles, and cored upside down to 83 mm [7]. Proposal: ReachBot, a boom-limbed climber, has a notional Mars lava tube mission concept with no hardware operated for that study [9]. Proposal with design targets only: LunarLeaper, a CubeSat-sized legged ETH Zurich robot aimed at a lunar pit, is at the thermal trade study stage, with a 32 kg mass and 164 W peak power as design estimates, and dust kick-off from walking legs unmeasured [16].

Hybrid legged and wheeled rovers. Field tested: DLR’s Scout rover uses a rimless-wheel spoke; a redesigned spoke was deployed near end to end in a Lanzarote lava tube in 2024, with durability and traction not yet quantified [18]. A preview paper described the planned integration of lidar-inertial SLAM for the October 2024 campaign and reports design targets, not results [10]. The program’s stated reason for walking matches Asagumo’s: the destination, not the landing site surface, sets the architecture [1].

Learned and model-based locomotion in low gravity. Ground tested with gravity offload: Magnecko, 15.65 kg, ran gravity-scaled reinforcement-learning controllers on flat mats with a spring offload at 0.4 m/s, with legs not offloaded [20]. A pronking controller for a 12.5 kg Unitree A1 was tested on a cable-offload and treadmill rig, with survival time as the only hardware result [21]. Simulation: reinforcement-learning policies for a 13.5 kg quadruped on a granular regolith contact model [11], and walking and jumping policies for the 14.5 kg Olympus under Mars gravity, of which only the attitude policy reached hardware, a 90 degree roll in 2.6 s against 0.96 s in simulation [12].

Proprioceptive terrain sensing. This is the family closest to Asagumo’s stated approach of navigating through communication loss from proprioceptive sensors and position data rather than camera and lidar input [1]. Simulation: a quadruped framework estimates elevation, slip and cost of transport at lunar gravity for permanently shadowed regions at 0.2 m/s [25]. Field analog: LASSIE legged robots estimate regolith strength from leg forces at White Sands and Mount Hood, with three geoscientists in the shared-autonomy test [24]. Field analog with arm payloads: three ANYmal quadrupeds mapped 1800 m2 at a space resources challenge final, locating seven of eight boulders over 358 m under simulated 5.0 s round-trip delay [23]. Field campaign: quadrupeds covered over 700 m on volcanic ash at Vulcano [22].

Link and coordination under non-line-of-sight. Terrestrial subterranean programs predict radio signal strength from lidar geometry, with up to 44 percent lower error than baselines [14], and drop relay radios when the measured link degrades [15]. Neither was qualified for flight. Field operations concepts for suited crews were exercised at a terrestrial lava tube in Oregon [13].

The numbers

Quantity	Value	Status	Source
Asagumo mass	1.3 kg	design	[1]
Asagumo envelope	10 x 10 x 10 cm, four legs	design	[1]
Planned traverse and duration	at least 10 m, up to 8 days	plan	[1]
Peregrine separation to attitude excursion	92 min	measured	[2]
Peregrine cislunar operation	10 days 14 hours	measured	[2]
SCAR-E foot sinkage, 0 degrees	12.98 mm	bench test	[8]
SCAR-E foot sinkage, 20 degrees	17.31 mm	bench test	[8]
SCAR-E foot sinkage, 35 degrees	36.19 mm	bench test	[8]
Microspine anchor hold	above 150 N	lab test	[7]
LIMBERO mass and climb speed	9.7 kg, 0.017 m/s	lab test	[19]
Magnecko mass and test speed	15.65 kg, 0.4 m/s	lab test	[20]
Olympus roll, hardware vs simulation	2.6 s vs 0.96 s	lab test	[12]
LunarLeaper mass and peak power	32 kg, 164 W	estimate	[16]
ANYmal field area mapped	1800 m2	analog field	[23]

What is not established

No mass breakdown, actuator count, gait, walking speed, step height, power budget, thermal limit, processor or radio parameter for Asagumo is published, and the only primary source is a conference abstract; Spacebit’s descriptive material is no longer online [1]. Whether the proprioceptive navigation approach was implemented in simulation or on hardware is not documented.

For the field as a whole: every locomotion result above is at 1 g or under spring or cable offload that does not reproduce leg dynamics, on mats, sand, ash or rock rather than lunar regolith [20][21][22][24]. The granular contact models are calibrated on poppy seeds, not simulant [11]. Sinkage was measured on a static bench load, with lunar gravity only in simulation [8]. No walking robot has been tested in vacuum, dust and thermal extremes, and the dust kick-off of legs is unmeasured [16]. Whether a CubeSat-scale legged rover can walk usefully inside a lunar lava tube remains open.

References