AstroAnt

Overview

A wheeled rover that lands on the Moon has to survive the descent, hold traction on loose regolith, and carry its own power and thermal control across a surface with no atmosphere to buffer either extreme. AstroAnt sidesteps all three problems by never touching lunar ground at all: it rides on the exterior of another spacecraft and inspects that spacecraft rather than driving on terrain. It is a magnetic-wheeled crawler carrying a thermopile, built by MIT Media Lab to ride on the top surface of Lunar Outpost’s MAPP rover and take contactless temperature readings of the rover’s radiator from several positions, so the thermal system can be monitored spatially rather than at fixed sensor points [1]. The concept behind it is a swarm of such small robots inspecting the external surfaces of larger spacecraft rather than the terrain those spacecraft stand on, and before flight it was tested on four NASA parabolic flights in 2021 that included lunar-gravity and zero-gravity parabolas, the only reduced-gravity test evidence held for the vehicle [1]. It flew mounted on MAPP, a 10 kg, 10 cm/s rover built by Lunar Outpost [3], itself a payload of Intuitive Machines’ IM-2 Athena lander.

The design space this occupies is well established on Earth, though not previously flown to the Moon: a small robot that trades a wheeled or legged undercarriage’s need for terrain traction for adhesion to a structure it was never meant to grip unaided. JPL’s line of limbed climbing robots answers that same problem with the microspine, an array of small hooked flexures that catches a rough surface’s own asperities rather than relying on friction or a preload the robot supplies. The four-limbed LEMUR IIb carried a gimbal-mounted ankle joint built to interface a JPL omnidirectional microspine gripper to the limb, giving the gripper passive compliance to an uneven rock surface and torque neutralization while stepping [10]. A microspine anchor mounted on LEMUR IIb was also shown decoupling a rotary percussive coring drill’s reaction load from the robot’s own mass entirely: the anchor gripped vesicular basalt at more than 150 newtons in any direction, survived over one hundred engage and release cycles, and cored 83 mm into basalt while inverted, upside down, a load path that a gravity-stabilized wheeled or legged rover cannot supply without carrying its own reaction mass [11]. The same hooked-tine principle was built into a rotary form, a wheel-shaped microspine array that let a small two-wheeled robot climb cinder-block walls at 25 cm/s, about five times faster than an earlier climbing robot, by holding each hook’s engagement angle between 30 and 45 degrees to maximize climbing force per unit of adhesion force; mounted on LEMUR IIb itself, the same rotary anchor supported more than 160 N against the robot’s 140 N weight, enough to walk inverted [12]. The same hooked-tine grip was also adapted to an asteroid gripper, where eighteen metal microspine suspensions generated close to 20 N of holding force against natural rock in a laboratory rig, aimed at surface gravity too weak for a wheel or foot to rely on friction at all [13]. JPL’s successor LEMUR 3 paired microspine grippers for a rock face inside a lava tube with gecko-adhesive grippers for a mock solar panel on the same four-limbed chassis, so the end effector, not the platform, was matched to the surface, though both demonstrations needed gravity offload to support the 35 kg robot [9]. Most recently, JPL has been remaking the microspine itself in bulk metallic glass, forming arrays of microscale metal spines and benchmarking their friction against acrylic, carbon fiber, aluminum and wood, work aimed at a spine that can survive vacuum, cold and radiation where a polymer flexure cannot [14]. A JPL summer-program prototype effort built a sixteen-mechanism omnidirectional microspine gripper and recorded its failure mode: an external load on the housing, as a drill would apply, changes the hook engagement angle and breaks an otherwise secure grip [15]. AstroAnt sits inside that same climbing-robot lineage but answers the attachment problem differently: rather than a spine array sized to catch a rock’s natural asperities, it uses magnetic wheels, because the surface it grips is not natural rock but a rover chassis its builder can choose to be ferromagnetic on purpose [1].

Nothing about the vehicle is autonomous in the way a free-ranging rover would need to be. It carries one sensor to several places on another machine that a fixed sensor could not reach, and the mission’s payoff is spatial coverage of a thermal measurement rather than any new locomotion capability applied to natural terrain. IM-2’s Athena lander touched down about 250 m from its intended site on 6 March 2025 and came to rest on its side in a crater at Mons Mouton near the lunar south pole; the solar panel orientation and the crater’s extreme cold meant the company did not expect the lander to recharge, and the mission was declared over the following day [5]. MAPP was trapped under that tipped lander with it, so AstroAnt’s own flight ended before its one instrument produced any published reading, the same outcome that also stopped the mission’s science drill from reaching regolith [5]. The design point AstroAnt represents, an inspection crawler for a structure a robot never needs to grip unaided, is therefore untested where a microspine gripper has been the answer on rock or an asteroid’s loose regolith [13].

Power and energy

Not published. AstroAnt carries a battery, but its capacity, charging method and recharge interface do not appear in either MIT program page describing the vehicle [1], [2].

Thermal

Not published, which is a notable gap for a robot whose task is to sit on a radiator in lunar daylight and whose one instrument is a thermometer.

Compute and avionics

Not published.

Autonomy

Not described beyond the fact that the robot and a central station on the host vehicle communicate over Bluetooth Low Energy, with the station providing the path back to the ground [1]. Unlike the JPL climbing robots it otherwise resembles, which were built to negotiate an unknown rock face or panel edge with a human operator picking each foothold [9], AstroAnt is commanded from a station rather than negotiating its own path, consistent with a working surface that is a known, flat host chassis rather than an unstructured one.

Communications

Bluetooth Low Energy between the robot and a central station carried on the host rover, with the station relaying through the host [1]. Data rate and range are not published. Lunar Outpost’s own account of the Voyage 1 mission describes its Stargate mission control software commanding MAPP within a total downlink allocation of about 6 megabits for transit and surface operations combined [6], the ceiling every payload’s telemetry, AstroAnt’s included, would have had to share.

Payload and instruments

A single thermopile on the underside of the robot, giving contactless temperature readings of whatever surface it stands over [1]. The measurement it was flown to make was the performance of MAPP’s thermal radiator, sampled at several points across the radiator rather than at the one place a fixed sensor would occupy, which is the argument for making the sensor mobile at all [1].

Lunar Outpost markets MAPP itself as a general lunar roving platform meant to carry different payloads across different missions rather than a single-purpose rover, though its own product page gives no mass, power, payload capacity or performance figures to back that claim [4]; the mass and speed figures above come from the Media Lab’s own description of its host, not from Lunar Outpost [3]. The same product line, at a scale suited to a different payload, was built as the small rover carrying the Lunar Vertex mission’s PRISM magnetometer and imaging instruments to Reiner Gamma, chosen because distinguishing the competing origins of that site’s magnetic anomaly needed vector field measurements and close-up imaging at more than one point, which a fixed lander payload cannot provide [7]. That rover chassis, a variant of MAPP, was integrated with a four-sensor fluxgate magnetometer on a mast and a multispectral microscope, environmentally tested, and delivered to Intuitive Machines in January 2024 for a mission designed to run for one lunar daylight period, 10 to 13 Earth days [8]. Other lunar rover work in the corpus addresses the surface rather than the vehicle’s own deck. RASSOR 2.0 is a 66 kg excavator whose counter-rotating bucket drums cancel horizontal reaction force so that digging does not depend on traction, and whose custom actuators came in more than 50 percent lighter than off-the-shelf equivalents [17]. Its descendant IPEx took a common actuator design through a four-test qualification sequence, with the arm actuator surviving 1,402,223 input revolutions against a 337,000 revolution requirement [18]. The actuator-level record reports the same arm actuator at 1,402,223 motor revolutions with no failure, over four times the mission requirement, under a 2-sigma load [19]. Neither carries an inspection payload, and both were tested on regolith simulant on Earth. AstroAnt’s contribution sits at the opposite end of that product family from those larger derivatives: not a bigger rover carrying more instruments, but the smallest possible add-on riding on one.

Modes of operation

Not published.

Ground operations

Not published beyond the communications path already described above. Lunar Outpost’s own account of the mission describes Stargate reaching a measured uptime of 99.998 percent while commanding MAPP within that same downlink allocation [6], the operating context AstroAnt’s own commanding depended on even though it never drove.

Technologies developed

The design point is the contribution: a small, magnetically adhering robot carrying a sensor, a radio and a battery, built to ride on and inspect another spacecraft rather than to traverse regolith [1]. Set against JPL’s climbing-robot work, the technology AstroAnt represents is narrow but distinct from what has flown before: microspine adhesion has been demonstrated on Earth in a linear form on a compliant ankle joint [10], as a drill anchor that reacts load independent of gravity [11], in a rotary wheel form on rough vertical walls [12], on an asteroid analog [13], on rock and a mock solar panel with an interchangeable end effector [9], and as microscale metal spines formed from bulk metallic glass [14], but none of that lineage has previously flown as a complete, self-contained inspection robot riding on the exterior of an operating spacecraft, which is what AstroAnt was built to be even though its flight opportunity did not let it operate.

What is not established

No peer-reviewed or conference account of AstroAnt’s own design exists; everything about the vehicle traces to two MIT Media Lab program pages, neither of which gives mass, size, battery capacity, drive speed, processor details, or thermal limits, and both report no results from the four parabolic flight campaigns cited as its only reduced-gravity testing [1], [2]. Because the host rover was never deployed, AstroAnt itself generated no flight data at all: everything above the level of its pre-flight design is unflown, and nothing in the corpus establishes how its magnetic-wheel adhesion would have performed in the lunar environment as opposed to on Earth. Reduced-gravity testing of the kind AstroAnt’s four parabolic flights represent is known to change results: parabolic flight tests of a wheel and soil both in lunar gravity showed net traction reductions of 20 percent or more and sinkage increases of up to 40 percent against Earth-based reduced-weight tests [16]. That work concerns wheel-soil interaction, which AstroAnt’s magnetic adhesion to a chassis does not involve, so it does not transfer to the vehicle’s own performance. NASA’s consolidated mobility test guidance likewise flags cone index gradient matching of lunar soil strength under a different gravity field as of uncertain validity, and is a set of recommendations rather than test data [20]. How that approach would compare on endurance, coverage or reliability to the microspine and limbed climbing approaches developed at JPL, none of which has flown either [9], [10], [11], [12], [13], [14], is not addressed by anything the corpus holds.