Infrastructure Pilot Excavator

NASA/Frank Micheaux. Public domain (NASA / US government work).

Overview

IPEx is a lunar regolith excavator and hauler built at NASA KSC. It digs with two counter-rotating bucket drums, carries the load in those same drums, and drives it to a delivery point, so one vehicle does the work that terrestrial practice splits between an excavator and a truck.

It is the flight-directed successor to RASSOR, the KSC bucket-drum testbed that established the counter-rotating architecture at TRL 4 [5]. The published literature calls the program the ISRU Pilot Excavator; the NASA program page now calls it the Infrastructure Pilot Excavator and keeps the IPEx acronym [12]. A parallel outreach project built Mini RASSOR, an educational scale model, at the same KSC lab that carried the wheel concept through several iterations on the way to the flight design [9].

Specifications

Parameter	Value	Source
Mass class	30 kg	[2]
Volumetric scale against RASSOR 2.0	50 to 70 percent	[4]
Locomotion	four-wheel skid steer	[1]
Drive speed, nominal	30 cm/s
Drive speed, demonstrated maximum	40 cm/s	[3]
Obstacle height	7.5 cm	[1]
Slope limit	15 degrees

As of mid 2024 the TRL 5 design was about 95 percent complete and 90 percent fabricated, with TRL 6 the next target [1]; a five-day continuous ground demonstration in August 2024 established TRL 5 [3]. The program target is a TRL 6 excavator ready for lunar demonstration.

Mission profile

No launch has been assigned.

Parameter	Value	Source
Regolith to be moved	10,000 kg in one lunar day	[1], [2], [12]
Surface duration	11 Earth days	[1], [2]
Haul distance	100 m	[2]
Traverse total	70 km	[1]
Excavation rate required	42 kg/h, which is near-continuous cycling over the 11 days	[1], [2]

To supply that much regolith the vehicle has to drive further, traverse the same ground more often, and excavate more than any planetary robot flown to date [1]. Prior lunar sample return collected tens of kilograms, against the 10,000 kg IPEx is sized for [12].

Architecture

IPEx uses two sets of bucket drums, one either side of the chassis, digging simultaneously in opposing directions, so the horizontal digging reactions cancel and excavation force is reacted through the structure rather than through traction [2], [4]. Regolith enters through scoops around the drum exterior and is retained by an internal baffle system; reversing drum rotation discharges it. In-situ lunar soil relative density is about 65 percent in the top 15 cm and exceeds 90 percent below 30 cm [8], so digging forces remain substantial at one sixth weight.

The vehicle both excavates and hauls, carrying regolith in the drums to a delivery point rather than requiring a separate transport vehicle [2]. The RASSOR patent covering the zero horizontal reaction force architecture is US 9,027,265, granted 12 May 2015 [5]. NASA’s 2020 public bucket-drum design challenge set constraints of comparable order to the flight drum: at most 175 mm of scoop engaged at once, 5 kg drum mass, 450 mm diameter, 360 mm length and at least 17.6 L of captured regolith, against the 437.1 mm diameter RASSOR 2.0 drum measured in the scaling study below [11].

Dust tolerance is the organizing constraint across the vehicle, with the thermal system, the charging system and the actuators identified as the three areas most exposed to it [2].

Mobility

Mobility is four-wheel skid steer at a nominal 30 cm/s, able to cross rocks up to 7.5 cm high and slopes up to 15 degrees [1]. Maximum sustained drive speed demonstrated on the test site was 40 cm/s [3]. Each mobility actuator is allocated about 0.85 kg against a 30 kg vehicle mass [2]. Motor speeds as low as 25 RPM are required during parts of the concept of operations, including docking to the wireless charger.

Early IPEx wheel concepts, developed through several design iterations at the KSC lab alongside the Mini RASSOR outreach model, worked through spoke and rim geometry before the design settled [9]. The wheel began as a modified version of the VIPER wheel, adapted for skid steer driving. Grousers, cleat surfaces and spokes are removable so that they can be changed as full scale testing progresses, and the side rings carry a large chamfered edge that acts as a ski during skid steer turns, keeping the wheel edge from digging into the regolith and raising lateral and torsional loads. Ten wheel geometries were run in a trade study, which found a tradeoff between slip rate and power use, no significant effect from cleat design, and a small advantage to tapered over square grousers across the test set [1]. RASSOR tilt table testing had already established that wheels rather than tracks suit a machine whose excavation reaction is internally canceled: on a 25 degree slope in BP-1 simulant a 17 inch four wheel configuration succeeded three times in three, against one in three for metallic link tracks [5].

Actuator internals for the TRL 5 build use a ThinGap LSI 75-12 frameless brushless DC motor [1] with Honeywell SS511AT Hall effect sensors, a Harmonic Drive CSF 14-80LW gearset, SKF 71809 angular contact bearings in a back-to-back arrangement on the output, and a Nomex dust seal [2]. Sealing follows a layered scheme: a labyrinth path outermost, then a felt seal, then a PTFE lip seal. Because manufacturer motor curves and gear efficiencies assume specific greases, temperatures, and Earth pressure, a Gen 1 actuator was characterized end to end to separate the inefficiencies of motor, motor bearings, harmonic drive, output bearings, dust seals, and resolver under representative conditions [2]. Vacuum-rated grease was applied to bearings and gearing, and the open questions examined were how grease plating changed bearing and gear performance across temperature, how the resolver behaved at low speed, and the drag torque contributed by the dust seals. Actuator characterization covered 25, 50, and 100 RPM and then 200 RPM intervals up to 1700 RPM, at 35 and 40 degrees C surface temperature and bus voltages of 47.6, 53.2, and 58.8 V, with a further campaign across a transient thermal profile from -35 to 40 degrees C [2].

Power and energy

IPEx is battery powered and recharges by docking with the lander. Recharging is wireless, using an antenna carried on the radiator cover, which removes the need for a dust-tolerant electrical connector, the component most likely to fail in a regolith environment [2]. The radiator cover provides a rotational degree of freedom used to align the charger during docking [1]. Battery capacity in watt-hours is not published. Duty cycle is set by battery depletion: the vehicle cycles between dig and dump sites until the battery is empty, then docks and recharges. On the test site this worked out to about 10 excavation and unload repetitions per charge, with 35 automated docking and recharge sequences across the demonstration [3].

Thermal

A continuously exposed radiator panel would be coated by excavation-generated dust and stop rejecting heat [1]. Lunar surface temperature at 85 degrees latitude ranges from a mean 182 K at local noon to a mean 61 K near 3 a.m. equivalent [7].

IPEx uses a buffered radiator. Heat is absorbed during operation by a phase change material rather than radiated, with the radiator kept closed under an actuated cover. When the PCM has fully melted the vehicle pauses, opens the cover, and radiates the stored heat to space until the PCM resolidifies [1]. On the test site a thermal reset was performed every 10 mission repetitions, paired with the docking and recharge sequence [3].

The PCM is n-hexadecane at 99 percent or better purity, melting at 18 degrees C. The installed charge stores a calculated 300.5 kJ of latent heat over that transition, which is the energy the vehicle can absorb between radiator openings [1]. It sits in a 6061 aluminum housing whose base conducts to the avionics enclosure and whose cavity is packed with thermal-epoxy bonded aluminum fins to distribute the load and move heat upward as the material expands and contracts. The cover doubles as the radiator surface and is coated with S13GLO, a low-outgassing white paint of zinc oxide in an RTV602 silicone binder. The cover itself is Cerakote coated and sealed with felt. The wider thermal subsystem comprises the radiator, cover and its actuator, PCM, heat spreader, heater strips, avionics enclosure, and multi-layer insulation.

Compute and avionics

Avionics are housed in a dedicated enclosure that forms the conductive sink for the PCM [1]. The avionics and software subsystems are deferred to a separate publication and are not published; the TRL 5 ground demonstration used a surrogate avionics setup derived from the final flight computer, together with larger externally mounted battery packs suited to terrestrial testing [3]. All localization and autonomy software ran onboard, with only a subset of imagery and telemetry downlinked. Software is built on ROS.

Payload and sensing

The excavation implement is the payload. Beyond it, IPEx carries a minimum of four cameras: one stereo pair, one side-facing, and one mounted between a set of bucket drums [1]. The stereo pair supports visual odometry and hazard identification, the side-facing camera tracks the lander for absolute pose, and the drum camera views the terrain and the excavator itself. The drum view covers the working face, where cut depth beyond about 50 percent of the scoop opening causes regolith to bridge the opening and reduce the mass collected per rotation [4].

Field of view is set against two opposing requirements: a wide field helps near-field hazard detection and visual odometry, a narrow one improves pose estimation against a distant lander [1]. A 5 megapixel sensor was chosen to balance visual fidelity against onboard compute, with images downscaled for the most demanding operations. Initial work used a Sony IMX264 with a 3.45 micron pixel and a COTS C-mount lens, but C-mount lenses proved unsuitable for vacuum: venting difficulties, outgassing from the grease in the adjustment mechanisms, and the complexity of variable aperture and focus [1]. The design moved to S-mount lenses, which have fixed aperture and no moving parts and are straightforward to vent and secure, and correspondingly to the smaller-pixel Sony IMX547 sensor.

Dust on the lenses is mitigated in three layers. Transparent electrodynamic dust shield covers use alternating current to drive dust off the glass [1]. Each cover is held by a hold-down release mechanism that can jettison it if the EDS fails. Testing confirmed dust-clearing performance adequate for continued operation, though it also revealed that illumination behind the same piece of EDS glass produces reflections and image artifacts. Monocular camera modules carry a three-LED unit each; the stereo modules have no onboard LEDs and are lit by a stereo LED unit mounted elsewhere on the chassis.

Autonomy

The excavation duty cycle leaves no room for an operator in the loop on individual digs, so the vehicle cycles autonomously and is supervised at the campaign level [2]. RASSOR’s own auto-dig routine, the precursor IPEx’s cycling replaces, is a PID loop on drum motor current with the digging arms as effectors; its derivative term goes unused in practice, and without an operator periodically raising the current setpoint the arms drift up until the drums stop engaging the soil, so the original design still assumed a monitoring human [10].

Localization uses no global positioning system. The vehicle estimates depth from stereo imagery and recovers absolute pose by recognizing the lander and its fiducial targets in the side-facing camera [1]. The traverse geometry is chosen to serve this: the path between dig and dump sites is an arc that keeps the lander continuously in the side camera’s view, which also lets a 15 m radius working area around the lander simulate the 100 m drive length that oxygen production teams specify as a minimum [2].

Drum fill state can be inferred without dedicated instrumentation: a free-spinning drum current model fitted to motor current reached about 7.4 percent mean percentage error on RASSOR hardware, and a neural network augmentation of it was integrated into the robotic system [6].

Modes of operation

The mission concept is cyclical [2]. After landing, IPEx deploys from the lander and performs a teleoperated slow-driving mapping routine over a half-circle of about 15 m radius, from which a dig site, a dump site, and a hazard-avoiding path between them are identified. It then drives that arcing path back and forth, excavating and unloading, until the battery is depleted; it docks, recharges wirelessly, and repeats, until 10 metric tons have been moved over 11 Earth days [2].

Three distinct modes therefore exist: teleoperated mapping, supervised autonomous excavation cycling, and docked recharge. To these the thermal design adds a fourth, a scheduled pause with the radiator cover open while the phase change material resolidifies [1]. On the test site the autonomous cycle was defined as three drives from the excavation zone to the dump zone, excavation and unloading, then three more drive cycles, averaging nine minutes per repetition [3].

Ground operations

The August 2024 TRL 5 demonstration is the only published operations exercise, in which IPEx ran 334 excavation and unload cycles over 77 hours 46 minutes of continuous operation, equivalent to 205 hours 18 minutes of mission time or about eight days, covering 58.1 km, and finished with 59 hours of operational margin [3], against a flight requirement of 70 km over 11 days [1].

It was operated by four rotating ground crews of four people each: a primary operator with command authority during nominal operations who led problem deliberation, a secondary operator providing support, performance tracking, and troubleshooting, a telemetry desk officer monitoring and analyzing incoming data, and a simulation coordinator [3]. The simulation coordinator role exists because the test is a simulation: that operator oversaw the run with real-time data and imagery, performed out-of-simulation actions such as swapping batteries to limit real downtime, and was the only role with access to the situational cameras, so the operating crew saw only what a lunar crew would see.

Ground control sat 183 m from the test site [3]. The lander-to-ground link was represented by reduced data transmission and a 14 second round-trip delay, which is far longer than lunar light time and reflects CLPS relay and processing overheads rather than propagation. Rocks were placed in the test area unknown to the operators, and low solar angle illumination was reproduced with stadium lighting. An OptiTrack motion capture system published ground-truth pose on a ROS topic, recorded alongside the vehicle’s own telemetry in a time series database, giving operators outside the simulation a reference against which to evaluate the onboard localization.

Technologies developed

The transferable results are the dust-tolerance mechanisms rather than the excavator as a whole: the buffered radiator with actuated cover and phase change material, which converts a dust-vulnerable continuous radiator into a duty-cycled one; wireless charging, which removes the dust-tolerant connector problem entirely; the layered labyrinth, felt, and PTFE lip actuator seal stack, characterized for drag torque rather than assumed [2]; and electrodynamic dust shield lens covers with jettisonable hold-down release backup [1].

The autonomy test site itself is a product: a controlled terrestrial environment with granular material, a rock distribution drawn from a power law fit to Surveyor VI data, a full-scale lander mock-up, low-solar-angle stadium lighting, and motion capture ground truth, together with published instance segmentation and photogrammetry datasets collected in it [3].

Scaling

Bucket drums resist classical blade and bucket force models because of their geometry, so three geometrically similar drums were built and measured on a gantry test stand in BP-1 simulant [4].

Drum	Diameter	Width	Scoop width	Scoop height	Mean fill per drum	Four-drum capacity
Small	237.5 mm	206 mm	51.6 mm	26.4 mm	3.80 kg	15.21 kg
Medium	295.1 mm	254 mm	63.5 mm	34.5 mm	7.30 kg	29.20 kg
Large (RASSOR 2.0)	437.1 mm	358 mm	90.4 mm	47.8 mm	24.98 kg	99.94 kg

Values from [4]. A 1.84 times increase in diameter from small to large gives a 6.6 times increase in captured mass.

Test variables were linear cut speeds of 10 and 30 mm/s and cut depths of 10 mm and 40 percent of scoop height, 12 unique cases repeated four times for 48 runs, with drum tangential velocity held at 8.5 times the linear cut speed to keep cut pitch constant across scales [4]. Force and torque were logged at 1 kHz through an NI-9237 module. Horizontal excavation force per unit drum width rises with both cut speed and cut depth, and mechanical excavation energy per kilogram collected falls with increasing cut speed, increasing drum size and decreasing cut depth. Lateral force was negligible throughout, and vertical force with the drum fully engaged equalled the weight of captured regolith. Cut depth is bounded above by bridging: digging at full scoop depth lets regolith bridge the opening, and limiting cut depth to at most 50 percent of the scoop opening collects more regolith per rotation [4].

The scaling study concluded that the small and medium drums collect enough regolith at the required rate for the IPEx concept of operations [4]. The bucket drum concept originates in Lockheed Martin work from 2008.