ATHLETE

Source: Heverly [2]. Public domain (NASA / US government work).

The problem

A lunar cargo and habitat transporter has to take a payload off a high lander deck, move it across soft regolith and rock fields, and do so within launch mass limits. The Altair descent stage carries large liquid hydrogen tanks, which puts the flat cargo deck more than 6 m above the surface, and walking cargo down from it was the task ATHLETE was built for [9], [8]. A conventional six-wheeled vehicle must be able to lift a wheel out of a hole using the rim thrust of the other five, so wheel and drive mass is set by the worst terrain. ATHLETE sizes wheels and drives for nominal terrain and uses the limbs to lift a stuck wheel, which is the mass trade the design rests on [6], [11].

The soil assumptions behind the trade are anecdotal. The 0.25 MPa per meter “4 sigma softest” lunar bearing figure is derived from the depth of Apollo boot prints, not from measurement [7]. First-generation ATHLETE accepted 20 to 45 kPa ground pressure because it can walk out of soft soil, against about 7 kPa for the Apollo rover [1].

What has flown

ATHLETE has not flown. All published results are terrestrial. The record is a sequence of ground prototypes.

First generation, 2005 to 2008. Two Software Development Models were built in a design, build and test cycle that began in March 2005, with the first vehicle operational that October [1]. Each is an 850 kg vehicle with a 2.75 m hexagonal frame and six 6-DOF limbs, carrying about 250 kg of payload on Earth; 36 joints share one 120 W motor and gearhead family, with overall ratios of 3,640:1 to 13,100:1 [1]. Mojave Desert and Meteor Crater tests confirmed power-efficient rolling and tool use from the wheel-driven quick-disconnect adapter [6]. At Moses Lake in June 2008 two vehicles carried graphite micro-habitat mockups 2.34 m in diameter and 3.66 m long, mated hatches to within a few millimeters in 25 to 40 minutes with a 5 to 10 s command delay, and accumulated almost 9 km of traverse with a longest single drive of 1.8 km [7].

Tri-ATHLETE SDM-T12, 2009 to 2010. The second generation is two 720 kg three-limbed vehicles that dock either side of a cargo pallet, 4.1 m tall at full extension, with 500 kg combined payload on Earth and a seventh thigh-pitch joint added to offload cargo from a 3.2 m mock lander deck [2]. In September 2009 a three-week trial at Black Point Lava Flow demonstrated that offload, walking over obstacles more than 1 m tall, backhoe excavation and more than 6 km of traverse [2]. The 2010 campaign at Hahamongna Watershed Park and Black Point covered 82.5 km in eight weeks, 23.1 km and 59.8 km at the two sites, and reached a simulated outpost in four days at over 5 km per day against a 20 km in seven days milestone [4].

Traverse was limited by operations. More than 30 percent of traverse time was idle on the best Black Point day and typically over 40 percent, with ground speeds mostly 0.5 to 3 km/h against a 3 km/h nominal design speed [4], [2]. Repeated heading commands and joystick jitter, short drives caused by steering difficulty with the handheld PortOps controller, accounted for more than half of all drive terminations; the author concludes the operator interface, not the hardware, was the binding constraint [5]. Active limb compliance held wheel normal forces evenly distributed across six limbs, using a force estimate with an error of about 100 N [4].

Low-gravity emulation, 2011 to 2013. Suspended from six computer-controlled winches with ankle force-torque sensing driving the winch dynamics, the vehicle performed a simulated 5 microgee hop of almost 2 m and an auger anchoring test in soft simulant that ended when the bin ran out of depth [11]. A later paper reports a 20 milligee landing and bounce on a mock asteroid in the same testbed [12]. Both are photographic or qualitative results [11].

What is being tried now

ATHLETE-derived concepts. Mobile in-situ resource utilization (ISRU) and construction on the limbed platform is a concept: a microwave volatile extraction plant, freeform additive construction and swappable excavation tools on the tool adapters, none built or tested [3]. The mass argument has been formalized in a parametric model that finds ATHLETE saves only 2 to 5 percent over rocker-bogie at the low end of the mass range and pulls ahead only above a few tonnes; it has not been checked against the mass of a built vehicle [12]. Funding context for the original Altair-era architecture has ended, so the lander-offload case is historical [8].

Dedicated excavators. The alternative to a walking transporter is a small rigid-wheeled vehicle that avoids the traction problem by changing the task. The ISRU Pilot Excavator (IPEx) uses one common actuator design for wheel, arm and bucket drum, and its Technology Readiness Level 5 qualification covered mobility, cameras, dust mitigation and thermal control, a ground campaign [15]. A 3D-printed bucket drum tool for the MoonBot modular rover measured 777.54 kg/h continuous excavation in quartz sand at Earth gravity, not regolith simulant [16]. Neither concept has flown.

Teach and repeat for cargo routes. Repeating a taught route between a lander and a habitat is the autonomy that would replace ATHLETE’s joystick driving. A Doppler lidar teach-and-repeat pipeline with a degeneracy-aware localizer completed all trials on featureless routes where the variants without it did not, on a wheeled terrestrial vehicle with three trials per route [24]. It is a ground test, not flight.

Legged and hopping robots in low gravity. Reinforcement learning policies for walking, vertical and forward jumping and in-flight attitude control were trained for a 14.5 kg quadruped under Mars gravity; in simulation they reach 3.1 m vertical and 3.9 m forward jumps, but only the attitude policy was transferred to hardware [19]. A proprioceptive hybrid internal model for continuous pronking under lunar gravity was tested on a gravity-offload rig [20]. A planar model and reaction-wheel controller for an underactuated bipedal hopper was evaluated only in simulation at lunar gravity [21]. A simulation study of a 24 kg quadruped on downhill ramps under lunar gravity finds torso sliding more efficient than walking above a crossover slope of 4 to 10 degrees [22]. A 9.7 kg limbed climbing robot with spine grippers has made teleoperated climbing trials on lava rock under 1 g [18]. The LASSIE program uses leg force sensing on legged robots to characterize regolith strength, verified in the lab and at White Sands and Mount Hood [17]. All of these are ground tests or simulation.

Wheel and terramechanics modeling. A resistive-force-theory model selected the grouser geometry of the LEV-1 hopping rover wheel, which then operated on the Moon after the January 2024 landing of SLIM, the one lunar-surface result in this list [27]. A variable-height grouser wheel was evaluated in 750 gantry trials on a single wheel geometry and load [26]. An open tradespace toolkit evaluates sub-50 kg lunar rovers in about 30 ms per mission and is simulation only [25]. NASA’s consolidation of mobility test practice sets out full-vehicle and single-wheel drawbar pull, slope and obstacle methods, with LTV as the case study; it reports recommendations, not new measurements [23].

Reduced-gravity testing of mobility. Parabolic flight campaigns on a single-wheel testbed found that matching wheel load alone overstates mobility, with drawbar pull to weight lower by 20 percent or more and sinkage up to 40 percent greater in lunar gravity at equal load [14].

The numbers

Parameter	Value	Source
First-generation vehicle mass	about 850 kg	[1]
First-generation frame	hexagonal, 2.75 m tip to tip	[1]
First-generation limb length, wheel diameter	1.8 m, 0.5 m (later Wilcox papers give 0.71 m)	[1], [6]
First-generation wheel motor	730 W	[1]
First-generation payload, Earth gravity	250 kg	[1]
Tri-ATHLETE half mass	720 kg	[2]
Combined Tri-ATHLETE payload, Earth gravity	500 kg	[2]
Tri-ATHLETE height, full extension	4.1 m	[2]
Nominal driving speed	3 km/h	[2]
Avionics battery	four cells in series, 51.2 V, 130 Ah, 78 kg	[2]
Processors	four 800 MHz PowerPC computers per half	[2]
Michelin Lunar Wheel	71 cm by 23 cm, 15 kg, 2500 N nominal and 4500 N maximum load, 10 km/h	[2]
Wheel life under load in liquid nitrogen	more than 5,000 km (supplier-run)	[2]
2010 traverse	82.5 km, 23.1 km and 59.8 km at the two sites	[4]
Idle fraction of traverse time	over 30 percent, typically over 40 percent	[4]
Mobility mass fraction, Mars and lunar gravity	20 to 25 percent and 12 to 15 percent	[12]
Lunar mobility mass relative to payload	15 to 20 percent (scaling estimate)	[2]

Joint capability, Tri-ATHLETE [2]:

Joint	Overall gear ratio	Harmonic drive	Ratchet torque	No-load speed
Hip yaw	6,591:1	CSG-40-80	3,600 Nm	0.16 rad/s
Hip pitch	6,480:1	CSG-58-80	10,000 Nm	0.11 rad/s
Thigh pitch	6,480:1	CSG-58-80	10,000 Nm	0.11 rad/s
Knee pitch	3,240:1	CSG-40-80	3,600 Nm	0.22 rad/s
Knee roll	3,514:1	CSG-32-50	1,200 Nm	0.30 rad/s
Ankle pitch	6,591:1	CSG-32-80	1,800 Nm	0.16 rad/s
Ankle roll	3,514:1	CSG-32-50	1,200 Nm	0.30 rad/s

Source: Heverly [2]. Public domain (NASA / US government work).

Each joint is a brushless motor with brake and incremental encoder, a planetary stage and a harmonic drive; the ratchet torque of the harmonic drive acts as the mechanical fuse, with structure sized above it [2]. Joint torque is inferred from the difference between motor-side and output-side encoders and a measured torsional stiffness, and the six torques resolve the wrench at the wheel [1]. Ankle pitch backlash corrupted this estimate on the second-generation vehicle until the joint was redesigned [4].

Source: Heverly [2]. Public domain (NASA / US government work).

What is not established

No mobility, traction or slope measurements exist for the first-generation vehicle; its 10 km/h top speed, 250 kg payload and 20 to 45 kPa ground pressure are design values, and the torque estimate accuracy is not quantified [1]. The Tri-ATHLETE field trial reports capability demonstrations, not drawbar pull, slip, sinkage or power curves, and the Michelin wheel life figure is a supplier test not independently verified [2]. The 2010 traverse was on graded dirt and lava flow at 1 g with no soil characterization, no walking, and mostly local operation with an operator alongside, so it does not represent remotely operated or lunar performance [4]. Remote operation was too rare to characterize behavior under communication delay, and the proposed interface fixes were not evaluated [5]. The 15 to 20 percent lunar mass fraction is a scaling estimate with no reduced-gravity test behind it [2], and the parametric mass model is not validated against a built ATHLETE [12]. The low-gravity winch hop is qualitative, with no repeatability or validation of the emulated dynamics [11]. The reduced-gravity wheel result [14] indicates 1 g field mobility figures do not transfer directly to the Moon. No thermal design values are published, and the 7075-T6 aluminum structure is acknowledged as unsuitable for the lunar thermal environment [2]. The legged and hopping results above are simulation or 1 g hardware tests, and the IPEx qualification is also 1 g and ground based [15].