ATHLETE

Source: Wilcox [5]. Public domain (NASA / US government work).

Overview

ATHLETE is a wheel-on-limb cargo and habitat transporter developed at the Jet Propulsion Laboratory, with six wheels carried on the ends of six articulated limbs [1], [6]. The design premise is a mass trade: wheels and wheel drive actuators are sized for nominal terrain rather than for the worst terrain the vehicle will ever meet, and when the rolling vehicle gets stuck it locks the wheels and uses them as feet to walk out [8]. The mass saved in the wheels and drive actuators more than pays for the limb actuators, while the limb structure weighs about the same as a conventional mobility chassis.

Prototypes were built in two generations. Quarter-scale Software Development Model vehicles were built and tested between 2005 and 2009, each about 850 kg with a 300 kg payload on Earth [2], [6]. The second generation, SDM-T12, was built in 2009 as a pair of triangular three-limbed vehicles called Tri-ATHLETEs that dock to a shared cargo pallet to form the six-limbed system [9], [5].

Specifications

Parameter	Value	Source
Limbs	six, each ending in a powered wheel	[9]
Degrees of freedom per limb	7, six for positioning plus one redundant pitch actuator for compact stowage
First-generation SDM frame	hexagonal, 2.75 m across	[4]
First-generation SDM limb height at the hip pitch axis	2.08 m maximum
Degrees of freedom per limb, first generation	6
Wheel diameter, first-generation SDM	0.71 m	[4], [6], [7]
Wheel power take-off	square key identical to a 1/2 inch socket drive, rotating with the wheel	[6], [7]
Tool interface	quick-disconnect adapter over the power take-off, with a fold-out high-definition stereo camera pair	[6]
Wheel motor rating	more than 750 W (1 horsepower)	[7]
First-generation SDM mass	about 850 kg each	[2]
First-generation SDM payload, Earth gravity	300 kg
Tri-ATHLETE half mass	720 kg	[5]
SDM-T12 maximum height	4.1 m
SDM-T12 payload, Earth gravity	450 kg	[9]
Combined Tri-ATHLETE payload capacity	500 kg	[5]
Nominal driving speed used to size joint rates	3 km/h
Structural material	7075-T6 aluminum

The two payload figures come from different papers on the same vehicle, 450 kg in the traverse analysis [9] and 500 kg in the mechanisms paper [5], and are quoted here as published.

Scale is deliberate. SDM-T12 is approximately half the size of the conceived flight vehicle, so that a payload can be retrieved from the top of a 6.4 m tall Altair lunar lander deck at half scale [5]. The half-scale choice was driven by the practical limit on transporting representative payloads in Earth gravity. Half-scale field testing covered cargo offloading from a lander mockup, multi-kilometer traverse, climbing over rocks more than 1 m high, and tool use [4]. In lunar gravity the payload to vehicle mass fraction improves sharply, and the whole ATHLETE mobility system was projected at 15 to 20 percent of the payload mass it carries.

Mission profile

ATHLETE has not flown. It was developed under the Human-Robot Systems Project, led by Rob Ambrose of NASA JSC, with funding from the NASA Office of the Chief Technologist, and JPL principal investigator Brian Wilcox [9], [4]. The assigned target was the Altair lunar lander, whose descent stage performs lunar orbit insertion and therefore carries large liquid hydrogen tanks; hydrogen density of 71 kg per cubic meter puts the flat top cargo deck more than 6 m off the ground, and walking cargo down from that deck was the problem ATHLETE was built to solve [2], [3]. The class of payload considered was up to about 15 metric tons off the lander deck.

Field demonstration in fiscal year 2010 set a milestone of a 20 km cross-country traverse in under seven days, which SDM-T12 completed in four days at an average of over 5 km per day, then drove a further 4.21 km to rendezvous with NASA JSC Space Exploration Vehicles [9]. Across eight weeks of testing at Hahamongna Watershed Park in Pasadena and the Black Point Lava Flow north of Flagstaff, Arizona, the vehicle covered more than 80 km.

Mobility

Each limb carries seven actuated degrees of freedom: hip yaw, hip pitch, thigh pitch, knee pitch, knee roll, ankle pitch and ankle roll [5]. Six of these position the wheel and the seventh is a redundant pitch actuator that lets the limb stow compactly [9].

Because the wheels sit on limbs rather than a passive suspension, the vehicle has little inherent compliance to terrain, and the limbs are run as an active suspension instead. An onboard algorithm adjusts limb position in response to estimated ground contact forces, and holds an even load distribution across the six limbs over rolling, rutted and sloped terrain [9]. Ground contact force is not measured directly: each joint carries an incremental encoder on the motor input and a 12-bit absolute encoder on the joint output, the difference between them gives the mechanical windup, and windup combined with a characterization of the joint torsional stiffness gives joint torque, from which the vehicle computes per-limb ground contact force and rebalances itself [5].

The sizing argument behind the architecture is quantitative. Rim thrust required scales with the cube of the linear dimension at constant wheel diameter, while wheel contact patch area scales with the square [7]. A conventional six-wheeled all-terrain vehicle must be able to lift a wheel out of a hole using the combined rim thrust of the other five, whereas ATHLETE lifts that wheel with its limb and needs only enough thrust to climb a moderate slope, taken as 27 degrees in 2-sigma soil [8]. Each wheel can be lifted by its limb and set on or over an obstacle rather than being dragged over it by thrust from the other wheels [1]. That lets the wheels be about half the diameter and one quarter the mass of a conventional design, with peak torque one tenth as great, from which about 90 percent of the drive gear mass is saved.

Joint capability, Tri-ATHLETE, from [5]:

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

Joint speeds were set by the requirement to servo the limbs actively over terrain undulations of up to 10 degrees while holding the cargo pallet level at the 3 km/h nominal driving speed [5]. Measured field speeds were lower: most distance at Hahamongna was covered between 0.5 and 1.5 km/h, and at Black Point between 1.5 and 3 km/h [9]. Before 2010, driving speeds rarely exceeded 1 km/h.

Source: Wilcox [5]. Public domain (NASA / US government work).

Every joint is the same architecture: a brushless DC motor with a power-off holding brake and an incremental encoder, a planetary gearbox of 40:1 to 81:1, and a CSG-style harmonic drive whose circular spline is the actuator output [5]. Only two motors are used across all seven joints, a low-power unit in hip yaw, knee roll, ankle pitch and ankle roll and a high-power unit in hip pitch, thigh pitch and knee pitch, which minimizes interfaces and makes joints interchangeable. The standard Oldham coupling in the harmonic component set is discarded to save mass and replaced with an involute spline bolted to a custom six-bolt pattern in the wave generator, which requires tight concentricity between the planetary output shaft and the harmonic axis. The harmonic ratchet torque is used as a mechanical fuse: overload causes a non-catastrophic ratchet event rather than a break, at the cost of reduced future torque capability.

Joints are sealed with a two-layer stack. The outer seal is Nomex felt of rectangular section with a height to width aspect ratio of 1.25 to 2.0, running on hard-anodized aluminum [5]. Compressing it 7 to 16 percent of nominal height sealed well but produced high drag torque and forced the seals out of their grooves, because felt hoop strength is very low; 5 to 10 percent compression is sufficient and the seal should be captured or bonded. The inner seal is a spring-loaded PTFE lip seal on a hard-anodized surface, whose radial spring keeps lip force nearly constant through thermal expansion and contraction. The same double-seal approach was used on the Mars Exploration Rovers.

Power and energy

Each Tri-ATHLETE half carries four Valence Technologies lithium iron magnesium phosphate batteries wired in series, a 78 kg pack at 51.2 V and 130 Ah, able to supply 300 A for 30 seconds [5]. Charging is by a pair of 1 kW, 60 V supplies switched by a Valence battery management system that monitors health and state of charge. Wheel drive motors are rated above 750 W each, which is what makes the power take-off useful as a tool drive [7].

Thermal

No thermal design values are published; the prototypes are terrestrial. The structural material choice is explicitly not flight-representative for this reason: 7075-T6 aluminum was selected for low cost, low mass and high strength, and the paper states it is likely inappropriate for a flight system given lunar thermal extremes [5]. 7075-T6 is about 1.8 times as strong as 6061-T6 at up to three times the raw material cost, and given machining time added 20 to 25 percent to total cost per part.

Compute and avionics

Avionics sit in a drop tray that lowers out of the main chassis for access [5]. Each Tri-ATHLETE half carries a 4U cPCI card cage with four Extreme Engineering XCalibur1000 800 MHz PowerPC computers, chosen because a space-flight-qualified computer of comparable capability exists. One computer runs vehicle motion and ground communications; the other three each handle stereo vision processing for four of the twelve camera pairs on the vehicle. Seven of those camera pairs are in the drop tray, along with a 1 W wireless router, a MEMS inertial measurement unit, GPS and a high-resolution tilt sensor [5].

The main computer sends via-point commands to each motor controller over a CAN bus at 8 Hz; the controller is local to each motor and closes a position loop at 2.7 kHz, updating its setpoint and blending smoothly into the new motion profile as via points arrive [5].

Autonomy

Driving was operator-commanded through a joystick interface with a deadman switch, not autonomous [9]. What ran onboard was the active terrain compliance algorithm and, from fiscal year 2010, a 7-DOF inverse kinematics solution built on the earlier 6-DOF solution so that all seven joints move in coordination. The vehicle computes its own per-limb ground contact forces from joint windup and redistributes load without operator input [5].

Two changes to the compliance algorithm were needed for long traverses. Pre-drive reinforcement of wheel pitch and yaw was disabled, because continuous force balancing during driving already reduced uneven loading and the pre-drive step was frequently stalling joints [9]. Unregulated continuous reinforcement also let wheels migrate away from their nominal drive positions, which had to be corrected in the normal course of the traverse.

Communications

Field operation used a 1 W wireless router carried in the drop tray [5]. Traversing beyond the direct communication link required operating the vehicle outside direct operator control [9].

Payload and instruments

ATHLETE carries payload rather than instruments. The cargo interface is the pallet that joins the two Tri-ATHLETE halves, to which the payload is docked; the halves can then undock, leaving pallet and payload behind, and drive off as two three-limbed robots to combine with a different pallet [5].

The tool interface is the second payload path. Each wheel has a quick-disconnect adapter that locks a tool over the rotating power take-off, so the wheel drive actuator powers the tool [7]. Power can be taken mechanically, as when drilling into natural terrain, or through a generator that supplies electrical power to a more complex tool. Tools drawn from a holster include grippers and drills, and the interface has been proposed as a mount for a dexterous anthropomorphic robot [8]. A fold-out high-definition stereo camera pair is exposed when the tool adapter opens to receive a tool [6].

Regolith construction and ISRU tools were studied as a later application: microwave hardware mounted on ATHLETE to heat regolith and drive off volatiles, either radiated from a horn across the surface or through coaxial cables into excavated material, with the vehicle used as a precision positioning platform for additive construction in which regolith is laid down in thin layers and sintered [11].

Modes of operation

Published modes are rolling drive on locked-suspension wheels with active compliance, walking with the wheels locked as feet for extraction from extreme terrain, cargo docking and undocking at the pallet, and tool operation with the vehicle stationary and a limb used as a positioner [8], [5], [7]. The design assumption is that the rolling vehicle gets stuck roughly once per day and walks itself out.

Ground operations

Driving was done through PortOps, a handheld unit whose joystick carries built-in tilt sensors and gives access to a limited command set, with the full ATHLETE command set available through a PortOps laptop; a local operator walks alongside the vehicle [10]. The interface has a loss-of-control safety feature and a deadman switch [9]. Remote operation was used infrequently and its main effect on efficiency was longer pauses between mobility commands while the remote operator worked out a safe course of action. Most commanded drives at Black Point Lava Flow ended in an off-nominal condition that stopped the vehicle and required a new command to be issued, and heading recommand and joystick jitter alone accounted for more than half of all drive terminations [9]. Operational efficiency was measured rather than assumed: excluding stops longer than 10 minutes, more than 30 percent of traverse time was spent idle even during the most efficient driving at Black Point Lava Flow, and idle time was generally over 40 percent. The causes recorded were spurious triggering of the loss-of-control feature, accidental release of the deadman switch when an operator shifted grip, actuator stalls and motor controller errors, and an interface so sensitive to small joystick variations that several command attempts were typically aborted before the intended heading was achieved. Efficiency improved across the campaign as operators gained experience and system faults were fixed.

Technologies developed

The transferable results are the wheel-on-limb mass argument and its supporting model, the force-sensing-free active suspension, and the wheel power take-off.

The mass model in [8] gives mobility subsystem mass fraction against total vehicle mass for ATHLETE against a rocker-bogie baseline, under stated assumptions: 6-strut kinematically determinate linkages in 7075 aluminum, harmonic drive gear set mass fitted as 0.001 times ratchet torque to the power 0.942 and outside diameter as 0.0121 times ratchet torque to the power 0.3075, 60 percent rolling resistance on long slopes in soft regolith, a characteristic density of 26 kg per cubic meter, a thigh fraction of 60 percent of limb length, 0.1 m/s cruise speed and a 300 km design life distance. Under those assumptions the mobility mass fraction in Mars gravity is 20 to 25 percent over a few hundred kg to a few tonnes of vehicle mass, and in lunar gravity 12 to 15 percent from under 1 tonne to several tonnes, with the ATHLETE advantage rising asymptotically above a few tonnes and, on the Moon, above 10 tonnes for mobile habitats. The Apollo lunar rover wheel is used as the wheel mass datum, 5.44 kg for 289 N of rated load [8].

The joint torque estimation method differences a motor-side incremental encoder against an output absolute encoder and applies a gearbox torsional stiffness characterized by applying a known load and reading the difference between the two encoders; the six resolved joint torques plus the limb kinematics give the full wrench at the wheel, so per-limb ground contact force is available without a force sensor, and the incremental encoder can be recalibrated by lifting the limb clear of the ground [11]. That estimate is what the active suspension runs on.

The structural approach was cost-driven and is documented as such: bonded and riveted joints in place of large machined geometries, with Hysol EA 9394 used as a liquid shim between concentric features and radial rivets carrying the entire analyzed load while the adhesive is credited with no shear strength; a riveted sheet-metal-over-rib closed box chassis; and bolted friction joints elsewhere, sized on bolt preload at 50 percent of fastener ultimate tensile strength, or 50 percent of insert pullout, or 57 percent of tapped hole capability, with an assumed friction coefficient of 0.2 and 1200 MPa black oxide socket head cap screws on all critical fasteners [5].

Source: Wilcox [5]. Public domain (NASA / US government work).