Phoenix Robotic Arm

NASA/JPL-Caltech/University of Arizona. Public domain (NASA / US government work).

The problem

A lander arm at the Martian north polar region has to dig, scrape and carry soil to instrument inlets, and it has to do so with no force-torque sensor, a UHF-only relay link and a thermal swing from -90 to -20 C in a CO2 atmosphere at pressures as low as 5 Torr [2]. The Phoenix arm had two requirements beyond those of earlier Mars lander arms. The first was hard icy soil: loose regolith at the site was only a few centimeters deep over ice-cemented material that a scoop blade could not cut, so the arm carried a powered rasp as its primary icy-soil tool [3]. The second was a link that gave the operators few opportunities to correct a bad command. Phoenix had no direct-to-Earth surface link and relied on orbiter relay passes for both commands and data [22]. Joint actuators had to be warmed before each use, and the motors that drove them came from a brush DC lineage whose life limit is brush wear [13].

What has flown

Phoenix arm, 2008. The arm was a four-degree-of-freedom backhoe manipulator descended from the design built for the canceled Mars Surveyor 2001 lander and the Mars Polar Lander MVACS arm [2]. NASA JPL built it with Alliance Spacesystems as arm subcontractor and Honeybee Robotics supplying the rasp [4]. Phoenix landed on 25 May 2008 at 68.16 N, 233.35 E, and the arm made its major operations through sol 150, with last contact on sol 157 [16][4]. It executed 53 dig sequences in 11 areas and delivered 17 samples to instruments: 5 to TEGA, 4 to MECA wet chemistry and 8 to the MECA optical microscope [4]. One of the TEGA deliveries was icy soil. The Snow White trench, about 25 cm wide by 75 cm long and 5 cm deep at most, took 7 digs and 12 scrapes, and Stone Soup was the deepest trench at about 15 cm near the lander [4]. Indurated material lay between about 2 and 10 cm below the surface across nearly the whole workspace, sloping deeper toward the lander, and the ice table in Snow White was about 3 cm down [4]. A rasped icy sample yielded about 2.5 cc from a 4 by 4 grid of 16 rasp holes [4]. The results paper reports 31 samples acquired and delivered over 152 sols, a count the flight operations paper does not match [1]. The arm’s results underpin the site’s habitability case: segregated ice, calcium carbonate at 4 to 5 percent by weight and perchlorate at about 0.5 percent by weight [11]. Modeled at Phoenix summer temperatures, magnesium perchlorate solutions at their eutectic remain thermodynamically stable for a few hours a day, the basis for treating transient brine as possible there [5].

Three of the first four icy-sample attempts failed, for reasons specific to icy material [4]. On the first two, the sample congealed in the scoop during delivery, attributed to solar illumination making it stick during the dump. The third succeeded by scraping rather than rasping, which gave a lower-ice sample that did not congeal, and TEGA confirmed water ice in it. The fourth was rasped with scoop motions sequenced to minimize solar energy, but delivery failed because the target TEGA cell doors did not open fully. Phoenix reached the ice-cemented ground but did not substantially penetrate it [11]. The corrections applied on Mars were operational: acquisition early in the morning, when the scoop is coldest, and sequencing to limit insolation [4]. The failure acted between acquisition and delivery, not during cutting.

InSight Instrument Deployment Arm, 2018 to 2022. The arm reused the Mars Surveyor 2001 design, with four degrees of freedom, 1.8 m reach and no force-torque sensor, and was scoped for 90 sols [19]. It operated for 1442 sols, including HP3 mole assistance by scoop pinning and back cap pushing, tether burial and seven solar array cleaning dumps, and it ran without degradation after sol 550 with joint motors unheated below -90 C, outside their -60 C qualification floor. It placed the wind and thermal shield over the seismometer to within 1.5 cm of the intended position, against a target offset of 5 cm, and contact forces reconstructed from motor current and calibrated in a testbed at matched poses were of order 55 to 77 N, varying strongly with arm extension [19]. With scoop friction and direct arm support the HP3 mole reached 40 cm, and the scoop interactions also gave soil properties [31]. A testbed campaign supported the attempted recovery of the stalled mole from sol 87 through sol 754 [32].

Curiosity arm, 2012. The five-degree-of-freedom MSL arm carries a turret of instruments and the rotary-percussive drill [6][10]. Over the first 200 sols the sol 180 mini-hole drill placement landed within about 2 to 3 mm of the target picked in a MAHLI image [14]. A 2004 JPL survey had placed fixed-base manipulators at the state of practice of the time [15].

What is being tried now

Arm telemetry as a soil instrument. Phoenix reconstructed backhoe forces from joint currents and kinematics. Soil above the icy interface grew stronger with depth, and on sol 116 a 70 N push against the undisturbed wall of the Dodo-Goldilocks trench did not cause failure, which bounds trench-wall cohesion from below [1]. The method has flown since: Mars Exploration Rover wheel trenching gave friction angle and cohesion from motor current and suspension angle, which the authors describe as model-derived rather than shear-cell values [23], and InSight back-calculated cohesion from scoop pushes near the HP3 pit [31]. A proposed Perseverance variant that uses the abrading bit for pressure-sinkage and shear tests recovered friction angle to about 3.7 degrees at 95 percent confidence on an Earth testbed, against a reported uncertainty of about 5 degrees on friction angle and 100 percent on cohesion at Phoenix; it had not been executed on Mars when published [26]. Drill telemetry has a ground-tested counterpart: torque, weight on bit and penetration rate from TRIDENT, the 1 m rotary-percussive drill design VIPER carries, located an ice-rock transition at about 62 cm in an Antarctic campaign [27]. This is a field test, not a flight result.

Cold-operable arms. COLDArm is a four-degree-of-freedom arm of about 2 m with a 40 N tip force, derived from the Phoenix and InSight arms and designed to run below -173 C without heaters, using bulk metallic glass gears and a heaterless motor controller [20]. Component tests reached 1.5 million motor revolutions in vacuum below 100 K, but the arm as tested used heritage harmonic gears and cryogenic system tests were future work [20]. A later project briefing describes it as TRL 6 and reports dynamometer, thermal vacuum, vibration, dust seal and regolith simulant tests [34]. It has been ground tested only. It removes the heater and thermostat burden of the Phoenix thermal design [9].

Autonomous icy-material sampling. The Europa Lander concept used a saw-type excavation tool followed by a separate collection tool [17][35]. Its ARIEL autonomy would choose dig sites on board, replacing the multi-cycle ground process Phoenix used, because the concept allowed only one ground-in-the-loop cycle before excavation [28]. A JPL field demonstration on the Matanuska Glacier in 2022 ran 58 tests, and a three-tier fault response, which reduces preload as spindle current rises and retracts to clear a balled bit, excavated a 27 cm plunge trench with 12 autonomous fault recoveries and no operator input [33]. Most of the mission-level autonomy was exercised in simulation, not on hardware [33]. Nothing here has flown.

Rasp-descended and alternative sampling tools. The Dual-Rasp system for an Enceladus lander uses two counter-rotating rasps that cut the top centimeter of ice and throw the cuttings into a collection cup [29]. A variant adds closed pneumatic transfer, avoiding the open scoop transfer in which Phoenix samples congealed [21]. It was at TRL 5 with end-to-end transfer not yet demonstrated, and its vacuum particle-path test ran in Earth gravity only [21]. The Icebreaker concept pairs a Phoenix-class lander and arm with a 1.2 m rotary-percussive drill, citing roughly 1 m/hr at about 100 W from chamber and Antarctic field tests; it is a proposal and its figures are secondhand [11]. Piezoelectric percussion in a related coring drill used about 220 Wh per meter at full duty against about 350 Wh per meter for rotary-only drilling in 40 MPa gypsum [36]. Frozen lunar simulant strength rises with ice content and falls with temperature, from about 17 MPa at 253 K to about 37 MPa at 77 K for 12 wt% water, which sets the design load for any cold-ice tool [37].

Arm-driven coring and caching. A five-degree-of-freedom arm supplied alignment, preload and feed for rotary percussive coring at 10 N of preload for hole start and 20 N during coring, in field tests at Mono Lake [38]. The first end-to-end robotic demonstration of the chain, from tube placement through coring, sealing and caching, was at TRL 4 [24]. Neither has flown.

Flight software and operations. Arm commands expanded on board into motion sequences, guarded moves and a surface accommodation algorithm are the Phoenix baseline [4]. InSight showed a different limit: the number of onboard files correlated with boot duration on the command and data handling computer, which made file count an operational constraint [30]. A catalog of 48 major anomalies across six Mars surface missions puts the average loss at 11 percent of prime mission surface duration and 6.6 percent over the full mission, with mechanical hardware the largest category at 13 anomalies [25].

The numbers

Specifications

Parameter	Value	Source
Degrees of freedom	4: shoulder azimuth, shoulder elevation, elbow, wrist pitch	[2]
Length	2.35 m	[16]
Link material	aluminum links, titanium end fittings	[2]
Joint actuators	brushed DC motors, planetary gears plus harmonic drive; wrist uses a bevel gear at the planetary output	[2]
Motor type	Maxon RE020 brush DC, copper/graphite/MoS2 brushes to a JPL formulation	[12]
Braking	motor leads actively shorted, rotor captured by magnetic detents	[2]
Position sensing	quadrature encoders at the motor shaft, potentiometers at the joint output	[2]
End effector	scoop with primary and secondary blades, powered rasp, TECP, RAC	[2][3]
Secondary blade	tungsten carbide, on the underside of the scoop, for scraping indurated material	[2]
Rasp drive	brushed DC motor, spur gears plus miter gear pair, 1.25:1 reduction motor to bit	[3]
Design environment	-90 to -20 C diurnal, CO2 atmosphere, pressure to 5 Torr	[2]

Mission profile

Parameter	Value	Source
Landing	25 May 2008, 68.16 N, 233.35 E	[16]
Primary mission	90 sols	[16]
Major arm operations	through sol 150	[4]
Last contact	sol 157	[16]
Dig sequences executed	53, in 11 areas	[4]
Samples acquired and delivered	17: 5 to TEGA, 4 to MECA wet chemistry, 8 to MECA optical microscope	[4][8]
Icy-soil samples delivered to TEGA	1	[4]
Snow White trench	about 25 cm wide by 75 cm long, 5 cm maximum depth	[4]
Icy sample yield by rasping	about 2.5 cc from 16 rasp holes	[4]
Rock push	0.5 m guarded move with the scoop front, sol 117	[4]
Relay passes	860, of which Odyssey provided 71 percent	[16]
Data returned	more than 38 Gb, an average of 242 Mb per sol	[16][22]
Forward link	574 Mb total, 97 percent through Odyssey	[16]

Comparable arms

Arm	Reach	Status	Source
Phoenix	2.35 m, 4 DOF	flown 2008, 157 sols	[16][2]
InSight IDA	1.8 m, 4 DOF	flown, 1442 sols	[19]
COLDArm	about 2 m, 4 DOF, 40 N tip force	ground tested	[20]
Curiosity	5 DOF	flown, placement within 2 to 3 mm over first 200 sols	[14]

Rasp qualification

Ground qualification of the engineering-model ISAD in a thermal vacuum chamber, using JSC Mars-1 with water ice, the same with embedded basalt fragments, and pure ice, measured 60 to 80 percent transfer efficiency [3] moving rasp cuttings from the rear to the front chamber. Average rasp cutting duration was 16 s in icy soil and grew to 37 s after 60 minutes of simulated operational life as the tungsten carbide bit wore [3].

Mechanism, thermal and avionics

The rasp bit sits in a pivoting, spring-loaded housing that retracts when the scoop is preloaded and protrudes during cutting, with a knurled contact plate that grips icy soil [3]. Driving the housing against its lower hard stop once per revolution makes the rasp a vibrator that moves cuttings from rear to front chamber [3]. The thermal requirement was to warm actuators to -55 C within one hour of use (three hours for the wrist), hold them below 85 C through 3.5 hours of daytime operation and tolerate an overnight minimum of -93 C; a mechanical thermostat guarded against a stuck-on wrist heater [9]. Arm software ran on the lander Command and Data Handling computer, with a payload electronics board for motor drive, heater drive, encoder counting and sensor digitization [2]. Processor part and memory size are not published.

Brush breakages in Phoenix qualification life testing at -80 C triggered an independent NASA Engineering and Safety Center review [12]. Environmental loads other than pyroshock were negligible for the brushes: the weakest brush tested had a quasi-static capability of 945 g against a 58 g landing qualification load, and relative brush movement needs peak accelerations above 200 g radial or 75 g axial [12]. Brush motors were chosen over brushless types for Sojourner and the Mars Polar Lander arm because they need no commutation electronics, at the cost of brush wear as the life limit [13]. Testing of 29 Maxon RE016 units at -70 to -80 C returned average lifetimes under 10 million revolutions against 30 to 40 million for Sojourner, which forced a change of motor and a silver-graphite brush material; high-load mechanism life then reached 102 million revolutions at -70 C in 1.0 kPa CO2 [13].

Control and operations

Arm software expands single commands such as dig a trench into motion sequences, tracks time and energy and ends an activity when allocations are exceeded. Contact is handled by guarded moves against motor current and computed joint torque thresholds, used for TECP insertion, rasp load plate preload, sample acquisition and trenching. Faults, meaning hardware failures, are distinguished from events, meaning any other cause of an incomplete command such as a rock in the path [4]. A surface accommodation algorithm adjusts the scoop trajectory when motion is impeded and, if necessary, dumps and re-executes the dig; autonomous recovery from these events was enabled in flight and saved sols that would otherwise have gone to recovery [4].

Ground planning is the other half of the arm’s autonomy. Operations were planned in the Rover Sequencing and Visualization Program, with target designation from panoramas, collision checking and duration estimation [2]. Each delivery used a Robotic Arm Camera image overlaid on a Payload Interoperability Testbed image so the operator could correct positioning before the dump [4]. Before terrain contact the camera imaged the three lander footpads [2], and it imaged the MECA and TEGA inlets before and after delivery [7].

What is not established

• Sample counts. The flight operations paper reports 17 samples delivered [4], while the results paper reports 31 acquired and delivered over 152 sols [1]. Neither reconciles the two.
• Quantified arm performance. The flight operations paper gives no placement accuracy, repeatability, force or torque figures, reports hardness as motion-impeded events rather than measured resistance, and gives no denominator of attempts for delivery success except for icy soil [4].
• Soil strength. Strength values are inferences from motor current and imagery, not penetrometer or shear-cell data, so their uncertainty is set by the force reconstruction [1][23].
• Brush motor root cause. The NESC review did not establish a root cause for two broken-brush motors found without shock exposure, and its 2600 G component-level shock figure comes from MER-era tests on a shared motor lot rather than Phoenix hardware [12].
• Ground-test fidelity for icy transfer. The rasp transfer efficiency and cutting duration figures [3] come from a thermal vacuum chamber. The congealing failure appeared only on Mars, and no cited ground test reproduced Mars pressure, sample temperature and solar flux together.
• Successor performance. Every figure cited for Icebreaker, Dual-Rasp, COLDArm, ARIEL and the Matanuska demonstration comes from proposals, development reports or Earth tests; none has been flown on an icy target [11][20][21][28][33].
• Processor and memory. The arm’s computing resources are not published [2].