InSight Instrument Deployment Arm

The problem

A lander arm has to move instruments from the deck to the ground and place them within centimeters of a chosen spot, in a thermal environment cold enough to stress its motors, and without a direct measurement of what its end effector is touching. The InSight arm placed the lander’s seismometer and heat-flow package on the surface of Homestead hollow, a site chosen over three years from an initial 16 candidate ellipses in western Elysium Planitia for its flatness and low rock abundance [4], and later georeferenced to 4.502384N, 135.623447E from post-landing HiRISE imagery [5]. Placement is a flight-software problem as much as a mechanical one: the onboard controller computes the gravity-induced tool-frame deflection from the arm’s own link stiffness and any dangling payload mass, then commands the inverse transform so gravity pulls the tool back to the intended pose, which let the same code run unchanged between Earth testbeds and Mars [1].

The arm began as the Mars Surveyor 2001 lander arm, a four-degree-of-freedom backhoe with graphite-epoxy links, a scoop-and-grapple end effector, and no force-torque sensor [2]. Its grapple was designed to tolerate about 7 mm of arm positioning error when lifting the Marie Curie rover off the deck in 3 cm steps [2]. The version refurbished for InSight replaced the links with titanium and raised the joint torques, and it kept the predecessor’s blind spot: end-effector force is inferred, not measured [18][6].

What has flown

InSight Instrument Deployment Arm, 2018 to 2022

The arm made the first precision deployment and release of instruments by a robotic arm on another planet [1], and the wind and thermal shield it stacked over the seismometer landed within 1.5 cm of its target, a figure obtained after bundle adjustment of the arm camera’s workspace mosaics cut the terrain mismatch between overlapping frames from an average of 5.0 mm to 1.9 mm [18]. The 1 cm positioning accuracy and 1.5 cm instrument position knowledge quoted for the deployment system are requirements it had to meet, not a demonstrated repeatability [6].

From sol 550 the joint heaters were not run at night, and the motors dropped below -90 C against a -60 C qualification floor with no degradation and no operational impediment through the rest of the mission. That is 30 C of margin below a ground qualification limit, established by flying it, on one vehicle over one mission [6].

No contact force the arm applied was measured directly. Every figure comes from reproducing the flight pose in a JPL testbed and reading a three-axis load cell in Earth gravity, and scoop pushes at the same site gave median forces that varied by a factor of three between a flat pose and a vertical-forearm pose [6]. The 80 N average downward force often quoted as the end effector’s capability is a design figure carried over from the predecessor arm [2].

The arm also cleaned the solar arrays. Seven times it dumped soil from its scoop onto the lander deck so wind would carry it across the arrays and abrade the dust off. The campaign gained about 80 W hr/sol in total, roughly 15 percent of array output in that period, and extended the mission by about 242 sols, with about 30 percent of the dumped soil dispersed by wind [6]. The seventh dump, 99.1 cm³ released over the east deck in low wind, dispersed nothing and gave no measurable change, so the gain is not attributable dump by dump [6].

The largest single use was the attempt to rescue the HP3 mole. From sol 87 through 754 the arm ran a testbed campaign to diagnose why the mole would not go deeper and to try pinning, back-cap pushing and scoop pressing [15]. The mole’s recoil-control design assumed hull friction from loose regolith [10]; before landing its hammer strokes had been proposed as a seismic source for mapping subsurface layering [9], and after landing the same strokes became the evidence that a cohesive duricrust was stopping it [16]. Two arm-assisted attempts made the mole reverse and back out, because pressing on the regolith beside the hull is not equivalent to pressing on the hull [17]. A catalog of major anomalies across NASA Mars surface missions places multi-year, never-fully-resolved cases like this one at the far tail of the cost distribution it measures in lost sols [8]. The recovery is covered on the InSight HP3 mole page.

Operations were constrained by the lander, not only the arm. Shortly after landing the team uplinked about 200 sequences and found that the onboard file count correlated with command-and-data-handling boot duration, so file count became a hard operational limit, managed with a high-fidelity onboard file-system model [28].

The arm’s cameras came from one family: an arm-mounted Instrument Deployment Camera with a 45 by 45 degree field of view and a body-mounted Instrument Context Camera with a 124 by 124 degree fisheye, both Mars Science Laboratory flight-spare Hazcam and Navcam units upgraded to color before flight [11]. Ground processing turned their images into mosaics, terrain meshes and placement products [12], and commanding against those products used the stereo and rendered-view visualization line JPL had built up since Sojourner [7].

Earlier and contemporary arms

The Viking Surface Sampler used a furlable boom, twelve feet of tape wound flat on a six-inch drum, carrying a scoop head over a 120 degree arc; every motion was commanded singly and preplanned because of light time, with no feedback in the digging [21]. The Phoenix arm, a 4 DOF 2.4 m arm in the Mars Surveyor 2001 and MVACS line, made 53 dig sequences across 10 trenching areas and delivered 18 samples, and the full 152-sol mission acquired 31 samples to the onboard instruments [19]; it found indurated icy layers 3 to 10 cm down from increased joint torque and motion-impeded events, a qualitative hardness only [20].

Curiosity’s arm carries a drill on its turret. In its first 200 sols the drill placement on the sol 180 mini-hole landed within about 2 to 3 mm of the target picked in an arm-camera image, with a nominal drill preload of 300 N, a position correction made by locating the stabilizer impressions of an earlier preload test [29]. The Perseverance arm departs from the InSight lineage with five degrees of freedom and a six-axis force-torque sensor, and logged 27 arm unstows in its first 210 sols [25].

What is being tried now

Measuring force at the wrist (flown). The InSight approach to the no-sensor problem was a testbed reproduction of flight poses, which bounds force but does not measure it on Mars [6]. Perseverance flew the sensor instead. Its calibration was the hard part: a sensor-level fit reached 3 sigma residuals below 67 N and 7.5 N·m, but bolting on the roughly 45 kg turret introduced scaling, offset and thermal errors, and only a two-stage model met the range requirements [24].

Force-controlled docking (flown). The Mars 2020 corer docks to a rotating dock ring on the arm by Force-Corrected Docking, an iterative loop that reads the force-torque sensor, makes a small correction and repeats until preload is on target and lateral loads are near zero. It is the first planetary docking of a robot arm with a rigid structure [23].

Risk rules for arm placement (flown). Curiosity ran its Slip Risk Assessment Process 1,258 times before putting the arm on the ground with a 93.6 percent go rate; half of 58 no-go outcomes came from a wheel perched on a rock [26].

Soil properties from arm contact (flown data, modeled). InSight’s arm-and-mole record became a soil model [17]; pre-landing laboratory tests on regolith simulants had bracketed friction angle and seismic velocity ranges [3], and a pre-landing compilation of orbital, lander and rover data had forecast a cohesive near-surface layer over looser sand [13]. The same inference has been made from wheel telemetry: motor current and suspension angle from Spirit and Opportunity trenching gave Mohr-Coulomb cohesion and friction angle at Gusev and Meridiani [31]. A method proposed for Perseverance would use the abrading bit for pressure-sinkage and shear tests; it was validated on Earth on a replica testbed in simulant, with 95 percent confidence intervals of about 3.7 degrees on friction angle and 3.3 to 5.2 kPa on cohesion, and has not been run on Mars [30].

Cold-operable lander arms (ground test). COLDArm is a four-degree-of-freedom, roughly 2 m, 40 N tip-force arm derived from the Phoenix and InSight arms and designed to operate below -173 C without heaters, using bulk metallic glass gears, ceramic bearings and a heaterless motor controller; its first results are component-level and system tests on the ground [27]. Project reporting places the arm at TRL 6 [32]. A thermal study of sampling-arm actuators built a baseline actuator to Mars Science Laboratory targets (208 N·m at 2.0 rpm) and correlated a thermal model to within 4 C, while the oscillating heat pipe that would improve it was only simulated [33].

On-orbit manipulators (not comparable). NASA’s catalog of in-space servicing, assembly and manufacturing systems lists free-flying and station-mounted manipulators [14], a different design problem from a one-shot surface placement.

The numbers

Parameter	Value	Source
Degrees of freedom	4: shoulder azimuth, shoulder elevation, elbow, wrist	[1]
Reach	1.8 m	[6]
Link material	titanium	[6]
Payload capacity	9 kg at 1.65 m extension in Mars gravity, equivalent to 33 N	[18]
Peak joint torque	35, 120, 65 and 10.5 N·m, joints 1 to 4	[18]
End effector	scoop with a 250 cm³ chamber, plus grapple	[18]
Qualified temperature range	-110 to +70 C structure and joints; -60 to +70 C motors	[6]

Every row is a design specification or capability, not a measured performance [18].

Soil property at Homestead hollow	Value	Source
Duricrust cohesion, arm campaign summary	2 to 15 kPa	[6]
Duricrust cohesion, cone penetration theory	4 to 25 kPa (friction angle unknown)	[17]
Duricrust cohesion, scoop blade at one spot	5.8 kPa	[17]
Surface bulk density	about 1200 kg/m³	[16]
Penetration resistance	0.5 to 1.2 MPa duricrust; 5.3 MPa below 30 cm	[17]
P-wave and S-wave velocity	114 and 60 m/s	[17]
Bulk modulus	9.84 MPa, uncertainty 6.54 MPa	[17]

The Viking Physical Properties Investigation reported cohesion of 0.7 to 3.0 kPa for drift material, 1.5 to 16 kPa for blocky material and 0.5 to 5.2 kPa for crusty to cloddy material [22]. The InSight duricrust cohesion overlaps the blocky-material range and exceeds the crusty-to-cloddy range at its upper end [22][17].

On sols 1074 to 1075 the arm pushed to four sequential commanded depths from a 1 cm standoff, with current and torque limits set for maximum downward force, and the imprint depth was read from digital elevation models built from IDC images [6].

What is not established

The flight software paper gives no placement accuracy, repeatability, deflection magnitude or trial counts [1]. The one measured placement is a single shield stacking within 1.5 cm [18], and the 1 cm accuracy and 1.5 cm knowledge figures are requirements [6]. No contact force was measured on Mars; every figure is a testbed reconstruction [6]. The warm-below-qualification result rests on one vehicle over one mission [6]. Duricrust cohesion depends on an unknown friction angle, and the scoop estimate is one spot at one site [17]. The mole never reached its target depth, so the soil results stand in for the heat flow result [17]. The wind-dependent array cleaning gain is correlational [6]. The COLDArm and abrading-bit results are ground results; neither has operated on another planet [27][30].