Spirit

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

The problem

A Mars rover has to cross unknown terrain, place instruments on rock and soil, and survive on solar power, with a radiation-hardened flight computer, a link to Earth that is mostly relayed through orbiters, and no possibility of repair. Spirit, formally MER-A, was designed for 90 sols and 600 m of traverse [6]. The environment attacks that requirement three ways. Loose soil and slopes turn commanded distance into wheel slip. Dust settles on the solar array and sets the energy available each sol. Mechanisms age: every actuator has a qualified revolution count, and the mission is long enough to exceed it.

What has flown

Spirit (2004 to 2010). Spirit launched on 10 June 2003 and landed in Gusev crater in January 2004 after an airbag landing at over 80 km/h and 40 g [1]. It was last heard from on sol 2210 after driving 7730.50 m, more than 12 times the planned 600 m, and returned 21.57 GB, 52.8 percent of it science images [13]. The rover is a six-wheel rocker-bogie vehicle of 185 kg on 26 cm wheels [28], [11], powered by a 1.33 m2 triple-junction array [19] on a 20 MHz RAD6000 with 128 MB RAM and 256 MB flash [23]. Its design cost, with Opportunity through the 90-sol prime mission, was $820 million [5].

Mobility defined its history. The right front drive actuator drew nearly 1 A against a typical 0.4 A before it stalled on sol 779, and Spirit then drove backward for years with the wheel dragging; the loss cut effective thrust by an estimated 17 percent [15]. The dragged wheel trenched the soil and exposed light-toned sulfate-rich and silica-rich material at eight locations, including a Gertrude Weise patch near 90 percent SiO2, the highest silica then reported on Mars [11]. Later spectroscopy placed opaline silica at 65 to 92 wt percent SiO2 around Home Plate [26]. After the failure the maximum climbable slope was 12 degrees, which limited the northerly tilt reachable for winter [13]. Drive actuators on Spirit accumulated up to 18.747 million motor revolutions against a 10 million qualification, where the prime mission was expected to use 2.5 million [13]. The rover was embedded in soft sand near Home Plate from about March 2009, and on 26 January 2010 NASA redesignated it a stationary science platform after extraction attempts failed [2].

Power set the end. Energy production at Gusev fell from 900 W-h/sol at sol 1 to 434 in the first winter and 278 in the second, with dust factors of 0.71, 0.53 and 0.35 and northerly tilts of 13.5, 11.5 and 28.8 degrees across the three winters; a wind cleaning near sol 420 raised output from about 400 to over 700 W-h/sol [13]. The project’s assessment was that dust factor times tilt sets the life of a solar rover, and no telemetry exists from the fourth winter [13].

Opportunity (2004 to 2018). The twin drove 45,161.04 m over 5111 sols, with a longest single-sol drive of 219.89 m on sol 410, and ended when the 2018 planet-encircling dust event cut array output below what communication needs [14]. That storm reached a peak optical depth of 10.8 to 11.1 in eight sols, against 4.72 in twenty-one sols in 2007 [20]. Opportunity’s minimum dust factor was 0.43 against Spirit’s 0.25, because Meridiani cleaned more often than Gusev [19].

Flown autonomy. Visual odometry converged 97 percent of the time on Spirit and 95 percent (828 of 875) on Opportunity, measured slip ratios up to 125 percent on a slope above 25 degrees, and took nearly 3 minutes per step on the RAD6000 with steps limited to about 75 cm [10]. The autonomous targeting software AEGIS picks rock targets in navigation images and points the Pancam; it was uploaded to Opportunity in December 2009 and ran within the same 20 MHz, 128 MB budget [23]. Ground-derived slip tables held for pure upslope and cross-slope travel but failed on mixtures of bedrock and loose sand [10].

Communications and operations. Through September 2005, 93 percent of Spirit’s data went through Mars Odyssey, 4 percent through Mars Global Surveyor and 3 percent direct to Earth [1]. Relay passes averaged about 56 Mb per sol per rover through Odyssey [16]. The operations architecture ran two shifted teams on Mars local time with a tactical cycle each sol and a two-week strategic plan [6]. Mission clocks drifted to 21 minutes 5 seconds on Spirit by February 2010, which corrupts sun-based attitude [22]. By 2014 the team had cataloged each failure with its workaround; dust obscured about 80 percent of Spirit’s Mini-TES signal after the 2007 storm, and the pre-drive actuator heating setpoint was raised from -20 C to 0 C to recover lubricated actuators [15]. In 2011 plan persistence moved to cloud services after saving a 300-tuple plan took almost 8 s for a remote user [35].

Successor rovers. Curiosity’s wheels, with 0.75 mm skin against MER’s 3.52 mm, had four broken grousers by sol 3195 and 26,256 m [28]. The mechanism is the all-wheels-same-speed control mode meeting pointed rocks, which loads a leading wheel to as much as twice its static load [32], [40]. Curiosity used visual odometry on 87 percent of its drives and averaged 46 m/hr in full VO mode [27]. Perseverance’s ENav, helped by an FPGA for stereo, used autonomous navigation for 90 percent of 32.1 km through sol 1312 at an average 92 m/h, against 17.3 m/h for Curiosity [29].

What is being tried now

Faster onboard autonomy (flown). MER stereo took 24 to 30 s for a 256-pixel-wide map and visual odometry 160 s on the RAD6000; a Virtex 5 FPGA implementation ran a 1024-pixel-wide stereo map in 0.082 s and was tested on a prototype rover at 6.44 cm/s [24]. The FPGA approach is what Perseverance flew [29]. Curiosity’s thinking-while-driving change runs VO in parallel with motion and was measured against a 51.5 m/hr Mars Yard drive rate without it [27].

Onboard global localization (flown). Spirit’s position error accumulated with distance. Perseverance repurposed the Ingenuity base station processor to match post-drive orthomosaics to HiRISE, with mean error of 0.36 m on 264 panoramas, after a 3-sol drive had grown 32.92 m of position uncertainty with no global fix [30].

Slip prediction from environment (flown data, ground model). On Curiosity, adding thermal inertia to a slope-only model cut slip error from 22.0 to 16.8 percent RMS over 135 samples at nine sandy sites, where measured slip ranged from 15.4 to 83.4 percent [31]. A survey reports support vector machine classification of slip into low, moderate and high from proprioceptive data as the best of the algorithms compared [37]; the corpus holds no flight use of it.

Wheel-soil modeling (ground-tested). Cohesion and friction were first derived on Mars from Spirit and Opportunity wheel-trenching telemetry: friction angles of 30 to 37 degrees and cohesion of 0 to 2 kPa from seven trenches [25]. Resistive force theory and continuum plasticity were benchmarked against forced-slip wheel experiments including Mars Mojave Simulant and agreed with drawbar pull to about 25 percent [3]. Discrete element simulation of a MER spare wheel matched drawbar pull to 6.1 percent at slip 0.7 but not at slip 0 [18]. A stochastic response surface propagates soil uncertainty to a confidence ellipse and ran 246 s against 17,526 s for Monte Carlo on a 13.4 kg test rover [8]. None flew on Spirit.

Parked-rover safety (flown). Curiosity’s Slip Risk Assessment Process cleared 93.6 percent of 1258 evaluations, and half of the 58 no-go cases were a perched wheel [34].

Power (flown). InSight’s arrays lost 0.2 percent per sol to dust and sat at a dust factor near 0.25 after one Mars year, the Spirit minimum [33], [19]. Dust optical depth on Spirit’s calibration target peaked at 1.5, equivalent to about 29 micrometers [17].

The numbers

Parameter	Value	Source
Planned surface life and traverse	90 sols, 600 m	[6]
Wheel diameter	26 cm	[11]
Mass	185 kg	[28]
Solar array	1.33 m2, triple junction	[19]
Flight computer	RAD6000, 20 MHz, 128 MB RAM, 256 MB flash	[23]
Last communication	sol 2210	[13]
Odometry	7730.50 m	[13]
Data returned	21.57 GB	[13]
Single event upset warnings	23	[13]
Drive actuator revolutions, maximum	18.747 million	[13]
Minimum dust factor	0.25	[19]
Data via Odyssey, to Sept 2005	93 percent	[1]
Microscopic Imager scale and depth of field	30 micrometers per pixel, plus or minus 3 mm	[7]
Pancam, Navcam, Hazcam resolution	0.28, 0.77, 2 mrad per pixel	[4]
Rock abrasion tool cut	5 mm deep, 4.5 cm wide	[4]

Energy by winter	W-h/sol	Dust factor	Tilt
Sol 1	900
Sol 90	600
Winter 1	434	0.71	13.5 degrees
Winter 2	278	0.53	11.5 degrees
Winter 3	227	0.35	28.8 degrees

All rows are from [13].

What is not established

Spirit’s loss is attributed to dust and tilt, not diagnosed, because no telemetry exists from the fourth winter [13]. The right front failure has a suspected cause, lubricant maldistribution, and no confirmed one [15]. The 17 percent thrust loss has no derivation in its source [15]. Wheel slippage in the Columbia Hills mission overview is described as excessive with no percentages [4]. The two silica figures, 90 percent at Gertrude Weise and 91 percent in an outreach deck, are from different sources and are not reconciled [11], [5]. Wheel-soil models disagree with measurement across the slip range, none ran onboard, and all tests cited are one-gravity tests on simulants [3], [18]. The soil-derived cohesion values rest on seven trenches [25]. Later-rover figures come from different vehicles, processors and terrain and bound what is possible without restating what Spirit did [29], [27].