Opportunity

NASA/JPL-Caltech/Cornell Univ./Arizona State Univ. Public domain (NASA / US government work).

The problem

A solar-powered, teleoperated vehicle had to cross Meridiani Planum on a 20 MHz processor [2] without a map of what lay ahead, and keep working through dust, cold and component wear for far longer than the 90 sols it was planned for [1]. Four environmental effects set the requirements.

Meridiani is flat and less rocky than Gusev, which allows long drives, but wind-formed ripples of loose sand are geometrically unremarkable and so invisible to a hazard detector that reasons about shape [11]. Aeolian sand has a cohesion at or below 1 kPa and a friction angle near 30 degrees, against 4 to 25 kPa for cemented material, so the bearing capacity differs by an order of magnitude with no difference in relief [9]. Two dust processes degrade the arrays: suspended dust reduces intensity with optical depth, and deposited dust obscures the cells cumulatively [4]. The temperature swing across a sol commonly exceeds 80 K at low latitude [7]. Data return depends on orbiter relay passes of a few minutes each [24].

What has flown

Opportunity, 2004 to 2018. It landed in Meridiani Planum on 25 January 2004 UTC [1], [16]. Its twin Spirit landed in Gusev crater and operated until 2010; through March 2005 the two had driven 3158 m and 4161 m respectively [2]. The NASA mission page dates the landing to 24 January, which is the Pacific date; dates here are UTC [16]. The end-of-mission report gives Opportunity alone 217,594 images, 41.96 GB over 5830 communication passes, and 45,161.04 m over 1403 driving sols, with last contact on sol 5111 (10 June 2018) [17]. The off-Earth wheeled odometry record is 45.16 km over 14.375 years [10]. Landing site candidates had been compared in 2002 by landing ellipse dimensions, and a revised navigation estimate led to no change in the Meridiani ellipse [32].

Purgatory ripple, sols 446 to 484. On sol 446 the rover executed 50 m of blind driving into what imagery showed as an unremarkable pile of sand; the wheels turned through 50 m while the vehicle advanced about 2 m [2]. Extraction took 39 sols [11]. Commanded 2 m motions with the wheels steered for a 23 degree heading change produced progress on the order of 1 mm on many sols [2]. Driving ended on sol 484 when visual odometry failed to converge three times, because the tracked features had left the field of view [2]. Through sol 2300 (13 July 2010) the rover had driven 21,760 m by wheel-turn odometry, 16,389 m of it after leaving the ripple at sol 511 [11].

Visual odometry and slip checks. MER flew visual odometry on another world for the first time. After each commanded 2 m of driving, it estimated actual motion; if little or none was confirmed more commands were sent, and once it measured non-trivial motion or failed to converge all driving stopped [2]. Blind segments over sand were then capped, initially at 5 m, followed by a 20 cm slip check measured by visual odometry, and checks halted drives before they dug in on sols 501 and 603, so driving resumed the next sol instead of costing 39 [2]. Curiosity later used visual odometry on 87 percent of its drives [27]. Convergence was 95 percent (828 of 875 pairs) on Opportunity and 97 percent (590 of 609) on Spirit, and visual odometry took up to three minutes per step on the RAD6000, so it was commanded selectively: for drives under about 15 m on slopes steeper than 10 degrees, with a dragged wheel, or in sand [2].

Hardware wear. The highest-use wheel drive motor accumulated 92.7 million revolutions against a 10 million cycle qualification, and the left front steering actuator stalled with that wheel toed out 33.3 deg [17]. The right front steering actuator failed on sol 433, leaving the wheel turned inward about 8 degrees, and the instrument deployment device shoulder joint was declared failed on sol 1542, after which the arm was carried in a forward “fishing stow” [11]. Rising right front wheel current, possibly from uneven lubricant, was handled by rotating the wheel back and forth, resting it, and driving mostly backward [11]. A catalog by the operations team records the workaround for each degradation: the Moessbauer cobalt-57 source decayed on a 271 day half-life, raising integration time from 8 hours at landing to 48 hours by 2007, and dust from the 2007 storm obscured all of Opportunity’s Mini-TES signal [19]. Flash sector degradation produced boot failures and data loss, handled by reformatting around bad sectors, until Opportunity ran without usable non-volatile storage [19].

Power and the 2018 storm. The 1.33 m2 array at Meridiani cleaned and dirtied in a seasonally repeatable pattern over eight Mars years and never fell below a dust factor of about 0.43, against about 0.25 at Gusev [18]. Wind gusts repeatedly cleared the arrays; that was not credited in the power budget and accounts for much of the mission duration. The 2018 planet-encircling event reached a peak measured optical depth of 10.8 to 11.1 in eight sols, against 4.72 in twenty-one sols in 2007, and began at Ls 182, the earliest in a decade [20]. Array energy fell from 667 Wh at an optical depth of 0.605 to 22 Wh at 10.8, with battery state of charge at 31 percent against a 40 percent flight rule [18]. The flight team names the absence of an undervoltage lockout between the batteries and the mission clock FPGA as the most likely cause of the loss, since a parasitic load kept draining the cells [20]. More than 1000 recovery commands were radiated before the search ended on 12 February 2019 [17].

The 2007 storm. The June to July 2007 planet-encircling dust storm forced the sequencing team to change planning and power-fault strategies so both rovers survived it; those strategies were then reused in Spirit’s third winter [35].

Science traverse. The Block Island iron-nickel meteorite, about 65 cm wide, was found at sol 1961 [11]. The traverse to Endeavour crater began with the rover about 11 km from the rim at sol 2239; the clay-mineral and hydrated-sulfate identifications on the rim at that date were orbital spectral inferences, not rover measurements [11]. Both rovers together returned more than 300,000 raw images [16]. Surface and orbital observations of fresh craters at Meridiani bracket the latest phase of granule ripple migration at about 50 to 200 ka [34]. Onboard target selection (AEGIS) was uploaded to Opportunity in December 2009 and had sixteen tabulated Mars runs through sol 2550, picking rock targets in navigation camera images and pointing the Pancam at the top-ranked one [33].

What is being tried now

Slip estimation and compensation. A survey of wheel slippage methods covers proprioceptive, exteroceptive and model-based estimation, and torque-based and velocity-based compensation, using the Lunar Roving Vehicle, Lunokhod, Sojourner, MER and MSL slip incidents as motivation [26]. The MER method, visual odometry with a slip check, is a fault-protection threshold on slip ratio, not a soil model; the strength contrast that produces the runaway is not observable from the surface the rover is standing on [5]. Visual odometry remains in service on later rovers: a Curiosity flight change that runs it during motion raised measured average drive rate in the Mars Yard from 51.5 to 77.5 m/hr [27]. Both are flown.

Statistical mobility prediction. This propagates soil parameter uncertainty through a wheel-soil model to bound where the rover ends up, with 2-sigma ellipses of 0.06 by 0.03 m in lunar simulant and 0.16 by 0.01 m in Toyoura sand on a 10 degree side slope; a stochastic response surface took 246 s against 17,526 s for 500-run Monte Carlo [6]. It was tested on a 13.4 kg laboratory testbed and has not run onboard any rover.

Alternatives to Bekker-Janosi terramechanics. Granular resistive force theory and continuum plasticity (material point method) were compared against forced-slip wheel experiments in poppy seeds, sand and Mars Mojave Simulant. Both beat the semi-empirical model on torque and sinkage in poppy seeds, but drawbar pull from all three landed within about 25 percent of measurement, with the traditional model the most accurate [5]. Discrete element simulation of a MER spare flight wheel was fit to single-wheel drawbar and sinkage tests at slip 0 to 0.5 and then used to predict slip of 0.7 to 0.99, where no test data exist [23]. A GPU physics simulator showed that gravity offload by mass reduction gives overoptimistic trafficability, because the soil grains still feel Earth gravity [30]. All of this is ground testing or simulation.

Operations and margin. A catalog of 48 major anomalies on NASA Mars surface missions puts the average loss at 11 percent of prime mission surface duration and 6.6 percent over the full mission [28]. Spirit’s end-of-mission report ties rover life to the product of array dust factor and achievable northerly tilt: production fell from 434 W-h/sol in the first winter to 227 W-h/sol in the third, and the sol 779 wheel failure capped climbable slope near 12 degrees [29].

The numbers

Specifications

Parameter	Value	Source
Mass	185 kg	[1]
Wheels	six driven, four corner wheels steered, 0.25 m diameter, rocker-bogie	[3]
Solar arrays	triple-junction GaInP/GaAs/Ge	[4]
Power subsystem	direct energy transfer bus, GaAs triple-junction arrays, Li-ion, Li-SO2 and thermal batteries, 118 pyro events	[31]
Rover batteries	Lithion 8 and 10 Ah lithium-ion cells	[25]
Processor	RAD6000, 20 MHz	[2]
IMU attitude drift	under approximately 3 deg/h	[2]
Enclosure	composite honeycomb exoskeleton lined with aerogel	[1]
Transponder and SSPA allowable flight temperature	50 C (protoflight 60 C)	[1]
Surface peak daytime temperature	-15 C	[4]

Mission record

Quantity	Value	Source
Planned mission	90 sols	[1]
Last communication	sol 5111, 10 June 2018	[17]
Total traverse	45.16 km over 14.375 years	[10]
Odometry through sol 2300	21,760 m	[11]
Images returned	217,594	[17]
Highest-use drive motor	92.7 million revolutions against 10 million qualified	[17]
Mini-TES	stopped responding sol 2257	[11]
Rock abrasion tool	38 rocks ground by sol 2300	[11]
Meridiani dust factor minimum	about 0.43	[18]

Visual odometry constraints

Constraint	Value	Source
Camera field of view / height	45 deg / 1.5 m	[2]
Overlap between stereo pairs	at least 60 percent	[2]
Maximum drive per step	75 cm	[2]
Maximum heading change per step	18 deg	[2]
Maximum slip ratio measured	125 percent	[2]
Smallest displacement resolved	2 mm	[2]
Drive rate with visual odometry running continuously	approx 10 m/h	[2]

Telecommunications

Link	Value	Source
X-band LGA	7.8125 bps uplink / 10 bps downlink minimum	[1]
X-band HGA	2 kbps / 28.8 kbps at shortest range	[1]
UHF return	128 or 256 kbps per pass	[1]
Relay passes above 20 deg	approx 1.8 per sol per orbiter, 2 to 8 min	[24]
Share of data volume by relay	89 percent at 90 sols, 97 percent by October 2005	[24]
Mars Global Surveyor relay	dropped for Opportunity after sol 171	[1]

Pancam calibration target images (more than 17,000 per rover) were used to derive deposited dust optical depth over the full mission at both sites [21], and the Pancam solar-imaging record gives atmospheric optical depth over 5 Mars years [22].

What is not established

The slip-check strategy is documented as a flight-validated threshold, but no held source gives the soil strengths at the sol 446 and sol 603 sites, so the mapping from observed slip to bearing capacity is not established [2], [9]. Wheel-soil models are compared only against laboratory data: the plane-strain continuum model and extrapolated response surfaces limit what transfers to a flown 0.25 m wheel [5]. Statistical mobility prediction was validated on three runs per soil on one 10 degree slope [6]. Mission-overview results from the extended traverse rest on opportunistic measurements taken after hardware had degraded [11]. The end-of-mission report is a project self-report, and the loss of contact has no root-cause analysis because the vehicle never responded [17]. The undervoltage lockout cause is the flight team’s most likely explanation, not a confirmed one [20]. Workarounds for Mini-TES, Moessbauer and flash degradation are cataloged from the operators’ side, not as measured instrument performance [19]. Dust factor absorbs all unmodeled loss, so cell degradation and dust are not separable in it [18].