Rosalind Franklin

ESA/S. Corvaja.

The problem

Rosalind Franklin is the ESA ExoMars rover, a solar-powered six-wheeled vehicle that carries a 2 m rotary drill and an internal analytical laboratory [13], [11]. It has to drive between sampling sites, acquire cores from depths of 0 to 2 m and deliver them, crushed, to onboard spectrometers and a mass spectrometer without exposing the material to the ambient environment [14], [15]. Mass is about 310 kg, of which 26 kg is instrument payload, a figure that excludes the drill and the sample processing mechanisms [13].

The environment sets four constraints. Lander accommodation capped the wheels at 28.5 cm diameter and 12.0 cm width, so traction on loose soil has to come from wheel compliance and gait rather than size [11]. Communication is two Trace Gas Orbiter relay passes per sol, so autonomy level is set by link availability [13]. Organic and volatile content has to survive acquisition, which puts thermal limits on the drill and on the crusher [11]. The vehicle also has to keep its sample path clean: the laboratory path is sealed at positive pressure until opened on Mars [15].

The rover has not been to Mars. Launch was suspended in 2022 when cooperation with Roscosmos ended and the Russian descent module was lost to the program [6]. ESA is building a European entry, descent and landing module reusing the qualified onboard computer, the radar Doppler altimeter and the parachute system, with NASA providing the launch service, elements of the landing propulsion and radioisotope heater units [6]. The 2004 Phase A concept run by an MDA-led team was a 240 kg rover with a different chassis and is not the flown configuration [2].

What has flown

Nothing of the rover itself has flown. The evidence is flight heritage of related hardware and software, and ground testing of Rosalind Franklin hardware and testbeds. Every item below is a different vehicle unless stated.

Schiaparelli, 2016. The ExoMars EDL demonstrator launched with the Trace Gas Orbiter in March 2016 and did not survive descent, so the landing was never verified in flight and arrival timing was checked against predicted rise time instead [17]. Its DREAMS package returned no surface data [7].

MER navigation, 2004 onward. The Field D* global planner was uploaded to the Mars Exploration Rovers in summer 2006 and checked out five times on Opportunity, the first on sol 1014 [24]. The rover’s PanCam descends from the MER Pancam, a stereo multispectral imager flown on Spirit and Opportunity [3].

Curiosity mobility, 2012 onward. Over 21,318.5 m and 738 attempted drives in seven years, visual odometry converged in 99.55 percent of attempts, with feature-poor sandy terrain as the dominant failure [25]. Curiosity’s wheel skin began cracking about 14 months into the mission; by sol 3195 and 26,256 m, four grousers were broken [21]. Traction control, a proprioceptive wheel-speed algorithm, reduced the resultant load on leading front wheels by 19 percent on average in ground tests [22]. Shedding the unconstrained inner wheel section and driving on the rim was tested on the Scarecrow rover, where the only statistically significant slip increase was 2.95 percent [23]. Thermal inertia from the rover ground temperature sensor cut slip prediction error on sand from 22.0 to 16.8 percent over 135 flight measurements [26].

Perseverance, 2021 onward. The ENav algorithm has flown on Perseverance since 2021, with FPGA-accelerated stereo, wheel-slip bounding-box expansion and verification in Monte Carlo simulation and the JPL Mars Yard [29]. It averaged 92 m/h through sol 1312, against 17.3 m/h for Curiosity [29]. Mars 2020 flight software is also tested in SSDEV, which runs it against commercial avionics so that testing is not limited by scarce flight-like testbeds [28]. During the Jezero rim ascent, wheel slip exceeded 94 percent on a 20 degree slope and the drive was abandoned [30].

Rosalind Franklin’s flexible, independently steered and deployed wheels carry loads along a different path from the Curiosity wheels, and no test of this wheel on sharp outcrop has been published.

What is being tried now

Wheel-walking. A coordinated gait drives the wheel motors and the deployment joints together, so each wheel train is advanced by the knee joint while the wheel turns [1]. Tests reported for the ExoMars configuration showed better dynamic stability during egress, better traction on loose soil and increased gradeability [11]. Status: ground tests and simulation only. A 2005 simulation of the earlier chassis gave 25 degrees climbed with wheel-walking against 18 with normal driving, and 23 against 16 on a mixed drift-cloddy soil [2]. A 2019 simulation on the ExoTeR testbed found normal driving always draws less current, and wheel-walking wins on energy per distance only on terrains with flat slip ratio of about 0.47 or more and rolling resistance coefficient of about 0.21 or more [1].

Locomotion mode selection and slip estimation. Wheel odometry cannot give slip during wheel-walking because the wheels do not roll continuously; an estimator that commands a wheel speed increment and reads wheel train acceleration was proposed and tested in simulation [1]. Slip prediction from Curiosity flight data adds thermal inertia as an input to a slope-only model [28]. Neither approach has been run on Rosalind Franklin.

Rover testbeds. ESA’s ExoTeR (2008 to 2010) and MaRTA (2017 to 2019) are built from commercial parts and are not flight representative; their wheel-walking results are reported only qualitatively, as encouraging or reduced slip [16]. Six force-torque sensors mounted above the MaRTA wheels were field-tested at Bardenas Reales at 1 to 5 cm/s. Drawbar pull could not be read from them because vibration swamped the signal, though terrain classification reached 95.6 percent against 85.8 percent for the chassis IMU, and the authors recommend motor current sensing for slip detection [8].

Soil and gravity for testbeds. The three testbed simulants ES-1, ES-2 and ES-3 were characterized by direct shear and pressure sinkage, and the compacted state of ES-1 reached 1.21 g/cm3 against a 1300 kg/m3 specification minimum, with instantaneous friction angles of 29.5 to 32.3 degrees against a 16 to 24 degree specification [18]. On gravity, parabolic-flight data showed that testing at lunar weight on Earth overstates mobility, with drawbar pull to weight of an ExoMars flexible wheel in ES-2 falling 20 percent on average in lunar gravity at equal load [20]. Granular scaling laws predicted lunar-gravity wheel performance from 1 g tests with a drawbar-pull error of 8.5 percent, against 44 percent for matching cone index [31].

Onboard navigation compute. FPGA stereo and visual odometry co-processors were prototyped on the Athena rover for continuous driving, with a 1024-pixel stereo map in 0.082 s against 24 to 30 s for 256-pixel stereo on the MER RAD6000 [27]. Rosalind Franklin’s onboard navigation design is not published to that depth: ground control designates targets from compressed stereo imagery, and the rover computes its own solution over the resulting digital terrain maps, using inclinometers, gyroscopes and Sun sensors for attitude and heading [13].

Operations rehearsal. The Atacama ExoMars-like field trial of February 2019 drove a rover emulator through a compressed reference surface mission; two cores were analyzed with a portable Raman demonstrator in the field and a Raman simulator in the laboratory [12].

The numbers

Parameter	Value	Source
Locomotion formula	6 x 6 x 6 + 6 (supported, driven, steered, deployed)	[11]
Suspension	triple bogie, no central differential	[11]
Steering modes	point turn, double Ackermann, crab	[11]
Wheel	28.5 cm diameter without grousers, 12.0 cm wide, flexible, grousered	[11]
Average ground pressure	about 10 kPa	[11]
Drill	rotary, no percussion, 0 to 2 m depth, rock up to 150 MPa	[11], [14]
Core sample	about 3 cm long by 1 cm diameter	[14]
Crushed particle size	few to about 500 um, median 250 um	[15]
Nominal surface lifetime	218 sols, 6 experiment cycles, about 1.5 km	[11]
Assumed traverse and drilling rate	about 100 m per sol, 50 cm per sol	[11]
Earliest launch, landing	October 2028, 2030	[6]
Landing site	Oxia Planum, Noachian Fe/Mg phyllosilicates under a dark resistant unit	[10]
Link	two TGO passes per sol	[13]

The reference surface mission is a contractual object: ESA specified that industry deliver a rover able to execute it within the nominal mission duration, with demonstration by simulation and test [13]. It begins with egress and a 10 sol commissioning phase, then a move of about 60 m out of the descent engine contamination zone before the drill and laboratory drawer open, and a drill blank run precedes any biosignature search [11]. An experiment cycle runs seven sols and takes two samples, one surface and one from 1.5 m; a vertical survey drills in 50 cm increments to reach depths of 0, 50, 100, 150 and 200 cm [11]. The drill requirement set counts 7 experiment cycles and at least 2 vertical surveys to 2 m, a minimum of 17 samples [14]. TGO session local time drifts about 30 minutes earlier per sol for three sols and then jumps about 2 hours forward [11].

Payload, from mast to laboratory drawer:

Instrument	Function	Key parameters	Source
PanCam	multispectral stereo panoramic and high-resolution imaging	two wide-angle cameras, 38.3 degree field of view, 11-position filter wheels, 150 mm stereo baseline; high resolution camera 4.8 degree field of view	[5]
ISEM	mast IR point spectrometer, coaligned with the PanCam high-resolution camera	records IR spectra of reflected solar light	[11]
WISDOM	ground-penetrating radar for drill site selection	step-frequency, 0.5 to 3.0 GHz, 3 to 10 m depth range depending on regolith dielectric properties, vertical resolution to 3 cm	[4]
ADRON	neutron detector for subsurface hydrogen	supports drill site selection	[11]
CLUPI	close-up imager on the drill box	focus from 10 cm to infinity, 7 um per pixel at 10 cm, 932 g, under 15 W, 2652 by 1768 pixel sensor	[19]
Ma_MISS	IR spectrometer integrated in the drill tool	images the borehole wall as drilling proceeds	[11]
MicrOmega	near-IR hyperspectral microscope on crushed sample	250 x 256 pixels by 320 spectral steps, 0.95 to 3.65 um, 20 um per pixel	[11]
RLS	Raman spectrometer on crushed sample	532 nm excitation, 50 um spot, about 150 to 3800 cm-1 shift	[11]
MOMA	mass spectrometry of organics, two modes	11.5 kg without margin; 32 single-use ovens; pyrolysis to 850 C; 266 nm UV laser; gas chromatograph 1.6 kg	[9]

MOMA runs either as gas chromatography mass spectrometry, with powder dosed into a single-use oven, or as laser desorption mass spectrometry on powder in the refillable container, and ovens are conserved where laser desorption can answer the question [9]. MicrOmega cube analysis on the rover computer selects grain targets for the two spectrometers without a ground cycle [11].

The drill tool is about 70 cm long, with three 50 cm extension rods that route Ma_MISS signals through every joint, and a backup tool without the spectrometer [14]. Sample transfer passes through a door in the front panel to the sealed laboratory drawer, where a jaw crusher and two piezovibrated dosing stations feed the refillable container or an oven [15]. A jammed drill string can be released by counter rotation, and an emergency ejection unit is the last resort, after which no further samples can be delivered [14]. In Mars chamber tests through ice-lens columns of 0 to 35 percent water content, continuous cutting raised temperature about 20 C, and a variable cutting law cut that to 5 C or less [11].

What is not established

No flight data exist: the rover has not launched, and the figures above come from design descriptions, requirements, chamber tests or simulation [11]. The 25 and 18 degree slope figures come from a 2004 simulation of an earlier chassis that the authors state is not validated against a full-rover test [2]. The wheel-walking switching thresholds were derived on a reduced-scale testbed in simulation with synthetic terrains and do not transfer to named Martian soils [1]. Launch date, landing module and mass budget were set before the 2022 suspension and are being re-baselined [6]. CLUPI figures are pre-flight design values [19]. The Curiosity wheel loads and slip results are for a different rover and wheel [23], [28]. The only EDL flight data of the program are Schiaparelli’s, which do not cover a landing [17]. Array area, peak generation, battery capacity, processor and memory for the rover are not published [13].