Pragyan

ISRO.

The problem

A lunar rover has to move a few tens of kilograms over regolith whose strength is known from a small number of sites, on a power budget set by a panel the size of the chassis, with no battery recharge through night and no hand to free a stuck wheel. The recommended engineering properties of the lunar surface are a compilation of in situ Surveyor, Luna, Lunokhod and Apollo measurements, laboratory tests on returned samples and remote sensing, and its shear strengths are the output of an inversion, not direct measurements of cohesion and friction angle [3]. The wider synthesis draws almost entirely on six Apollo sites and three Luna sites within about 27 degrees of the equator [4], so the 69.4 S Chandrayaan-3 landing site lies outside the range the data cover. The Soviet cone penetrometer record from Lunokhod 1 was reduced to unit penetration resistance for four terrain units and compared with the Apollo 14 simple penetrometer [2]. Gravity changes it again: static cone penetration in CUMT-1 simulant on a magnetic-levitation rig cut average penetration resistance over the top 150 mm by 29.1 percent at 1/6 g and 40 percent relative density, and by 8.8 percent at 76 percent relative density [37].

Dust and temperature add to the soil uncertainty. Dust adhesion on lunar hardware is governed by electrostatic, charge-patch and van der Waals forces, and no direct lift-off force measurement for real lunar dust exists [14]. The coldest south polar surfaces measured from orbit reach 29 K in permanently shadowed craters [16]; the Chandrayaan-3 site is sunlit, and the rover was designed for one lunar day, not for night survival [23].

What has flown

Precedents on the Moon. Lunokhod 1 was a 756 kg, 2.2 m vehicle with eight independently driven 510 mm wheels, each with its own motor, brake and explosive jettison [1]. The Apollo lunar roving vehicle carried a woven wire-mesh wheel whose features each replaced an earlier failure [8]. Candidate lunar wheels had been compared in fine terrestrial sand at five strength levels before that selection, at 1 g and in air [9].

Precedents on Mars. Sojourner and the Mars Exploration Rovers use rocker-bogie linkages. The Sojourner design was chosen by solving the friction coefficient needed to climb an axle-high vertical wall, near 0.8 for six wheels and between 0.78 and 1.4 for four wheels depending on which end meets the wall [5]; the analysis tool was a quasi-static planar model built at JPL [7]. The Mars Exploration Rovers folded and latched the linkage for stowage, with six latches that all engaged on Mars [6]. Sojourner’s wheel dug 0.8 to 1.2 cm after seven revolutions in soft material, and was triboelectrically charged and picked up dust in ground tests [18]. Driving cost on the Mars Exploration Rovers was dominated by autonomy overhead: blind driving ran about 120 m per hour, hazard avoidance 10 to 35 m per hour, visual odometry 10 m per hour and both together 5 m per hour [11].

Other recent lunar landers and microrovers. SLIM landed on 19 January 2024 with a position error of 3.4 m on its first hover image against a 100 m requirement, then ended 60 m east of the target with a 90 degree attitude deviation after one main engine lost thrust, which left the solar cells facing west [33]. The SLIM project review [34] organizes the mission results around precision landing and lander mass reduction, the latter through a 500 N class ceramic combustion chamber and nozzle main engine and flexible thin film solar cells bonded to the exposed structure [34]. The vision-based navigation system on SLIM matched craters during descent and ran obstacle detection at 50 m altitude, and the flight telemetry of 19 to 20 January 2024 is reported alongside Monte Carlo and hardware-in-the-loop validation [35].

Pragyan on Chandrayaan-2 (2019). The first flight unit was a six-wheeled rover specified at 27 kg and 50 W with a 500 m design range, solar powered and able to communicate only with the lander [19]. Lander separation was on 2 September 2019 from a 100 km polar orbit; descent was reported as normal to an altitude of 2.1 km, after which communication was lost and the rover never reached the surface [22]. The 500 m range was never exercised [19].

Pragyan on Chandrayaan-3 (2023). The second unit, 26 kg and 50 W, was carried inside the Vikram lander on LVM3-M4, launched on 14 July 2023 [24], [21]. Vikram landed at 69.367621 S, 32.348126 E on 23 August 2023, and the rover was deployed down a primary and secondary ramp mechanism onto the surface on 24 August [23], [24]. The lander’s design response to the 2019 loss included four throttleable 800 N engines, a Laser Doppler Velocimeter, a Ka-band altimeter and a hazard detection and avoidance camera, while the rover configuration was carried over essentially unchanged [23]. Powered descent began at a 30 km altitude with the four 800 N engines and eight 58 N attitude thrusters [24]. A decoupled quaternion attitude law that separates thrust-axis roll from thrust pointing, proposed so that the hazard camera can be oriented, has been tested only in the authors’ simulator with a 150 degree commanded rotation [36].

Mobility is six grousered wheels; the published primitives are forward drive, reverse drive and in-place rotation about the rover center. Wheel centers are 680 mm apart front to rear and the outer wheel faces are 750 mm apart, so a rotation in place disturbs a circular patch about 1012 mm across [25]. Total distance driven was about 101 m, against the 500 m design range of the Chandrayaan-2 unit [19]. The rover was placed in sleep mode on 4 September 2023 ahead of lunar night, and attempts to re-establish communication around sunrise on 22 September did not succeed [24].

Science: LIBS confirmed sulfur and APXS detected minor elements in the first surface results [24]. APXS excites with alpha particles as well as X-rays, which raises sensitivity to low-Z elements by more than two orders of magnitude relative to X-ray-only excitation [29]. LIBS is a 1.1 kg instrument drawing 1.19 W in steady state and firing 3 to 4 mJ pulses at 205 mm standoff, an order of magnitude below the mass and power of standoff LIBS instruments flown on Mars [26].

Control was open-loop: numbered mobility commands (mobility-ids) were uplinked through the lander by ISRO through ISTRAC, Bengaluru, with the deep-space link at the Indian Deep Space Network [23], [24]. No onboard hazard detection or autonomous path selection is published for the rover; hazard detection and avoidance is attributed to the lander [23].

What is being tried now

Orbital reconstruction of rover traverses (applied to Pragyan, flown hardware). Rover tracks are not directly visible in Chandrayaan-2 Orbiter High Resolution Camera imagery at about 25 cm nadir resolution, but the in-place turn patches raise the local gray count by 15 to 20 counts in a difference image built from October 2021 and April 2024 acquisitions at 9.1 and 10.8 degrees solar elevation [25]. Eight in-place turn locations, carrying mobility-ids M-14 through M-34, were matched to the uplinked commands with positional accuracy within 1 m [25]. A 0.30 m per pixel digital elevation model of the site resolves the rover as a discrete feature 8.3 m from the lander [30]. This is a published result on flown hardware, made with Chandrayaan-2 orbiter imagery. LRO’s narrow angle camera, whose preflight calibration is documented, is the other orbital imager used for placing lunar surface assets [17].

Wheel and soil modeling (ground tests, not flown). A two-dimensional discrete element model selected an 18-lug, 250 mm wheel for a Japanese lunar micro rover concept and was validated on a sloped bed of FJS-1 simulant [10]. Scarab, a 280 kg four-wheel skid-steered prototype for polar cold traps, cut downhill slip from 37 percent to 15 percent with active body roll control in field tests on GRC-1 simulant [12], [13]. All of this is Earth-gravity work.

Terramechanics simulation for lunar wheels (ground, simulation). A multibody dynamics framework with a physics-based terramechanics model has been applied to analyze Yutu mobility performance [31]. A physics-based continuum model of the soil is offered as a faster alternative to semi-empirical wheel-soil models [32]. Both are simulation work; neither is validated against Pragyan telemetry.

Reduced-gravity testing (ground, parabolic flight and levitation). Two Falcon 20 parabolic campaigns found that granular scaling laws predicted lunar-gravity wheel performance within 8.5 percent mean squared percentage error on drawbar pull, against 44 percent for matching cone index gradient in a loosened 1 g soil; the GRC-1 simulant is weaker in lunar gravity than the Apollo in situ soil [40]. Each parabola gives 20 to 30 seconds, so runs are transient [40].

Wheel-soil sensing (designed, never flown). VIPER, a 450 kg four-wheeled rover canceled in 2024, was to measure wheel normal force with torque transducers in its suspension joints and sinkage with an encoder pattern on the wheel rim; the paper lists anticipated data products only [38]. Pragyan carries no such instrumentation in the published record.

Visual-inertial odometry for night driving (ground, field trials). A bird’s-eye-view feature-matching front end was tested on a half-scale prototype of the Endurance rover over five trajectories totaling under 550 m at 0.24 m/s, and lowered the minimum reliable image rate from 0.40 Hz to 0.25 Hz on one night trajectory, a 76.6 percent mean reduction across trajectories; one night run is simulated [39].

Layout tradespace (analysis and open toolkit, not flown). A tradespace analysis of sub-50 kg lunar rovers finds that rigid four-wheel layouts Pareto-dominate six-wheel rocker-bogie designs on smooth regolith, with six wheels entering the Pareto set only once an obstacle-navigation requirement is imposed [27]. Neither [27] nor the ISRO pages publish a Pragyan suspension analysis.

Onboard motion planning (ground and flight-software development). CRESCENT, the trajectory optimizer developed for JPL’s CADRE lunar rovers, runs a two-level planner onboard at 1 Hz and 5 Hz to execute directives from multi-agent autonomy without a per-move uplink [20]. Pragyan flew none of this.

Learned perception on Pragyan imagery (ground processing of flown data). A 2026 pipeline applies CLAHE-enhanced stereo, YOLOv11n object detection and a small neural network for distance estimation to Chandrayaan-3 NavCam stereo imagery, and reports a median depth error of 2.26 cm within 1 to 10 m against ISRO’s CAHV triangulation utility [28]. The reference is another algorithm, not a surveyed ground truth, and the detector was trained on Apollo imagery, not on Chandrayaan-3 images [28]. It has not run onboard.

Landing hazard detection (lander side). ALHAT-style lidar hazard detection was sized by Monte Carlo studies on synthetic terrain [15]; Chandrayaan-3 instead flew a hazard detection camera whose performance is not published [23].

The numbers

Specifications

Parameter	Value	Source
Mass, Chandrayaan-3 unit	26 kg	[23], [21]
Mass, Chandrayaan-2 unit	27 kg	[19]
Dimensions	917 x 750 x 397 mm	[21]
Wheels	six, grousered	[25]
Wheel center spacing, front to rear	680 mm	[25]
Track width, outer wheel faces	750 mm	[25]
In-place turn circle diameter	1012 mm	[25]
Electrical generation	50 W	[23], [21]
Power source	solar	[19]
Payload count	2	[21]
Communications	rover to lander only	[23]
Design range, Chandrayaan-2 unit	500 m	[19]

Mission profile

Parameter	Value	Source
Launch	LVM3-M4, 14 July 2023	[24]
Landing	23 August 2023	[24]
Ramp-down onto the surface	24 August 2023	[24]
Landing site	69.367621 S, 32.348126 E	[23]
Prime landing ellipse	4 km x 2.4 km	[23]
Surface mission life	one lunar day, about 14 Earth days	[23], [21]
Distance driven	about 101 m	[25]
Sleep mode entered	4 September 2023	[24]
Recovery attempts ended	22 September 2023	[24]

Vikram carried a mass of 1749.86 kg including the rover and generated 738 W at winter solstice [21]. The Chandrayaan-3 lander touchdown requirements are vertical velocity at or below 2 m/s, horizontal velocity at or below 0.5 m/s and slope at or below 12 degrees [23].

A 0.30 m per pixel digital elevation model covering 2.18 x 2.24 km around the lander gives a mean slope of 3.61 degrees with a standard deviation of 2.93 degrees and a 95th percentile of 9.25 degrees; within a 250 m buffer the mean falls to 3.10 degrees and the 95th percentile to 7.24 degrees, with 13.7 m of local relief [30].

Thermal and avionics

LIBS is qualified for an operating range of -20 to +55 degrees C and a storage range of -40 to +70 degrees C [26]. The APXS silicon drift detector must be held near -35 degrees C by an integral thermoelectric cooler with an active Peltier controller that maintains the setpoint for ambient temperatures below +30 degrees C and reaches a stable temperature about two minutes after switch-on; the detector is coupled to the metallic portion of the payload assembly and to a radiator plate [29]. A common rover FPGA-based electronics unit services the payloads: for APXS it generates the peak-detector control signals on an event trigger, reads the 12-bit serial ADC output and assembles packets of about 1 KB of X-ray events, and LIBS takes its telecommands over a 16-bit serial interface [29], [26].

Payload

Instrument	Function	Source
APXS	X-ray fluorescence elemental composition, excited by a 244Cm source emitting both X-rays and alpha particles	[29]
LIBS	Laser-induced plasma emission spectroscopy over 220 to 800 nm for Mg, Al, Si, K, Ca, Ti, Fe and volatiles	[26], [21]

APXS uses a silicon drift detector of 30 square millimeters active area and 450 micrometer thickness behind an 8 micrometer beryllium window [29]. A deployment mechanism on the rover bottom chassis steers it from a stowed 0 degree position to a deployed 90 degree position; deployed, the detector stands about 55 mm above the surface and sees a circular field of view about 12 cm across. A calibration target plate on the chassis covers the stowed front face and gives in-flight gain and offset monitoring [29]. The source capsule has a 6 mm active diameter inside an 8 mm disc and a 10 mm encapsulation, shielded by a 2 mm gold cup inside a 1 mm stainless steel outer layer [29].

LIBS parameters, all from [26]:

Parameter	Value
Laser	diode-pumped solid state, 1540 nm
Pulse energy	3 to 4 mJ
Pulse duration	7 ns
Beam diameter	0.8 mm
Instrument to surface distance	205 mm plus or minus 8 mm
Spot size at that distance	about 100 micrometers
Angular field of view	plus or minus 3 degrees
Focusing lens	N-BK7, 200 mm focal length
Collection optics	two-fold lens-mirror, 37 mm entrance pupil, optical gain 0.65
Spectrograph	flat-field, f/2.2, aberration-corrected concave holographic grating
Slit	50 micrometers by 500 micrometers
Detector	UV-enhanced linear CCD, 2048 pixels, 14 by 200 micrometers
Spectral range	220 to 800 nm, resolution better than 1 nm
Mass	about 1.1 kg
Steady state power	1.19 W
Peak power	10.42 W for 20 shots at 5 Hz
Envelope	180 x 150 x 80 mm
Structure	Mg-alloy base, Ti-alloy opto-mechanical mounts

What is not established

No rated drive speed, gradeability, obstacle height, battery capacity, thermal design values for the rover, flight processor or radio link parameters are published for either rover unit [23], [21], [19]. The 500 m range belongs to the Chandrayaan-2 unit, whose lander was lost [19]; the 101 m driven by the Chandrayaan-3 unit was set by the single lunar day, not by a demonstrated mobility limit [25]. ISRO figures disagree across pages (rover 26 kg versus 27 kg, lander 1749.86 kg versus 1752 kg), and the 2.1 km altitude at which Chandrayaan-2 communication was lost has no published cause [22]. No wheel wear, dust or soil-strength measurement from the Pragyan wheels is published; the soil context comes from Earth-based tests, Mars measurements and Apollo and Luna data from other latitudes [3], [4], [18]. The APXS and LIBS composition results are reported by ISRO [24]; their uncertainties are not part of the sources read. Lander verification campaigns (helicopter cold test, tower-crane hot test, leg drop on simulant) are named without data [23].