Mars Science Helicopter (MSH)

Updated 10 July 2026

Mars Science Helicopter (MSH) is a family of autonomous Mars rotorcraft designed for extended aerial science missions, accessing challenging terrains and operating under thin-atmosphere conditions.
MSH studies evaluate diverse vehicle configurations—from mid-air deployed coaxial designs to jetpack-delivered hexacopters—focusing on optimized rotor aeromechanics, payload delivery, and safe landing strategies.
Integrated systems feature innovative terrain-relative odometry, map-based localization, and behavior tree–driven autonomy that together address navigation, environmental sensing, and mission-phase control challenges.

Searching arXiv for papers on the Mars Science Helicopter and closely related Mars rotorcraft navigation, autonomy, and EDL concepts. The Mars Science Helicopter (MSH) is a family of future Mars rotorcraft concepts studied as autonomous aerial science platforms for long-range reconnaissance, access to terrains that are difficult or impossible for landed systems, and operation within Mars-specific entry, descent, landing, navigation, and autonomy architectures. In the literature, the term encompasses multiple related designs rather than a single frozen configuration: a mid-air deployed coaxial science helicopter for highland exploration, a larger jetpack-delivered multirotor concept sized to fit within a heritage aeroshell, and associated onboard navigation, rendering, and autonomy frameworks intended to support fully autonomous operation under thin-atmosphere, delayed-communication conditions (Delaune et al., 2020, Delaune et al., 2022, Pierno et al., 2 Sep 2025).

1. Conceptual scope and mission objectives

MSH studies consistently frame the vehicle as an aerial science asset intended to extend Mars surface exploration beyond the reach of ground-based systems. Early mid-air deployment work defined regional exploration goals of more than $100$ km traverse in the southern highlands, with science centered on stratigraphic access, paleomagnetism, crustal composition, and paleoenvironment. That study specified a minimum “Package A” of $540$ g, consisting of a magnetometer, a near-IR camera, and an Acousto-Optic Tunable Filter spectrometer, and an extended “Package B” of about $2$ kg adding X-ray fluorescence, Mössbauer, and environment and soil sensors (Delaune et al., 2020).

Later work broadened the operational framing from deployment and flight mechanics to persistent onboard decision-making. In the hybrid autonomy study, MSH mission objectives include autonomous search, evaluation, and revisit of science waypoints while maintaining at least one safe landing site “in reserve,” and demonstrating deep-space aerial autonomy under multi-sol mission constraints. That same study reports sorties up to about $1.2$ km total flight distance in Monte Carlo trials, a nominal cruise envelope of $10$–$50$ m AGL, and low-altitude landing-site surveys at about $5$ m (Pierno et al., 2 Sep 2025).

The term “Mars Science Helicopter” therefore denotes a mission class characterized by aerial science, autonomous navigation, and access to terrain such as cliff bases, sediment layers, fracture networks, crater rims, steep slopes, and high-elevation sites. A plausible implication is that MSH is best understood as a systems-level research program spanning vehicle design, EDL, guidance and control, terrain-relative navigation, map-based localization, and onboard autonomy rather than as a single airframe.

Study focus	Representative configuration	Stated objective
Mid-Air Deployment	Coaxial rotorcraft, gross mass $4.14$ kg	Highland science and soft touchdown
Mid-Air Helicopter Delivery	6-rotor-arm hexacopter, $31.2$ kg dry mass	Larger rotors and $5$ kg science payload
Hybrid autonomy framework	Future MSH with stereo camera, laser-rangefinder, onboard computer	Long-range autonomous reconnaissance

2. Vehicle configurations and rotorcraft aeromechanics

Published MSH studies examine materially different rotorcraft embodiments. The mid-air deployment concept proposes a coaxial vehicle with two $540$0-blade rotors, each of radius $540$1 m, gross mass $540$2 kg, rotor solidity $540$3, design thrust-to-solidity ratio $540$4 with maximum $540$5, blade tip Mach number up to $540$6, disk loading about $540$7 kg/m$540$8, cruise speed about $540$9 m/s, range about $2$0 km, hover endurance about $2$1 min, and battery capacity about $2$2 Wh (Delaune et al., 2020). By contrast, the MAHD study uses a stowed $2$3-rotor-arm hexacopter with $2$4 kg dry mass inside a $2$5 m aeroshell and a deployed rotor span of $2$6 m using $2$7 $2$8 m rotors with $2$9 blades each; in that comparison, MAHD is associated with rotor radius $1.2$0 m, range $1.2$1 km, hover $1.2$2 min, and payload $1.2$3 kg, versus an airbag alternative with rotor radius $1.2$4 m, range $1.2$5 km, hover $1.2$6 min, and payload $1.2$7 kg (Delaune et al., 2022).

The aeromechanical difficulty is set by the Martian atmosphere. The design guide on rotor RPM restates the hover condition as $1.2$8 and actuator-disk momentum theory as

$1.2$9

so that

$10$0

It then derives a rotor-frequency scaling model,

$10$1

and the Earth-to-Mars ratio

$10$2

Under the stated example, a helicopter hovering at $10$3 RPM on Earth would require about $10$4 RPM on Mars (Blanco, 2021).

The same guide gives a representative MSH numerical example with $10$5 kg, $10$6, $10$7, chord $10$8 m, $10$9 m, $50$0 kg/m$50$1, and $50$2 m/s$50$3. It computes $50$4 N, $50$5 m$50$6, disk loading $50$7 N/m$50$8, induced velocity $50$9 m/s, induced power $5$0 W, rotor frequency $5$1 Hz, rotor speed $5$2 RPM, $5$3 rad/s, tip speed $5$4 m/s, and $5$5 assuming $5$6 m/s (Blanco, 2021).

A separate full-scale test-environment report formulates the same physical constraint in lift-equation form,

$5$7

and, for a prototype with rotor diameter $5$8 m, $5$9 m, and $4.14$0, solves for $4.14$1 m/s and $4.14$2 rad/s $4.14$3 rpm while keeping tip Mach below $4.14$4 (Afman et al., 2019). This does not define a universal MSH rotor; rather, it illustrates how different concept studies choose different geometry and operating points while remaining governed by the same low-density aerodynamic scaling.

3. Entry, descent, deployment, and Earth-based validation

Two principal Mars EDL architectures recur in the MSH literature: Mid-Air Deployment (MAD) and Mid-Air Helicopter Delivery (MAHD). MAD omits a propulsive or airbag landing system. It uses a $4.14$5 sphere-cone heat shield and a conventional disk-gap-band parachute, reaching a parachute-phase terminal descent speed of about $4.14$6 m/s rather than about $4.14$7 m/s for past missions. At an altitude trigger of about $4.14$8 km MOLA, a linkage extends the helicopter below the backshell skirt, spins up the rotors to design tip Mach number, and releases the vehicle under active control. End-to-end DARTS simulations reported full vertical deceleration in about $4.14$9 m altitude loss, hover around $31.2$0 m MOLA, no recontact with the backshell, and predicted landing accuracy no worse than tens of meters, with operation validated up to $31.2$1 km MOLA (Delaune et al., 2020).

MAHD replaces the lander with a jetpack. In that architecture, the backshell separates and the jetpack decelerates the helicopter from roughly $31.2$2 m/s at about $31.2$3 m AGL to hover at $31.2$4 m AGL, after which the helicopter trims its rotors, the jetpack drops away, and the rotorcraft flies to the surface under its own lift. The stated advantage is packaging efficiency: only MAHD leaves enough room in a $31.2$5 m aeroshell for the largest rotor option, enabling a stated $31.2$6 science payload increase to $31.2$7 kg versus $31.2$8 kg. The same study gives a total EDL stack mass of $31.2$9 kg, with $5$0 kg for MSH dry mass including avionics and margins, $5$1 kg for the jetpack, and $5$2 kg for aeroshell, backshell, and parachute (Delaune et al., 2022).

The MAHD jetpack uses $5$3 Aerojet Rocketdyne MR-107N monoprop thrusters, each with maximum steady thrust $5$4 N and specific impulse about $5$5 s, plus $5$6 propellant tanks of $5$7 L each for about $5$8 kg of fuel. The control system is fully implemented on MSH avionics, with an inner attitude-and-heave loop at about $5$9 Hz crossover and $540$00 phase margin, and an outer lateral-translation loop at about $540$01 Hz crossover and $540$02 phase margin. In $540$03 Monte Carlo runs at $540$04 Hz, fuel use was about $540$05 kg over a $540$06 s powered phase, $540$07 attitude rates at the end were below $540$08/s, and altitude and lateral errors were $540$09 m and $540$10 m respectively (Delaune et al., 2022).

Because Mars flight cannot be reproduced directly on Earth, the literature also proposes a full-scale atmospheric flight experimental environment combining reduced-gravity parabolic flight with an onboard vacuum chamber. That report describes the only known Earth-based test capable of simultaneously reproducing Martian surface gravity $540$11, Martian air density of about $540$12, and full-scale free-flight dynamics. During each “Martian g-maneuver,” a standard freighter would provide about $540$13–$540$14 s at $540$15, while a modular chamber in the cargo bay would be regulated to approximately $540$16–$540$17 mbar. The proposed instrumentation includes an IMU with $540$18 m/s$540$19 acceleration accuracy and $540$20/s gyro accuracy, pressure transducers with $540$21 mbar accuracy, optical motion capture with $540$22 mm position accuracy, an RPM encoder with $540$23 rpm, and $540$24 kHz synchronized logging. Expected uncertainties include gravity drift of $540$25, density variation of $540$26, and chamber edge-effects producing $540$27–$540$28 thrust error (Afman et al., 2019).

A recurring misconception is that Mars rotorcraft development reduces to static vacuum testing. The cited studies instead treat deployment, free-flight separation, rotor–jet interaction, attitude transients, and low-density control authority as central system-level problems requiring dynamic validation.

MSH navigation research separates into terrain-relative odometry during EDL and drift-bounding map-based localization during flight. For EDL, “Structure-Invariant Range-Visual-Inertial Odometry” extends xVIO by tightly fusing a monocular camera, IMU, and a downward-pointing $540$29D laser rangefinder. The target mission segment runs from parachute jettison at about $540$30 km above ground to a $540$31 m hover under near-constant-velocity descent of about $540$32 m/s and lateral winds up to $540$33 m/s. The estimator is designed explicitly for terrain with elevation variations up to $540$34 m and without any planar terrain assumption (Alberico et al., 2024).

The EKF state is partitioned into inertial and visual components, with

$540$35

and SLAM features represented in inverse depth. The key design choice is to initialize a “range-feature” when the corner score at the LRF boresight peaks, using the range observation as a direct depth constraint. This is intended to remove monocular scale ambiguity even under low IMU excitation. In a $540$36-run Monte Carlo evaluation over Valles Marineris terrain, visual-only xVIO diverged in about $540$37 of runs with mean $540$38-velocity error $540$39 m/s, xVIO plus range-facet with a planar model diverged in about $540$40 of runs with $540$41 m/s error, and the range-feature formulation had $540$42 divergence with $540$43 m/s mean error and peak error at or below $540$44 m/s (Alberico et al., 2024).

For long-range flight after EDL, the main problem is unbounded drift in onboard VIO. The proposed remedy is Map-based Localization (MbL), which registers a query image to a georeferenced orbital map. Geo-LoFTR builds on LoFTR by injecting local geometry from a DEM. Given query image $540$45, map crop $540$46, and depth image $540$47, geometry-aware map features are formed by cross-attention between appearance and depth. At runtime, a rough prior with about $540$48 m 2D uncertainty is used to extract a $540$49 km$540$50 search area, split into $540$51 windows with $540$52 overlap; Geo-LoFTR produces up to $540$53 matches per window, retains those at at least $540$54 confidence, back-projects map points to $540$55D, and solves PnP in RANSAC to recover the refined pose (Pisanti et al., 13 Feb 2025).

The quantitative gains reported for Geo-LoFTR are specifically directed at Mars illumination variability. Under matched lighting, the model achieves $540$56 @$540$57 m and exceeds $540$58 @$540$59 m in its best case; under full $540$60 azimuth sweeps it maintains @$540$61 m in the range $540$62–$540$63; across altitude bands from $540$64 m to $540$65 m, the @$540$66 m rate drops by only $540$67 points; and during a simulated Martian day, it leads for nearly the entire interval from LMST $540$68 to $540$69, with only the very low-elevation case at $540$70 favoring fine-tuned LoFTR slightly (Pisanti et al., 13 Feb 2025).

These methods depend on synthetic data because Mars lacks large annotated aerial image corpora. MARTIAN is the rendering framework introduced to close that gap. It ingests a HiRISE stereo DTM at $540$71 m/post and an ortho-projected image at $540$72 m/px over Jezero Crater, imports them into Blender, drapes the orthoimage via UV mapping, assigns a Principled BSDF material in Cycles, and places cameras above terrain using a BVH-based “snap to terrain” ray cast. The world frame is east-north-up with origin at the map center, and each pose is stored as a $540$73 homogeneous transform $540$74. The physically correct rendering equation is stated as

$540$75

with sun irradiance set to about $540$76 W/m$540$77, sun angular diameter about $540$78, and camera positions sampled over altitude $540$79–$540$80 m. Pose annotation is described as having sub-millimeter fidelity in simulation, with no added noise and validation error much less than $540$81 cm between ray-cast and analytic terrain height (Pisanti et al., 28 May 2026).

The framework has already been used to pre-train LoFTR for Ingenuity MbL, where omitting MARTIAN caused an $540$82 drop in localization Acc@$540$83 m, while the MARTIAN-enabled system reached $540$84 Acc@$540$85 m and $540$86 Acc@$540$87 m over $540$88 flights, outperforming SIFT and ORB baselines. The same summary reports that Geo-LoFTR yielded up to $540$89 improvement in Acc@$540$90 m under varied illumination versus vanilla LoFTR, sustained performance across altitudes $540$91–$540$92 m and LMST $540$93–$540$94, and $540$95 successful global poses in zero-shot transfer to Mars2020 LCAM-CTX localization during descent from $540$96 km to $540$97 m (Pisanti et al., 28 May 2026).

A common misunderstanding is that synthetic Mars imagery is useful only for qualitative visualization. The reported role of MARTIAN is substantially more technical: it provides controllable illumination, exact pose annotations, and data already used to bootstrap matchers evaluated on real Mars imagery. For MSH specifically, the framework is stated to adapt directly to operations up to $540$98 m AGL by restricting sampling to $540$99–$2$00 m and generating maps at expected local mean solar times such as morning, noon, and afternoon (Pisanti et al., 28 May 2026).

5. Autonomy architecture and onboard execution

The autonomy study treats MSH as a fully autonomous deep-space vehicle, not as a teleoperated aircraft. The operational driver is Earth–Mars one-way light time of $2$01–$2$02 minutes, combined with thin-atmosphere control sensitivity, uncertain terrain roughness and slopes, and limited onboard energy with nonlinear battery discharge. The proposed architecture is a deterministic Finite State Machine (FSM) combined with Behavior Trees (BTs) (Pierno et al., 2 Sep 2025).

The high-level FSM is defined as

$2$03

with

$2$04

Events include internal BT outcomes such as $2$05 and $2$06, health alerts such as $2$07 and $2$08, and mission conditions such as $2$09. Undefined events produce self-transitions. Within each state, a dedicated BT executes fine-grained actions; the example takeoff BT combines health checks, offboard mode, arming, and a takeoff action on the primary path, with descend–land–disarm as a failover path (Pierno et al., 2 Sep 2025).

The implementation is explicitly sized for flight-qualified embedded autonomy. The paper reports a VOXL2-based payload with stereo camera, laser-rangefinder, and onboard computer, using about $2$10 MB RAM, with VOXL2 CPU load around $2$11, control loop time about $2$12 ms, and event-to-state transition latency about $2$13 ms. Monte Carlo validation in Gazebo Garden plus PX4 and Mars-like terrain comprised $2$14 autonomy-module trials with random health-event injections and $2$15 healthguard trials with random anomalies. The autonomy module completed $2$16 end-to-end, the healthguard achieved $2$17 correct anomaly detection with about $2$18 ms event-propagation latency, and no crashes were reported, though the emergency-land rate increased with mission distance (Pierno et al., 2 Sep 2025).

Field validation used a ModalAI Sentinel drone with VOXL2, Pixhawk 6C Mini, stereo/RGB camera, Lightware LRF, and telemetry link. In $2$19 tethered trials, each health event was sent manually five times and produced $2$20 correct transitions. In free flight over desert-like terrain, a $2$21 m square mission at $2$22 m altitude with $2$23 waypoints was tracked within less than $2$24 m RMS error in $2$25–$2$26 and $2$27–$2$28, and a battery-low event triggered landing-site search followed by safe descent and disarm (Pierno et al., 2 Sep 2025).

This architecture is not itself a navigation algorithm; rather, it is the mission-phase supervisor intended to orchestrate guidance, landing-site search, health monitoring, and contingency handling. A plausible implication is that MSH autonomy research assumes perception modules such as range-VIO and MbL will be embedded inside a broader event-driven control framework, rather than operating as isolated estimators.

6. Atmospheric interactions, environmental uncertainties, and science opportunities

MSH research also treats the helicopter as an atmospheric probe. Wind profiling based on Ingenuity attitude data uses trimmed-flight force balance,

$2$29

together with attitude and ground-speed telemetry to infer ambient wind vectors from $2$30 m to $2$31 m altitude. The resulting estimates are the first winds at such altitudes on Mars. For flights $2$32–$2$33, directions agreed with MEDA when available, especially during in-place hovers, but inferred speeds were larger by factors of $2$34–$2$35. For Flight $2$36, inferred winds reached $2$37–$2$38 m/s with directions near $2$39–$2$40, whereas MCD predicted $2$41–$2$42 m/s from $2$43 and MarsWRF $2$44–$2$45 m/s from $2$46. For Flight $2$47, inferred winds were $2$48–$2$49 m/s on ascent and $2$50 m/s on descent, again with direction deviations up to $2$51 from models (Jackson et al., 2024).

Those discrepancies are treated cautiously. Using Monin–Obukhov similarity, the retrieved winds imply friction velocities of about $2$52–$2$53 m/s and roughness lengths of at least $2$54 m for Flights $2$55–$2$56, which the paper describes as implausibly large for the smooth delta plain. The stated explanations are overestimated wind from steady-state aerodynamic modeling, transient gust response, or sensitivity to rougher upwind fetch hundreds of meters away. The study therefore recommends improved transient flight-dynamics modeling, multi-altitude hovers of at least $2$57 s, and additional meteorological instrumentation such as hot-wire or ultrasonic wind sensors, fast-response temperature and pressure probes, and retention of at least $2$58 Hz attitude and ground-speed telemetry (Jackson et al., 2024).

Electrical interactions constitute a second environmental issue. The triboelectric-charging analysis models the blade leading edge as a linear source of charge current due to collisions with suspended dust grains. The total blade current is

$2$59

Using $2$60 m, $2$61 rad/s, $2$62 mm, and $2$63 fC per grain, the paper estimates about $2$64 nA per blade for ambient dust and $2$65 nA per blade for enhanced dust. Modeling the rotorcraft as a single-node circuit with atmospheric resistance and capacitance gives $2$66 F, $2$67, and $2$68 s (Farrell et al., 2021).

With Martian pressure near $2$69 Torr and a $2$70 cm gap, Paschen-law estimates give $2$71 V and $2$72 V/m, while laboratory measurements cited in the paper suggest $2$73 kV/m. Under ambient dust, the field would grow at about $2$74 kV/m per second and would cross the breakdown threshold in order $2$75 s if there were no dissipation; under enhanced dust, threshold could be reached in about $2$76 s. The paper concludes that takeoff and landing in a rotor-created particulate cloud could trigger local atmospheric electrical breakdown, but that the resulting $2$77A-level currents are not large enough to create a hazard to avionics. It instead frames such flights as a possible in situ experiment on Martian near-surface electrical properties (Farrell et al., 2021).

These environmental studies sharpen two points often obscured in simplified presentations of Mars rotorcraft. First, the atmosphere is not merely thin; it is dynamically uncertain, terrain-coupled, and imperfectly modeled even at meter-to-tens-of-meters altitudes. Second, rotorcraft can serve as atmospheric and electrostatic measurement platforms, not only as transport systems. A plausible implication is that future MSH mission design will need to co-optimize flight safety, navigation robustness, and atmospheric science return.