In-Space Systems: A Practical Reference

Rendezvous and proximity operations means flying one spacecraft up to another and holding it there without hitting it. Orbital mechanics makes the intuitive approach wrong: thrusting forward raises your orbit and slows you down, so a chaser closes on its target by flying below it and letting the shorter orbit period do the work. The drift is worth memorizing, because it sets every timeline. A chaser one kilometer lower gains about 9.4 km of along-track distance per orbit, since the along-track drift over one revolution is 3π times the altitude difference. ESA's ATV was injected roughly 100 km below the International Space Station, phased up to a waypoint 39 km behind and 5 km below it, and then flew a sequence of ground-authorized hold points down to contact at 7 cm/s with 1.5 cm accuracy. The last stretch is flown inside a cone centered on the target's docking axis: Shuttle approaches used an 8-degree corridor that tightened to 5 degrees, with hold points where the vehicle could stop and wait for sunlight, a ground station pass, or a decision from the crew. Everything is built around one rule, called passive safety. At every point on the approach, if the vehicle stopped maneuvering entirely, its unforced motion has to stay outside the keep-out volume for the next 24 hours.

Passive safety is nearly free to design in, because it is a choice about where to put the waypoints rather than hardware you have to buy, and it converts the worst realistic failure from a collision into a slow drift-away. That is why ISS approaches are routine now instead of exceptional. What it costs is time and propellant. Hold points, lighting constraints, and the requirement to be in view of a ground station stretch a rendezvous that is physically achievable in one orbit into hours or days, and the vehicle has to carry propellant for retries it will usually not need. The rule also only covers the failures that were modeled, and the case that breaks it is a thruster that fires and stays on, which no amount of trajectory design will make passively safe. The answer is separate hardware: ATV carried a collision-avoidance sequence on a control chain fully independent of its main navigation system, so a fault in the primary computer could not disable the escape. The target pays too, because a station has to hold a specified attitude and stop maneuvering during the approach window.

If the target is attitude-controlled and carries retroreflectors or a docking target, approach along the R-bar from below, since gravity gradient decelerates you and you spend less propellant braking and put less thruster plume on the target's solar arrays. If plume impingement is not a concern and you want a shorter approach, a V-bar approach along the velocity vector is simpler to fly but needs active braking the whole way. In GEO, hold station on a natural-motion safety ellipse rather than a fixed offset, because the relative motion carries you around the target with no station-keeping burns and a stuck thruster leaves you circling instead of closing. If the target is a crewed station you do not get to design your own approach rules, so budget from the start for the visiting-vehicle requirements, including an abort path on independent hardware. If the target is tumbling and has no cooperative aids, the assumptions here stop holding and the problem becomes the one in entry 3.

Along-track drift of about 9.4 km per orbit per km of altitude difference · ATV phasing waypoint 39 km behind and 5 km below the ISS · 8-degree approach corridor narrowing to 5 degrees on Shuttle approaches · contact at 7 cm/s to 1.5 cm accuracy · passive safety required for 24 hours after an abort · roughly 58 m/s to close a 100 km altitude gap (derived here, not published).

The Shuttle program approved the direct +R-bar approach in April 1994, first flew it on STS-66 that November, and used it for every Mir docking, with rendezvous and docking on flight day 3. ATV flew five fully automated approaches to the ISS between 2008 and 2014 using relative GPS and an independent collision-avoidance system; Soyuz and Progress now use a two-orbit profile that reaches the station about three hours after launch; and Northrop Grumman's MEV-1 ran the same discipline in GEO in 2020, approaching Intelsat 901 on a slow, ground-supervised profile.

The cost of rendezvous is not in the hardware. A lidar and a camera set are tens of kilograms and tens of watts on a vehicle that weighs a tonne or more, and the money goes into guidance software, the Monte Carlo and hardware-in-the-loop campaigns that qualify it, and the flight controllers who staff the approach. Propellant is the other line item, and most of it goes into the climb rather than the docking. Closing ATV's 100 km altitude gap costs about 58 m/s, which is half the orbital speed times the fractional change in orbit radius (7.75 km/s ÷ 2, times 100 km / 6,690 km); that arithmetic is done here rather than looked up. On top of it sits the reserve held for aborts and retries, which is mass that would otherwise have been cargo. Almost all of it is paid by agencies today: NASA's first Commercial Resupply Services contracts were $1.6B to SpaceX for 12 missions and $1.9B to Orbital Sciences for 8, roughly $130M and $240M per delivery, and rendezvous is the capability that turns a launch into a delivery. The cost falls with repetition rather than with technology, since the ISS visiting-vehicle requirements are published and the fifth vehicle to meet them spends far less on verification than the first. The commercial version of this question is being reopened in GEO, where a servicer that can approach a satellite it did not design is selling the approach itself rather than the payload it carries. What that service costs and whether enough customers exist is covered at the mission level on the space-launch-and-satellites sheet.

6 minGetting to the Space Station - RendezvousSimply Space · 10k+ views

18 minRendezvous and Proximity Operations Fundamentals - Tech TalkAnsys Government Initiatives (AGI) · 10k+ views

History of Space Shuttle Rendezvous (NASA Johnson Space Center) · ATV flight phases (ESA)

Relative navigation is the chain of sensors that tells a chaser where the target is, and no single sensor covers the whole approach. It starts with GPS: both vehicles fix themselves to roughly ten meters, exchange those fixes over a direct radio link, and difference them, which cancels most of the common error and gives a relative position good to a meter or better from tens of kilometers in. ESA's ATV flew this way down to 249 m, then handed over to a videometer that pulsed a laser at retroreflectors on the Russian docking port and a telegoniometer that measured bearing, and closed the last stretch at 7 cm/s to 1.5 cm accuracy. Outside low Earth orbit, or against a target with no GPS receiver, the chain instead runs camera to lidar to camera again: a visible camera gives bearing only from hundreds of kilometers, a lidar gives range and then a three-dimensional point cloud from a few kilometers in, and inside roughly 50 m the job becomes six-degree-of-freedom pose estimation, which means matching a model of the target to what the sensor sees. NASA's Raven experiment flew exactly this stack on the ISS from 2017, carrying a visible camera, a long-wave infrared camera at 8–14 µm, and a flash lidar that pulses a 1572 nm laser and times the return on a 256 by 256 detector array at up to 30 Hz, all on a two-axis gimbal. Over two years it watched about 50 approaches and departures by Progress, Soyuz, HTV, Cygnus and Dragon, running model-based pose estimation on camera imagery and iterative closest point matching on the lidar point cloud, and fusing both in a navigation filter.

Against a cooperative target the problem is close to solved, because retroreflectors and a published docking target turn pose estimation into geometry with known landmarks, and the accuracy needed at contact is not extreme: the International Docking System Standard accepts 0.10 m of lateral misalignment and 4 degrees of angular error at first touch. Against an uncooperative target every one of those aids disappears and the sensor has to work out both what the object is and where it is. Lighting is the first problem, since a low Earth orbit spends roughly a third of every 90-minute revolution in eclipse and a passive camera sees nothing there, which is why Raven carried an infrared camera that images the target's own thermal emission and a lidar that supplies its own light. Glare is the second, because multi-layer insulation is specular and crumpled, so a camera looking at a sunlit satellite gets moving highlights rather than stable features. The third is that the machine learning methods that do best on this problem are trained on synthetic renderings and lose accuracy on real imagery, a gap that is the main obstacle to flying them.

If both vehicles are yours and both carry GPS receivers, use relative GPS for everything outside a few hundred meters, because it is cheap, works in eclipse, and needs no line of sight. If you are approaching a target you did not build, plan on a lidar, since it is the only sensor that gives range directly and works regardless of sun angle, and accept that it will be the most expensive item in the suite. Add a thermal camera if any part of the approach happens in eclipse and you cannot afford to wait for sunrise, and add a visible camera regardless, because operators want to see what the vehicle sees. Do not plan a mission around monocular pose estimation on an unknown target unless you can tolerate losing the solution and backing out, since the published accuracy figures come from synthetic test sets. Above GEO or in cislunar space, GPS is not available at all, so the chain starts at the camera and the whole budget shifts onto optical.

GPS fixes to roughly 10 m each, differenced to a meter or better relative · ATV handover from relative GPS to optical at 249 m · contact at 7 cm/s to 1.5 cm accuracy · IDSS capture envelope of 0.10 m lateral and 4 degrees at first contact · Raven flash lidar at 1572 nm on a 256 by 256 array at up to 30 Hz, plus an 8–14 µm infrared camera · about 50 ISS approaches and departures observed over two years.

ATV's videometer and telegoniometer pair, flown five times to the ISS between 2008 and 2014; NASA Goddard's Raven, hosted on the ISS ExPRESS Logistics Carrier as part of STP-H5 from 2017; the TriDAR scanning lidar flown as a Shuttle detailed test objective; Astroscale's ADRAS-J, which switched from angles-only navigation to model-matching navigation as it closed on an H-2A upper stage in 2024; and Starfish Space's Otter, which navigates on cameras and software rather than carrying a lidar.

Cameras are cheap and lidar is not, and that single split drives the sensor budget. A visible or thermal camera can be a modified commercial part, which is what Raven did, reusing flight-spare cameras and a flight-spare flash lidar alongside a commercial infrared camera and gimbal to hold the cost of a technology demonstration down. A space-qualified lidar is a different item: it carries a pulsed laser whose lifetime is counted in shots, radiation-tolerant timing electronics, and a detector array with almost no commercial equivalent, and the supplier base is a handful of companies. NASA's response was to write a common sensor specification in 2014 and fund industry to build to it, on the argument that one specification amortized across servicing, asteroid capture, and landing missions removes the non-recurring engineering that each mission was otherwise paying separately. The commercial bet running against that is software: if pose estimation on a plain camera becomes reliable enough, the lidar comes off the vehicle and the capability moves into code that costs nothing to copy onto the next spacecraft. That bet has not been settled, and the reason is the synthetic-to-real gap rather than anything about the cameras.

5 minTech on Deck: Raven (Ep. 4)NASA's Exploration and In-space Services · under 1k views

Raven: An On-Orbit Relative Navigation Demonstration Using International Space Station Visiting Vehicles (NASA Goddard Space Flight Center) · A Survey on Deep Learning-Based Monocular Spacecraft Pose Estimation: Current State, Limitations and Prospects (arXiv)

A non-cooperative target is an object that will not help you catch it: no grapple fixture, no retroreflectors, no working attitude control, and usually no telemetry about what it is doing. What decides whether it can be caught at all is its rotation rate, because a rigid capture requires the chaser to match that rotation first and then absorb the target's angular momentum through the arm. The standard breakdown treats anything under 5 deg/s as slow tumbling, which is roughly the maximum relative rate a current robotic manipulator is expected to handle; 5 to 18 deg/s as medium; and 18 deg/s and above as fast, the rate beyond which matching the motion at all is considered very difficult. Rate matching is expensive for reasons that have nothing to do with the arm: a chaser spinning at the target's rate has its solar arrays and antennas sweeping past the Sun and the ground, so power and communications drop in and out, and its own reaction wheels saturate against a disturbance that never stops. Real targets sit across the whole range. ESA's Envisat, the roughly 8-tonne reference target for most European debris-removal work, rotates at about 2.19 deg/s with a 164.5-second period, which is slow enough to be interesting and heavy enough to be hard. Once contact happens, the momentum has to go somewhere, and since a spacecraft reaction wheel typically stores tens of newton-meter-seconds while a tumbling multi-tonne body carries hundreds, the difference comes out of thrusters and therefore out of propellant.

The case for solving this is that it is the only way to reach the objects that matter, since the derelict rocket bodies and dead satellites driving collision risk are exactly the ones nobody prepared. The problem is that every capture method has a narrow window. A manipulator with a gripper is the preferred answer when the target has a dedicated grappling feature and is stable to medium tumbling, and it becomes markedly more complex and less safe without one. Clamps around a launch adapter ring suit rocket bodies, which have few appendages to snag; nets and harpoons work from a stand-off distance and suit flat surfaces on payloads rather than curved rocket skins, but they are single-shot and a miss leaves the target spinning and the servicer with nothing. Contactless methods (ion plumes, eddy-current brakes, electrostatic tractors, laser ablation) are the only candidates above 18 deg/s and none of them has flown. The honest status of the whole category is that as of 2026 no spacecraft has captured a fully uncooperative object, and every orbital capture so far has involved a target that was either purpose-built for it or still under attitude control.

If the target has a grapple fixture or a docking plate, use an arm, since that is the case the hardware and the flight heritage actually cover. If it is a spent upper stage with no fixture, plan on a clamp around the payload adapter ring or the interstage, because the geometry is known from the launch vehicle user's guide even when the object is not. If it is a payload with solar arrays and booms, expect the appendages, not the tumble rate, to be the constraint that kills a clamp approach. If measured rates exceed roughly 5 deg/s, budget a separate detumbling step before capture rather than assuming the arm will absorb it, and treat any concept that skips this step as unproven. Above 18 deg/s, stop planning a contact capture. And if you are building the satellite rather than chasing one, put a docking plate on it, because a few kilograms at build time is far cheaper than any of the above.

Slow tumbling below 5 deg/s, the rate a current manipulator is expected to handle · medium tumbling 5–18 deg/s · above 18 deg/s rate matching considered very difficult and contact capture unproven · Envisat at about 2.19 deg/s on a 164.5-second period, roughly 8 tonnes · ADRAS-J closed to 15 m on a 3-tonne, 11 m long H-2A upper stage without capturing it · ESA's €86M ClearSpace-1 service contract, with launch now planned for 2029.

Flown: RemoveDEBRIS fired a net and a harpoon in 2018, both at targets it carried up and released itself; Astroscale's ELSA-d magnetically captured a client satellite in 2021 that was built with a ferromagnetic docking plate, and its fully autonomous attempt was cut short by thruster problems; Northrop Grumman's MEV-1 docked to Intelsat 901's apogee engine nozzle in 2020, a satellite never designed for docking but still holding attitude; and Astroscale's ADRAS-J worked its way in on an H-2A upper stage through 2024, reaching 50 m in May, flying around it in July, and closing to 15 m in November while photographing the payload adapter fitting, without touching it. Not flown: ADRAS-J2, planned for fiscal 2027, which will try to grab that same adapter fitting with an arm; ESA's four-arm ClearSpace-1, retargeted from a Vega adapter to the 95 kg PROBA-1 satellite and now scheduled for 2029; and every contactless detumbling method proposed so far.

How many objects actually need removing, and what a removal ought to cost per object, is covered on the space-launch-and-satellites sheet; the question here is whether the grab works at all. This is agency-funded work, and the buyers are visible: ESA bought ClearSpace-1 as an €86M service contract rather than a procurement, and JAXA funded ADRAS-J as Phase I of its Commercial Removal of Debris Demonstration with a capture mission to follow. The cost sits in verification more than in hardware, because contact dynamics between two free bodies in six degrees of freedom cannot be tested at full scale on the ground, so programs pay for air-bearing tables, robotic hardware-in-the-loop rigs that fly two manipulators against each other, and long simulation campaigns, and they still fly a one-shot mechanism the first time it ever meets a real target. Capture hardware is also mission-unique, since a clamp sized for one launch adapter fits nothing else, which prevents the amortization that made rendezvous sensors cheaper. The arithmetic that everyone eventually reaches is that prevention costs almost nothing by comparison: a docking plate adds a few kilograms and a small fraction of a satellite's build cost, and it moves a future client from the uncatchable column to the routine one. Astroscale sells such plates, and constellation operators buying them is a better indicator of where this market goes than any single removal contract.

3 minTackling Space Junk with a High-Tech Harpoon and NetThe Wall Street Journal · 5k+ views

2 minNets and harpoons: how to clear up space junk | Transformative TechFT Tech 4.0 · 5k+ views

ClearSpace-1 (ESA) · Space Debris Ontology for ADR Capture Methods Selection (arXiv / Acta Astronautica)

Docking is when an active vehicle flies into a mating interface under its own power, as opposed to berthing, where a robotic arm places an inert module into the interface for it. Two mechanism families do the flying case. Probe-and-drogue puts a probe on one vehicle and a receiving cone on the other, which is light and simple and has flown on Soyuz and Progress for decades, but the two halves are not interchangeable, so a probe vehicle can never mate with another probe vehicle. Androgynous designs give both sides identical hardware that can act as either the active or the passive half, which costs mass and complexity and buys the ability to mate any port to any port. The International Docking System Standard, published and maintained by the ISS partners and now at Revision F, is the androgynous case written down: three inward-pointing guide petals on a soft capture ring, twelve pairs of hooks for the structural mate, and a defined capture envelope. Docking happens in two stages under that standard. Soft capture catches the vehicle and kills the residual motion, using a ring mounted on six electric linear actuators arranged as a Stewart platform, which in the NASA Docking System actually extends toward the incoming vehicle to meet it rather than waiting to be pushed; hard capture then drives twelve structural hooks that pull the two tunnels together, compress an elastomeric seal, and turn the joint into a pressure-tight structural interface, after which motorized umbilical connectors mate power and data.

The reason to split capture into two stages is that they are sized for completely different loads. The soft capture system is rated for about 3,900 N of tension and 2,800 N·m of bending at its mating plane, while the hard capture interface mated to the ISS carries 17,700 N axially and 68,700 N·m of bending, so the ring is a shock absorber and the hooks are the structure. Making the soft capture ring actively reach out is what lets a light vehicle dock to a heavy station gently, because earlier mechanisms relied on the chaser's thrusters to push the ring into engagement and therefore needed enough closing energy to drive the latches. The cost of that design is a new failure mode: the actuators are efficient enough that unpowered they offer almost no resistance, so a power or avionics loss in the middle of docking leaves the chaser drifting toward the station with no way to stop it, which is why the NASA Docking System carries a redundant avionics string that can be brought up mid-dock. The wider weakness is that a docking port is expensive in a way a grapple fixture is not: it needs a structural load path, a seal, hooks with specified stiffness, connectors, guides that work across the whole capture envelope, and a way to undock after a failure.

If crew or pressurized cargo has to pass through, you need a docking or berthing port with a hatch and a seal, and for anything visiting the ISS or Gateway that means building to IDSS. If the payload is cargo, you already have a robotic arm, and the module is large, berth instead of docking: an arm places the module with no closing velocity, so the structural loads are lower and the hatch can be much larger than a docking tunnel allows. If you are mass-constrained and will only ever mate with vehicles you also build, probe-and-drogue is lighter and the heritage is unmatched, but accept that no one else's vehicle can help yours. If you are designing a satellite that might be serviced later, do not fit a docking port, because a servicing client needs a grapple fixture or a docking plate rather than a pressurized mechanism, which is a far smaller and cheaper thing. And if you are approaching an existing satellite that has neither, this entry does not apply and entry 3 does.

IDSS initial contact conditions of 0.05–0.10 m/s closing, 0.04 m/s lateral, and 0.20 deg/s in pitch, yaw and roll · capture envelope of 0.10 m lateral and 4 degrees angular misalignment at first contact · three guide petals and twelve hook pairs, with up to 24 hooks engaged on a fully androgynous pair for extra load capacity · soft capture rated to 3,900 N tension and 2,800 N·m bending · hard capture mated to the ISS at 17,700 N axial and 68,700 N·m bending · design cases spanning visiting vehicles up to 25 tonnes mating with a 350-tonne station.

Soyuz and Progress still fly probe-and-drogue with the Kurs radio system; APAS was the androgynous mechanism used for Shuttle-Mir and Shuttle-ISS dockings, and the IDSS traces much of its geometry to it. NASA's own IDSS implementation, the NASA Docking System Block 1, completed qualification in 2017 and was the first docking system NASA had developed since the Apollo-Soyuz Test Project; the two International Docking Adapters installed on the ISS in 2016 and 2019 converted Shuttle-era ports to the standard, and Crew Dragon and Starliner use them. Europe's International Berthing and Docking Mechanism is a separate IDSS-compatible implementation built by Sener and QinetiQ under ESA contracts. The berthing counterexample is the Common Berthing Mechanism, which Canadarm2 uses to attach Cygnus and the Japanese HTV.

The mechanism is a small share of a vehicle's mass and a large share of its schedule risk, which is why NASA went from Apollo-Soyuz in 1975 to a qualified new docking system in 2017 without building one in between. Most of the money goes into qualification rather than parts, because contact between two free bodies cannot be tested at full scale in 1 g, so programs build six-degree-of-freedom robotic docking rigs that fly one mechanism against another on industrial robot arms and then correlate the models against them. Publishing the standard is the economically interesting move. An interface definition document that anyone can build to means a station operator is not locked to one vehicle supplier and a vehicle builder is not locked to one station, which turns docking hardware into something closer to a catalog part and is the main reason the ISS partners published IDSS instead of keeping bilateral interface agreements. Whether that holds is the open question for commercial stations: if the new platforms all accept IDSS ports, the qualification cost is paid once and spread across every vehicle, and if each builds its own interface, every operator pays that bill again. Today the buyers are still agencies, and the only commercial demand is from vehicles whose customer is an agency-funded destination.

7 minHow the NASA Docking System WorksSimply Space · 100k+ views

11 minWhy The Docking Adapters On The Space Station Are Shaped OddlyScott Manley · 1m+ views

International Docking System Standard (IDSS) Interface Definition Document (IDD) Revision F (NASA) · Reference Guide to the International Space Station, Utilization Edition (NASA Johnson Space Center)

A space manipulator is a long, slow, very strong arm that moves payloads, holds spacewalkers, grapples arriving cargo vehicles, and increasingly does fine work with tools. Canadarm2 is the reference design: 17 m long, 1,497 kg, seven degrees of freedom with three joints in the shoulder, one in the elbow and three in the wrist, and every joint able to turn 270 degrees in each direction. It has an identical Latching End Effector at both ends, so either end can be the base, and it walks end-over-end across the station by latching onto power data grapple fixtures that supply it with power, a data connection and video. It handles loads up to 116,000 kg, which sounds impossible for a 1.5-tonne arm until you notice the speeds: 37 cm/s unloaded, 15 cm/s carrying a load while supporting a spacewalk, and 2 cm/s carrying a load under ground control. Nothing has weight up there, so what the joints have to do is arrest momentum rather than hold up a mass, and moving 116,000 kg at 2 cm/s gives 2,320 kg·m/s, which takes only about 230 N at the tip to stop over ten seconds (that arithmetic is done here, not published). The other thing that separates a modern arm from the original 410 kg, six-degree-of-freedom Shuttle Canadarm is sensing: force-moment sensors give Canadarm2 a sense of touch and let it back off when it feels something it should not, and it runs automatic collision avoidance against a model of the station.

An arm is the general-purpose option, which is why it survives: one Canadarm2 berths cargo vehicles, moves spacewalkers, carries Dextre for fine work, and has been repaired in orbit twice, once when a wrist roll joint was replaced in 2002 and again when both end effectors were swapped out in 2017 and 2018. The costs are mass, power, and delay. At 1,497 kg an arm of that class is more than half the dry mass of a typical commercial servicing vehicle, which is why free-flying servicers carry arms of a few meters and tens of kilograms instead. Delay is the harder constraint: reflecting contact forces back to a ground operator, which is what would make an arm feel like a hand tool, becomes unstable at even small round-trip delays, so ground control falls back to supervisory or model-mediated schemes where the operator sets goals and the arm executes them. The penalty is measurable. In a controlled test of a teleoperated cutting task, a 4-second delay cut the mean speed from 2.04 mm/s to 1.76 mm/s and raised the path error from 0.92 mm to 1.44 mm, and Canadarm2's own loaded speed drops from 15 cm/s with spacewalkers helping to 2 cm/s under ground control.

If the object you need to grab has a grapple fixture, use an arm, because that is the case forty years of flight heritage actually covers: the end effector closes snare wires around a known pin, the fixture tells the arm where it is, and it also supplies the power and data the arm needs. If the surface is unprepared, budget for machine vision, force-torque sensing, and compliance control, and expect the gripper to be built for one specific feature such as a launch adapter ring or an engine nozzle rather than being general. If the task is a single mating operation you will repeat, a purpose-built mechanism beats an arm on mass, power, and risk, so do not put an arm on a vehicle whose only job is to dock. If the operator will be on the ground with seconds of round-trip delay, design the task as a sequence of autonomous steps with defined abort conditions rather than as continuous joystick control, and plan on a fraction of the speed a crew member would achieve. If crew are present and can supervise directly, you get roughly seven times the loaded speed for free.

Canadarm2 at 17 m, 1,497 kg, seven degrees of freedom, joints turning 270 degrees each way · payloads up to 116,000 kg · tip speed 37 cm/s unloaded, 15 cm/s loaded during spacewalks, 2 cm/s loaded under ground control · the original Canadarm at 15 m, 410 kg, six degrees of freedom and no force sensing · a 4-second delay slowing a teleoperated cut from 2.04 to 1.76 mm/s and raising path error from 0.92 to 1.44 mm · roughly 230 N at the tip to arrest 116 tonnes in ten seconds (derived here).

Canadarm2 and Dextre on the ISS, joined by the European Robotic Arm on the Russian segment and the Japanese JEM remote manipulator on Kibo; the original Canadarm, first flown in 1981 and used through the end of the Shuttle program; and Canadarm3 for Gateway, at 8.5 m and an estimated 715 kg, designed to run primarily autonomously because the crew will usually not be there. The heritage point for free-flying servicers is JAXA's ETS-VII in 1997, a 2,860 kg pair of satellites in a 550 km orbit and the first uncrewed spacecraft to carry a robotic arm. It docked its chaser to its target three times under automatic and remote-piloted control, and it did the first space robot work teleoperated from the ground through a relay satellite, including handling small parts and a propellant replenishment experiment. Orbital Express followed in 2007, and Northrop Grumman's Mission Robotic Vehicle, launched in 2026 with arms developed under a DARPA program, is the current commercial case.

Space robotics is unusual in that the price is negotiated between agencies rather than set by a market: Canada has supplied ISS and now Gateway robotics for decades in exchange for astronaut flights and research access, MDA builds the hardware, and the Canadian Space Agency is the customer. Inside the arm, cost concentrates in the joints, each of which is a motor, a gear stage, a brake, redundant windings, and position sensing qualified for radiation and for thermal cycling between sunlight and deep space, and then in the requirement that every module be replaceable on orbit, which adds connectors, latches, and mass to a part that would otherwise be bolted in. That design choice paid off, since Canadarm2 has now outlived two of its own joints and both of its hands. The commercial servicing market inverts the whole calculation, because a 1,497 kg arm cannot fly on a two-tonne servicer, so the product people are actually building is a two- to three-meter arm weighing tens of kilograms with a mission-specific gripper. That arm gets cheap the same way rendezvous software does, by flying repeatedly on the same vehicle design, and the operator that flies ten servicing missions with one arm design will have a cost structure nobody funding a one-off demonstration can match.

16 minOh, Canadarm - Why NASA Calls On Canada for Robot Arms IN SPACEScott Manley · 100k+ views

4 minHow Canadarm Grabs - Grapple Fixtures and Latching End EffectorsSimply Space · 10k+ views

Canadarm, Canadarm2, and Canadarm3 – A comparative table (Canadian Space Agency) · Teleoperation and Visualization Interfaces for Remote Intervention in Space (NASA Goddard Space Flight Center)

A life-extension vehicle is a tonne-class spacecraft that flies to a geostationary satellite with an empty tank, grips it, and does the station-keeping on its behalf. It can grip an arbitrary client because geostationary satellites all carry the same two leftovers from launch: a liquid apogee engine with an open nozzle pointing aft, and the ring that bolted the satellite to its launch vehicle adapter. Northrop Grumman's Mission Extension Vehicle begins its capture from about 80 m, closes in, pushes a probe up the apogee engine nozzle until it is inside the throat, expands the probe so it cannot pull back out, then winches the client in until stanchions seat on the adapter ring and the joint goes rigid. Neither of those features was designed as a docking port, and that is the whole trick. Once mated, the client's own propulsion stops mattering: the MEV's electric thrusters hold attitude and fly the combined stack, supplying the roughly 50 m/s per year of north-south station-keeping that lunisolar perturbations cost at geostationary altitude plus a few m/s a year east-west. At the end of the contract the servicer pushes the client to a graveyard orbit, undocks, and moves to the next one. Whether enough clients exist to make a market is covered by Satellite servicing and life extension on the space-launch-and-satellites sheet; this entry is about the mechanism and its limits.

The strength is that it works on satellites nobody planned to service. MEV-1 docked Intelsat 901 on 25 February 2020 in a graveyard orbit above the belt, brought it back down to a working slot, and served it for about five years; MEV-2 docked Intelsat 10-02 in April 2021, that time inside the GEO belt itself. Because the servicer takes over attitude control as well as translation, it also rescues a client with a failed thruster or a dead reaction wheel, not only one with an empty tank. The weaknesses are structural. One MEV serves one client at a time, so a tonne-class spacecraft is tied up delivering a few hundred m/s to a satellite of similar mass, which is an expensive way to buy delta-v. The nozzle-and-ring trick also only works where those features exist, so an all-electric bus with no apogee engine, a satellite in low orbit, or anything with hardware closing out the aft end is not a candidate. And the realistic bad outcome during an approach is a collision that destroys two satellites and salts one of the most valuable orbits in use with debris, which is why every approach is slow, ground-supervised, and rehearsed for months.

Buy life extension when three things are true at once: the payload still has paying customers, the orbital slot is worth keeping, and propellant is the only thing that has run out. If a transponder or a power system has failed instead, extension does not help, because nothing flying today repairs electronics. Decide early. A servicer usually needs about a year of electric orbit raising to reach GEO and then more time to drift to your longitude, so this is a decision made three years before the tank empties, not a rescue. If you need more than about five or six years, buy a replacement satellite, because that is the horizon one docking or one pod covers and buying two services in sequence costs more than a new bus. For a defense customer the calculation is different and usually better: the product is sustained maneuver and the ability to relocate an asset, which is worth paying for whether or not the payload brings in revenue.

MEV-1 docked Intelsat 901 on 25 February 2020, MEV-2 docked Intelsat 10-02 in April 2021 · capture initiated from about 80 m · roughly five years of service per MEV docking, up to six years per Mission Extension Pod · about 50 m/s per year of north-south station-keeping · service reported near $13M a year against $300–600M for a replacement satellite and its launch · MEV in the tonne class on a 15-year design life

Northrop Grumman's SpaceLogistics is the only operator with paying commercial customers: MEV-1 with Intelsat 901, MEV-2 with Intelsat 10-02, and the Mission Robotic Vehicle launched on 21 July 2026 carrying three Mission Extension Pods, each rated for up to six years of orbit control once bolted to a client's apogee engine, with the robotic arms built by the Naval Research Laboratory under DARPA's RSGS program and Optus D3 announced as the first pod client. Starfish Space is the main challenger, holding a $54.5M US Space Force contract for a geostationary Otter.

Public reporting put MEV-1's service near $13M a year, so five years runs to roughly $65M against $300–600M to build and launch a replacement geostationary satellite, which is 10–20% of the alternative — arithmetic done here, not a published figure. That is a good trade for the client and an awkward one for the operator, because an MEV costs a few hundred million to build and launch and can bill only one customer at a time, so the operator needs most of a 15-year design life sold to pay it back. The Mission Extension Pod architecture is the answer to that: put the expensive parts, meaning the robot arms, the sensors, and the vehicle that flies to each client, into one reusable Mission Robotic Vehicle, and make the consumable a small pod that costs a fraction of an MEV. The binding constraint is the qualifying client population, which is a few dozen satellites worldwide at any moment and shrinking as high-throughput satellites make fifteen-year-old wideband payloads uncompetitive. Defense buyers are the steadier customer because they will pay for the maneuver capability itself, and the US Space Force's $54.5M award to Starfish Space is worth more than several years of commercial life-extension revenue. What would change the unit economics is refueling: until a servicer can be refilled it is a consumable, and every mission carries the full cost of a vehicle.

10 minA New Satellite Is Preparing To Repair An Old SatelliteScott Manley · 100k+ views

3 minThe Future of Satellite Servicing in SpaceNorthrop Grumman · 10k+ views

In-Space Servicing, Assembly, and Manufacturing (ISAM) State of Play - 2025 Edition (NASA) · Robotic Servicing of Geosynchronous Satellites (DARPA)

On-orbit repair means opening up a spacecraft that is already flying and replacing something inside it. Astronauts have done this well and robots have barely done it at all. The serious record is Hubble: five Space Shuttle missions between 1993 and 2009, all crewed, on a telescope deliberately built to be serviced, with a grapple fixture for the Shuttle arm, handrails, foot restraint sockets, and instruments packaged as orbital replacement units on captive bolts. Servicing Mission 1 in 1993 installed the COSTAR corrective optics and Wide Field and Planetary Camera 2 to compensate for the primary mirror's figure error, and Servicing Mission 4 in 2009 replaced three rate sensor units, all six battery modules, and a fine guidance sensor, and swapped in Wide Field Camera 3, across five spacewalks. SM4 also carried out two repairs nobody had designed for: the Advanced Camera for Surveys and the Space Telescope Imaging Spectrograph were opened and had circuit boards replaced, which meant taking an aluminum front panel off in a pressure suit by removing 111 fasteners of several head types, many of them thread-locked, behind a purpose-built capture plate that trapped each screw as it came free. Robotically the record is much thinner. Japan's ETS-VII flew the first robotic arm on an uncrewed spacecraft in 1997, and DARPA's Orbital Express transferred and installed an orbital replacement unit between two spacecraft in 2007, but those two spacecraft were designed together to make it work.

Repair is the only servicing option that restores capability rather than just extending life, and Hubble is the proof: a telescope that cost roughly $1.5B to build was a partial failure for three years and became the most productive observatory ever flown after one spacewalking crew installed corrective optics. Astronauts are good at this because they diagnose, improvise, and feel what they are doing. The limits are geography and cost. Crew can reach low Earth orbit and nothing else, so every satellite in GEO is out of reach of the only repair method with a track record, and each Shuttle servicing flight cost on the order of $1B. Robotic repair runs into four problems at once: client spacecraft have no grapple fixtures or visual fiducials, their fasteners are staked or lockwired and a single escaped screw becomes debris in a valuable orbit, thermal blankets have to be cut open before anything underneath can be reached, and the operator cannot close a force loop from the ground. Round-trip light time to GEO is about 0.24 s before you add relay hops and ground processing, while contact dynamics need control at tens to hundreds of hertz, so the arm has to be compliant and largely autonomous at the point of contact and the ground can only supervise. That is why every robotic task demonstrated so far was rehearsed for months against a full-scale mockup, and why improvisation is not on the menu.

Repair is worth attempting when one spacecraft carries a large share of a mission's value, the failure is a replaceable box rather than a design flaw, and the orbit is reachable. If you are running a large constellation of cheap satellites, replace the failed unit and do not think about repair, because the cost of a servicing mission exceeds the cost of the satellite. If the problem is an empty propellant tank rather than a broken box, buy life extension instead; it is an operational product and repair is not. If you are designing a spacecraft today and want the option later, spend the mass now on a grapple fixture, a few visual fiducials, and captive robot-compatible fasteners over the boxes most likely to fail, because none of that can be added after launch and its absence is what makes unprepared servicing so expensive. And if you are counting on robotic repair at GEO for a business plan, note that no vehicle has performed one yet; the Mission Robotic Vehicle is the first one in position to try.

Five crewed Hubble servicing missions between 1993 and 2009 · Servicing Mission 4 replaced three rate sensor units, six battery modules, and a fine guidance sensor over five spacewalks · 111 fasteners removed behind a capture plate to reach one STIS circuit board · Hubble built for roughly $1.5B, with each Shuttle servicing flight on the order of $1B · GEO round-trip light time about 0.24 s against contact control loops needing tens to hundreds of hertz · one robotic orbital replacement unit transfer flown, on Orbital Express in 2007

Hubble Servicing Mission 1 in 1993 and Servicing Mission 4 in 2009 are the flight record for crewed repair; Japan's ETS-VII in 1997 and DARPA's Orbital Express in 2007 are the robotic ones. The Robotic Refueling Mission experiments on the ISS between 2011 and 2015 used Dextre and custom tools to cut lockwire and work fill and drain valves designed for a technician on the ground, NASA's OSAM-1 would have refueled and relocated the unprepared Landsat 7 before it was canceled in 2024, and Northrop Grumman's Mission Robotic Vehicle carries the first dexterous arms sent to GEO.

Crewed repair is priced by the ride, and the ride is the whole cost. A Shuttle servicing flight ran on the order of $1B, which was clearly worth it once for a $1.5B telescope that was not working and is not worth it for anything routine. Robotic repair has no price because it has no product: the funded vehicles are government-backed, with DARPA paying for the Mission Robotic Vehicle's arms through RSGS and NASA canceling OSAM-1 in 2024 after cost growth. The cost structure is dominated by non-recurring engineering per task rather than by hardware, since an unprepared repair means designing a tool, building a full-scale mockup, and rehearsing the sequence for months, and none of that amortizes onto the next client unless the next client has the same interface. That is the whole argument for building servicing interfaces into satellites while they are still on the ground, which is the subject of the next entry. Until then the buyers are civil and defense agencies paying for capability demonstrations, there is no commercial repair customer, and the honest read is that this is a research program with two flight results, not a market.

5 minHubble’s Servicing Mission 1NASA Goddard · 50k+ views

7 minHubble’s Servicing Mission 4NASA Goddard · 10k+ views

Astronaut Missions to Hubble (NASA Science) · Fastener Capture Plate Technology to Contain On-Orbit Debris (NASA Technical Reports Server)

Satellite upgrade means changing what a spacecraft already in orbit does, either by replacing its payload with a newer one or by bolting an additional payload onto a bus that is already up there. It has never been done. The motivation is a mismatch in lifetimes: a geostationary bus is designed for 15 years or more while the payload technology it carries turns over much faster, so operators regularly retire satellites whose power, propulsion, and structure are all healthy because the transponders are a generation behind. A spacecraft that could actually be upgraded would need a specific list of things no operational satellite has. The payload has to separate at one structural interface with robot-compatible captive fasteners instead of forty bolts and a wiring harness. Power and data need blind-mate connectors that demate and remate in vacuum after a decade of thermal cycling. The thermal joint has to be breakable and remakeable and still conduct every watt the new payload dissipates into the bus radiator, which a bolted, gap-filled joint does far better than any separable one. Alignment has to repeat well enough that an antenna or an optical bench comes back pointing where the star tracker thinks it does, and the bus has to carry spare power, data rate, and heat rejection reserved for a payload nobody has designed yet. All of that is mass and money spent at build time to buy an option that only pays off if a servicer exists ten years later.

The prize is real. Payload is roughly half the cost of a geostationary communications satellite and the bus plus launch is most of the rest, so reusing a working bus would in principle halve the cost of a capability refresh and remove about three years of build time. Everything else is a weakness. There is no flight heritage: the closest anything has come is Orbital Express in 2007, which moved a battery and a computer between two spacecraft that had been designed together for that demonstration, and the ISS external payload platforms, where payloads genuinely are attached and removed robotically, but only because the station was built around standard interfaces with a long arm, a dexterous manipulator, and a crew on hand. The bus being reused is also old: geostationary solar arrays typically finish 15 years at 75–85% of their beginning-of-life power and the batteries have faded, while a newer payload usually needs more power rather than less. Standards are the other blocker, since Sierra Space's SPDP, iBOSS's iSSI, and SpaceWorks's FuseBlox all exist as servicing interfaces and none of them flies on an operational client, so a servicer built for one is useless against a satellite carrying another.

Do not build a mission plan around this today, because there is no vehicle and no interface with flight heritage behind it. If you need new capability on orbit soon, buy a hosted payload slot on somebody else's satellite before it launches; that is the working version of this idea and it is purchased routinely. If you are designing a geostationary bus now and expect a 20-year asset life, the cheap hedge is the servicing basics rather than a swappable payload: a grapple fixture, a few visual fiducials, and an unobstructed apogee engine nozzle cost a handful of kilograms and keep both life extension and future robotic work available. If you are building a defense system where the threat changes faster than the satellite does, put the reconfigurability in software and in a phased array rather than in a mechanism that has to survive a decade and then work once. And if you are funding interface standards, be clear that the payoff depends on a fleet converging on one of them, which is a coordination problem more than an engineering one.

No payload swap has ever flown · payload roughly half the cost of a geostationary communications satellite, bus and launch most of the rest · bus design life 15 years or more against much faster payload turnover · geostationary solar arrays typically end 15 years at 75–85% of beginning-of-life power · DARPA Phoenix satlets about 7 kg each · three competing servicing interfaces in development, none on an operational client

DARPA's Phoenix program set out to make geostationary hardware reusable and finished without flying a swap, leaving behind a roughly 7 kg modular satlet and the Payload Orbital Delivery canister, a standardized container that rides to GEO on a commercial communications satellite; NovaWurks carried the satlet work forward. Orbital Express's 2007 orbital replacement unit transfer is still the only robotic component handover in space, the ISS external platforms (the ELC sites, the JEM Exposed Facility, and Airbus's Bartolomeo) are the only operational payload-exchange service anywhere, and NASA's AXCIS project is developing the interface and tool components a future servicer would use.

The arithmetic that makes this attractive is simple and the arithmetic that kills it is everything else in the mission. A geostationary communications satellite and its launch run $300–600M together, and roughly half the satellite cost is payload, so on paper a swap saves the bus and the launch. Against that you have to put a servicing vehicle in geostationary orbit, get the new payload up to 36,000 km anyway (a comsat payload is hundreds of kilograms, so that is most of a launch either way), pay the non-recurring engineering for a robotic task nobody has performed, and then accept a ten-year-old bus with a degraded array. Nobody has closed that case. The build-side cost is the part that gets underrated: designing for swap adds mass, connectors, and structure to every satellite in a fleet, and the operator pays that on all of them to hold an option on a service that does not exist. Falling launch prices push the same way, because the cheaper a whole new satellite is to fly, the less a saved bus is worth. The plausible first customers are defense programs that value refresh speed over cost, and the plausible first step is a servicing interface adopted widely enough that one servicer has more than one client.

8 minDARPA Phoenix Satellite ServicingDARPAtv · 50k+ views

Phoenix (DARPA) · Advancement of eXploration Components for In-Space Servicing (AXCIS) Early Career Initiative (ECI) Project (NASA Technical Reports Server)

An inspection spacecraft flies close to another object and photographs it, and two very different jobs share that name. The cooperative case is a small free flyer released from your own vehicle to look at your own hardware. NASA's Seeker 1.0 is the clearest example: a 4.2 kg free flyer with twelve cold gas nitrogen thrusters, a 35 Wh battery good for about an hour, and 5.8 m/s of delta-v in total, released from a departing Cygnus in 2019 to show that a machine-vision guidance loop can find a spacecraft and hold position on it without any cooperative markers. Its inspection requirement, set with the people who would use the images, was to resolve a 64 mm feature at 8 by 8 pixels from 10 m away. The non-cooperative case is flying up to somebody else's satellite or to a derelict and characterizing it, which Astroscale's ADRAS-J did through 2024 against an H-IIA upper stage that had been in orbit since 2009, closing to roughly 5 m. Optically the close-range job is easy and the constraints are elsewhere: the target is lit only by the sun and by Earth so half of every orbit is dark, and the binding resource is propellant, because an inspector in geostationary orbit spends weeks drifting to each new longitude, and getting there faster costs far more delta-v.

Inspection is the cheapest useful thing to do with proximity operations, because nothing has to touch anything. There is no capture mechanism, no docking load path, and no way to damage the client, which is why a 4.2 kg CubeSat can do a useful version of it while a servicer cannot weigh less than a tonne. It also answers questions no ground sensor can: a 1 m telescope working at 550 nm has a diffraction limit near 0.67 µrad, which at geostationary range is about 24 m, so the whole satellite lands inside one resolution element, while an inspector 100 m away with a 10° field across 1,024 pixels gets roughly 0.17 mrad per pixel, or 1.7 cm on the target (both figures derived here from the aperture and the field of view). The weakness is not technical. A close-approach inspector is indistinguishable in hardware from a co-orbital weapon, since the propulsion, the sensors, and the approach profile are the same and only the intent and the last few meters differ. No treaty forbids approaching another operator's satellite, there is no minimum separation and no right of way, so every case is handled diplomatically after the fact: France's defense minister publicly accused a Russian Luch/Olymp satellite of espionage in 2018 after it maneuvered near a Franco-Italian military communications satellite, and the US Space Force objected publicly in 2020 when a pair of Russian satellites shadowed an American reconnaissance satellite, one of which later released an object at high relative speed. That ambiguity is the reason inspection missions attract political attention out of proportion to their size.

Fly a small free flyer when you need to know why your own large spacecraft is misbehaving and you cannot see the relevant surface from onboard cameras; at a few kilograms it is the cheapest anomaly-resolution tool available. If the inspection is a precursor to servicing, do not build a separate inspector, put the sensors on the servicer, because it has to make the same approach anyway. If you want persistent knowledge of somebody else's assets in geostationary orbit, you need a dedicated maneuvering spacecraft with real propellant and a program to pay for it, which in practice means a defense customer. If the question is where an object is and what it is doing rather than what it looks like, buy ground and space-based tracking instead, which is far cheaper than flying to it. And do not plan a close approach to a satellite you do not own without talking to the operator first: nothing legally stops you, but the diplomatic cost is real and consent is how the commercial operators are handling it.

Seeker 1.0 at 4.2 kg, 35 Wh, about an hour of operation and 5.8 m/s of total delta-v · inspection requirement of a 64 mm feature at 8 by 8 pixels from 10 m · ADRAS-J closed to roughly 5 m on a derelict upper stage in 2024 · Mycroft started proximity operations from 3,500 m · a 1 m ground telescope resolves about 24 m at geostationary range against roughly 1.7 cm for an inspector at 100 m (both derived) · GSSAP operational since 2014

NASA's Seeker 1.0 flew from Cygnus in 2019, following AERCam Sprint, a 15.9 kg sphere flown from the Shuttle payload bay on STS-87 in 1997, and Aerospace Corporation's AeroCube-10 pair, in which one 1.5U CubeSat imaged the other. On the defense side, AFRL flew ANGELS and then Mycroft, which began proximity operations from 3,500 m and inspected a defunct AFRL satellite in geostationary orbit, and the US Space Force's GSSAP spacecraft have operated near the geostationary belt since 2014. Astroscale's ADRAS-J is the commercial non-cooperative case, and the Space Force's Tetra-5 mission is the next scheduled multi-spacecraft demonstration.

The cheap end of this category is genuinely cheap. Seeker was built from mostly commercial parts under a streamlined NASA class that fixed cost and schedule up front, and at 4.2 kg it rides as a secondary payload, so the marginal cost of adding inspection to a mission that already exists is small. The expensive end is a defense program with maneuvering spacecraft in near-geostationary drift orbits, and those are not priced publicly. The unit economics of a dedicated free-flying inspector are poor because propellant is the product: once it is spent the spacecraft has nothing left to sell, and there is no refueling service to restock it. That pushes the commercially viable versions toward two shapes, either a hosted sensor that images other spacecraft as they pass and never maneuvers at all, or an inspection capability bundled onto a servicer so one vehicle sells two things. The most likely paying commercial customer is anomaly resolution and insurance: an operator with a $300M satellite that failed to deploy an array will pay for a photograph, and today there is usually no way to get one.

20 minHow Real Satellites Dogfight - Proximity Operations In Space ExplainedScott Manley · 100k+ views

3 minADRAS-J | Active Debris Removal by Astroscale - Japan | Concept of OperationsAstroscale · 10k+ views

Seeker 1.0: Prototype Robotic Free Flying Inspector Mission Overview (NASA Technical Reports Server) · 2026 Global Counterspace Capabilities Report (Secure World Foundation)

An orbital transfer vehicle is a propulsion stage that rides up with its payloads and then takes them where the launch vehicle did not go. The engineering choice is chemical or electric, and the delta-v budget together with how long the customer can wait decides it. An impulsive chemical transfer from a low orbit at 28.5° inclination to geostationary costs about 4.24 km/s in two burns and takes roughly five hours, since half a geostationary transfer orbit is about that long; at 320 s of specific impulse, meaning an exhaust velocity of 3.14 km/s, the rocket equation gives a mass ratio of e^1.35 = 3.9, so 74% of everything that leaves low orbit is propellant. Electric propulsion cannot burn impulsively, so it spirals, thrusting continuously while the orbit rises, and that costs more delta-v: NASA's high-power solar electric study budgets just under 6 km/s from low orbit to geostationary against about 3 km/s starting from a geostationary transfer orbit. But at 2,600 s of specific impulse the mass ratio is only e^0.235 = 1.27, so propellant is 21% of departing mass instead of 74% (both mass fractions are arithmetic done here from the rocket equation). What the customer pays for that is time, and the same study puts the low-orbit-to-geostationary spiral at 6–8 months and the transfer-orbit version at about 5 months. So the two classes sell different products: slow electric tugs do last-mile work inside low orbit, shifting altitude and phase over days to weeks and releasing each customer into its own slot, while chemical kick stages sell a single high-energy delivery in a day. Prices, market size and who is winning are covered by Orbital transfer vehicle on the space-launch-and-satellites sheet.

The strength of an electric tug is that it moves the same mass on a fifth of the propellant, which is why it fits as a secondary payload, and that altitude changes inside low orbit are nearly free: a few m/s buys tens of kilometers, and differential drift then separates satellites in right ascension at no propellant cost at all. Four things work against it. Trip time comes first, because the payload is out of service for months with no revenue during any of it. Radiation comes second and gets underestimated: a slow spiral crosses the Van Allen belts thousands of times rather than a handful, so trapped-proton dose that a chemical transfer absorbs in hours accumulates over months, the solar arrays lose output to displacement damage, and every box has to be qualified to a total dose a direct injection would never impose. NASA's high-power solar electric study treated that as a design driver, deploying its geostationary payloads first specifically to get them out before the dose piled up, and then parking the vehicle in a circular 8,000 km orbit as a deliberate radiation soak equivalent to a full low-orbit-to-geostationary transfer. Third, most electric tugs do not thrust in eclipse, because batteries sized to run the thrusters through shadow cost more mass and money than the schedule is worth, which stretches the trip further. Fourth, plane changes stay unaffordable at any specific impulse, since a 1° change in low orbit costs about 134 m/s (2 × 7.7 km/s × sin 0.5°, derived here), so no tug fixes an orbit you should have launched into. The failure mode that matters commercially is simpler than any of these: the tug sits between a working satellite and its orbit, and a tug anomaly strands the customer.

Buy last-mile transfer when the rideshare goes to roughly the right orbit and you need a specific altitude, a specific phase, or separation from the hundred satellites you launched beside, and you can accept weeks in transit. If you need a different plane or a different local time of ascending node, buy a dedicated launch, because that is an inclination change at about 134 m/s per degree and no tug can afford it. If your satellite already carries electric propulsion, skip the tug entirely, since onboard thrust does the same altitude and phasing job over the same weeks for the price of a thruster you already bought. If the payload is radiation-sensitive, or its design life is short enough that months of transit matter, pay for chemical transfer or direct injection and take the hours instead of the months. For high-energy delivery to geostationary or a lunar trajectory, chemical propulsion is the only thing that arrives in a day, but nothing at the multi-tonne scale has flown yet, so wait for a demonstration flight before putting anything expensive on one.

Chemical low-orbit-to-geostationary transfer about 4.24 km/s in roughly five hours, needing 74% of departing mass as propellant at 320 s specific impulse (derived) · electric spiral just under 6 km/s over 6–8 months at 21% propellant and 2,600 s (derived) · geostationary transfer orbit to geostationary about 3 km/s and roughly 5 months electrically · inclination change about 134 m/s per degree in low orbit (derived) · SMART-1 at 367 kg took about 13.5 months from an elliptical Earth orbit to lunar orbit · 22 ION Satellite Carrier missions flown by March 2026

D-Orbit's ION Satellite Carrier is the flight record for last-mile delivery, with 22 missions by March 2026; Momentus's Vigoride and Exolaunch's Reliant serve the same niche, Firefly's Elytra and Moog's Orbital Maneuvering Vehicle are the launch-adjacent versions, and Impulse Space's roughly 300 kg Mira is the flown chemical vehicle, with the multi-tonne Helios stage still ahead of it. ESA's SMART-1 is the clearest demonstration of what a slow electric spiral actually looks like: a 367 kg spacecraft carrying 19 kg of instruments, launched into a telecom-style elliptical orbit on 27 September 2003 and captured into lunar orbit on 15 November 2004, using lunar resonances and swing-bys to save xenon along the way.

Last-mile transfer is the one part of in-space logistics with real revenue today, and it is still small: deployment sells for roughly $0.5–3M on top of a rideshare seat against $7.5–8.5M for a dedicated small launch, and 22 ION missions by March 2026 make D-Orbit the clear leader. The structural cost problem is that the vehicle is expendable, so its full build cost amortizes across a single mission's customers, and the fix, reuse, requires refueling that does not exist yet. Electric tugs are squeezed from below, because a satellite that carries its own electric propulsion does the same altitude and phasing work for the price of a thruster, so the operators who actually buy tugs are the ones flying buses with no propulsion at all or who need to be on station before commissioning starts. Chemical high-energy transfer has better economics per kilogram delivered, since the alternative is a heavy-lift launch rather than a cheap thruster, but nothing at the tonne scale has flown and a $300M geostationary satellite will not be a new stage's first payload, so that half of the category has to buy its way onto orbit with demonstration flights. Falling launch prices cut both ways: cheaper rideshare enlarges the set of payloads that need a tug, and cheaper dedicated launch takes the top of the market away.

9 minSpace Tug Or NotEager Space · 5k+ views

9 minHow Do Ion Engines Work? The Most Efficient Propulsion System Out ThereFraser Cain · 1m+ views

LEO to GEO (and Beyond) Transfers Using High Power Solar Electric Propulsion (HP-SEP) (NASA Technical Reports Server) · SMART-1 overview (ESA)

Storable propellants are the ones that stay liquid at ordinary spacecraft temperatures with no cooling: hydrazine, MMH, the nitrogen tetroxide oxidizers, and green monopropellants such as high-test peroxide. Moving them between two docked spacecraft is a plumbing and interface problem rather than a fluid mechanics problem. The propellant is already held at the tank outlet by a diaphragm, a bellows, or a surface-tension propellant management device that was fitted for the engine feed system, so it does not have to be settled first; pressurant gas or a pump drives it through a mated coupling into the receiver, and the receiver vents its own ullage gas to make room. The difficult parts are the joint and the bookkeeping: mating two halves of a coupling in vacuum without spilling a toxic fluid, leak-checking the joint before the valves open, and knowing how much propellant actually moved when there is no gravity to make a fuel gauge work. This has been done many times. Russian Progress vehicles have transferred propellant to a station since Salyut 6 in 1978 and have kept the ISS supplied since 2000, at roughly 850 kg of propellant per flight; the Shuttle's Orbital Refueling System moved hydrazine between two tanks six times on STS-41G in 1984; DARPA's Orbital Express transferred hydrazine from the ASTRO servicer to the NEXTSat client at about 492 km in 2007; and in 2025 China's Shijian-25 reportedly docked with Shijian-21 in geostationary orbit and transferred about 142 kg of hydrazine.

The strength is that the fluid itself is undemanding. Hydrazine sits at a low vapor pressure at room temperature, so a sealed line holds it for years, there is no chilldown, no boiloff, and no reason the transfer has to be quick. That is also what makes a storable depot conceivable: propellant launched in 2026 is still propellant in 2031. The weaknesses are handling and measurement. Hydrazine and NTO are toxic and corrosive, so a leak at the coupling contaminates both spacecraft, and active leak monitoring at the interface during a transfer is a real gap, with the ISS ammonia leak detector the only flown hardware that does anything like the job. Quantity gauging is the other one, since propellant mass in a microgravity tank is inferred from pressure, temperature, and integrated flow rather than measured, which matters when someone is buying by the kilogram. And most satellites already in orbit carry a fill-and-drain valve that was safety-wired shut on the ground and never meant to be reopened by a robot.

If you are moving propellant to a spacecraft designed to receive it, treat storable transfer as an integration job, not a research program: the subsystems are mature, and industry gap assessments put what is left in leak detection, gauging, high-pressure compressors, and liquid-free venting rather than in the transfer itself. Fit a standard port at build time (see the refueling interfaces entry) instead of expecting a robot to defeat a safety wire; the robotic route works, and NASA's Robotic Refueling Mission demonstrated every step of it on the ISS, but it adds an arm, a tool set, and hours of teleoperation to a task a port reduces to minutes. If the propellant is liquid oxygen, methane, or hydrogen, none of this carries over, and you should read the cryogenic transfer entry instead, because the fluid mechanics change completely. If the goal is a small delta-v top-off on a satellite you have not launched yet, do the arithmetic first, because launching the extra propellant with the spacecraft is usually cheaper than buying it in orbit. Whether refueling beats replacing a satellite at all is the mission-level question, and the `space-launch-and-satellites` sheet covers it.

Roughly 850 kg of propellant per Progress flight to the ISS · station refueling in continuous use since Salyut 6 in 1978 · six hydrazine transfers on the Shuttle's Orbital Refueling System in 1984 · Orbital Express transferred hydrazine between two free flyers at about 492 km in 2007 · about 142 kg reportedly moved between Shijian-25 and Shijian-21 in GEO in 2025 · 50 kg of hydrazine planned for the first Tetra-5 transfer · 1.7 liters of ethanol moved by RRM-1 in 2012.

Russian Progress and ESA's ATV refueling the ISS; DARPA's Orbital Express in 2007, using VACCO's refueling coupler between ASTRO and NEXTSat; NASA's Robotic Refueling Mission 1 and 2 on the ISS, which cut lock wire, removed a safety cap, and pumped 1.7 liters of ethanol into a mock tank; Orbit Fab's Tanker-001 Tenzing depot; the US Space Force's Tetra-5, which plans to move 50 kg of hydrazine from an Orbit Fab Kamino depot into a client satellite through a RAFTI port.

Almost none of the cost is in the fluid path. A RAFTI service valve is a 0.52 kg part, and the propellant itself is cheap on the ground; what costs money is the vehicle that carries it, the rendezvous and docking stack that gets it there, and the flight qualification that lets an operator risk an eight-figure satellite on the operation. That means unit economics improve with reuse rather than with volume of propellant: a tanker that serves one customer per launch repeats the whole expense every time, while a depot that is filled once and drained many times spreads it. Every buyer today is a government. The Space Force is funding Tetra-5 and Tetra-6, DIU is funding the RAPIDS depot, and the Space Force's Elixir contract with Northrop Grumman pays for a GEO refueling payload, so the commercial revenue in this category is currently government contract revenue with a commercial label. The signal to watch is not another successful demonstration but a published price per kilogram delivered that a satellite operator can put into a fleet model, plus a second customer willing to go after the first. Until an operator can buy propellant the way it buys launch, spacecraft will keep being designed as if refueling does not exist, which keeps the addressable fleet small.

2 minDextre tops off the tank: The Robotic Refueling MissionCanadian Space Agency · 10k+ views

3 minRobotic Refueling Experiment Heads to SpaceNASA Johnson · 5k+ views

In-Space Servicing, Assembly, and Manufacturing (ISAM) State of Play - 2025 Edition (NASA) · A Cross-Disciplinary Study of On-Orbit Refueling for Geostationary Satellites (COSMIC)

Cryogenic transfer moves liquid oxygen, methane, or hydrogen between vehicles, and unlike storable transfer it is a fluid management problem rather than a plumbing problem. At 25 psi, liquid oxygen boils at 95.6 K, methane at 118.4 K, and hydrogen at 22.3 K, so every one of these fluids is sitting at its boiling point inside the tank and any heat that gets in turns liquid into gas. In microgravity the liquid does not settle at the outlet; surface tension pulls it onto the walls and leaves the vapor bubble wherever it happens to be, so before anything flows you either fire small thrusters to settle the propellant under a fraction of a g or you fit a screen-channel liquid acquisition device that holds liquid at the outlet by capillary action until the pressure difference exceeds the screen's bubble point. The receiving tank then has to be chilled down, because a 300 K tank flashes incoming liquid straight to gas and pressurizes itself; the standard methods are charge-hold-vent cycling and spray cooling, and every gram vented during chilldown is propellant you no longer have. Ullage pressure has to be managed the whole time, since the receiver's vapor has to be condensed or vented as liquid comes in, and venting in microgravity risks throwing liquid overboard along with the gas. Nobody has transferred bulk cryogen between two spacecraft. SpaceX moved about 10 t of liquid oxygen between two internal tanks on Starship's third flight in March 2024, RRM-3 stored 42 liters of liquid methane on the ISS at essentially zero boiloff before a cryocooler failure canceled its transfer demonstration, and SHOOT moved 152 liters of superfluid helium between tanks on STS-57 in 1993 using properties no other cryogen has.

The reason to do this at all is performance: liquid oxygen with hydrogen or methane is the only chemical propellant combination with enough specific impulse to make lunar and Mars architectures close, and refilling a stage in orbit lets it launch nearly dry and then leave with a full tank. The weakness is that the losses are set by hardware and elapsed time rather than by how much propellant you deliver. Chilling an aluminum receiver tank from 300 K to 20 K removes roughly 170 kJ per kilogram of tank structure, and the hydrogen doing the cooling carries away between about 1,600 and 4,300 kJ/kg depending on how warm the vent gas leaves, which works out to roughly 0.04 to 0.11 kg of hydrogen spent per kilogram of receiver tank (derived here from published enthalpies, not a published rate). Real chilldowns land at the bad end of that range because up to 85% of the process sits in film boiling, where a vapor blanket insulates the wall from the liquid, and both tanks keep boiling off the entire time the operation runs. Put those together and a slow transfer of a small quantity can consume more propellant than it delivers, which is why the architecture question is always whether you can move a large load quickly rather than whether the transfer works at all. Measurement is the other soft spot: the best flown cryogenic mass gauging, on RRM-3, had about 2% uncertainty, and a 2% error on a 100 t transfer is 2 t of propellant nobody can account for.

Use cryogenic transfer only when the mission genuinely needs cryogenic performance, which in practice means a lunar lander, a Mars stage, or a nuclear thermal vehicle, and plan for one large fast transfer rather than a series of small ones. If the receiver is small relative to the amount you want to move, stop and check the chilldown arithmetic, because the cost scales with the receiver's dry mass and can swallow the delivery. Choose settling over a liquid acquisition device if you have thrust and propellant to spare, since settled operations let you reuse mature upper-stage practice and give expulsion efficiencies above 99.5%; choose an acquisition device when a settling burn would cost more delta-v than the propellant it saves. If the spacecraft only needs a few hundred meters per second and can accept a lower specific impulse, use storable propellant and the storable transfer entry instead, because none of these problems arise. Budget active cooling from the start for anything that has to loiter, and see the boiloff control entry for what that costs in watts and kilograms.

Boiling points at 25 psi of 95.6 K for liquid oxygen, 118.4 K for methane, and 22.3 K for hydrogen · roughly 170 kJ per kg of aluminum tank removed in a 300 K to 20 K chilldown · about 0.04–0.11 kg of hydrogen spent per kg of receiver tank structure (derived here) · up to 85% of line chilldown spent in film boiling · about 10 t of liquid oxygen moved between internal Starship tanks in March 2024 · 42 liters of methane held at zero boiloff by RRM-3 with about 2% gauging uncertainty · expulsion efficiency above 99.5% under settled operations.

SpaceX's March 2024 Starship flight, which moved roughly 10 t of liquid oxygen between a header tank and the main tank under NASA's Tipping Point program; NASA's Robotic Refueling Mission 3 on the ISS, which stored liquid methane but lost its cryocooler before the transfer; Eta Space's LOXSAT-1 liquid oxygen fluid management mission; NASA's Reduced Gravity Cryogenic Transfer project, which tests line and tank chilldown with coated tubes and pulsed flow; the SHOOT superfluid helium transfer on STS-57 in 1993.

This is entirely agency-funded technology development with one commercial vehicle riding on it. NASA's Cryogenic Fluid Management Portfolio Project pays for the component work, the four 2020 Tipping Point awards bought flight demonstrations from industry, and the paying customer for the capability is the Artemis program, which cannot land a crew without it. The cost that decides the architecture is not the transfer hardware but the number of tanker flights: a lunar lander needing a full propellant load in low Earth orbit implies several to many launches per mission, and every percent of propellant lost to chilldown and boiloff adds directly to that count. That is why the money goes into insulation, cryocoolers, and low-loss chilldown rather than into bigger pumps, since reducing losses is what removes launches. There is no service market here and probably will not be one soon, because the only buyers are the two or three organizations building cryogenic exploration vehicles, and each is solving the problem inside its own architecture rather than buying propellant from anyone. Whether refueling makes commercial sense at all is a mission-level question the `space-launch-and-satellites` sheet covers.

2 minSpaceX CRS-4 Launch - Fuel Slosh, internal fuel tank camera (5G to 0G)malu5531 · 100k+ views

9 minDemonstrating Rocket Fuel Transfer in SpaceNASA Johnson · 500k+ views

NASA's Developments in Cryogenic Fluid Management Technology (NASA) · Nitrogen flow boiling and chilldown experiments in microgravity using pulse flow and low-thermally conductive coatings (npj Microgravity)

A depot is a tank that stays in orbit, is filled by tankers over weeks or months, and dispenses propellant to customers on demand. The tank is the easy part. Around it you need power, thermal control good enough that the contents survive the wait, attitude control and station-keeping to hold an orbit for years, a docking mechanism and a fluid interface that arriving and departing vehicles both speak, and gauging accurate enough that a buyer knows what was delivered. Thermal design dominates the layout for cryogens: ULA's 2008 depot study put a conical sunshield up to 80 feet long in front of a liquid oxygen tank, parked it at 1,300 km to cut atomic oxygen erosion and drag, and chose a near-zero beta angle for the longest possible eclipse, which brought absorbed sidewall heat flux to about 0.5 BTU per hour per square foot and boiloff below 0.1% of a full tank per day. Storable depots need none of that, because hydrazine and high-test peroxide do not boil off, which is why the only depot ever flown carries a storable. Orbit Fab's Tanker-001 Tenzing went up in June 2021 with high-test peroxide and RAFTI ports and has never transferred fluid to anything. The follow-on, Kamino, is funded through the Defense Innovation Unit's RAPIDS program and is meant to supply 50 kg of hydrazine to a Tetra-5 satellite in geostationary orbit.

A depot separates the rate propellant arrives from the rate it is used, which is the entire point: you can accumulate a load with many small launches and then hand it over in one operation, and a customer can plan a mission that no single launch supports. It also amortizes the expensive parts, since the avionics, thermal system, docking mechanism, and station-keeping propellant are launched once and used many times, where a tanker flown per customer repeats all of it every trip. The weakness is that a depot loses money and propellant while it waits. There is no revenue during an idle period, station-keeping still burns propellant, and for a cryogenic depot the boiloff rate sets a hard shelf life: at 0.1% per day a full load is 3% lighter after a month and 30% lighter after a year. That makes utilization the binding design variable, and utilization depends on customers who do not yet exist.

Treat a depot as a business decision, not an engineering one. Build one if you have a committed buyer with a schedule, because the whole case rests on throughput and an unsold depot is worse than no depot. If you have one customer and one delivery, fly a tanker instead and skip the station-keeping, the shelf life, and the years of on-orbit risk. If the propellant is storable, the depot can wait years for a buyer, which is why every near-term commercial depot is a hydrazine or peroxide depot rather than a cryogenic one. If the propellant is cryogenic, size the thermal system against the expected time between fill and sale rather than against a nominal mission duration, and read the boiloff control entry before assuming passive insulation is enough. The `space-launch-and-satellites` sheet covers whether the refueling market exists at all; this entry is about what the hardware has to do if it does.

Boiloff below 0.1% of a full tank per day in ULA's shielded liquid oxygen depot design · roughly 0.5 BTU per hour per square foot of absorbed sidewall heat flux · a sunshield up to 80 feet long at a 1,300 km orbit · 3% of a load lost per month and about 30% per year at 0.1% per day · one depot flown to date, Tanker-001 Tenzing in June 2021, with zero fluid dispensed · 50 kg of hydrazine planned from the Kamino depot to Tetra-5 · roughly $20M for 100 kg of hydrazine delivered in GEO, which is $200,000 per kilogram, as the only published price target.

Orbit Fab's Tanker-001 Tenzing, the first and so far only propellant depot in orbit, and its Kamino successor funded through DIU's RAPIDS program; ULA's cryogenic depot design based on Centaur flight practice, launched on a single medium EELV with no orbital assembly; the Starship propellant depot in NASA's Artemis lander architecture, which is a depot inside one company's mission rather than a facility anyone else can buy from; NASA's long line of unbuilt depot studies going back to the 1970s.

The honest summary is that this is a commercial problem rather than an engineering one. Nothing about the hardware is beyond what has flown, and every serious study since the 1970s has concluded roughly that, but no customer has committed to buying propellant on a schedule, so no depot gets financed and no spacecraft gets designed assuming one exists. The only published price target in the business is Orbit Fab's roughly $20M for 100 kg of hydrazine delivered in geostationary orbit, which is $200,000 per kilogram and far above the cost of simply launching a fuller tank, and it stays there until a depot is filled once and drained many times. Government is doing what it usually does when a market will not form on its own: DIU is paying for the RAPIDS depot, the Space Force is paying for the Tetra-5 demonstration that will buy from it, and both sides of the first transaction are the same customer. Commercial satellite operators say plainly that they intend to be fast followers, using infrastructure somebody else builds, which is a rational position and also the reason nobody builds it. The thing to watch is the first purchase order from a non-government buyer with a delivery date on it. Another successful demonstration would not move this category; a signed order would.

19 minStarship Orbital Propellant DepotEager Space · 10k+ views

A Practical, Affordable Cryogenic Propellant Depot Based on ULA's Flight Experience (NASA) · The Economics of Orbital Mass: How Depots Can Work in LEO and Where to Put Them (COSMIC)

A cryogenic tank holds liquid at its boiling point, so every watt that gets in turns liquid into vapor, and the vapor has to be vented before it bursts the tank. Passive control slows the leak: multilayer insulation, a sunshield, low-conductivity struts, thermal coatings, and vapor-cooled shields that route the escaping cold gas through the structure so it intercepts heat on the way out. Active control removes the heat instead of resisting it, using a cryocooler connected to the tank either as tubing bonded to the wall or as a cooled shield buried inside the insulation, an arrangement NASA calls broad area cooling. The two are usually stacked, because a 90 K cooler is far more efficient than a 20 K one, so a 90 K shield intercepts most of the incoming heat and a smaller 20 K machine finishes the job for hydrogen. Scale matters more than anything else here: heat leak grows with surface area while capacity grows with volume, so small tanks lose a much larger fraction per day than big ones, and hydrogen loses roughly eight times the volume fraction that oxygen does because its latent heat per unit volume is about 32 MJ/m³ against oxygen's 243 MJ/m³. NASA's own analysis puts an 8.4 m diameter liquid hydrogen tank at a few hundred watts of heat leak without a 90 K shield and over 1,000 W if the surrounding structure is aluminum, which on the roughly 22 t such a tank holds works out to about 0.26% per day at 300 W and 0.88% per day at 1,000 W (arithmetic derived here from hydrogen's 446 kJ/kg latent heat, not a published rate).

Passive measures are cheap, weightless in operating terms, and cannot fail, but they only slow the loss. Ground testing on NASA's SHIIVER rig cut the boiloff rate of a large hydrogen tank by about 46% at half-full by adding multilayer insulation to the domes, and a 90 K broad-area-cooling shield with cooled penetration straps cut it by about 60% in the reduced-boiloff tests, which still leaves a rate that accumulates over a mission measured in hundreds of days. Only a cryocooler reaches zero, and it is paid for in power and mass. Flight cryocoolers available today lift less than 1 W at 20 K and about 20 W at 90 K; NASA's development articles are more than an order of magnitude beyond that, and the Creare reverse turbo-Brayton machine built for liquid hydrogen delivered 19.2 W at 20 K for about 1.8 kW of input power, at a flight-projected 106 kg and 5.5 kg per watt of lift. That 1.8 kW has to come from a solar array and leave through a radiator, so the real system mass is well above the cooler itself, and the machine is a single point of failure that takes the propellant with it: RRM-3 held 42 liters of liquid methane at essentially zero boiloff on the ISS for four months and then lost its cryocooler, which ended the mission's transfer demonstration.

If the cryogenic phase of the mission lasts days, insulate and accept the loss, because a cryocooler and its power system will mass more than the propellant they save. If it lasts months, add a 90 K shield with broad area cooling, which is the cheapest large reduction available and is efficient enough to be worth its power. If it lasts a year or more, or if the propellant is hydrogen, plan for full zero-boiloff from the start, because passive insulation cannot get the rate to zero and losses accumulate linearly with time while insulation improvements do not. Size the 90 K stage first and let it knock the load down before the 20 K stage sees it, since a 20 K watt costs several times what a 90 K watt costs. Budget the radiator and array alongside the cooler rather than after it, and see the radiators and the two-phase loops entries for what rejecting a couple of kilowatts actually takes.

Hydrogen's latent heat is about 32 MJ/m³ against oxygen's 243 MJ/m³, roughly an 8:1 boiloff penalty by volume · a few hundred watts of heat leak on an 8.4 m hydrogen tank, over 1,000 W with aluminum structure · about 0.26%/day at 300 W and 0.88%/day at 1,000 W on a 22 t load (derived here) · dome multilayer insulation cut boiloff about 46% in SHIIVER testing, a 90 K cooled shield about 60% · today's flight cryocoolers lift under 1 W at 20 K and about 20 W at 90 K · 19.2 W at 20 K for 1.8 kW input, 106 kg flight projection and 5.5 kg per watt · 42 liters of methane held at zero boiloff on the ISS for four months by RRM-3.

NASA Glenn's SMiRF facility reduced-boiloff and zero-boiloff tests, which achieved steady-state zero boiloff on a 1.4 m³ hydrogen tank with a tube-on-tank cryocooler; the SHIIVER rig, which tested dome insulation and vapor-cooled skirts on an upper-stage-scale tank; Creare's 20 W at 20 K reverse turbo-Brayton cryocooler and the parallel 150 W at 90 K developments for oxygen and methane; RRM-3's zero-boiloff methane storage on the ISS; NASA's planned large integrated flight demonstration, which pairs 20 K and 90 K broad area cooling on a 261 kg liquid hydrogen load.

The currency here is mass and watts rather than dollars, because there is no catalog price for a space cryocooler and the units are effectively hand-built to order. What a program pays is roughly 5.5 kg per watt of 20 K lift plus the array and radiator to serve 1.8 kW, and what it buys back is propellant that would otherwise be vented, so the trade closes as soon as the mission is long enough that the vented mass exceeds the cooling system mass. That is why hydrogen architectures carry coolers and short-duration oxygen stages do not. Development is funded almost entirely by NASA's Cryogenic Fluid Management portfolio, spread across two vendors and two cycles, Brayton from Creare and Stirling from Converter Source, which is a deliberate second-source strategy in a market too small to support competition on its own. The one thing that could widen that market is in-situ propellant production, since liquefying oxygen or methane on the Moon or Mars needs the same 90 K machines at similar capacity, and liquefying 0.3 kg of hydrogen an hour needs 150 to 300 W at 20 K. If those two demand sources arrive together, the per-unit cost falls for both; if neither arrives, this stays a portfolio of technology contracts.

9 minWhy Is It So Difficult to Store Rocket Fuel in Space? - SpaceX & Blue Origin's Biggest HeadacheSpace Beetle · 1k+ views

NASA Cryocooler Technology Developments and Goals to Achieve Zero Boil-Off and to Liquefy Cryogenic Propellants for Space Exploration (NASA) · Cryogenic Fluid Management (CFM) (NASA)

A refueling interface is the fluid coupling where the propellant actually crosses between two spacecraft, and it decides whether a servicer can refuel a client it did not design. The client half has to be passive, since a satellite low on propellant may have little power and no working attitude control: no actuators, no commands, just a valve that opens when the servicer's active half engages it. It also has to seal a toxic propellant at flight pressure, tolerate the misalignment a docking approach leaves behind, protect its sealing faces from atomic oxygen and contamination for the years before it is used, and then break cleanly without spilling. Orbit Fab's RAFTI service valve is the most mature: it replaces the spacecraft's ordinary fill-and-drain valve so the same port serves ground fueling and orbital refueling, weighs 0.52 kg in its Class 2A form, holds 650 psig for low-pressure propellants and 3,000 psig in the high-pressure variant, leaks less than 10⁻⁶ scc/s externally through a triple seal, carries a 15-year rating, and accepts docking of masses up to 1,000 kg at 0 to 25 mm/s with ±10 mm and ±5 degrees of misalignment. It was flight-qualified in 2024 and no propellant has yet passed through one in orbit. Northrop Grumman's Passive Refueling Module is the other candidate, a government-owned design that Space Systems Command selected in January 2024 as its first preferred refueling interface and placed under its own configuration control, with the intent of publishing the specification so any builder can make compatible hardware. The mechanical docking ring is a separate problem covered by the docking mechanisms entry; this one is about the plumbing that runs through it.

The strength is how cheap the option is. Half a kilogram and a valve change on a spacecraft that was going to carry a fill-and-drain valve anyway buys the possibility of being refueled, and if it is never used it costs almost nothing. The weakness is that there are two candidate interfaces and no agreed one, so a servicer built around RAFTI cannot refuel a client fitted with a PRM, which splits an addressable fleet that is already small. Standards work exists but stops short of the part that matters: ANSI/AIAA S-157, released in early 2025 and drawing on CONFERS recommendations, is a voluntary consensus standard that says what an interface must do rather than specifying form and fit, and few missions have made it a compliance document. The alternative to a standard port is a robot that defeats a legacy valve, which works and has flown, but it turns a minutes-long connection into hours of teleoperation with an arm and a tool set. And leak monitoring at the joint during transfer is still an open gap, so the interface's most important failure mode is the one nobody can watch in real time.

Fit a port on any spacecraft you are building now that might plausibly be worth refueling, because the mass and cost are trivial against the option it preserves, and the US Space Force's RG-XX solicitation, released in January 2026, requires that solutions can be refueled in orbit. Choose PRM if the customer is a US national security program, since Space Systems Command owns the design and intends to publish it, and choose RAFTI if you need flight-qualified hardware today or want the same port to do ground fueling. If you are building a servicer rather than a client, plan on carrying adapters for both and price the second one in, because betting on a single interface winning is a bet on somebody else's procurement. Only fall back on robotic access to a legacy fill-and-drain valve when the client is already in orbit and has no port, and budget the arm, the tools, and the operator time when you do. Do not build a business whose margin depends on owning the interface, because the government has already moved one of the two candidates into public configuration control.

RAFTI Class 2A at 0.52 kg · 650 psig for low-pressure propellants and 3,000 psig in the high-pressure variant · external leakage below 10⁻⁶ scc/s through a triple seal · 15-year rated life in LEO and GEO · docking of masses up to 1,000 kg at 0 to 25 mm/s with ±10 mm and ±5 degrees of misalignment · flight-qualified in 2024 with zero orbital fluid transfers through it · PRM selected by Space Systems Command in January 2024 as its first preferred interface · ANSI/AIAA S-157 released in early 2025 as a functional, not form-and-fit, standard.

Orbit Fab's RAFTI service valve and its GRIP active counterpart, flown as the fill and drain interface on Tanker-001 Tenzing and planned for the Tetra-5 transfer; Northrop Grumman's Passive Refueling Module and Active Refueling Module, first flying on the Mission Robotic Vehicle and exercised by the Space Force's Elixir payload; VACCO's refueling coupler, which transferred hydrazine between two free-flying spacecraft on Orbital Express in 2007; OSAM-1's Cooperative Servicing Valve, flight-qualified for hydrazine before the mission was canceled in early 2024; NASA's Robotic Refueling Mission and the RROxiTT oxidizer test, which showed a robot can work through the safety wire and caps on a valve never meant to be reopened.

Nobody makes money selling half-kilogram valves, so what companies are competing for here is position rather than revenue: a servicer can only sell to clients that fitted its interface. A client operator's cost is a rounding error, so adoption gets decided in procurement rather than on price, and the buyer with enough volume to settle it is the US government. Space Systems Command has already acted like it knows this, taking the PRM design into government configuration control and planning to publish the specification, which removes the licensing revenue an interface owner would otherwise expect and turns the value into servicer contracts instead. The cost of the standoff falls on servicer operators, who have to carry hardware for two interfaces or address half the market, and on satellite builders, who must commit years before any servicer exists. Industry assessments put the useful window for settling this at roughly three to five years before other national programs establish competing interfaces, after which the choice is inherited rather than made. The practical read for a startup: build to whichever interface your first government customer names, keep the fluid path adaptable, and do not price a business plan around a standard you do not control.

United States ISAM Interface Ecosystem 2025 Report (COSMIC) · Propellant Transfer Technologies (NASA)

A spacecraft solar array is a set of triple-junction cells carried on a structure that has to survive launch folded and then open exactly once. The cells are gallium indium phosphide on gallium arsenide on a germanium substrate, roughly 30% efficient, and at Earth's distance the Sun delivers 1,361 W/m², so a packed array makes about 400 W/m² at the start of life. What separates one design from another is the structure, and the number that matters is watts per kilogram. Cells bonded to the spacecraft's own skin give 20–40 W/kg and need no mechanism at all. Rigid deployable panels, usually aluminum honeycomb with composite face sheets on a hinged yoke, give 30–70 W/kg. Flexible blankets give the most: Redwire's Roll-Out Solar Array stores strain energy in two composite booms that unroll the blanket with no motor driving the deployment, and reaches about 100 W/kg. The six roll-out arrays retrofitted to the ISS produce over 20 kW each and take the station from 160 kW to 215 kW. NASA's survey of flown missions found they cluster near 30 W/kg overall, with the best array ever flown around 200 W/kg, and nothing below 1 W/kg.

Solar is by a wide margin the cheapest watt available in space, it consumes nothing, and it fades slowly enough to predict, so a triple-junction array in GEO is usually sized to lose 1–2% of its output a year and specified to hold 75–85% of beginning-of-life power at 15 years. The weakness is the deployment. Anything folded has to unfold once with no repair, through hinges, latches, tie-down cutters, and a synchronizing cable that cannot be fully tested in the configuration it flies, because on the ground the structure is carrying its own weight. Lucy launched in 2021 with one of its two circular fan arrays unlatched; after a year of attempts NASA stopped at over 98% deployed and committed to flying the 12-year mission in that state, and Mars Global Surveyor flew with a panel a broken damper had left short of its latch, which forced a gentler and much longer aerobraking plan. Output also falls with the square of distance, to roughly 50 W/m² at Jupiter, and area costs something even close to Earth, since a large array in low orbit adds drag and a solar-pressure torque the attitude system has to trim. Strings fail individually over a long life as well: on the ISS each failed string costs about 300 W.

If the load is under a few hundred watts and you would rather have no deployment failure mode at all, mount the cells on the body and accept 20–40 W/kg. If you need kilowatts, use a deployable, and pick rigid panels when stowed volume is not the binding constraint and a roll-out or fan-fold blanket when it is, since a blanket stows in a fraction of the volume and roughly doubles the watts per kilogram. Size the array at end of life, with degraded cells and the worst sun angle, because a GEO array sized at beginning of life is 15–25% short in the year that matters. Past roughly 3–5 AU stop asking the question, because the flux is too low and radioisotope power in entry 18 is the alternative. On a surface with a long night, the array is not what sizes the system; the storage is, which is the arithmetic in entries 17 and 19.

About 30% triple-junction cell efficiency and roughly 400 W/m² at Earth distance · 20–40 W/kg body-mounted, 30–70 W/kg rigid deployable, about 100 W/kg for a roll-out blanket · flown missions cluster near 30 W/kg, best about 200 W/kg · six iROSA arrays at over 20 kW each, taking the ISS from 160 kW to 215 kW · 1–2% loss a year in GEO, so 75–85% of beginning-of-life power at 15 years · about 50 W/m² of sunlight at Jupiter · order $500–1,000 per watt installed

NASA's ROSA flight experiment deployed on the ISS in June 2017, and six ISS Roll-Out Solar Arrays were installed by spacewalk from 2021 onward over the original arrays, which were designed for 15 years and had been flying since December 2000; each new array produces over 20 kW and the set takes the station from 160 kW to 215 kW. DART carried two ROSA wings to its impact with Dimorphos, Lucy flies two circular fan-fold arrays, and Juno reached Jupiter on three rigid panels holding roughly 60 m² of cells.

An installed array runs on the order of $500–1,000 per watt, roughly five thousand times what a terrestrial PV module costs per watt, and the cell is not where the money goes. Space triple-junction cells are grown by metal-organic vapor phase epitaxy on germanium wafers, in low volume, by three or four suppliers worldwide, so they are expensive; the panel substrate, hinges, dampers, harness, and above all the qualification and deployment test campaign are the larger share. That is why constellation volume moves the price more than a better cell does. A program buying identical panels by the thousand amortizes one qualification campaign across all of them, while a one-off science mission pays for its own. Supply is concentrated: Rocket Lab bought SolAero for about $80M to own a cell and panel line rather than buy from one, Azur Space and CESI cover most of Europe, and Boeing's Spectrolab supplies much of the rest. Redwire sells ROSA as a catalog product, which makes it one of the few places on this sheet where a real specific-power improvement has been bought rather than promised. The buyer is almost always the spacecraft prime rather than the operator, so array cost is negotiated as part of a bus, not as a line item the end customer sees.

8 minRoll-Out Solar Array Experiment (ROSA) Deploys on International Space StationNASA Johnson · 50k+ views

1 minImpact Story: Roll-Out Solar ArraysNASA Video · 10k+ views

New Solar Arrays to Power NASA's International Space Station Research (NASA) · On Orbit Flight Testing of the Roll-Out Solar Array (NASA)

The battery carries the spacecraft through eclipse and through any load the array cannot meet, and the power management electronics decide how the array, the battery, and the loads are connected. Lithium-ion has displaced nickel-hydrogen almost everywhere since about 2005, and a flight pack delivers 100–130 Wh/kg once cells, structure, harness, balancing electronics, and the thermal interface are counted, against 150–270 Wh/kg for the bare commercial cells inside it. The duty cycle is what makes spacecraft batteries unusual. A satellite in low Earth orbit passes through shadow for roughly 35 minutes of every 90-minute orbit, 16 times a day, which is about 5,500 charge and discharge cycles a year and something like 27,500 over a five-year mission. Around that sits the power management: a regulator that either dumps the array's excess (direct energy transfer) or tracks its peak power point, a charge controller that holds each cell inside a narrow voltage window, and a bus that is 28 V on small spacecraft and 100 V or higher on large ones, because at 20 kW a 28 V bus would carry over 700 A of current. Li-ion also has to be held near room temperature, so the battery is usually one of the tightest thermal boxes on the vehicle and draws heater power during the same eclipse it is discharging into.

Li-ion roughly doubled the usable energy per kilogram over nickel-hydrogen and shrank the volume more than that, which is why the ISS replaced its 48 nickel-hydrogen orbital replacement units with Li-ion units two for one starting in December 2016. What it gave up is tolerance. Nickel-hydrogen shrugged off overcharge and deep discharge and delivered tens of thousands of cycles; Li-ion does not, so life is bought by keeping the depth of discharge low. The ISS nickel-hydrogen batteries were designed for ten years at a 35% maximum depth of discharge, and Li-ion packs in low orbit are typically held near 20–30%, which means the pack is sized three to five times larger than the eclipse energy alone would require. The failure mode is also worse: a shorted Li-ion cell can go into thermal runaway and take its neighbors with it, so flight packs carry per-cell fusing, bypass paths, and a management system whose job is mostly to prevent one cell from being driven outside its window. And the qualification is slow, because proving 27,500 cycles at a given depth of discharge and temperature means running a real cycling campaign for years, which is a schedule item rather than an engineering one.

Size the pack from the eclipse energy, then divide by the depth of discharge you can defend for the mission life, not the other way around. If you are in low Earth orbit, plan on 20–30% depth of discharge and accept a pack three to five times the naive size, because at 5,500 cycles a year an automotive-style 80% discharge would be worn out inside a year. If you are in GEO the arithmetic is much kinder, since eclipses happen only in two seasons around the equinoxes and run at most 72 minutes, giving on the order of 90 cycles a year, so 60–80% depth of discharge is normal and the pack is far smaller for the same load. If the mission is short and the cost target is aggressive, use screened commercial 18650 or 21700 cells rather than space-qualified ones: NASA has flown Sony, LG, and Panasonic consumer cells on Ingenuity, PACE, and Europa Clipper. If the load has to run through a 354-hour lunar night, stop sizing batteries and read entries 18 and 19, because at 100 Wh/kg the storage alone is about 3.5 kg per watt of continuous load.

100–130 Wh/kg at the pack, 150–270 Wh/kg at the commercial cell · about 35 minutes of eclipse per 90-minute orbit, 16 orbits a day, roughly 5,500 cycles a year and 27,500 over five years · 20–30% depth of discharge in low orbit against 60–80% in GEO · ISS nickel-hydrogen units designed for ten years at 35% maximum depth of discharge, replaced two for one by Li-ion from December 2016 · GEO eclipses at most 72 minutes, on the order of 90 a year · about 700 A on a 28 V bus at 20 kW, which is why large spacecraft run 100 V or more · roughly 100 kg of pack for a 5 kW low-orbit satellite (derived below)

The ISS Li-ion orbital replacement units built around GS Yuasa cells are the largest flight batteries ever used on a crewed vehicle; Saft's VES16 and EaglePicher's space cells cover most Western science and defense spacecraft; and screened consumer cells have flown widely, with Sony US18650 VTC4 on the Ingenuity Mars helicopter, LG ICR18650 B3 on Europa Clipper, and LG INR18650-MJ1 on PACE.

Two costs dominate, and neither is the cell chemistry. The first is qualification: lot acceptance testing on every cell batch, plus a life-cycle test that has to demonstrate the mission's cycle count at the mission's depth of discharge, which for a low-orbit satellite means years of continuous cycling in a chamber before the design is trusted. The second is mass, since the depth-of-discharge rule multiplies the pack. Work an example: a 5 kW low-orbit satellite draws 5 kW for about 0.58 hours of eclipse, so 2.9 kWh comes out of the pack each orbit; at 25% depth of discharge the pack has to hold 11.7 kWh, and at 120 Wh/kg that is roughly 100 kg, of which the eclipse actually uses a quarter. That arithmetic is done here rather than looked up, and at $2,000–6,000/kg to low orbit the launch cost of the margin alone runs into six figures. The clear cost-down move over the past decade has been screened commercial cells in place of space-qualified ones, which is what constellation operators do and what NASA now does on cost-capped missions. It works because 18650 and 21700 cells are made by the hundreds of millions for cars and tools, so the space buyer inherits a production line nobody in this industry could afford to build. Space-grade suppliers stay in business on the missions where a 15-year life or a crewed vehicle makes the screening argument unacceptable.

3 minSpacecraft PowerNASA Jet Propulsion Laboratory · 50k+ views

State-of-the-Art of Small Spacecraft Technology: 3.0 Power (NASA) · International Space Station Lithium-Ion Battery Status (NASA)

A radioisotope thermoelectric generator turns the decay heat of plutonium-238 into electricity with no moving parts. The fuel is plutonium dioxide pressed into ceramic pellets, clad in iridium, and stacked into graphite General Purpose Heat Source modules built to survive a launch failure and reentry intact. The current unit, the Multi-Mission RTG, holds about 4.8 kg of oxide making roughly 2,000 W of heat, which is about 0.42 W per gram, and wraps it in lead telluride thermocouples that convert around 6% of that to 110 W of electricity at beginning of life. The generator weighs about 45 kg, so the specific power is roughly 2.4 W/kg, which is an order of magnitude worse than a solar array in sunlight and the reason nobody uses one where sunlight is available. Output falls about 2% a year, from the 87.7-year half-life of the plutonium plus slow degradation of the thermocouples, so a 14-year mission plans on roughly three quarters of its launch power. Smaller radioisotope heater units, about 1 W of heat each, are used separately to keep components warm without spending electrical power on heaters.

It works where nothing else does: through a 354-hour lunar night, through a Mars dust storm that can cut solar output for weeks, and past Jupiter where sunlight is down to about 50 W/m². It also runs for decades with no mechanism to fail, which is why the Voyagers are still transmitting. The binding weakness is fuel supply, not engineering. The United States stopped making plutonium-238 in the late 1980s and restarted at Oak Ridge decades later; GAO reported in 2017 that the restarted program had produced 100 grams in total against a target of 1.5 kg a year by 2026, and output since has climbed into the hundreds of grams a year rather than kilograms. One MMRTG needs 4.8 kg of oxide, which is roughly 4.2 kg of plutonium, so a single generator is close to three years of the entire national production target and new fuel has to be blended with legacy inventory to reach the required isotopic spec. The practical consequence is blunt: the supply supports something like one MMRTG every two or three years, which is a handful of missions per decade, and units are allocated by NASA rather than bought. The other weakness is conversion efficiency, since 6% means 94% of an extremely scarce material is thrown away as heat.

Use radioisotope power when sunlight is unavailable or unreliable and the load is under a few hundred watts, which in practice means the outer planets, a Mars surface mission that has to survive dust storms and winter, or a lunar site that has to run through the night. Do not plan a mission around one until you have confirmed the allocation, because the constraint is fuel rather than money and NASA states in its announcements of opportunity whether a unit is available for that competition. If the load is more than a few hundred watts, an RTG is the wrong tool and fission in entry 19 is the answer, since specific power goes from 2.4 W/kg to roughly 7 W/kg and scales with size rather than against it. If you only need to keep a component warm rather than powered, ask for radioisotope heater units instead, which use grams rather than kilograms. And if the mission can be flown on solar with a larger array, fly it on solar, because that decision frees a unit for a mission that has no alternative.

About 4.8 kg of plutonium dioxide per MMRTG, roughly 2,000 W thermal at 0.42 W per gram · about 6% thermal-to-electric conversion, 110 W at beginning of life · roughly 45 kg per unit, about 2.4 W/kg · output falling about 2% a year against an 87.7-year half-life · production target of 1.5 kg a year by 2026, against 100 grams produced in total as of GAO's 2017 report · radioisotope heater units about 1 W of heat each · about 50 W/m² of sunlight at Jupiter

The MMRTG powers Curiosity, landed 2012, and Perseverance, landed 2021, and is the baseline for Dragonfly at Titan; the earlier GPHS-RTG design flew on Galileo, Ulysses, Cassini, and New Horizons; the Voyagers have been running on RTGs since 1977. The Advanced Stirling Radioisotope Generator, which would have converted about a quarter of the heat instead of 6% and cut the fuel per watt by roughly a factor of four, was canceled in 2013 before flight.

This is an allocation problem wearing the costume of a market. There is one producer, the Department of Energy, one buyer, NASA, and no price at which more fuel appears, because the constraint is neptunium-237 target fabrication and irradiation capacity at Oak Ridge and Idaho rather than budget in any single year. The cost that shows up in a mission's books is the reimbursement to DOE for the generator and its fuel, and it is large enough that a Discovery-class proposal usually cannot carry one; the cost that does not show up is the opportunity cost, since a unit given to one mission is a unit no competing mission can have for years. Delivery is expensive too: 45 kg on the lunar surface at CLPS prices of $0.5–1.2M/kg is $22–54M of transport alone for 110 W, which is arithmetic done here rather than a published figure. The one real lever is conversion efficiency. A dynamic converter at 25–30% would stretch the same plutonium across four times as many missions, which is why NASA keeps restarting Stirling work after canceling it. Nothing here has a commercial buyer, and the handful of startups proposing radioisotope units for lunar landers face the same fuel queue as everyone else.

4 minHow Does a Radioisotope Thermoelectric Generator Work? The Seebeck EffectJPLraw · 50k+ views

3 minPrepping the Perseverance Power SourceIdaho National Laboratory · 100k+ views

Space and Defense Power Systems (Department of Energy) · Space Exploration: DOE Could Improve Planning and Communication Related to Plutonium-238 and Radioisotope Power Systems Production Challenges (GAO)

A fission surface power system is a small reactor that makes tens of kilowatts of electricity continuously, regardless of sunlight. NASA's specification for the lunar unit is at least 40 kWe for ten years under about 6,000 kg of launch mass, which works out to 150 kg per kilowatt, or roughly 7 W/kg. The design approach that came out of NASA's Compass studies is a heat-pipe reactor fueled with high-assay low-enriched uranium and moderated with yttrium hydride, coupled to free-piston Stirling convertors and deployable radiators derived from ISS hardware, with power sent to the users at ±2,800 VDC over about a kilometer so the reactor's radiation shadow does not have to cover the base. Conversion runs around 25–30%, so 40 kWe means roughly 140–160 kW of reactor heat and radiators sized to shed three to four times the electrical output. The only US space reactor ever flown was SNAP-10A in 1965; the closest recent hardware is KRUSTY, a 2018 ground test at Nevada that ran a real fission core against Stirling convertors for 28 hours. This entry is about the power system as a mission element; the reactor physics and the terrestrial microreactor lineage are covered by Space & Heat-Pipe Reactors on the nuclear reactors sheet.

The case for it is one piece of arithmetic that nothing else answers. A lunar night is 354 hours, so carrying a 40 kW load through it on batteries takes 40 kW times 354 hours, or 14,160 kWh; at 100 Wh/kg of pack that is about 142 tonnes, and even at a full depth of discharge nobody would allow. The reactor is roughly 6 tonnes for the same job, which is a factor of about 24 on mass, and the gap widens as the load grows because a reactor's mass scales far better than a battery's. Against that, three weaknesses are real. Nothing in this class has flown, so every performance number is an analysis rather than a measurement. The radiators are large and have to deploy, which puts a single-point failure on the heat rejection path rather than on the reactor. And launch authorization is a schedule item measured in years: under the 2019 presidential memorandum on launching space nuclear systems, a low-enriched fission system falls in Tier II, which requires a mission safety analysis report and an Interagency Nuclear Safety Review Board evaluation before the agency head can authorize, while anything not low-enriched falls in Tier III and needs the President's own authorization. That distinction is why the current designs are built around HALEU rather than the highly enriched uranium a compact reactor would otherwise prefer.

Choose fission when you need kilowatts continuously through a long dark period and the load is above a few kilowatts, which on the Moon means anything past a small science station and certainly means in-situ resource production. Below about 1 kW, radioisotope units in entry 18 are simpler and have flight heritage, though the plutonium supply may not have a unit for you. Between those, run the battery arithmetic honestly at the depth of discharge you can defend and see where the crossover lands for your night length and load; on the Moon it usually falls in the low single-digit kilowatts. If you pick fission, budget the launch approval from the start rather than treating it as paperwork, and design to Tier II by using low-enriched fuel unless there is a compelling reason not to. Check the radiator layout early as well, because at 25–30% conversion the heat rejection hardware is physically larger than the reactor and it is what constrains where the system can sit.

At least 40 kWe for ten years under about 6,000 kg, which is 150 kg per kilowatt or roughly 7 W/kg · 25–30% Stirling conversion, so 140–160 kW of reactor heat and radiators shedding three to four times the electrical output · 354-hour lunar night, so 14,160 kWh for a 40 kW load and about 142 tonnes of battery at 100 Wh/kg, a factor of about 24 on mass (derived here) · power distributed at ±2,800 VDC over about a kilometer · KRUSTY ran 28 hours in 2018; SNAP-10A in 1965 is the only US space reactor flown · low-enriched systems in Tier II of the 2019 launch authorization memorandum, everything else in Tier III

NASA and the Department of Energy selected three fission surface power design concepts in June 2022, from teams led by Lockheed Martin, Westinghouse, and Intuitive Machines with X-energy, targeting a lunar demonstration in the early 2030s. KRUSTY, run at the Nevada National Security Site in 2018, is the ground demonstration the concepts descend from. Historic flight hardware is Soviet: about thirty BES-5 reactors on RORSAT ocean surveillance satellites, and two TOPAZ units flown in 1987.

This is a government program with no commercial buyer, and the first-unit cost is not public. The design concept awards were roughly $5M each for twelve months of work, which is study money rather than build money, and the real cost lands when someone funds fuel fabrication, a ground qualification article, and a nuclear-certified launch campaign. The value case is easiest to see from the alternative. Delivering 142 tonnes of battery to the lunar surface at CLPS task-order prices of $0.5–1.2M/kg is $70–170B, which is several years of NASA's entire budget, so past a few kilowatts of continuous night load there is no competing option at any price. That does not make a reactor cheap, only unavoidable, and the history of this field is that programs die before flight rather than over budget: SP-100 was canceled in 1994, Prometheus in 2005, and DRACO wound down in 2025. The exposure an investor can actually buy is in the supply chain rather than the mission, since HALEU fuel fabrication, refractory cladding, Stirling convertors, and compact shielding all cross over into terrestrial microreactors, and that crossover is where the same components have a second customer.

7 minNASA's New Space Reactor Is Powered by Nuclear FissionSeeker · 1m+ views

15 minWhy We’re Putting Nuclear Reactors on the MoonStarTalk · 1m+ views

A Deployable 40 kWe Lunar Fission Surface Power Concept (NASA) · Presidential Memorandum on Launch of Spacecraft Containing Space Nuclear Systems (The White House)

Nothing convects in vacuum, so every watt a spacecraft consumes has to leave as infrared radiation from a surface, and the size of that surface follows the Stefan-Boltzmann law rather than any design choice. Radiated flux goes as the fourth power of temperature, so a panel at 300 K with an emissivity of 0.85 sheds about 390 W/m² into empty space. In flight that drops to roughly 100–250 W/m², because the radiator also absorbs sunlight at 1,361 W/m², Earth's infrared at around 237 W/m², albedo, and heat from the rest of its own spacecraft. Getting the absorbed part down is the job of the coating: optical solar reflectors, which are thin quartz tiles silvered on the back, and silvered Teflon and white paints all aim for high infrared emissivity with the lowest solar absorptance available, typically starting near 0.08–0.2. Heat reaches the radiator through conduction, embedded heat pipes, or a pumped loop rather than through the air, which is entry 21. The layout follows from the numbers: a geostationary communications satellite puts its radiators on the north and south faces, which barely see the Sun at that orbit, and a 20 kW satellite needs on the order of 100 m² at 200 W/m², which is why a high-power comsat is a modest box with two large wings for power and two large flat faces for heat.

A radiator is passive, cheap hardware with essentially no failure mode of its own, and its performance is calculable to within a few percent, which is rare on this sheet. Two things make it hard anyway. The first is that the area scales with power and area has to be carried, deployed, and pointed away from the Sun, so above roughly 10 kW the radiators start dictating the spacecraft's shape and the deployment mechanism becomes a single-point failure on the heat rejection path. The second is the temperature exponent. The same surface that sheds 390 W/m² at 300 K sheds about 5 W/m² at 100 K and about 0.15 W/m² at 40 K, so cooling a detector to 40 K takes roughly 2,600 times the area per watt as cooling electronics at room temperature, and that ratio is why cryogenic instruments need a sunshield rather than a bigger radiator. Coatings also degrade: ultraviolet and atomic oxygen raise solar absorptance over a mission, so a radiator sized on day-one properties is undersized at end of life. And a radiator sized for the hot case overcools in the cold case, so the spacecraft burns survival heater power to undo its own thermal design, which is what louvers and variable-conductance heat pipes exist to reduce.

Size the radiator at end of life, at the hottest environment the orbit produces, with the coating's degraded absorptance, and then check that the cold case does not cost more heater power than the mission can spare. If the dissipation is under a few hundred watts, use body-mounted panels on faces that avoid the Sun and skip the mechanism entirely. If you are above roughly 5–10 kW, plan on deployable radiators and treat the deployment as you would a solar array deployment, because it fails the same way. If the environment swings hard between hot and cold cases, as it does for a lunar surface system or a vehicle that goes from low orbit to deep space, look at variable-geometry or louvered radiators before you look at a bigger fixed one. If the requirement is a cold detector rather than warm electronics, do not try to solve it with radiator area alone: get as far as you can passively with a sunshield and then use a cryocooler, which is entry 21.

About 390 W/m² at 300 K with emissivity 0.85, and 100–250 W/m² in practice after environmental heating · absorbed fluxes of 1,361 W/m² solar and around 237 W/m² Earth infrared · coating solar absorptance starting near 0.08–0.2 and rising over the mission · about 5 W/m² at 100 K and 0.15 W/m² at 40 K, roughly 2,600 times the area per watt against a 300 K radiator (derived here) · on the order of 100 m² for a 20 kW satellite at 200 W/m² · roughly 5,000 m² for a 1 MW payload at the same flux (derived here)

Geostationary communications platforms such as the Boeing 702 and Airbus Eurostar families carry north and south radiator panels with embedded heat pipes and deployable extensions for the high-power versions; the ISS rejects its external loads through two ammonia loops feeding deployable radiator wings; and JWST reaches roughly 40 K on the cold side of a five-layer sunshield about the size of a tennis court rather than through any radiator that would fit on a spacecraft.

Radiator hardware is among the cheapest things on a spacecraft, and it is expensive anyway, because what it costs is area and mass that would otherwise have been payload. Aluminum honeycomb panels, embedded constant-conductance heat pipes, and optical solar reflector tiles come from a small set of suppliers at prices measured in thousands of dollars per square meter, against a launch cost of $2,000–6,000/kg to low orbit and $15,000–30,000/kg to GEO for whatever the panel weighs. The trend that matters commercially is that satellite power keeps rising, so heat rejection is a growing share of the bus: a 20 kW communications platform needs about 100 m² of radiator, and the vendors who can deploy that reliably sell the high-power end of the market. The same arithmetic is what stalls the proposals for large computing payloads in orbit. A one-megawatt compute load needs roughly 5,000 m² of radiator at 200 W/m², which is a structure larger than the ISS solar array wings and has to be deployed, pointed, and kept clear of its own spacecraft, so the thermal system rather than the launch cost is the part of those business plans that has not closed.

7 minSpacecraft thermal systemReflective Layer · 10k+ views

10 minRadiators In Realistic Sci-Fi (And Why You Need Them)Spacedock · 100k+ views

State-of-the-Art of Small Spacecraft Technology: 7.0 Thermal Control (NASA) · Design and Modeling of a Variable Heat Rejection Radiator (NASA)

Radiators only work if the heat reaches them, and these are the two ways of moving it. A loop heat pipe does it passively: a sintered nickel or titanium wick with pores a couple of micrometers across sits only in the evaporator, and the capillary pressure it develops drives vapor down one smooth tube to a condenser on the radiator and pulls liquid back down another, so the loop carries hundreds of watts over several meters and around bends that a plain heat pipe cannot manage. Capillary pumped loops use the same physics with the compensation chamber separated from the evaporator, which gives finer temperature control but needs an active priming step before startup, and that is largely why loop heat pipes displaced them. Below roughly 120 K a passive path stops being enough and the heat has to be pumped uphill by a mechanical cryocooler, usually a Stirling or a pulse tube for 40–150 K, a Joule-Thomson or turbo-Brayton stage for 20 K and below. What a cryocooler costs is input power, all of which comes back out at the warm end and has to be radiated. At around 100 K a good space cooler needs roughly 10–15 W of input per watt lifted, and a small tactical unit closer to 25–40 W/W; at 40 K it is on the order of 40–80 W/W; and the JWST MIRI cooler draws about 145 W in normal science operation to hold an optical bench below 7 K. A recent NASA and Creare 20 K machine reached 24.4 W of lift at 21 K at 17.68% of the Carnot limit, and since the Carnot coefficient of performance rejecting to about 300 K is 20 over 280, or 0.071, that comes to roughly 80 W of input per watt lifted.

Loop heat pipes have no moving parts, no power draw, and decades of flight use, and they solve the problem that killed simple heat pipes on large spacecraft, which is transporting kilowatt-scale heat across a deployed joint or several meters of structure. Their limits are startup behavior and gravity: a loop that will run fine in orbit can be awkward to test on the ground in the orientation it flies, and startup can oscillate under low heat loads. Cryocoolers are the opposite trade. They work, and the flight record is better than most people expect, since the pair of pulse tube coolers on AIRS has run continuously since 2002 and the drive level needed to hold the focal plane at 55 K rose only 2–3% over nineteen years. But they burn power that becomes heat at the reject interface, they wear, and they vibrate. The compressors run near 30 Hz and their harmonics, which is exactly the band a precision-pointed instrument cares about, and that is a common reason an instrument team asks for passive cooling instead. JWST needed design mitigations plus a dedicated ground jitter campaign before flight because the near-infrared instruments sat near MIRI's cooler, and on-orbit measurements then showed no detectable effect, which is the outcome you get only by budgeting for it early.

If you need to move a few hundred watts to a few kilowatts more than about half a meter, or across a hinge, use a loop heat pipe rather than a constant-conductance heat pipe, and accept that the startup case has to be analyzed. If the load is tens of kilowatts, a mechanically pumped single-phase loop is the honest answer despite the pump, which is what the ISS does with ammonia. For a cold detector, get as far as possible passively first with a sunshield and a dedicated radiator, since a passive path costs no power and nothing wears; add a cryocooler only for the last stretch the geometry cannot reach. If you do add one, budget three things up front: the input power, the same number again as heat into the radiator sizing in entry 20, and the exported vibration against the pointing stability of every instrument nearby. If the instrument cannot tolerate the vibration at all, look at a turbo-Brayton machine, whose continuous flow avoids the low-frequency reciprocating disturbance, and pay for it in efficiency at the smaller heat loads.

Loop heat pipe wick pores a couple of micrometers across, carrying hundreds of watts over several meters passively · passive cooling runs out around 120 K · roughly 10–15 W of input per watt lifted at 100 K for a good space cooler, 25–40 W/W for a small tactical unit · on the order of 40–80 W/W at 40 K · a NASA and Creare 20 K cooler reached 17.68% of Carnot and 24.4 W of lift at 21 K, which works out near 80 W in per watt (derived above) · JWST's MIRI cooler draws about 145 W to hold its bench below 7 K · AIRS pulse tube coolers running since 2002 with the drive level up 2–3% in nineteen years · compressor disturbance near 30 Hz and its harmonics

Loop heat pipes have flown on GOES weather satellites, ICESat, and Aura, and are the transport path for the VIPER lunar rover's thermal system; small Stirling and pulse tube coolers from Lockheed Martin, SunPower, AIM, and Ricor now fit CubeSats at 0.35–1.2 kg for 1–2 W of lift near 105–150 K; Northrop Grumman's pulse tube pair on AIRS has run continuously since 2002; and the MIRI instrument on JWST uses a four-stage hybrid pulse tube and Joule-Thomson cooler to reach below 7 K.

The two halves of this entry have completely different economics. Loop heat pipes are hardware: a handful of suppliers build them for tens of thousands of dollars a unit, they are qualified once per design, and the cost of using one is engineering analysis rather than procurement. Cryocoolers are a small, concentrated, high-reliability market, with Northrop Grumman, Lockheed Martin, Ball, Creare, Thales Cryogenics, and Air Liquide covering most of the flight units in the West, and a flight-qualified cooler with its drive electronics is a multi-million-dollar item where a tactical cooler adapted for a smallsat is two orders of magnitude less. What you pay for at the top end is life. A ten-year continuous cooler runs on flexure or gas bearings so that no surface touches another, and that plus the qualification campaign is most of the price. The interesting demand shift is that the buyers are no longer only infrared astronomy and missile warning. Zero-boiloff storage of liquid hydrogen at 20 K, covered in entry 14, needs coolers an order of magnitude larger than any science instrument has asked for, and NASA has been funding that through SBIR work with Creare precisely because no commercial product existed at that capacity.

2 minEngineers with Markers | Loop Heat Pipe Technology OverviewAdvanced Cooling Technologies Inc. · 1k+ views

5 minPulse Tube CoolerOperational Facts · 10k+ views

A comparative analysis of loop heat pipe based thermal architectures for spacecraft thermal control (NASA) · AIRS Pulse Tube Coolers Performance Update – Twenty Years in Space (International Cryocooler Conference)

In-space assembly means launching a large structure as separate pieces and joining them together in orbit. The argument for it is not that rockets cannot lift enough mass, because they can; it is that the payload has to fit inside a fairing about 5 m across, and that number has barely moved in forty years. The Ariane 5 fairing that carried the James Webb Space Telescope was 5.4 m in diameter and 13 m tall, which is why a 6.5 m mirror had to be folded. Exactly one large structure has ever been assembled in orbit: the International Space Station, 419,725 kg and 109 m end to end around a 94 m truss, delivered on 42 assembly flights, 37 on the Space Shuttle and 5 on Russian Proton and Soyuz rockets, with crew aboard continuously from November 2000. Every structural joint on it was made by an astronaut in a pressure suit or by the Canadarm2 with an operator watching. Robotic assembly is a much shorter list: the SAGE III instrument and its support truss were bolted onto an ISS platform by Canadarm2 in 2017, GITAI's robotic arm assembled a four-panel solar array mockup inside the pressurized Bishop Airlock in 2021 and worked fasteners and connectors outside it in 2024, and the SPIDER payload that was to assemble seven elements into a 3 m communications antenna with a 5 m arm was canceled along with OSAM-1 in 2024.

Assembly breaks the link between aperture size and fairing diameter, and it does it with components that are individually small, individually testable, and individually replaceable if one arrives broken. It also removes the single irreversible deployment: a structure that is bolted together can be stopped, inspected, and continued, where a folded one gets one attempt. The weakness is that the assembler has to exist, and above low Earth orbit it does not. Crew can only reach low orbit, and crew time is expensive even there: NASA prices ISS crew time to commercial users at $130,000 an hour, so a two-person six-and-a-half-hour spacewalk is roughly $1.7M of crew time at that rate, which is arithmetic done here rather than a published cost. Robots avoid that but hit the control problem instead, because contact forces during a join need a loop running at tens to hundreds of hertz and the ground is 0.24 s away at GEO. The deeper problem is precision: an optical telescope needs its segments aligned to a small fraction of a wavelength, and the ISS truss was assembled to millimeters, so the demonstrated skill and the needed skill are several orders of magnitude apart.

If the structure fits inside a fairing, launch it whole and stop there. If it does not fit but can be folded, deploy it, because deployment has flown hundreds of times and robotic assembly has not. Reach for assembly when the aperture is past what folding can plausibly reach, when the structure has to be extended or reconfigured after it is built, or when one deployment failure would cost the whole mission and you would rather have a sequence that can be paused. If you are planning assembly beyond low Earth orbit, assume no crew and price the robot and its operations as the main cost, not the structure. And if you are designing any spacecraft that might be assembled or extended later, put the grapple fixtures, fiducials, and standard structural joints on it now, because nothing flying today has them and their absence is what makes every assembly concept start from scratch.

Fairing diameter about 5 m, Ariane 5 at 5.4 m by 13 m · ISS 419,725 kg, 109 m end to end, 94 m truss · 42 assembly flights, 37 Shuttle and 5 Russian, crewed continuously from November 2000 · SPIDER planned 7 elements into a 3 m antenna using a 5 m arm, canceled 2024 · ISS crew time priced at $130,000/hr, so a two-person 6.5-hour spacewalk is roughly $1.7M of crew time (derived here) · GEO round-trip light time 0.24 s against contact control loops needing tens to hundreds of hertz

The ISS Integrated Truss Structure is the only flight example at scale; the EASE and ACCESS experiments on STS-61B in 1985 measured how fast and how tired a suited astronaut gets while bolting truss bays together. Robotic work is thinner: SAGE III on the ISS in 2017 with Canadarm2, GITAI's arm inside and then outside the Bishop Airlock in 2021 and 2024, the canceled SPIDER arm on OSAM-1, DARPA's NOM4D program with a Caltech truss-assembly mission planned for low Earth orbit in 2026, and ground demonstrations at NASA Langley (PASS, SAMURAI) and NASA Ames (ARMADAS).

Everything here is paid for by civil and defense agencies, and there is no commercial buyer for orbital assembly at any price today. The cost structure is unusual because the structure is the cheap part: standard struts and nodes are ordinary aerospace hardware, while the assembler, the interfaces, and the operations campaign carry almost all of the money and all of the schedule risk. OSAM-1 shows what that risk looks like, since NASA canceled it in 2024 after cost growth without flying the assembly payload it carried. Launch cost cuts help less than they appear to, because launch is typically 10–25% of a program and assembly does not reduce the expensive part, which is engineering labor and qualification. The case closes only where the alternative is impossible rather than merely expensive, which today means apertures beyond roughly 10–15 m and structures that have to grow after they are built. Until a customer buys a second assembly mission with its own money, treat this as a research program with two modest robotic results and one enormous crewed one.

19 minBuilding the Space Station Truss (and the Solar Panel Tear)Jared Owen · 1m+ views

4 minBuilding in Space! On-orbit Servicing, Assembly, and Manufacturing (OSAM)NASA Langley Research Center · 10k+ views

International Space Station Facts and Figures (NASA) · In Space Assembled Telescope (ISAT) Study Preliminary Findings (NASA Technical Reports Server)

Additive manufacturing on orbit means running a 3D printer inside a spacecraft so the crew can make a part instead of waiting for one to be launched. The physics is friendly, because fused-filament printing holds its melt with surface tension and bonds each layer by conduction, neither of which needs gravity, so the hard parts are the housekeeping: the machine has to be sealed against fumes in a closed cabin, fit a rack, and run on a few hundred watts. Made In Space flew the first one to the ISS in 2014 inside the Microgravity Science Glovebox, with a build volume of 6 cm by 12 cm by 6 cm, and it printed 55 acrylonitrile butadiene styrene parts between 2014 and 2016. The demonstration everyone remembers is the ratchet wrench of December 2014: designed on the ground, reviewed and approved in under a week, uplinked as a file, and printed in four hours as 104 layers, 11.4 cm long. It was then returned to Earth for testing rather than used. Redwire's Additive Manufacturing Facility replaced it in 2016 as a commercial service with a 14 cm by 10 cm by 10 cm build volume in ABS, high-density polyethylene, and PEI-PC, and it is still operating. Metal arrived much later: ESA's Airbus-built Metal 3D Printer was installed in Columbus in January 2024, melts stainless steel wire with a near-infrared laser, and its first complete sample came back to Earth in February 2025 to be cut apart and compared against an identical part printed on the same machine before launch.

The real product is schedule, not mass. A design file weighs nothing and arrives in minutes, where the same part on a cargo flight waits for a resupply slot months out, and that is genuinely useful for the long tail of small non-structural items nobody thought to manifest. It is also cheap to try, since a printer is a locker-sized box rather than a vehicle. The weaknesses are that the build volumes are small, the material list is short, and no printed part has yet been qualified and installed as a flight-critical spare, so the demonstrated capability is tools, clips, brackets, and containers. Feedstock still launches from Earth, which means the mass saving only exists if you print fewer spares than you would otherwise have flown, and nobody has shown that on a real manifest. The interesting version is closing the loop by turning the station's plastic waste back into filament, and that has been tried once: Tethers Unlimited's Refabricator went up in 2019 to recycle ULTEM 9085 parts into feedstock, hit an anomaly in the recycling system at startup, and never met its objectives.

If you are provisioning a crewed vehicle in low Earth orbit with regular resupply, a printer is worth its rack space for the tail of the spares list, not for the items you already know you need. Print tools, jigs, containers, and adapters; do not plan on printing anything structural, pressure-bearing, or in the load path until someone has qualified one, which has not happened. If you need metal, treat it as a technology demonstration and not a supply chain, since exactly one metal printer has flown and it produced test coupons. If the mission is Mars or a long lunar stay where resupply is not an option, the calculation flips, because the value of a part you cannot otherwise get is unbounded. And if you are funding this area for the mass benefit rather than the schedule benefit, fund recycling rather than printing, because feedstock from Earth is what caps the saving and the one recycler flown so far failed at startup.

First ISS printer 2014, build volume 6 cm by 12 cm by 6 cm, 55 ABS parts printed 2014–2016 · ratchet wrench designed and approved in under a week, printed in four hours as 104 layers, 11.4 cm long, then returned to Earth · Additive Manufacturing Facility from 2016, 14 cm by 10 cm by 10 cm, ABS, HDPE, and PEI-PC · ESA metal printer installed January 2024, stainless steel wire melted by laser, first sample returned February 2025 · Refabricator flown 2019 to recycle ULTEM 9085 into filament, anomaly at startup · NASA prices ISS upmass to commercial users at $20,000/kg

Made In Space's 3D Printing in Zero-G demonstration in 2014 and the ratchet wrench that followed it; Redwire's Additive Manufacturing Facility, on the ISS since 2016 and sold as a commercial print service; Tethers Unlimited's Refabricator in 2019; Redwire Regolith Print in 2021, which extruded a regolith simulant and thermoplastic blend but hit production problems that stopped the returned coupons being analyzed; ESA and Airbus's Metal 3D Printer in Columbus from 2024; and OSAM-2, formerly Archinaut One, which would have printed a 10 m and a 6 m beam outside the spacecraft before it was canceled.

This is a small business dressed as a large one. The Additive Manufacturing Facility is a commercial service with a price list, but the customers are NASA, the ISS National Lab, and researchers, so the money is agency money routed through a commercial operator. The mass arithmetic does not carry it on its own: NASA prices ISS upmass to commercial users at $20,000/kg, so a 50 g printed bracket displaces about $1,000 of cargo, which will not repay a rack, a crew procedure, and a qualification campaign. What can repay it is avoiding a whole resupply slot for a mission-stopping small part, and that value shows up rarely and unpredictably, which makes it hard to underwrite. The clearest commercial path is not printing at all but recycling, since feedstock launched from Earth caps every mass saving and station packaging waste is already up there and free. Until someone flies a working recycler and installs a printed part in a critical position, the honest read is that on-orbit printing saves a resupply slot rather than enabling anything that was previously impossible.

9 minMade in Space: 3-D Printing Could Change the Way Astronauts Travel | Short Film ShowcaseNational Geographic · 50k+ views

6 minHow This Next Generation Satellite Will 3D Print Itself in SpaceSeeker · 100k+ views

Space Station 3-D Printer Builds Ratchet Wrench To Complete First Phase Of Operations (NASA) · A close-up look at the first metal part made in space (ESA)

A deployable aperture is a mirror, antenna, or shield that is folded to fit a fairing and unfolds once, on its own, after launch. The mechanism set is always the same: non-explosive release devices that fire to free a stowed item, hinges and latches that set the final geometry, motors and cables that pull membranes into tension, and springs that supply the motion. The James Webb Space Telescope is the extreme case and the reason this entry is the counter-argument to assembling structures in orbit. Its 6.5 m primary mirror is 18 hexagonal beryllium segments 1.32 m flat to flat, each 20 kg bare and about 40 kg as a full segment assembly with its actuators, and its sunshield is five Kapton layers measuring 21.197 m by 14.162 m that hold a 299 °C temperature drop across them. All of that folded into an Ariane 5 fairing 5.4 m across and 13 m tall, then unfolded into a 21 m by 14 m by 8 m observatory over about 14 days with nobody touching it. Deployment ran to more than 50 major steps driven by 178 release mechanisms, with more than 300 single points of failure and no repair possible a million miles from Earth. It worked, and the segments were then aligned to a small fraction of a wavelength to act as one mirror.

Deployment is the only method that has actually put an aperture larger than a fairing into space, and it does so at very low mass: Webb's mirror is about one tenth the mass per unit area of Hubble's, and unfurlable wire mesh reflectors are lighter still, which is why they are standard on communications and radar satellites. Nothing has to be operated, so there is no crew, no robot, and no light-time problem. The weakness is that every one of those mechanisms is a one-shot item with no second attempt, and a single stuck pin can cost the mission. Galileo's high-gain antenna failed to open in 1991 and the spacecraft flew its entire Jupiter tour on a low-rate backup. The other cost is verification: a structure light enough to fly is usually too weak to hold its own shape in 1 g, so ground testing needs gravity-offload rigs and partial deployments that never exercise the full sequence, and that test campaign is what dominates the last years of a program. Webb finished commissioning 197 days after launch against a 180-day plan, which is close, but the reason it was close is that the team rehearsed contingencies for years.

If the aperture fits in the fairing, do not deploy anything, because every mechanism you delete is a failure mode you delete. If you need roughly 5–15 m and the surface can be a wire mesh or a segmented reflector, deploy it; that is the proven path and there are suppliers who sell the reflector as a product. Count the single points of failure explicitly and early, because on a folded design the number grows faster than the aperture does, and be honest about which ones you can back out of and which end the mission. If the deployment cannot be fully exercised in 1 g, budget for a gravity-offload test campaign and treat its cost as part of the aperture, not overhead. Only look at assembling the structure in orbit instead when folding stops working, which is somewhere past 15–20 m or when the structure has to change after it is built, and weigh that against the fact that no aperture has ever been assembled robotically in space.

JWST 6.5 m primary from 18 beryllium segments 1.32 m flat to flat, 20 kg bare and about 40 kg per segment assembly · sunshield 21.197 m by 14.162 m in 5 Kapton layers holding a 299 °C drop · folded into an Ariane 5 fairing 5.4 m by 13 m, deployed to 21 m by 14 m by 8 m · more than 50 major deployments, 178 release mechanisms, more than 300 single points of failure · about 14 days of deployment, commissioning complete 197 days after launch against a 180-day plan · Webb's mirror roughly one tenth the mass per unit area of Hubble's

The James Webb Space Telescope is the reference case for a segmented mirror and a tensioned membrane sunshield. NISAR, launched on 30 July 2025, deployed its radar reflector and boom during commissioning and entered science operations in January 2026. Unfurlable mesh reflectors from L3Harris and Northrop Grumman Astro Aerospace are a catalog product for mobile communications and radar satellites. Galileo, launched in 1989, is the standing example of what one stuck deployment costs: its high-gain antenna would not open in 1991 and never did.

Deployable apertures split into two very different businesses. Mesh reflectors are a real commercial product: a handful of suppliers sell them to satellite primes, the design repeats between programs, and the buyer is a commercial or government operator who prices the reflector as a line item. One-off science apertures are the opposite. Webb cost about $10B, and the money went into non-recurring engineering, mechanism qualification, and a test campaign that could never fully rehearse the thing it was testing, none of which amortizes onto anything else. That is the real economic argument for deployment over assembly today: the deployable path has a supply chain and a price list at the small end, and the assembly path has neither at any size. It is also the argument for keeping apertures inside the fairing whenever the science allows, since the cost of an aperture scales with mechanism count and verification effort rather than with the area of the reflecting surface.

2 minJames Webb Space Telescope Deployment Sequence (Nominal)James Webb Space Telescope (JWST) · 1m+ views

12 minJames Webb Space Telescope Launch and DeploymentNorthrop Grumman · 1m+ views

Launching and Deploying the James Webb Space Telescope (NASA Technical Reports Server) · Webb's Sunshield (NASA Science)

In-space production means making a material in orbit and selling it on Earth, on the argument that removing gravity removes buoyancy-driven convection, sedimentation, and the defects that come with them. Three product families carry the field. ZBLAN, a heavy-metal fluoride glass of ZrF4, BaF2, LaF3, AlF3 and NaF, is drawn into mid-infrared optical fiber, and the pitch is up to two orders of magnitude lower insertion loss than silica because microgravity suppresses the crystallization and phase separation that otherwise nucleate as the fiber is drawn. Crystals grow more slowly and align better without convection, which is why protein and small-molecule crystallization has flown repeatedly, including ritonavir crystals grown aboard Varda's W-1 capsule between June 2023 and February 2024. And thin biological films can be built up layer by layer without settling, which is how LambdaVision makes bacteriorhodopsin artificial retinas, 200 layers on its most recent flight. Everything produced so far has been small: the largest single run reported is 5 km of heavy-metal fluoride fiber made by Flawless Photonics aboard the ISS in 2024, and before that the largest reported ZBLAN run was over 90 m. No product has yet been manufactured in orbit and sold at volume on Earth.

The physical argument is sound and has been demonstrated repeatedly at laboratory scale, and for a few products the microgravity version really is better than anything a 1 g process can do. The problem is the freight bill, which is published and which most of these ventures have not yet had to pay. NASA's own reimbursable rates are $20,000/kg to send passive cargo up, $40,000/kg to bring it back, $90,000/kg for a conditioned round trip, $120,000/kg if the payload needs power on the way, and $130,000 an hour for crew time. So a product returned from the ISS has to be worth more than $60,000/kg before it has covered its ride, and more like $120,000/kg for anything that has to stay powered, with the facility, the crew time, and the development on top. The second problem is that the terrestrial competitor keeps improving while an orbital process is being qualified, and terrestrial fluoride glass drawing has improved a great deal since the ZBLAN argument was first made in the 1990s. The third is scale: 5 km of fiber is a good technical result and a rounding error as a production quantity.

Start with the freight bill. If your product is not worth well over $100,000/kg delivered on Earth, at a volume that would justify a dedicated facility, the business does not close and no amount of process improvement fixes it. If the benefit you are chasing is knowledge rather than material, a better crystal structure or a formulation insight, fly the experiment and manufacture on the ground, because that is much cheaper and it is where most of the realized value in this field has come from. If the benefit really is the material itself, write down what the terrestrial process will look like in five years, since that is the competitor and it improves while you qualify. Ask who is paying for the ride today: NASA's In Space Production Applications program had put more than $60M into more than 20 awards by spring 2023 and supplies upmass, downmass, crew time, data and power to awardees at no cost, so a demonstration flown under it has not tested the economics at all. And if you need production volume, note that the ISS is scheduled to retire around 2030 and its commercial replacements are still designs.

NASA reimbursable rates: $20,000/kg upmass, $40,000/kg downmass, $90,000/kg conditioned round trip, $120,000/kg powered round trip, $130,000/hr crew time · so a returned product has to be worth over $60,000/kg to cover freight alone, and over $120,000/kg if it needs power in transit · largest run to date 5 km of heavy-metal fluoride fiber in 2024, against over 90 m from the earlier ZBLAN facility · 200 protein layers per artificial-retina flight · ZBLAN pitched at up to two orders of magnitude lower loss than silica · more than $60M across more than 20 InSPA awards, with transport supplied free · no product manufactured in orbit and sold at volume on Earth

Flawless Photonics' 5 km fluoride-glass run in 2024, funded by NASA, ESA and the Luxembourg Space Agency; Mercury Systems' Fiber Optic Production module and FOMS' SpaceFORM and Space Fibers, both drawing ZBLAN; Redwire's Industrial Crystallization Facility, which grew the first potassium dihydrogen phosphate crystals aboard the ISS in 2021, and its PIL-BOX pharmaceutical crystallization platform, operating since 2023; Redwire's BioFabrication Facility, which printed a partial human meniscus in 2020 for a Department of Defense program; LambdaVision and Space Tango's protein retina films; and Varda Space Industries' W-Series capsule, whose W-1 flight grew ritonavir crystals and returned them.

This is the part of the sheet where the arithmetic is least closed. Development capital is venture money and agency money together, the transport is currently free to the companies that fly under InSPA, and the revenue is zero at volume, so no one has yet run the full loop of buying a ride, making a product, and selling it at a price a customer set. The products that could clear $60,000–120,000/kg are a short list: active pharmaceutical ingredients, specialty optical fiber, some optical crystals, and possibly semiconductor material. Even on that list the quantities are the difficulty, because a market worth entering usually needs tonnes and the flown runs are grams to hundreds of grams. Varda's approach of flying its own free-flying capsule instead of renting ISS resources is the one structural attempt to attack the cost side, since it removes crew time and station integration from the bill. Two things would change the picture: a commercial platform that sells a round trip for a fraction of NASA's rate, and one product actually sold on Earth at a price the buyer chose. Until then, treat every business plan here as a claim about future launch prices rather than a claim about materials science.

14 minHow Space Manufacturing Could Be RevolutionaryScott Manley · 100k+ views

2 minMaking Fiber Optics in SpaceNASA Johnson · 10k+ views

Commercial and Marketing Pricing Policy (NASA) · In-Space Servicing, Assembly, and Manufacturing (ISAM) State of Play - 2025 Edition (NASA Technical Reports Server)

A commercial space station is a crewed platform in low Earth orbit that a company owns and operates and sells access to, with NASA as one customer rather than the owner. The category exists because the ISS is scheduled to retire around 2030 and NASA does not intend to build a replacement. What is being replaced is large: 419,725 kg, 1,005 m³ of pressurized volume, and 75–90 kW from eight solar arrays, costing NASA roughly $3B a year, of which about $1.1B is operations and maintenance and most of the rest is crew and cargo transportation. NASA's procurement is phased. In 2020 it gave Axiom Space a firm-fixed-price contract worth up to $140M over seven years for at least one habitable module attached to the ISS. In December 2021 it signed $415.6M of Space Act Agreements for free-flying station designs: $160M to Nanoracks, $130M to Blue Origin, and $125.6M to Northrop Grumman, which later withdrew and joined the Starlab team instead. Phase 2 switched from certification contracts to funded Space Act Agreements with a three-year base period, milestones running up to five years, and payment on milestones that include critical design review readiness and an initial crewed demonstration in space; NASA planned to award several of those in early 2026. Phase 3 would buy services under ordinary contracts, with formal certification attached.

The cost argument is the strong one. A purpose-built station launched in one or two pieces avoids three decades of accreted hardware and the maintenance load that comes with it, and the ISS's $1.1B a year of operations and maintenance is exactly what an operator hopes to undercut. Fixed-price and milestone-based agreements also move overrun risk onto the supplier, which is why NASA is using them. The weakness is demand. NASA has said it wants to be one of many customers, and today it is the only customer of any size: non-NASA demand is private astronaut missions, sovereign astronaut flights, and research, all of it small next to a station's operating cost. That shows up in the estimates, which do not converge. NASA-sponsored studies of destination development, operations and transportation came out anywhere between $5.1B and $37.5B over 15 years, a spread of seven times, and its own inspector general concluded the numbers would stay unreliable until the designs matured. Schedule is the other weakness: Commercial Crew took eight years from early design maturation to its first crewed flight, and a habitable station is a harder job than a capsule.

If you are a research or manufacturing customer planning payloads, assume the ISS through 2030 and treat every commercial destination date as likely to slip, because the program has already slipped several times and the hardest milestone, a crewed demonstration, is still ahead. If you are underwriting one of these companies, get the size of NASA's commitment in writing, model the business at that number alone, and then ask specifically who the second customer is and what they buy at what price; a plan whose second customer is a category rather than a name is a plan with one customer. If you are choosing between attaching to the ISS first and going straight to a free flyer, the attached route buys revenue and crew access earlier at the cost of depending on a station that is being retired. And if the only buyer turns out to be NASA at a price NASA sets, treat the arrangement as a cost-plus program with a commercial label, and price it that way rather than as a market.

ISS 419,725 kg, 1,005 m³ pressurized, 75–90 kW from eight arrays · roughly $3B a year to NASA, about $1.1B of it operations and maintenance · CLD Phase 1 $415.6M total in December 2021: Nanoracks $160M, Blue Origin $130M, Northrop Grumman $125.6M · Axiom firm-fixed-price contract worth up to $140M over 7 years · 15-year program estimates spanning $5.1B to $37.5B · private astronaut mission fees to NASA of $4.8M for integration and basic services plus $5.2M of crew time · ISS retirement around 2030

Starlab, whose 2021 agreement was signed with Nanoracks and which Northrop Grumman joined in 2023 after withdrawing its own design; Orbital Reef from Blue Origin and Sierra Space, including Sierra Space's expandable Large Integrated Flexible Environment habitat; Axiom Station, which starts as modules berthed to the ISS and is planned to comprise a Payload Power Thermal Module, Habitat 1, an airlock, Habitat 2, and a Research and Manufacturing Facility before detaching; and Vast's Haven-1 and Haven-2, whose trace contaminant control hardware NASA tested in the environmental chamber at Marshall that was used for the ISS life support system. NASA also holds unfunded agreements signed in 2023 with Blue Origin, Northrop Grumman, Sierra Space, SpaceX, Special Aerospace Services, ThinkOrbital, and Vast.

The whole business rests on one number nobody has fixed yet, which is how much NASA will commit to buy each year after 2030. The ISS costs NASA about $3B a year and the commercial successors are pitched at a fraction of that with far less pressurized volume, so a large share of the saving comes from being smaller rather than from being run better. Development money so far is a mix of agency milestone payments and private capital: $415.6M of Phase 1 agreements plus the Axiom contract on the public side, with Vast building Haven-1 under an unfunded agreement that supplies NASA expertise and test facilities rather than money. Revenue outside NASA exists and is thin. Private astronaut missions have flown and the fees NASA charges for one, $4.8M for integration and basic services plus $5.2M of crew time before the ride itself, show how much of that revenue leaves again. The classic anchor-tenant loop applies: an operator needs customers to justify the station and customers need a station to exist before they design experiments and products for it. Watch for two things that would break it, a firm multi-year NASA services commitment with a dollar figure, and a non-government customer buying a second mission with its own money.

16 minWhy Commercial Space Stations Are The Future NASA WantsScott Manley · 100k+ views

13 minNASA's New Space Station Has A Big Problem! (Axiom)The Space Race · 100k+ views

Commercial Space Stations (NASA) · NASA's Management of the International Space Station and Efforts to Commercialize Low Earth Orbit (NASA Office of Inspector General)

A lunar lander carries payload from a translunar trajectory to a controlled stop on the surface. Nothing brakes it but its own engine, so the descent from low lunar orbit costs about 1,900 m/s of delta-v with gravity and steering losses included, and the vehicle ends up mostly propellant tank. The sequence runs to completion once it starts: a braking burn from orbital speed, a pitch-over that brings the site into the cameras' view, terrain-relative navigation against a stored map, a hazard-avoidance divert, then a vertical descent at a few meters per second. Touchdown is where these vehicles fail. Slope and bearing strength at the actual footpad are known only from orbit, the vehicle is usually tall and narrow because tanks stack vertically, and a landing that is slightly fast or slightly sideways puts one leg down first and rolls the stack onto its side. Of the four landers NASA has flown under Commercial Lunar Payload Services, one landed upright and worked, two landed and tipped over, and one never reached the Moon.

The strength is price, and NASA bought it deliberately. CLPS task orders are firm-fixed-price awards to a vendor pool that has run at 13 to 14 companies, so the supplier carries the risk of a failed landing, and NASA's own estimate was about $1M per kilogram delivered when the initiative started in 2019, rising to about $1.2M/kg on the task orders it issued in 2023. Firefly's Blue Ghost Mission 1 shows what that buys when it works: it landed upright in Mare Crisium on 2 March 2025, ran ten NASA payloads, and operated through 16 March, a full lunar day. The weakness is that a fixed price does not make landing easier. Intuitive Machines' IM-1 landed on 22 February 2024 at 80.13°S, came to rest on its side, still operated all of its active payloads, and ceased on 29 February, nearly a week after landing; IM-2 landed at Mons Mouton on 6 March 2025, also on its side, and lasted 13 hours. Astrobotic's Peregrine had a propulsion failure shortly after separation on 8 January 2024 and spent 10 days in space without attempting a landing. NASA accepted that record in exchange for the price and paid no more when the landers failed, which is the trade a firm-fixed-price contract is for.

Buy a CLPS delivery if your payload costs less than the ride, weighs under a few hundred kilograms, and the science still has value if the lander is lost. That is the case the initiative was designed for, and it is the reason NASA's early payloads were deliberately low-cost and non-critical. If the instrument costs more than the delivery, do not fly it on a first flight of a new lander: VIPER is the worked example, a 500 kg rover whose delivery risk was part of why the project was canceled in 2024 and then reinstated three years later on a vehicle that will have flown once first. If you need a specific site, check what the vehicle's navigation has actually demonstrated rather than what it is rated for, since both tip-overs happened at the south pole on terrain that orbital maps had already flagged as rough. If you need to survive the 14-day night, assume you do not get it: every CLPS mission so far has ended at or before sunset, and night survival is a separate purchase. And if the measurement can be made from orbit, make it from orbit, because an orbiter costs a fraction of a landing and does not have to stay upright.

About 1,900 m/s of propulsive descent from low lunar orbit · roughly $1M per kilogram delivered at 2019 estimates, about $1.2M/kg on 2023 task orders · 13 to 14 eligible vendors, firm-fixed-price task orders · 4 CLPS landings attempted, 1 upright and working, 2 tipped over, 1 that never landed · Blue Ghost Mission 1 operated 2–16 March 2025, a full lunar day · IM-2 lasted 13 hours · delivery task orders grew from $781.4M to $984.3M by February 2024, an average delay of 14 months

Firefly's Blue Ghost Mission 1 is the only fully successful commercial landing; Intuitive Machines' Nova-C landers Odysseus (IM-1) and Athena (IM-2) both landed and tipped; Astrobotic's Peregrine failed in transit. In work as of mid-2026: Blue Origin's Blue Moon MK1 "Endurance" for the Shackleton Connecting Ridge, the Griffin-1 lander carrying Astrolab's FLIP rover, Firefly's Blue Ghost Mission 2 stacked on the Elytra orbiter for the far side, and Intuitive Machines' IM-3 "Trinity" for Reiner Gamma. Outside CLPS, China's Chang'e series and India's Chandrayaan-3 have landed successfully, JAXA's SLIM landed within tens of meters of its target and then tipped nose-down, and both of ispace's HAKUTO-R landers crashed.

CLPS is the clearest test in this field of whether fixed-price contracting works for something that has never been routine. NASA planned two deliveries a year against a $2.19B budget through 2028 and got four flown missions in six years, three of which reached the surface, with the eight delivery task orders rising from $781.4M to $984.3M by February 2024, a 26% increase, and seven of eight slipping by an average of 14 months. Two vendors left the program entirely: Masten Space Systems was paid $66.1M in milestones and filed for bankruptcy in July 2022, and Orbit Beyond returned a $97M award two months after winning it. Individual task orders show the same shape, with Astrobotic's Griffin delivery growing from $199.5M to $322.8M. What NASA got for the money is real: a delivered kilogram costs roughly $1.2M on recent awards, and the agency paid nothing extra when landers failed, which is the risk transfer working as designed. The problem the OIG identified is that the market is still almost entirely NASA, so the vendors have no second customer to absorb a bad year, and the larger awards now landing (Blue Origin's $188M with a $280.4M option to deliver rovers, and $190M for VIPER) concentrate the program in fewer, more expensive missions rather than the cheap high cadence it was sold as.

20 minWhy the US is Struggling to Land on the MoonReal Engineering · 1m+ views

2 minExploring the Moon with NASA's Commercial Lunar Payload ServicesNASA · 50k+ views

NASA's Commercial Lunar Payload Services Initiative (NASA Office of Inspector General) · Summary of the Contracted Deliveries of NASA Payloads to the Moon via Commercial Lunar Payload Services (CLPS) (NASA Technical Reports Server)

A surface rover is a chassis, a wheel and suspension system, a power supply, and some way of deciding where to put the wheels next. Vacuum removes the two things terrestrial vehicles rely on: tires cannot hold pressure and grease cannot stay in a bearing, so wheels are metal or composite and every joint is a dry or vacuum-rated lubrication problem. The Apollo Lunar Roving Vehicle used woven piano-wire mesh wheels with titanium chevron treads for exactly this reason. Traction comes from loose regolith with low bearing strength, so slope and wheel slip set the limit rather than motor torque: the LRV was designed for 20–23° in favorable conditions with a 25° maximum, and current lunar rovers sit in the same band. Speed is set by what a driver can see and react to. Apollo 15 and 16 averaged 2.0 and 1.7 m/s made good across their traverses against a design capability of 9–13 km/h, and the Apollo 16 mission report records a peak of 14 km/h. Dust is the other constant, because with no air to slow it, regolith thrown by a wheel flies a long ballistic arc, which is why fenders matter and why losing a fender extension on Apollo 16 and 17 covered the crew and the equipment.

Mobility does more for the return on a landing than almost anything else of the same mass. The three Apollo rovers turned walking distance into about 90 km of traverse across Apollo 15 to 17, and Chang'e-4's Yutu-2 has spent years crossing the far side because it can drive to a new site each lunar day rather than measuring one spot forever. The costs are mass, power, and wear. The vehicle has to carry its own energy through a 14-day night or shut down and be revived, the wheels take irreversible damage from sharp rock (Curiosity's aluminum wheels were holed within a few kilometers of driving over bedrock), and dust works into bearings, gears, and seals faster than anything designed on Earth expects. Reliability is not theoretical either: the Apollo 15 rover's front steering did not work at all, and the crew drove the whole mission on rear steering. Light time is the last constraint, and it is what makes the Moon a different product from Mars. A 2.6 s round trip lets an operator drive in a stop-and-look loop, while a 6 to 45 minute round trip to Mars forces the vehicle to plan its own path.

Put wheels under an instrument when the science needs more than one place, and leave them off when it does not. If the measurement is of the landing site itself, a fixed instrument or an arm on the lander is far cheaper than a rover and does not have to survive driving. If you need to cover hundreds of meters to a few kilometers within one lunar day, a small solar rover in the 10–100 kg class is usually enough, and you should size it around the traverse you can complete before sunset rather than around top speed. If the route goes into a permanently shadowed region, budget power and heat for operating in the dark from the start, because a solar rover that drives into a cold trap is on a countdown. Plan on teleoperation on the Moon and autonomy on Mars: 2.6 s is workable if you accept a slow duty cycle, and anything faster than that has to run onboard. And check slope against slip, not against the motor rating, since the failure that ends a traverse is usually a wheel digging in on a 20° regolith slope rather than a lack of torque.

LRV designed for 20–23° slopes, 25° maximum · design speed 9–13 km/h, 14 km/h peak recorded on Apollo 16 · actual speed made good 1.7–2.0 m/s on Apollo 15 and 16 · about 90 km of total Apollo traverse across three rovers · 2.6 s round-trip light time to the Moon against 6–45 minutes to Mars · NASA's first Lunar Terrain Vehicle awards were $219M to Astrolab and $220M to Lunar Outpost, with $188M plus a $280.4M option to Blue Origin for delivery · Astrolab's crewed vehicle about 900 kg and rated above 9.5 km/h

Three Apollo Lunar Roving Vehicles (1971–72), the Soviet Lunokhod 1 and 2, which were driven from Earth in a stop-and-go loop, and Chang'e-4's Yutu-2, the longest-lived lunar rover. On Mars, Sojourner through Perseverance, which plans its own routes because nobody can drive it. Currently funded: NASA's Lunar Terrain Vehicle Services awards to Venturi Astrolab (CLV-1, derived from its FLEX architecture, about 900 kg and rated above 9.5 km/h) and Lunar Outpost (Pegasus, rated for up to a year of operation and manual, autonomous, or teleoperated driving), with delivery contracted to Blue Origin under CLPS task order CX-2; Astrolab's FLIP rover rides the Griffin-1 lander, and Lunar Outpost's MAPP rover flew on IM-2.

NASA buys rover services rather than rovers, the same structure as CLPS, so the vendor builds and owns the vehicle and the agency pays fixed-price milestones for its use. The first phase of Lunar Terrain Vehicle Services went out in May 2026 as $219M to Astrolab and $220M to Lunar Outpost, which is roughly what a mid-size science spacecraft costs and is the largest commercial commitment to surface mobility ever made. Delivery is the other half of the bill and it scales the opposite way from intuition: NASA awarded Blue Origin $188M with a $280.4M option across two task orders to land the vehicles, and if that covers two vehicles of roughly 900 kg each, $468M works out near $260,000 per kilogram, about a fifth of the $1.2M/kg that small CLPS deliveries cost. That division is done here, not published. Bigger landers are cheaper per kilogram, and that is the main reason surface mobility at this scale is affordable now and was not five years ago. The commercial demand is thin. Beyond agency contracts, the paying customers so far are payload hosts renting space on a small rover, such as the Nokia LTE experiment carried by Lunar Outpost's MAPP, and those are marketing budgets rather than operations budgets. The realistic path to a second customer runs through construction and prospecting work, where a rover doing site preparation or hauling regolith is billed by the hour, and none of that exists yet.

16 minHow did the Lunar Roving Vehicle work? (NASA)Jared Owen · 1m+ views

15 minSpacecraft with Wheels: The Lunar Roving Vehicle (archival film)NASA's Marshall Space Flight Center · 100k+ views

Best Practices for the Testing of Planetary Roving Vehicle Mobility Systems and Tires (NASA Technical Reports Server) · NASA Provides Update on Moon Base Rovers, Landers, Missions (NASA)

Water at the lunar poles is the resource the whole cislunar propellant argument rests on, and nobody has yet measured it in place. The evidence is remote sensing. Lunar Prospector's neutron spectrometer found the strongest hydrogen signal at Cabeus crater, but its footprint was about 40 km across, so the apparent peak works out to only about 0.25 wt% water-equivalent hydrogen; reconstructing the signal at higher resolution suggests roughly 1 wt% inside permanently shadowed areas. When LCROSS deliberately crashed a spent upper stage into Cabeus in 2009 and a trailing spacecraft flew through the plume, the ejecta implied 3 to 10 wt% water at that one point. Those two numbers differ by an order of magnitude, which means the ice is buried, or patchy at scales far below the orbital footprint, or both. What is not known matters more than what is: the depth profile, the lateral variability, and above all the physical form, since loose frost, ice filling pore space, and regolith cemented into something with the strength of concrete are three completely different mining problems. Extraction concepts assume one of those and get designed around it, which is why prospecting has to come first.

The prize is real. Water gives both liquid oxygen and liquid hydrogen, it is the only lunar resource that yields hydrogen at all, and oxygen is 78–85% of the propellant mass in a LOX/hydrogen or LOX/methane stage. The counterargument is grade. At 1 wt%, producing one kilogram of water means heating about 100 kg of regolith, and regolith's specific heat near 0.8 kJ/kg·K over a 200 K rise is about 4.4 kWh of heat before any of it goes into subliming ice (that arithmetic is done here). The energy cost is therefore set by the ore grade rather than by the water, which is exactly why the concentration measurement is load-bearing and why a factor of three in grade is a factor of three in plant power. End to end, extraction plus electrolysis runs roughly 8–11 kWh per kilogram of water, 2–5 to liberate it and about 6 to split it at 55 kWh per kilogram of hydrogen. Outside the cold traps the resource is absent rather than marginal: sunlit-surface water measured from an airborne telescope runs at a few hundred parts per million, thousands of times too dilute to process. Add to that the fact that the shadowed regions where the ice is sit near 40 K with no sunlight, so a prospecting rover that drives into one is running on stored energy from the moment it enters.

Treat this as a prospecting problem until somebody publishes a measured grade at depth, and price extraction plants accordingly. If you are designing a plant now, size it against the 1 wt% case rather than the LCROSS case, since the low number is what the wide-area data supports and the high number came from one impact. If your architecture only needs oxygen, do not mine ice at all: oxygen is 40–45% of ordinary regolith everywhere on the Moon and the excavation happens in sunlight, which is a far easier engineering problem than working at 40 K in permanent darkness. Mine ice when you specifically need hydrogen, or when the same plant also supplies life-support water and the two demands together justify the power. If you are underwriting a business, ask what would change if the grade came back at 0.5 wt%, because a plant sized for 5 wt% at that grade needs ten times the excavation rate and does not close. And treat the storage end as part of the plant, since liquid hydrogen at 20 K is harder to keep than it is to make.

Apparent 0.25 wt% water-equivalent hydrogen at Lunar Prospector's roughly 40 km footprint · about 1 wt% after higher-resolution reconstruction inside permanent shadow · 3–10 wt% inferred from the LCROSS plume at one point in Cabeus · about 4.4 kWh of heat per kilogram of water at 1 wt% grade, derived here from regolith specific heat · roughly 8–11 kWh per kilogram of water end to end, 2–5 to extract and about 6 to electrolyze · a few hundred ppm on the sunlit surface · cold traps near 40 K · VIPER canceled July 2024, reinstated September 2025 on a $190M CLPS task order for delivery in late 2027

Lunar Prospector's neutron spectrometer (1998) and LCROSS's Cabeus impact (2009) are still the two measurements the whole field argues over. VIPER is the mission built to settle it: three instruments and a one-meter drill on a 100-day traverse at Mons Mouton, canceled by NASA on 17 July 2024 over funding, budget risk, and lander delays, then reinstated on 19 September 2025 as CLPS task order CS-7 to Blue Origin, with the delivery option exercisable only after Blue Origin's first Blue Moon MK1 flight. PRIME-1 flew on IM-2 in March 2025 with the TRIDENT drill and the MSolo mass spectrometer; the drill exercised all of its operational functions, but with the lander on its side the mass spectrometer ended up measuring propellant vented from Athena's own tanks rather than lunar volatiles. ESA's PROSPECT drill package is manifested on a later Intuitive Machines delivery to Mons Mouton.

The economics of lunar water are a resource-assessment problem wearing an engineering costume. NASA's own VIPER mission description said the distribution of the water is not understood well enough to evaluate economic models, and that was written in 2020 and is still true. What the resource competes against is delivered mass at roughly $1.2M per kilogram on recent CLPS task orders, which sounds like an enormous margin until you count what has to be landed first: an excavator, a thermal extraction system, an electrolyzer, a liquefier, cryogenic storage, and the tens of kilowatts of continuous power to run all of it through a 14-day night. The capital arrives years before the first kilogram of product, and the customer for that product is a propellant depot that does not exist. Government money is what funds every step today, and it is unstable money: VIPER was built, canceled, and reinstated inside 14 months without the hardware changing. There is one useful precedent on the property-rights side, which is NASA's 2020 contracts to buy small quantities of lunar regolith from commercial collectors, one of them awarded at $1, written specifically to establish that a company can sell material it gathers in space. Until a grade is measured in place, the honest description of this business is exploration, and exploration is priced very differently from mining.

9 minSearching For Ice In The Moon’s Shadowed CratersFraser Cain · 10k+ views

VIPER – Volatiles Investigating Polar Exploration Rover: Mission Overview (NASA Technical Reports Server) · Current Theories of Lunar Ice (arXiv)

Lunar regolith is 40–45% oxygen by mass, locked into silicates and oxides everywhere on the Moon rather than concentrated at the poles, so any route that breaks those bonds turns ordinary dirt into propellant oxidizer and breathing gas. Three routes are seriously studied. Hydrogen reduction passes hydrogen over regolith at 900–1,000 °C, strips oxygen from the iron in ilmenite as water, and electrolyzes the water; it runs cool and it is the most ground-tested, but it only touches one mineral, so a NASA plant model gives about 1% overall yield and needs 50,000 kg of regolith a year to make 500 kg of oxygen. Carbothermal reduction runs far hotter, around 1,600 °C, using methane in a recycled loop to pull oxygen out as carbon monoxide; the same NASA model gives about 10% yield, so 5,000 kg of regolith a year for the same 500 kg of oxygen. Molten regolith electrolysis skips the reagent entirely and passes current directly through molten silicate, releasing oxygen at the anode and leaving an iron-silicon alloy at the cathode. Energy per kilogram of oxygen is the number that decides whether any of this is worth landing, and NASA's own plant sizing of 1.63 kg an hour on 25.8 kW works out to about 16 kWh/kg.

The strength is that the feedstock is everywhere and sunlit, which makes this a much easier place to work than a 40 K permanently shadowed crater, and oxygen is most of the mass of a propellant load in the first place. The routes also separate cleanly on what they cost you. Hydrogen reduction has the mildest reactor and the worst yield, so its 200 kg of regolith a day puts the burden on the excavator; carbothermal moves ten times less regolith, about 20 kg a day for the same output, but needs a reactor at 1,600 °C and a working methane recovery loop. Molten regolith electrolysis has the best thermodynamics and gives structural metal as a byproduct, and its unsolved problem is the anode, because nothing cheap survives molten silicate at 1,600 °C for years and yttria-stabilized zirconia degrades in test. The weakness they share is energy. The thermodynamic floor for splitting silicate is roughly 5–7 kWh per kilogram of oxygen, worked out here from the free energy of silica dissociation, so a real plant at 16 kWh/kg is spending most of its input on heating regolith and on losses, which is also why the three routes land closer together in practice than their chemistry suggests.

Make oxygen locally only when there is a customer on the surface consuming tonnes a year, which describes no mission funded today. If the need is breathing gas for a small crew, bring it: a person consumes about 0.9 kg of oxygen a day, so a four-person crew for a month is 108 kg, and landing that costs far less than landing a plant and its reactor. If the need is ascent propellant, the arithmetic changes, because a crewed ascent stage needs tonnes and the oxidizer is most of it. Pick hydrogen reduction if you are flying a first demonstration and want the lowest reactor temperature and the most test heritage, and accept that your excavator does most of the work. Pick carbothermal if excavation and regolith handling are the constraint rather than reactor temperature. Pick molten regolith electrolysis if you also want metal, and only after somebody shows an anode with a multi-year life. In every case size the power system first, because 25 kW continuous through a 14-day night is a fission reactor question, not a solar array question.

Regolith is 40–45% oxygen by mass · about 16 kWh per kilogram of oxygen at NASA's plant sizing of 1.63 kg/h on 25.8 kW · hydrogen reduction about 1% yield, 50,000 kg of regolith a year for 500 kg of oxygen, roughly 200 kg a day · carbothermal about 10% yield, 5,000 kg a year, roughly 20 kg a day · hydrogen reduction at 900–1,000 °C, carbothermal and molten electrolysis near 1,600 °C · thermodynamic floor roughly 5–7 kWh/kg of oxygen, derived here · a 500 kg/yr plant modeled at about 260 kg total mass including excavator and hoppers

NASA's Carbothermal Reduction Demonstration ran an automated reactor in a vacuum chamber at Johnson Space Center with Sierra Space and extracted oxygen from simulant. Molten regolith electrolysis work at NASA and in industry, including Lunar Resources in Houston and Helios in Israel, is at laboratory scale and largely focused on anode life; NASA testing of zirconia-based anodes is published. On another planet the precedent already exists: MOXIE on Perseverance made about 122 g of oxygen across 16 runs from Martian atmosphere rather than regolith, which is the only in-situ oxygen production anyone has demonstrated off Earth. ESA's PROSPECT package will drill and measure at Mons Mouton, which is the prospecting step this depends on.

No paying customer exists, and that is the honest starting point: every dollar in this field is a civil agency's. The value proposition is delivered mass, since a kilogram of oxygen made on the surface displaces a kilogram that would otherwise cost roughly $1.2M to land on a CLPS task order. A 500 kg/yr plant therefore displaces something like $600M a year of delivered mass, against a modeled plant mass near 260 kg, which lands for about $300M. That looks like a six-month payback and it is not, because the plant number excludes the thing that actually dominates: 25 kW continuously through a 14-day night means a fission surface power unit in the tonnes, plus radiators, plus the spares for machinery that has to run unattended for years in abrasive dust. The other missing piece is demand. Oxygen is only worth making at the rate somebody burns it, and nothing planned this decade consumes tonnes a year on the surface, so a plant sized for a real customer is being designed ahead of the customer by roughly a decade. What would change the picture is a reusable lander refueling on the surface, because that single customer would consume tens of tonnes a year and would justify the reactor on its own.

3 minNASA Extracts Oxygen From Lunar Soil SimulantNASA Johnson · 10k+ views

Component and System Sensitivity Considerations for Design of a Lunar ISRU Oxygen Production Plant (NASA Technical Reports Server) · Hybrid lunar ISRU plant: a comparative analysis with carbothermal reduction and water extraction (arXiv)

Lunar dust is the near-term problem on this list, and Apollo established how fast it does damage. With no wind or water to round them, grains stay sharp, and micrometeorite impacts weld them into glassy agglutinates, so the material behaves like an abrasive powder that also carries an electrostatic charge from solar ultraviolet and sticks to everything. NASA's catalog of Apollo mission reports sorts the effects into nine categories, including clogging of mechanisms, abrasion, thermal control failures, seal leakage, and inhalation of the fraction below about 5 µm, which is the part that reaches the lungs. The numbers are blunt. Apollo 12's suits were more worn after 8 hours on the surface than the training suits were after 100, worn through the outer layer into the insulation above the boot, and one or two more excursions could have caused a pressure failure. The same suits went from sealing tight to leaking 0.15 psi/min after the first excursion and 0.25 psi/min after the second, against a 0.30 psi/min safety limit. Dust on radiator surfaces ran the Apollo 12 magnetometer about 68 °F hotter than designed, and the Apollo 16 and 17 rover batteries went over temperature because brushing the dust off did not work. Construction from regolith is the longer-term half of this entry, and it is still at the sintered-test-coupon stage.

Dust mitigation has one demonstrated technology and a long list of concepts. NASA's Electrodynamic Dust Shield, developed at Kennedy Space Center, runs a traveling electric field across a surface to lift and sweep charged grains off it, and it removed regolith from glass and thermal radiator samples on the lunar surface during Firefly's Blue Ghost Mission 1 in March 2025. That is the first on-surface demonstration of anything in this field, and it works on the surfaces that matter most, meaning radiators, solar cells, camera windows, and visors. What it does not address is the inside of a mechanism or a seal, which is where Apollo's worst failures were, and the answer there is still tolerance rather than removal: dust-rated bearings, labyrinth seals, and keeping suits outside the habitat entirely. On the construction side, sintering regolith with microwaves, lasers, or concentrated sunlight produces bricks with compressive strengths reported up to 45 MPa, comparable to ordinary structural concrete, so the material is adequate. The gap is the machine, because nothing has been built that sinters at construction rates in vacuum, in one-sixth gravity, on power that has to come from the same reactor as everything else.

Design for dust on every surface mission and treat construction as optional. If your hardware has a radiator, a solar array, an optical window, or a visor, plan for dust removal or for a large end-of-life derating, since the Apollo record shows accumulation you cannot brush off. If it has a bearing, a gear, or a seal, plan for tolerance rather than cleaning, and qualify it against a simulant with agglutinates rather than a crushed basalt that has none. Keep dust out of the pressurized volume by architecture, using suitports or a dedicated dust lock, because Apollo's approach of bringing suits inside contaminated the cabin and the crew's lungs on every mission. On the construction side, the first thing worth building is a landing pad, because a lander's plume throws regolith on ballistic trajectories that reach a long way (Apollo 12 landed under 200 m from Surveyor 3 and pitted it), and the second is berms and shielding, where you need bulk mass rather than precision. Leave printed habitat walls for later: if the structure has to hold pressure, a launched shell is far more predictable than a sintered one, and regolith is best used as the shielding piled on top.

Apollo 12 suits more worn after 8 hours on the surface than training suits after 100 · suit leak rate 0.15 psi/min after the first excursion and 0.25 after the second, against a 0.30 psi/min limit · Apollo 12 magnetometer about 68 °F above design because of dust on thermal surfaces · particles below about 5 µm are the inhalation concern · Electrodynamic Dust Shield demonstrated on the lunar surface in March 2025 on Blue Ghost Mission 1 · sintered regolith simulant reported up to 45 MPa in compression · ICON's NASA construction award $57.2M through 2028

The Electrodynamic Dust Shield flew as one of ten NASA payloads on Blue Ghost Mission 1 and cleared regolith from glass and radiator surfaces. SCALPSS, the stereo camera set that measures how a descent engine erodes the surface underneath it, has flown on Blue Ghost and is manifested on Blue Origin's first Blue Moon MK1. For construction, NASA's Moon to Mars Planetary Autonomous Construction Technologies project at Marshall awarded ICON a $57.2M SBIR Phase III contract running through 2028 for its Olympus system, aimed at landing pads, roads, and habitat structures. Sintering research runs on standardized simulants such as JSC-1A and the LHS and LMS highland and mare series.

Dust is a cost that shows up as shortened hardware life rather than as a line item, which is why it is chronically underfunded until something fails. The Apollo evidence is that a few days of surface activity consumed a meaningful fraction of a suit's life, so on a program planning 30-day stays the replacement cost of soft goods, seals, and bearings is a recurring bill paid at $1.2M per delivered kilogram. That gives dust mitigation an unusually clean business case for a technology at this maturity: a shield that costs a few kilograms and a few watts and doubles a radiator's usable life pays for itself in avoided resupply. Nearly all of the money is agency money, with defense interest in the same technologies for terrestrial sand and dust, and the Electrodynamic Dust Shield was funded through NASA's Game Changing Development program rather than by a customer. Construction is at an earlier and more speculative stage, where ICON's $57.2M over six years is the largest single commitment and is small relative to the problem. The economics only work when there is enough surface activity to amortize a construction machine, since one landing pad built by a machine that had to be landed at $1.2M/kg is a very expensive pad, and the second and tenth pads are what make it cheap.

8 minApollo, the Lunar Dust and NASA's Dirty ProblemCurious Droid · 500k+ views

9 minThis is why moon dust is such a problem for NASAVerge Science · 50k+ views

The Effects of Lunar Dust on EVA Systems During the Apollo Missions (NASA Technical Reports Server) · Overview of lunar dust toxicity risk (npj Microgravity)

A habitat is a pressure shell, a thermal system, and a set of machines that keep the air and water inside it usable. The number that describes how good those machines are is closure, the fraction of each consumable recovered rather than resupplied. NASA's baseline figures put one crew member at 0.895 kg of oxygen consumed and 1.085 kg of carbon dioxide produced per day, about 1.56 kg of food as eaten, and roughly 3.2 kg of water for drinking and food preparation, which is about 5.7 kg a day of consumables with no recycling at all. The International Space Station closes those loops very unevenly. Its water recovery system now runs at roughly 96–98% closure, including a brine processor that recovers what the urine processor leaves behind. Oxygen closure sits near 50%, because the Sabatier reactor that recombines carbon dioxide with hydrogen only recovers part of it and the rest vents as methane. Food closure is zero, and every calorie eaten in orbit for 25 years has been launched.

Closure is worth buying and it gets steeply more expensive as it approaches 100%. Water is the easy win because it is the largest single mass flow, so recovering 96–98% of 3.2 kg a day removes roughly 3 kg of daily resupply per crew member for a fixed hardware and power cost. Oxygen is harder, because the recovery step is chemistry rather than filtration, and going past 50% needs a Bosch reactor or plasma pyrolysis that NASA has been developing for decades without flying operationally. What closure costs is mass, power, and spares. Making 0.9 kg of oxygen a day means electrolyzing about a kilogram of water, which at realistic efficiency is roughly 6 kWh, or about 250 W of continuous power per crew member for breathing gas alone (that figure is derived here). The other cost is maintenance, and the station's recycling hardware is among the most repaired equipment on board, so spares mass has to be counted against the resupply mass saved. Food is the loop nobody has closed, because growing a person's calories takes tens of square meters of crop area at a few hundred watts per square meter, which is kilowatts per person against 250 W for oxygen.

Decide closure by mission duration and by what a kilogram of resupply costs where you are going. If the mission is weeks and resupply is routine, carry the consumables: open-loop hardware is lighter, simpler, and has almost nothing to break. If the mission is months to years in low Earth orbit, buy water recycling first, since it is the largest flow and the technology is operational. Add oxygen recovery next, and treat anything past about 50% closure as development rather than procurement. Do not plan on growing food to save mass on any mission now funded, because the power and area cost more than the food weighs; grow plants for crew morale, fresh produce, and research, and be honest that those are the reasons. On the shell, use a launched rigid or expandable module wherever the schedule matters, and consider regolith piled on top for radiation shielding on the surface rather than pressurized structure built in place. The market question for commercial stations is covered by Commercial space stations and platforms on this sheet; this entry is about what runs inside them.

0.895 kg of oxygen consumed and 1.085 kg of carbon dioxide produced per crew member per day · about 1.56 kg of food and roughly 3.2 kg of water for drinking and food preparation · about 5.7 kg a day of consumables with no recycling · ISS water recovery 96–98%, oxygen recovery about 50%, food recovery zero · roughly 6 kWh, or about 250 W continuous per crew member, to electrolyze oxygen, derived here · tens of square meters of crop area per person at a few hundred watts per square meter · ISS operations and resupply averaged about $1.1B a year across FY2016 to FY2020

The ISS environmental control and life support system is the only long-duration regenerative system with a real operating record: the Carbon Dioxide Removal Assembly, Oxygen Generation Assembly, Water Processing Assembly, Urine Processing Assembly, and the Brine Processing Assembly that pushed water recovery toward 98%. Shells span rigid aluminum modules such as Destiny, the roughly 1.4 tonne BEAM expandable module that added about 16 m³ to the station in 2016, and larger expandable designs such as Sierra Space's LIFE habitat. On the exploration side, NASA's shift from an orbital lunar strategy to a surface one has redirected Gateway's HALO power and avionics hardware into surface technology demonstrations, including systems meant to survive the lunar night.

Life support is where the operating cost of human spaceflight actually sits. ISS operations and resupply averaged about $1.1B a year from FY2016 through FY2020, and the reason closure is worth engineering is that every kilogram not resupplied is a kilogram of cargo capacity freed at whatever the launch and delivery price is. In low Earth orbit that price is low enough that closure is a nice-to-have and open-loop systems remain viable for short missions; on the lunar surface, where a delivered kilogram runs about $1.2M, the calculus inverts completely and a system that saves 3 kg a day per crew member removes about 4,400 kg of annual resupply for a four-person crew, which at that delivery price is billions of dollars a year on paper. The catch is that the savings are only real if the hardware does not need its own mass back in spares, which is the calculation the equivalent-system-mass method exists to make, and it is why NASA evaluates a recovery technology on mass plus power plus consumables plus spares rather than on recovery percentage. Development is entirely agency-funded and has been for forty years, with commercial station operators inheriting the technology rather than inventing it. That inheritance is the interesting commercial fact here: a private station buys ECLSS built on ISS heritage, and the failure rates and spares demand come with it.

2 minWater Recycling on the ISSCanadian Space Agency · 5m+ views

4 minHow Does The ISS Get Oxygen?Astro Bytes · 100k+ views

Life Support Baseline Values and Assumptions Document (NASA Technical Reports Server) · Plant and microbial science and technology as cornerstones to Bioregenerative Life Support Systems in space (npj Microgravity)

A reentry vehicle turns orbital kinetic energy into heat in the air ahead of it and keeps the rest out of the structure. Entry interface is conventionally taken at 400,000 ft (122 km), and the speed at that point sets the whole problem: a return from low Earth orbit crosses it near 7.8 km/s, and a return from the Moon near 11 km/s. Convective heating at the stagnation point scales roughly with the cube of velocity, so the step from Gemini's 7.9 km/s to Apollo's 11.1 km/s raised the heating environment about fourfold rather than by 40%. Apollo 4 flew that case in November 1967, entering at 11.14 km/s and measuring a peak heating rate of 425 Btu/ft²-s (about 480 W/cm²) and a total load of 37,522 Btu/ft² (about 43 kJ/cm²), against 164 Btu/ft²-s and 6,889 Btu/ft² on the first test flight, AS-201, which entered at 8.07 km/s. The shape that makes this survivable is blunt, because a detached bow shock standing off a domed face puts most of the energy into the air instead of the vehicle. The remaining freedom is the entry corridor: too shallow and the vehicle skips back out, too steep and deceleration exceeds what the crew and the structure will take. Apollo sized its heat shield against exactly those two bounding trajectories, with the steep boundary fixed by a 20 g biomedical limit, and Artemis I flew the modern version of the same case, crossing 400,000 ft at 10.99 km/s and a flight path angle of -5.66 degrees, skipping back up to 87.6 km, and steering 3,177 nautical miles to its splashdown target.

A capsule is the cheapest shape that works: no wings, no landing gear, and one heat shield geometry to qualify, which is why every crewed vehicle flying today is one. Zero-lift ballistic entry is the robust fallback, because it needs no guidance and no working attitude control, but peak deceleration runs around 8–9 g and the landing point can be hundreds of kilometers off target; Soyuz has reverted to it in flight more than once. Offsetting the center of gravity so the capsule trims at an angle of attack buys a lift-to-drag ratio near 0.3, which cuts peak deceleration to roughly 4 g from low orbit and 6–7 g on a lunar return and lets guidance fly to a point, and Apollo 4 landed within 10 miles of its predicted spot after 1,951 nautical miles of entry range. The cost of that is a vehicle depending on a working inertial unit and reaction control system, so the graceful-degradation path is back to ballistic with the g-load and dispersion that implies. The heat shield itself has no redundancy and no abort once entry begins, which is why qualification, not manufacturing, dominates a capsule program. Water landing adds a recovery fleet, which is a standing operating cost rather than a one-time one.

If you need to bring people or hardware back intact, start with a capsule and leave it only for a requirement it cannot meet. If the payload cannot take 8–9 g, or you need a runway landing, look at a lifting body such as Dream Chaser and expect to pay for wings, landing gear, and a much larger area of thermal protection. If the return is from beyond low orbit, do not scale a low-orbit heat shield up; size it against the lunar case from the start, because the heating environment is several times worse and entry 34 covers how the material choice changes with it. If your requirement is downmass rather than a particular landing site, accept a ballistic entry and a wide dispersion, since that removes the guidance system and most of the qualification cost. If you are flying commercially from the United States, put the FAA reentry license under Part 450 on the schedule early, because several operators have said the review takes longer than they planned for. Before committing to any of it, check who buys the return trip, because outside NASA's resupply contracts the demand for downmass is currently a few hundred kilograms a year.

Entry interface at 400,000 ft (122 km) · about 7.8 km/s returning from low orbit against about 11 km/s from the Moon · Apollo 4 peak heating 425 Btu/ft²-s (480 W/cm²) and 37,522 Btu/ft² total (43 kJ/cm²) · L/D near 0.3 cuts peak deceleration from 8–9 g ballistic to roughly 4 g from low orbit and 6–7 g on a lunar return · Apollo's 20 g undershoot limit on the entry corridor · Artemis I at 10.99 km/s and -5.66 degrees, skipping to 87.6 km · Cargo Dragon returns roughly 2,500–3,000 kg per flight

Soyuz has been bringing crews home since 1967 and still lands its descent module under parachutes and retrorockets in Kazakhstan; SpaceX's Crew and Cargo Dragon have been the only vehicles returning significant cargo from the ISS since the Shuttle retired in 2011; Orion flew the lunar case on Artemis I in December 2022; Boeing's Starliner and China's Shenzhou use the same architecture. At the small end, Stardust's sample capsule entered at 12.9 km/s in January 2006, the fastest Earth entry flown, and Varda Space Industries recovered its first W-series capsule at the Utah Test and Training Range in February 2024.

Capsule programs are dominated by development and qualification rather than by unit cost. NASA's 2014 Commercial Crew awards were $2.6B to SpaceX and $4.2B to Boeing for one vehicle each, and most of that went into structural, thermal, and abort qualification rather than into building flight articles. Reuse helps only if the heat shield and the structure can be checked quickly, so an ablative shield replaced between flights is often cheaper in practice than a reusable one inspected piece by piece, which is the argument in entry 34. The revenue side is the weak part. Downmass is the one thing only a reentry vehicle provides, and the buyer is almost entirely NASA: two or three Cargo Dragon returns a year at roughly 2,500–3,000 kg each, under resupply contracts priced near $130M a mission. Retiring the ISS removes most of that demand, so the commercial case rests on in-space manufacturing sending back small high-value payloads, and that has to cover a dedicated capsule plus a recovery operation for a few kilograms of product. Defense demand is separate and steadier, because a vehicle that can return from orbit on short notice to a chosen point is a capability a defense customer will fund whether or not a cargo market exists.

8 minBallistic Reentry vs Aerodynamic ReentryScott Manley · 100k+ views

4 minTalk Techy: How Do You Use Physics to Re-Enter Earth’s Atmosphere?Lockheed Martin · 10k+ views

Commercial Space Launch and Reentry Regulations: Overview and Select Issues (Congressional Research Service) · Orion Artemis I Entry Performance (NASA)

A thermal protection system holds the structure below its limit while the outer surface runs at 1,200–2,800 °C, and it does that in one of two ways. An ablator pyrolyzes: the resin decomposes, the gas blows outward into the boundary layer and blocks convective heat from reaching the wall, and the char left behind radiates and slowly recedes. Avcoat 5026-39 is the classic one, a quartz and glass fiber filler with phenolic microballoons in an epoxy-novolac resin at 0.50 g/cm³, used across Apollo and now on Orion; PICA is a carbon fiber preform impregnated with phenolic resin at 0.26 g/cm³, flown on Stardust, Mars Science Laboratory, and Crew Dragon as PICA-X; Boeing's Lightweight Ablator is a silicone with silica microballoons at 0.32 g/cm³ on Starliner. A reusable system gives up no material and keeps its shape: a coated silica tile insulates and re-radiates, and the Shuttle's LI-900 was 99.6% pure amorphous silica fiber at 0.14 g/cm³ under a 0.25–0.50 mm reaction cured glass coating, with reinforced carbon-carbon at 1.44–1.60 g/cm³ on the nose cap and wing leading edges where a tile could not survive. Peak heat flux picks the material: SLA-561V is good for roughly 100–200 W/cm², Avcoat about 900 W/cm², PICA about 1,400 W/cm², carbon-carbon up to 10,000 W/cm² for brief exposures, and carbon phenolic up to 30,000 W/cm². Areal mass is the price, and Apollo's shield weighed about 680 kg, over 10% of the command module, at 2.7 inches thick at the stagnation point and 0.7 inches on the lee side.

Ablators are forgiving in the way that matters: they need no mechanism, they work at fluxes nothing else survives, and adding margin means adding thickness. The cost is that they recede, so the outer mold line changes during flight, they are consumed in one entry, and they are sized against the worst trajectory in the corridor rather than the flown one, which makes most of the mass dead weight. Apollo shows the size of that penalty. Across every mission including the lunar returns, less than 20% of the available ablator was used, and engineers later concluded that half the thickness would still have left an acceptable margin. Reusable tiles are far lighter per unit area and survive a hundred flights, but they are brittle, they absorb water unless waterproofed, and each one is bonded to a strain isolation pad because the aluminum airframe expands and the silica does not. Damage is the failure mode rather than wear, which is why the Shuttle carried 30,759 tiles on Columbia at the start of the program and 24,177 by the end, with blankets replacing tiles wherever the heating allowed.

If the vehicle flies once and enters above roughly 8 km/s, use an ablator and pick it by peak heat flux: a low-density silicone like SLA-561V below about 200 W/cm², Avcoat or PICA in the hundreds to low thousands, carbon phenolic only when the flux runs into the tens of thousands. If the vehicle is meant to fly many times, do not choose tiles on material cost; choose them on how many hours of inspection per square foot the operation can absorb, because that is the recurring bill. If the answer is "not many", a replaceable ablator on a reusable structure is usually cheaper than a reusable surface, which is the trade a capsule operator makes every flight. If you are designing for a lunar or interplanetary return, size the shield against that entry from the start rather than scaling a low-orbit design, since the heating environment is several times worse (entry 33). And if launch price per kilogram is falling for your program, prefer the thicker, duller, cheaper-to-process option, because the mass penalty is the part of the trade that is getting less expensive.

Avcoat 5026-39 at 0.50 g/cm³, PICA at 0.26 g/cm³, LI-900 tile at 0.14 g/cm³, reinforced carbon-carbon at 1.44–1.60 g/cm³ · peak flux capability roughly 100–200 W/cm² for SLA-561V, 900 for Avcoat, 1,400 for PICA, up to 10,000 briefly for carbon-carbon · Apollo shield about 680 kg and over 10% of the command module, 2.7 in thick at the stagnation point · less than 20% of Apollo's ablator consumed on any flight · 30,759 Shuttle tiles falling to 24,177 by program end · LI-900 at $1,160/ft² purchase, 91 hr/ft² to install and 2.10 hr/ft² to inspect and repair per flight

Avcoat flew on every Apollo command module and returns on Orion, changed from Apollo-style honeycomb cells filled with injection guns to machined blocks bonded to a composite skin for producibility; Artemis I came home with more char loss than predicted, which NASA traced to gas trapped in the material during the skip entry. PICA was developed at NASA Ames and flew on Stardust, Mars Science Laboratory and, as PICA-X, on every Dragon. On the reusable side, the Shuttle's LI-900 and AETB-8 tiles, AFRSI blankets, and reinforced carbon-carbon leading edges remain the only flown reusable system of that scale, and Orion reuses AETB-8 tiles on its back shell.

Material price is the small term and labor is the large one. A NASA cost study of thirteen TPS options put LI-900 tile at $1,160/ft² to buy, 91 hr/ft² to install, and 2.10 hr/ft² to inspect and repair on every flight; at the study's $100/hr labor rate that is about $208/ft² per flight of recurring inspection against roughly $225/ft² of fabrication amortized over a 100-flight life. Reinforced carbon-carbon costs ten times as much to buy, at $12,000/ft², and needs 0.14 hr/ft² of inspection, so it costs about $14/ft² per flight to keep. The expensive material is cheap to operate and the cheap material is expensive to operate, and that is the whole reusable-TPS trade. Scale it to the orbiter and the point gets sharper: 24,177 tiles at roughly 0.44 ft² each is on the order of 10,000 ft², so 2.10 hr/ft² is roughly 20,000 technician-hours of inspection and repair per flight, arithmetic done here rather than looked up. The largest single line in that study was neither of these, though. It was payload displacement, $705–1,500/ft² per flight at an assumed $1,000/lbm to orbit, which is the cost of the mass the TPS takes from the payload. That term falls directly with launch price, so cheaper launch shifts the optimum toward heavier systems that cost less to fabricate and inspect, and it is one of the few places on this sheet where falling launch cost changes an engineering answer rather than just a budget.

6 minWhat is a Spacecraft's Heat Shield?Kennedy Space Center Visitor Complex · 10k+ views

3 minArtemis I’s Heat Shield Had an Unexpected ProblemNOVA PBS Official · 10k+ views

History and State of the Art in Advanced Thermal Protection Systems (Oak Ridge National Laboratory) · Thermal Protection Systems Technology Transfer from Apollo and Space Shuttle to the Orion Program (NASA)

Disposal has three engineering answers and the orbit picks which one applies. In low orbit you either let drag finish the job, spend propellant to drop perigee into the atmosphere, or fly a targeted reentry that puts the surviving debris inside a chosen ocean box; in geostationary orbit you raise about 300 km into a graveyard, which costs roughly 11 m/s. The delta-v for a simple deorbit burn is a one-line calculation, the difference between circular speed and the speed at apogee of an ellipse whose perigee is 100 km: about 128 m/s from 550 km, 194 m/s from 800 km, and 290 m/s from 1,200 km, which is arithmetic done here rather than a published table. A targeted reentry costs considerably more, because the entry has to be steep enough to keep the debris footprint small and the burn has to be broken into pieces a thruster can actually deliver. NASA's first controlled deorbit, the 14,910 kg Compton Gamma Ray Observatory in June 2000, burned 900 kg of hydrazine in four maneuvers, and the resulting debris footprint was still 2,638 km long and 117 km wide at three sigma. The rules driving all of this are a residual lifetime limit and a casualty limit. The IADC set 25 years in 2002, the FCC cut it to 5 years for US-licensed spacecraft in 2022, ESA's 2023 standard cut it to 5 years plus a cumulative collision probability under 10⁻³, and the widely adopted safety threshold is a 1-in-10,000 expected casualty count per reentry, counting every surviving fragment carrying 15 J or more.

Designing for disposal is cheap in mass. At 550 km, 128 m/s on a 220 s monopropellant system needs a propellant fraction of 1 − e^(−128/2158), which is 5.8%, so a 500 kg satellite reserves roughly 29 kg; from 800 km it is 8.6% and about 43 kg. The weakness is that a propulsive plan only works on a spacecraft that is still alive and still pointing, and the satellites that most need removing are exactly the ones that failed early. Passive devices avoid that failure mode because they keep working after the bus dies: a drag sail multiplies area-to-mass and pulls a spacecraft at 600–700 km down in a few years instead of decades, and a conductive tether generates drag against the geomagnetic field with no propellant at all. Both add area while the spacecraft descends, which raises its conjunction rate on the way down and puts a deployment mechanism in the critical path. Compliance is the real weakness. ESA's environment report finds that 86–99% of sub-tonne payloads reaching end of life in low orbit since 2020 were already in naturally compliant orbits, while only 57% of payloads above a tonne are set to clear within 25 years. Of the payloads that did reach end of life in a non-compliant orbit over the last decade, 20–85% attempted to comply with the 25-year limit and 5–85% succeeded, with 5–65% meeting the 5-year threshold; the yearly figures range that widely because the trend is rising and constellations dominate the recent years.

Pick the disposal method at design time, by altitude. Below about 400 km, natural decay handles it and you need nothing. Between 400 and 700 km, either reserve the propellant or fit a passive device, and prefer the passive device on a small satellite, because it still works after the spacecraft stops responding. Above roughly 700 km, drag is too weak to rely on, so budget real propulsion and protect the reserve in the operations plan rather than only in the mass statement, since the standard failure is an operator flying extra months of service and then finding the tank too low to reach the disposal orbit. If the spacecraft has a dry mass over about a tonne, or carries dense components like reaction wheels, titanium tanks, and optics, run a survivability analysis: if the casualty estimate exceeds 1 in 10,000, either design for demise by substituting materials that break up, or budget a targeted reentry with the attitude control and multi-burn sequence it needs. In geostationary orbit, take the 11 m/s graveyard raise early. Whether anyone will pay to remove objects that skipped all of this is covered by Debris removal and end-of-life disposal on the space launch and satellites sheet.

About 128 m/s to deorbit from 550 km, 194 m/s from 800 km, 290 m/s from 1,200 km, and 11 m/s to the GEO graveyard · a 5.8% propellant fraction at 550 km on a 220 s system, roughly 29 kg on a 500 kg satellite · CGRO burned 900 kg of hydrazine out of 14,910 kg and still left a 2,638 by 117 km debris footprint · 25-year IADC limit cut to 5 years by the FCC in 2022 and by ESA's standard in 2023 · a 1-in-10,000 casualty threshold per reentry, counting fragments above 15 J · only 57% of payloads above a tonne set to clear low orbit within 25 years · 5–65% successful compliance with the 5-year limit over the last decade

Large constellation operators run the highest volume of disposal actually happening, lowering satellites under their own propulsion at end of life; the Compton Gamma Ray Observatory was NASA's first intentional controlled deorbit in June 2000, and NASA awarded SpaceX about $843M in 2024 to build a US Deorbit Vehicle for the roughly 420-tonne International Space Station. On the passive side, the ADEO family of drag sails has flown as bolt-on disposal hardware, and Tethers Unlimited's Dragracer experiment in 2020 flew two identical cubesats of which the one carrying a 70 m conductive tape reentered within months while its twin stayed up. Enforcement is real but small: the FCC settled with DISH for $150,000 in 2023 over EchoStar-7, which was retired below its agreed disposal orbit.

The propellant is a rounding error and the opportunity cost is not. Reserving 5.8% of wet mass at 550 km costs a 500 kg satellite about 29 kg, which is either payload or several years of drag makeup at that altitude, and the operator feels the loss at the end of the mission rather than at the start, which is why the reserve historically got spent on extra service. A drag sail is a few kilograms of membrane and a deployment mechanism with no tank, plumbing, or propellant qualification behind it, which is why it undercuts a small propulsion system on a satellite whose whole disposal budget is tens of kilograms. Controlled reentry is a different budget entirely: it needs a propulsion system sized for one large burn sequence, attitude control that has to work at end of life, and in the ISS case a purpose-built $843M vehicle. The economics are the classic externality. The operator pays, everyone in that orbit shares the benefit, and nothing in the market corrects it, which is why the binding force is licensing rather than price. Enforcement so far has been light enough to price as a rounding error, since a $150,000 penalty against a geostationary satellite worth a few hundred million is not a deterrent. If regulators ever attach disposal to a bond or an insurance requirement, the calculation changes for the operators with the worst compliance records, which today are the ones flying large payloads above 700 km.

8 minThis Is Where We’re Gonna Bury The ISSSciShow · 100k+ views

2 minADEO – Deorbit Sailing on Angel WingsHPS GmbH - The team to trust · 5k+ views

Process for Limiting Orbital Debris, NASA-STD-8719.14 (NASA Technical Standards) · ESA's Annual Space Environment Report (European Space Agency)

Active debris removal means flying to an object you did not build, which is not cooperating and is probably tumbling, holding onto it, and bringing it down. Target selection is the settled part: NASA's LEGEND modeling picks the objects with the highest mass times collision probability, and a 2008–2009 study concluded that good post-mission disposal plus removal of about five such objects a year would stabilize the low Earth orbit population over 200 years, with anything above five improving it. Those five are large intact bodies, mostly spent rocket stages, not the fragment clouds that dominate the object count. Capture is where the concepts diverge. A rigid clamp or a set of arms grips a known feature such as a launch adapter ring, which is what ClearSpace-1 does with four arms on a 95 kg PROBA-1 measuring 0.6 by 0.6 by 0.8 m; a magnetic docking plate works only on a client fitted with one before launch; a net or a harpoon tolerates an unknown shape but leaves the chaser attached to a flexible, hard-to-control stack; and ion-beam shepherding never touches the target at all, at the cost of months of thrusting and a large power system. Then you still have to move it. Deorbiting from 800 km costs about 194 m/s applied to the combined mass, so a three-tonne upper stage plus a 300 kg chaser on a 220 s system needs roughly 280 kg of propellant for the final burn alone, before any of the phasing and rendezvous, which is arithmetic done here rather than a published figure. That is why every mission flown or funded so far removes exactly one object.

Removal reaches the part of the problem that mitigation cannot, because a spacecraft that died before it could deorbit itself will not comply with any rule written afterwards. The engineering strength is leverage: removing five well-chosen objects a year does more for the environment than sweeping up thousands of fragments, and those targets are large, tracked, and well-characterized. The weakness is that nobody has removed anything yet. Astroscale's ELSA-d captured a prepared client with a magnetic plate in 2021 and suffered thruster failures during the harder non-cooperative approach; ADRAS-J flew to within 15 m of a Japanese H-IIA upper stage in 2024, imaged it and characterized its motion, and completed operations without attempting capture; RemoveDEBRIS fired a net and a harpoon in 2018 at targets it had deployed itself. Tumble rate is the binding constraint, since above a few degrees per second the chaser has to spin up to match, which spends propellant and attitude authority and puts the capture mechanism at risk. And the failure mode is self-defeating: a botched grab on a multi-tonne body produces far more debris than it removes, so every mission is slow and cautious, which keeps the cost per object high.

Do not build a business plan around selling removal as a service today, because the customers are agencies buying single demonstrations of their own countries' debris. If you are that agency and you are responsible for a large object in a congested shell, buy the removal as a program with a named target and expect a program price. If you are building a spacecraft that will fly above 700 km, fit a capture interface at build: a docking plate or grapple fixture costs a couple of kilograms and converts a research problem into a procedure, which is the single highest-leverage thing anyone can do for this field. When you do have to capture something uncooperative, choose by tumble rate and geometry: arms or a clamp when the target is slow and its interface is known, a net when the shape is irregular, a contactless method when the object is too massive to grab safely. And never target another state's object without written consent, because Article VIII of the Outer Space Treaty leaves jurisdiction and control with the state of registry indefinitely and the Liability Convention puts the consequences of a failed attempt on the remover's launching state.

About five objects a year with the highest mass times collision probability stabilizes the low orbit population, on top of good post-mission disposal · roughly 194 m/s to deorbit from 800 km, so about 280 kg of propellant for a three-tonne stage plus a 300 kg chaser at 220 s · ADRAS-J approached to 15 m in 2024 with no capture attempt · ClearSpace-1 at €86M for one 95 kg target, launch planned 2029 · ADRAS-J2 at around ¥13.2B for one object · objects removed by anyone to date: zero

Astroscale is the most advanced operator, with ELSA-d demonstrating magnetic capture of a prepared client in 2021, ADRAS-J performing rendezvous and fly-around inspection of an H-IIA upper stage in 2024 under Phase I of JAXA's Commercial Removal of Debris Demonstration, and ADRAS-J2 contracted under Phase II at around ¥13.2B to go back and remove it. ESA contracted ClearSpace-1 in 2020 for €86M, changed targets in 2024 after the original one was struck by debris, and now aims a four-armed capture vehicle at the 95 kg PROBA-1 with OHB SE leading the industrial team and launch planned for 2029. The Surrey Space Centre's RemoveDEBRIS tested a net, a harpoon, vision-based navigation and a drag sail in 2018. The only object anyone has actually moved is Chinese: Shijian-21 docked with a defunct BeiDou navigation satellite in 2022 and towed it above the geostationary belt.

Cost per object removed is currently the whole story, and it sits near $80–100M: €86M contracted for ClearSpace-1 to remove one 95 kg satellite, and around ¥13.2B for ADRAS-J2 to remove one upper stage. Nothing works commercially at that price, because the benefit is spread across every operator sharing the orbit while the cost falls on one payer, which is the standard shape of a problem markets do not fund. Getting to single-digit millions per object needs several objects per mission, and the propellant arithmetic blocks that directly: carrying the delta-v to deorbit five three-tonne bodies means carrying five times 280 kg of propellant plus the phasing between them. The architecture that escapes this is a chaser that attaches a small deorbit package to each target and leaves it behind, so the vehicle carries kits rather than delta-v, and after that it needs refueling to keep going. The buyer today is a national program removing its own debris, and that is also the legal ceiling, since removing anyone else's object requires that state's consent. The dual-use problem cuts both ways commercially. A vehicle that can rendezvous with and grip an uncooperative satellite is operationally a co-orbital anti-satellite weapon, which makes an open service market politically hard and at the same time gives the same hardware a defense customer with a much larger budget, which is where several of these companies actually get paid.

2 minRemoveDEBRIS’s net captures space debrisSciNews · 100k+ views

3 minRemoveDEBRIS Active Debris Removal demonstration missionSSTLTV · 5k+ views

ClearSpace-1 (European Space Agency) · Threats to U.S. National Security Interests in Space: Orbital Debris Mitigation and Removal (Congressional Research Service)