Same three, different job
A Robotic Outpost Needs Energy, Water and Compute for Different Reasons
Reading a crewed base's requirements onto an uncrewed one is the most common way these projections go wrong.
Crew and robots both run on the triple substrate, but on different duty cycles and in very different proportions. With crew, energy and water keep people alive and compute helps them; without crew, energy and water serve the machines and the product, and compute does the work the crew would have done.
| Substrate | With crew | Without crew | Rated |
|---|---|---|---|
| Energy | Life support, habitat thermal control and ISRU at tonnage scale. Continuous; an interruption is fatal within hours to days | Instrument duty cycles, survival heat for mechanisms and electronics, recharging between drives or burns. Peaky; an interruption is usually survivable | Modeled |
| Water | First and not negotiable: drinking, breathing oxygen by electrolysis, agriculture, the thermal loop and storm-shelter mass | Propellant, thermal working fluid and the solvent for wet refining. Less of it, for different reasons, never zero at an industrial site | Modeled |
| Compute | Assists. Humans on site hold final authority and Earth stays in the loop where light time allows | Replaces the crew. Once the round trip is longer than the failures that matter, autonomy is the mission | Modeled |
Energy
Crew Raise the Firm Load About Eightfold, and Interruption Becomes Fatal
A crewed lunar site protects about 25 kW that can never pause; the same site without crew protects about 3 kW.
| Load that cannot pause | With four crew | Without crew | Rated |
|---|---|---|---|
| Life support: air revitalization, water processing, pressure control | 15 kW | 0 | Modeled |
| Thermal control | 6 kW (habitat at 20°C) | 1-2 kW (survival heaters only) | Modeled |
| Avionics and communications | 4 kW | 0.5-1 kW | Modeled |
| Baseline compute | Included above | About 1 kW | Modeled |
| Firm load | 25 kW | About 3 kW | Modeled |
Firm load matches the illustrative 150 kW bus on The Triple Substrate. Survival heat keeps lubricants, seals, batteries and electronics above their lower limits; mechanisms are designed to sit cold, not to be warm.
The difference is time to harm. Air revitalization off, four people in 100 m3 push CO2 to 1% in about 11 hours and to a dangerous 4% in about two days. A robot that loses power usually loses a schedule, not itself: Yutu-2 sleeps through every lunar night on radioisotope heater units, and JAXA's SLIM lander, not built for night, woke after three of them. When robots do die from an interruption it is usually dust on the arrays, as with Opportunity in 2018 and Zhurong in 2022.
| Case | What fails first | Time | Rated |
|---|---|---|---|
| Four crew in a 100 m3 habitat, air revitalization off | CO2 reaches 1% (about 1 kg per person per day) | About 11 h | Modeled |
| Same habitat | CO2 reaches 4%, immediately dangerous | About 2 days | Modeled |
| Yutu-2 rover · CNSA | Nothing: sleeps through every 14-day lunar night on radioisotope heater units | Operating since January 2019 | Observed |
| SLIM lander · JAXA | Not designed for night; woke after three lunar nights | 2024 | Observed |
| Opportunity rover · NASA | Dust storm cut solar power; never woke | June 2018 | Observed |
| Zhurong rover · CNSA | Entered planned hibernation; dust on the arrays kept it from waking | May 2022 | Observed |
Water
Crew Water Is Small Each Year and Absolute in Reserve
At 98% recovery four people draw about 1.5 t a year; what cannot be negotiated is the protected tank and the shelter.
| Claim | With four crew | Without crew | Rated |
|---|---|---|---|
| Drinking and hygiene make-up at 98% recovery | About 0.1 t a year | 0 | Observed |
| Breathing oxygen by electrolysis | About 1.4 t a year | 0 | Modeled |
| Storm shelter, 20 g/cm2 around a 2 m room | About 5 t, held | 0; electronics use spot shielding | Modeled |
| Thermal-loop working fluid | Under 1 t, held | Under 1 t, held | Modeled |
| Propellant oxygen, 50 t a year | About 56 t a year | About 56 t a year | Modeled |
| Wet refining solvent and reagent | Process-dependent | Process-dependent | Modeled |
Four crew need about 1.5 t of water a year once recovery runs at ISS levels - breathing oxygen alone takes more than drinking-water losses. The hard requirements are standing mass: about 5 t of water around a storm shelter at 20 g/cm2, a protected reserve that propellant production can never draw down, and the loop inventory. Past resupply range, a place without water is not a destination for humans; it is somewhere robots visit.
Without crew the water job changes rather than disappears. Propellant oxygen and hydrogen are the largest draw either way, at about 56 t of water for 50 t of oxygen a year. Wet refining - leaching, electrowinning, washing - needs water as a solvent, and thermal loops need a working fluid. Dust removal does not: in vacuum water flashes and freezes, and dust is cleared electrostatically, as NASA's Electrodynamic Dust Shield showed on Firefly's Blue Ghost lander in March 2025.
Compute
Once the Round Trip Is Longer Than the Failure, Autonomy Is the Mission
The threshold is set by the event, not by a fixed light time: Mars landings already run with no one on Earth in the loop.
| Round trip | Where | What Earth can do | Example | Rated |
|---|---|---|---|---|
| Under 0.3 s | Earth orbit | Fly it directly | Teleoperated robotic arms on the ISS and Tiangong | Observed |
| 2.6 s | Moon | Drive it with lag; supervise everything | Soviet Lunokhod drives (1970s); Yutu-2 | Observed |
| 6-44 min | Mars | Plan once per sol; never react | Perseverance AutoNav; Tianwen-1 autonomous landing | Observed |
| 30-62 min | Ceres | Plan per day or week | Dawn ran sequences uploaded days ahead | Observed |
| 66-106 min | Callisto | Approve plans; set rules | Juno runs fully sequenced science | Observed |
Entry, descent and landing at Mars takes about seven minutes; the one-way light time is 3-22 minutes, so every Mars landing - Perseverance in 2021, Tianwen-1 with Zhurong in 2021 - has been fully autonomous. With crew on site, people are the general-purpose fault managers, and compute assists them. Without crew, compute must do perception, planning and fault management locally, so an uncrewed site carries more on-site compute than a crewed one, not less. Earth's role moves from operator to rule-setter: approving plans, updating models and deciding what the site may do when nobody can answer in time.
Proportions
The Same Site Runs Flat With Crew and in Campaigns Without
Crew make demand steady and protected; robots make it peaky and deferrable.
| Measure | With crew | Without crew | Rated |
|---|---|---|---|
| Firm load as a share of the bus | High - about 25 of 150 kW | Low - about 3 of 150 kW | Modeled |
| Peak-to-average power | Low; life support runs flat | High; drives, drills and furnaces in campaigns | Modeled |
| Water held in reserve before arrival | Several tons, protected | Propellant stock only | Modeled |
| On-site compute | Assists people; modest | Perception, planning and fault management; larger | Modeled |
| Who cleans the solar arrays | A person with a brush | Wind, electrodynamic dust shields or nobody | Modeled |
| Cost of a missed day | A safety event | A schedule slip | Modeled |
A robotic site can run its industrial ladder in campaigns - mining for a week, refining the next - because nothing on site minds waiting. A crewed site cannot schedule life support, so most of its flexibility lives in the industrial loads above the firm line.
Errors
Five Ways Crewed Assumptions Distort Uncrewed Plans
Each produces an outpost sized for the wrong job.
| Error | What it produces | Rated |
|---|---|---|
| Sizing a robotic precursor to the crewed firm load | An outpost carrying five to eight times the protected power it needs | Modeled |
| Assuming robots need no water | No propellant, no wet refining and no working fluid at an industrial site | Modeled |
| Treating crew water as a large annual mass | Missing that the hard requirement is the protected reserve and the shelter, not the yearly draw | Modeled |
| Treating autonomy as an option at Mars | Plans that assume Earth can respond to failures faster than the 6-44 min round trip | Modeled |
| Assuming a crew site is just a robotic site plus people | No reserve built before the first crew lands | Modeled |
The transition is the hard part. The day crew arrive, the firm load steps up about eightfold and a protected water reserve has to exist already. The uncrewed phase therefore has to build the crewed reserves before the first landing; the Presence Gate sets out which destinations reach that point.
China
China Has Flown Both Sides of the Line
Robotic far-side, polar and Mars operations, a crewed station, and a crewed lunar landing targeted before 2030.
| Mission | Crewed or not | What it shows | Rated |
|---|---|---|---|
| Yutu-2 · CNSA | Uncrewed | Radioisotope heater units carry a rover through every lunar night since 2019 | Observed |
| Chang'e-5 · CNSA | Uncrewed | First robotic rendezvous and docking in lunar orbit, December 2020 | Observed |
| Chang'e-6 · CNSA | Uncrewed | Far-side sample return in 2024 through the Queqiao-2 relay | Observed |
| Zhurong · CNSA | Uncrewed | Dust ended a solar-powered Mars rover with no one to clean it | Observed |
| Tiangong · CMSA | Crewed since 2022 | Regenerative life support with water recovery and electrolysis oxygen | Observed |
| Crewed lunar landing · CMSA | Crewed | Mengzhou and Lanyue, targeted before 2030 | Projected |
Contested
Deliberately Not Claimed
The load-bearing uncertainties, tagged with the confidence they actually earn.
The firm-load figures are illustrative. Real habitat power depends on the life-support architecture; ISS-style physicochemical systems and bioregenerative systems differ by several times.
Shelter mass depends on the dose limit. 20 g/cm2 is a working figure for solar particle events; galactic cosmic rays need far more and are not stopped by any practical shelter.
98% recovery has only run in low Earth orbit. The ISS Water Recovery System reached it in 2023 with resupply as a backstop.
Uncrewed compute grows with the job. On-site compute is sized here for operations and fault management; manufacturing and design loops at a robotic site add load that is not yet characterized.
Record
Provenance and Data Sources
- Confidence
- Observed for flown, fielded or measured - Building for hardware and programs in development - Projected for announced plans - Modeled for original analysis
- Status date
- Q4 2026
- Analysis
- First-principles analysis; The Triple Substrate; The Machine Shop MVP
- Physical data
- CO2 production about 1 kg per person per day; NASA ISS Water Recovery System (2023); light times from mean orbital distances
- Missions
- CNSA Yutu-2, Chang'e-5, Chang'e-6, Zhurong; CMSA Tiangong; JAXA SLIM; NASA Opportunity, Perseverance, Electrodynamic Dust Shield on Firefly Blue Ghost Mission 1
- Weakest link
- Habitat firm load and shelter sizing, both architecture-dependent