Light time
Light Time Decides Who Is in Control
One-way signal time sets how much a machine has to decide for itself.
| Destination | Light time, one way | Control mode | Flown by | Rated |
|---|---|---|---|---|
| Earth orbit | Under 0.15 s | Direct teleoperation | Canadarm2; Tiangong's arm | Observed |
| Moon | 1.3 s | Teleoperation with lag | Yutu-2 driven from Beijing | Observed |
| Mars | 3-22 min | Daily plans; the rover drives itself between them | Perseverance AutoNav; Zhurong | Observed |
| Ceres | 15-31 min | Plans per day or week | Dawn | Observed |
| Callisto | 33-53 min | Rules and approvals only | Galileo, JUICE | Observed |
| Pluto | 4-7 h | Full autonomy | New Horizons flyby sequence | Observed |
At the Moon, 1.3 seconds allows a person on Earth to drive a machine, at about a quarter of on-site speed. At Mars, 3-22 minutes does not; every Mars landing has been fully autonomous. Past Mars, machines carry out goals and ask only for approvals. Round-trip figures are on The Substrate Tables.
The yardstick
The Comparison Is a Haulage Supervisor, Not a Driver
Rio Tinto has run autonomous haul trucks in the Pilbara since 2008, supervised from Perth about 1,500 km away. That is the yardstick for an off-Earth fleet: one person already supervises many machines.
| Measure | Value | Basis | Rated |
|---|---|---|---|
| Fleet hours per crew-hour | 500-1,000 | 160 machines working 20 hours a day for six months | Modeled |
| Against a 2025 haulage supervisor | 30-120× | Rio Tinto's Pilbara operators already supervise many trucks | Modeled |
| Unplanned interventions | One per machine per 170-680 h | What the fleet has to achieve | Modeled |
| Dust-exposed seals, 160 humanoids | About 6,400 | 30-50 per humanoid | Modeled |
| Servicing break-even | About 5.5 h per machine per tour, half an hour a month | Above it, the fleet costs more crew than it saves | Modeled |
| Data off a site | 3.2 Mb/s against 1.28 Gb/s | A 400-fold cut by sending what the machines understood, not raw stereo video | Modeled |
Perfect autonomy removes less than one crew member; a suitport removes about six. Autonomy buys output, not people. The fleet's value is moving work from outside the pressure boundary, where an hour costs a person 2.5 hours and dose, to inside, where it costs one.
Past the Moon, Earth drops out of the loop, and the people on site become governors of intent: setting goals, keeping systems running, resolving new failures and deciding what the machines try next. Machines working hours of light time from their owners have to recognize when conditions leave their model and stop safely; that capability is labor infrastructure in the same sense as a suitport.
The garage
Machines Need Suits Too
A humanoid working on a dusty surface has the same problem as a person in a suit.
Robots with many joints need covers for hands, feet and joints, and a garage where they are serviced out of the dust. The garage stays unpressurized, so servicing there is outside work for the crew, at the same overhead as any EVA.
Below about half an hour of human hands per machine per month the fleet saves crew; above it, the fleet costs more crew than it replaces. The servicer at the end of that chain does not have to move: an industrial robot in a powered, dust-free garage is the clean stopping point. Garage and boundaries are on Habitats & Life Support.
These are the surface vehicles that never carry a person: excavators, haulers, drones and humanoids, run autonomously or teleoperated from the site. None is human-rated, which is what lets them be swapped and replaced. Vehicles a person rides in are on Crew Transport.
| Machine | Class | Why | Rated |
|---|---|---|---|
| Small drone | Consumable | Replace it; repair later or never | Modeled |
| Humanoid, hundreds of kg | Swap the unit | Modules serviced indoors | Modeled |
| Excavator, several tons | Service | Too heavy to duplicate casually | Modeled |
| Pressurized rover, reactor, process plant | Maintain in place | The budget buys one | Modeled |
| World | Mobility | Why | Rated |
|---|---|---|---|
| Titan | Fly | Dense atmosphere, low gravity; NASA Dragonfly | Building |
| Mars | Drive, and fly with large rotors | Thin air needs low disc loading; Ingenuity | Observed |
| Moon and airless bodies | Drive or hop on propellant | Nothing for a rotor to push on | Observed |
Flown
Autonomy Already Lands, Drives, Flies and Docks
Every landing and most driving beyond the Moon are already autonomous.
| System | Data point | Rated |
|---|---|---|
| Chang'e-3 to Chang'e-6 · CNSA | Landers hover, map hazards and pick a landing spot on their own | Observed |
| Perseverance AutoNav · NASA | Drives itself over most of its route, planning as it goes | Observed |
| Ingenuity · NASA | 72 flights on Mars, each flown without real-time control | Observed |
| Tianzhou · CMSA | Cargo craft docks with Tiangong autonomously in about two hours | Observed |
| Crew Dragon · SpaceX | Docks with ISS autonomously | Observed |
| Three-Body Computing Constellation · ADA Space and Zhejiang Lab | 12 satellites with onboard AI, flown 2025 | Observed |
| Optimus · Tesla | Humanoid robot; no space qualification | Building |
Onboard and orbital AI compute are on Orbital Compute.
China
China Flies Autonomy at Every Scale
Landers, rendezvous, rovers and an AI constellation.
| Program | Data point | Rated |
|---|---|---|
| Chang'e landers · CNSA | Hazard-avoidance hover and autonomous landing, four times since 2013 | Observed |
| Zhurong · CNSA | Mars rover with autonomous hazard detection, 2021-2022 | Observed |
| Tianzhou · CMSA | Autonomous fast rendezvous and docking with Tiangong | Observed |
| Three-Body Computing Constellation · ADA Space and Zhejiang Lab | 12 satellites, up to 744 TOPS each, May 2025 | Observed |
Contested
Deliberately Not Claimed
The load-bearing uncertainties, tagged with the confidence they actually earn.
Goal-level autonomy fails by geography. If it does not arrive, the Moon keeps working through teleoperation at 1.3 seconds and nothing past it does.
Servicing rates are guesses. No fleet has worked a dusty surface for months; the half-hour-a-month break-even is the test a garage trial on Earth or the Moon has to pass.
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
- Data
- Rio Tinto AutoHaul; NASA Perseverance, Ingenuity and Dragonfly; SpaceX Crew Dragon; Tesla Optimus
- China
- CNSA Chang'e and Zhurong; CMSA Tianzhou; ADA Space and Zhejiang Lab Three-Body Computing Constellation
- Weakest link
- Fleet servicing hours on a dusty surface