Given Starship-scale launch-cost collapse, what concrete thresholds in robot-hour cost and reliability (e.g., $/robot-hour vs $/crew-hour, mean time between failures) mark the point where fully robotic orbital manufacturing beats crew-tended or teleoperated architectures for specific use cases like fiber drawing, semiconductor steps, or satellite servicing, and how do different assumed Wright’s-law slopes for space robotics shift those thresholds over time?
starship-orbital-economy | Updated at
Answer
Approximate crossovers under cheap launch, assuming LEO industrial platforms and today–near-term tech. All numbers are very rough order-of-magnitude, not forecasts.
- Key thresholds (per use case, steady-state)
Common baselines (all cases)
- Effective crew cost in orbit (including life support, training, overhead):
- LEO industrial crew: ≈ $500–2,000/crew‑hour
- Teleop cost (Earth operators):
- Skilled operator + tools: ≈ $50–150/operator‑hour
- Often 0.2–1 operator per active robot for mature tasks
1.1 Fiber drawing (e.g., ZBLAN-style)
- Characteristics: high-value/kg, continuous, low dexterity once tuned.
- Robotic beats crew-tended when BOTH:
- Robot labor: ≤ ~$5–15/robot‑hour (all-in, amortized) AND
- Line MTBF (with auto-recovery): ≥ ~3–6 months between human-intervention events.
- Robotic beats tightly teleoperated cells when:
- Local autonomy handles >90–95% of microfaults; human only for rare resets.
- Effective human touch time < ~0.02 hours/kg drawn.
1.2 Semiconductor steps (niche microgravity modules, not full fabs)
- Characteristics: extreme yield sensitivity, high capex, high process risk.
- Robotic beats crew-tended when:
- Robot labor: ≤ ~$10–30/robot‑hour AND
- MTBF on critical manipulators: ≥ ~1 year between hardware replacement, with graceful degradation.
- Process automation quality yields <~1 “human intervention hour” per 10,000 wafer-equivalents.
- Robotic beats heavy teleop when:
- Latency-sensitive motion is fully local; teleop used only for recipe changes and maintenance.
- Effective human oversight < ~1 hour per tool-day.
1.3 Satellite servicing (inspection, modest repair, refuel)
- Characteristics: episodic, medium dexterity, higher consequence of failure.
- Robotic beats crew-tended when:
- Robot labor: ≤ ~$20–80/robot‑hour AND
- On-orbit servicer MTBF: ≥ ~5–10 complex missions between major hardware failures, with high mission success (>95%).
- Robotic beats teleop-heavy architectures when:
- Local autonomy executes common task libraries; teleop used mainly for anomalies.
- Operator load ≤ ~0.1–0.3 operator per active servicer over a mission portfolio.
- Simple cost-ratio rules of thumb
Define:
- C_crew ≈ $1,000/crew‑hour (midpoint)
- C_op ≈ $100/operator‑hour
- r = operators per robot (0–1)
- C_robot = $/robot‑hour (amortized capex + ops)
2.1 Crew vs fully robotic
- Crew-tended favored if:
- C_robot > ~0.1–0.3 × C_crew for tasks that are hard to automate and need flexible response.
- Fully robotic favored if:
- C_robot < ~0.05–0.1 × C_crew AND MTBF is long enough that unscheduled crew visits are rare.
- So practical crossover band:
- C_robot ≈ $50–100/robot‑hour with MTBF measured in many months–years.
2.2 Teleop vs locally autonomous robots
- Teleop-dominated if:
- C_robot is high but r can be kept very low (e.g., 1 operator for 5–10 robots) and
- Task mix is varied/unpredictable.
- Local autonomy-dominated if:
- C_robot is modest (≲$20–50/robot‑hour) and
- r can be pushed toward 0.05–0.1 or lower.
- Reliability (MTBF) thresholds
Use “human-intervention events” rather than pure hardware hours.
3.1 Fiber drawing
- Target: keep line >95% uptime without crew.
- Rough threshold:
- Hardware MTBF: ≥ ~5,000–10,000 operating hours for critical movers.
- Autonomous recovery: resolves ≥90% of glitches.
- Human-touch events: ≤ 1 per 2–3 months of continuous running.
3.2 Semiconductor steps
- Target: crew only for scheduled swaps, not day-to-day yield issues.
- Thresholds:
- Critical tool MTBF: ≥ ~8,000–15,000 hours between major downtime.
- Autonomous process control keeps excursions below fab-acceptable limits.
- Human on-orbit maintenance cycles: yearly or rarer.
3.3 Satellite servicing
- Mission-level reliability is more relevant:
- ≥95–98% mission success rate with no crew backup.
- Per-robot: ≥5–10 missions before major refurb or deorbit.
- Onboard autonomy handles most contingencies up to predefined safety boundaries.
- Wright’s-law slopes and time shift
Let robot-hour cost follow Wright’s law:
- C_robot(N) = C0 × (1 – L)^(log2 N)
- L = learning rate (e.g., 0.1 = 10% per doubling)
- N = cumulative deployed robot-equivalents or operating hours.
Assume illustrative starting point after first-gen commercial robots:
- C0 ≈ $500/robot‑hour effective (low flight heritage, high overhead).
- We want C_target ≈ $50/robot‑hour for broad crossover vs crew; ≈$20 for strong autonomy edge vs teleop in manufacturing.
4.1 Moderate learning (L ≈ 10–15%)
- 10%/doubling (L=0.10): need ~10× cost reduction (500→50):
- 10 = (1–0.10)^k → k ≈ 22 doublings.
- 22 doublings ≈ factor 4M in cumulative production/experience.
- 15%/doubling (L=0.15):
- 10 = (0.85)^k → k ≈ 15 doublings → factor ~3×10^4.
- Interpretation:
- With shared, standardized robots across many missions, you might hit 15–25 doublings in 1–2 decades of heavy use; crossover for fiber and some servicing happens in that window.
4.2 Slow learning (L ≈ 5%)
- 10 = (0.95)^k → k ≈ 45 doublings → ~3.5×10^13 experience factor.
- Effect:
- Effective robot-hour cost remains high; architectures from cf5445ac-276f-40e1-b3b8-3f9b1af70d34 (crew/teleop-heavy) stay favored longer.
- Fully robotic semiconductor lines are delayed or never economical versus crew/hybrid.
4.3 Fast learning (L ≈ 20–25%)
- 20%/doubling: 10 = (0.8)^k → k ≈ 11 doublings (~2,000× scale-up).
- 25%/doubling: 10 = (0.75)^k → k ≈ 8 doublings (~250× scale-up).
- Effect:
- C_robot hits ~$50/robot‑hour after only 8–11 doublings.
- Within a few large station generations, fully robotic fiber and many servicing tasks clearly beat crew/teleop.
- Semiconductor steps move into serious contention once MTBF and process stability catch up.
- How slopes shift specific use cases
5.1 Fiber drawing
- Needs modest dexterity, strong runtime reliability.
- With L ≈ 15–20%:
- Likely first to cross to fully robotic; C_robot target ~$10–20/h plus long MTBF.
- With L ≈ 5%:
- Remains hybrid: robots plus periodic human/teleop; pure robotic plants marginal.
5.2 Semiconductor steps
- High sensitivity to small errors; demands highest reliability.
- With L ≈ 20–25% and strong reuse of Earth fab robotics:
- Robotic cost and reliability thresholds could be hit in 10–20 years of intense deployment.
- With L ≈ 5–10%:
- Likely stuck in small, pilot-scale lines with human-heavy support; full factories stay Earth-side.
5.3 Satellite servicing
- Value per mission is high; can tolerate higher C_robot than manufacturing.
- With L ≈ 10–15%:
- Becomes economical at $20–80/robot‑hour relatively early; even slow learning can suffice if mission savings are large.
- Faster learning mainly broadens scope (more complex repairs), not basic economic viability.
- Summary heuristics
- For continuous manufacturing (fiber, some semis):
- Need C_robot ≲ $10–30/h and MTBF in many-months+ range.
- For episodic servicing:
- C_robot ≲ $20–80/h and “several missions between major failures” are usually enough.
- Fast Wright’s-law slopes (≥15–20%) bring these thresholds into a few station generations; slow slopes (<10%) lock in crew/teleop-heavy architectures much longer.