Investment ResearchTopic research, evidence curves, and cycle signals
HomeSign inEN

Research library · updated 2026-06-16 · public

Robotics S4→S5 evidence upgrade checklist — morning review artifact

Date: 2026-06-16 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe research artifact for morning review Visibility: PUBLIC Target use: morning review; optional slide/site checklist after Hugo review Related files:

  • knowledge/robotics-research-harness.md
  • knowledge/robotics-why-now-catalyst-evidence-cards-2026-06-16.md
  • knowledge/robotics-tesla-optimus-public-evidence-card-2026-06-16.md
  • knowledge/robotics-morning-review-tesla-figure-unitree-evidence-split-2026-06-16.md
  • knowledge/robotics-dexterity-benchmark-gate-v1.md
  • knowledge/robotics-insurability-liability-gate-v1.md

Public-safety: yes. No Hugo private portfolio data. No buy/sell/hold recommendation. No private channel checks. No paid-report excerpts.

0. One-line answer

截至 2026-06-16，机器人研究库已经有足够 S3/S4 证据说明“为什么现在值得跟踪”：AI/model stack、低价硬件、客户现场 KPI、产线/产能指标、上市公司 filing-backed revenue/unit/margin、dexterity benchmark、safety/insurability gate 都在出现。下一步早审最重要的不是再找一个更热的视频，而是用同一把尺判断哪些证据能把公司或产业从 S4 推向 S5：accepted robots、runtime/intervention distribution、repeat orders、ROI/payback、gross margin/service burden、warranty/liability/insurance data、cash-flow materiality。🟢/🟡 public sources captured in related artifacts through 2026-06-16; 🟠 Charlie synthesis.

1. Core question

现有证据里，什么只是“值得继续跟踪”的 S4，什么才是“商业化经济性开始成立”的 S5？

Short answer: S4 证明“项目、产品、产线、客户现场或收入科目真实存在”；S5 需要证明“可重复、可付费、可盈利、可承担风险、可规模化”。

2. Upgrade checklist

Evidence categoryS4 evidence we already trackS5 upgrade requirementCurrent gap as of 2026-06-16Source gradeMorning-review action
Tesla Optimus capacity intentTesla Q1 2026 Form 8-K Exhibit 99.1 says Fremont first-generation Optimus line is designed for 1M robots/year and Texas second-generation line is designed for long-term annual capacity of 10M robots.Actual quarterly output, accepted units, line yield/utilization, deployed robot count, factory task KPI, intervention rate, cost saving/payback, external paid deployment or robot revenue.Public filing evidence supports capacity intent, not actual production or economics.🟢 Tesla SEC filing; 🟠 synthesisKeep as S4 capacity-intent; do not let Codex label 1M/10M as production.
Figure customer deploymentFigure official BMW post dated 2025-11-19 disclosed 11 months, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, and contribution to 30,000+ BMW X3 vehicles.Robot count at site, uptime/intervention distribution, contract value, customer-confirmed ROI/payback, repeat order, multi-site expansion, gross margin/service burden.Strong customer-site KPI, still missing economics and repeatability.🟢 Figure official; 🟠 economics classificationTreat as one of the strongest S4 deployment signals, not S5.
Figure manufacturingFigure BotQ post dated 2026-04-29 disclosed 350+ Figure 03 robots delivered, cadence from 1/day to 1/hour, >80% EOL FPY, 99.3% battery-line FPY, 500+ battery packs, 9,000+ actuators across 10+ SKUs.Sustained quarterly output, accepted deployments, cost curve, scrap/rework trend, warranty/service burden, backlog/customer demand and revenue/margin.Strong production-process evidence, but not customer economics.🟢 Figure official; 🟠 synthesisUse as manufacturing KPI card; separate from deployment ROI.
Unitree cost/accessUnitree official product pages captured in prior artifacts list R1 AIR from US$4,900 / R1 from US$5,900, G1 from US$13.5K, H2 at US$29,900; H2 Plus / G1-D connect to NVIDIA/Isaac and data/model tooling.Shipment volume by product line, active developer/customer base, industrial uptime, repeat orders, warranty/support cost, gross margin and software/service revenue.Low price and developer access are not reliability or margin proof.🟢 Unitree official pages; 🟠 value-migration implicationKeep as cost-access catalyst; avoid “low price = winner”.
UBTECH filing-backed revenueUBTECH 2025 annual report disclosed 2025 revenue RMB 2.001bn (+53.3% YoY), full-size embodied humanoid products/services revenue RMB 820.557m (+2,203.7% YoY; 41.1% of revenue), 1,079 full-size embodied humanoid units sold, and >6,000 annualized full-size humanoid capacity at end-2025.Revenue quality, customer mix, recurring vs one-off sales, gross margin by humanoid line, uptime/intervention, customer ROI/payback, warranty/service cost and cash conversion.Filing-backed commercialization is stronger than demos, but S5 needs unit economics and repeat customer proof.🟢 UBTECH annual report; 🟠 synthesisUse as “filing evidence has begun” catalyst, not blanket sector proof.
绿的谐波 / Leaderdrive supplier filing绿的谐波 2025 annual report disclosed revenue RMB 570.714m (+47.31% YoY), net profit RMB 124.367m (+121.42% YoY), harmonic reducer sales 425,158 units (+72.48%), and industrial/embodied-intelligent robot component revenue RMB 422.528m with 34.88% gross margin.Named humanoid OEM/order, humanoid-specific revenue split, value content per robot, ASP trend, margin durability, customer concentration, replacement/substitution risk.Strong filing-backed supplier relevance; still not named humanoid customer proof.🟢 绿的谐波 2025 annual report; 🟠 sector implicationKeep public case study sanitized; no portfolio data.
Dexterity benchmarkDexBench / Isaac Lab-Arena / GR00T / RobOmni evidence shows public measurement infrastructure emerging: 5 dexterity domains, 18 atomic tasks, 4096 simulation variations/task in one Isaac Lab-Arena benchmark example, and quantified GR00T trajectory needs.Real-world validated benchmark results, task/failure distribution, autonomy boundary, sim-to-real delta, customer productivity link, safety-compatible manipulation and repeat deployment.Benchmark infrastructure is S4; it does not prove customer economics.🟢 NVIDIA/RLWRLD/arXiv; 🟡 WEF/Robot Report; 🟠 synthesisAdd a “hands/manipulation gate” to evidence ladder.
Safety / liability / insurabilityEU Product Liability Directive 2024/2853 includes software/AI systems as products; EU Machinery Regulation 2023/1230 applies from 2027-01-20; OSHA frames robot safety as application-specific; IFR flags cyber/liability as 2026 robotics trends.Insurance-backed deployments, liability allocation, warranty reserves, incident logs, cyber-physical controls, underwriting criteria, product-liability claims history.Deployment KPI can still be blocked by risk-transfer requirements.🟢 EU/OSHA/NIST; 🟡 IFR/insurer/legal commentary; 🟠 synthesisAdd insurability as bridge gate before S5 economics.

3. The S4→S5 rule Hugo can use in morning review

A robotics signal should not be promoted to S5 unless it answers at least four of these seven questions with primary or customer-side evidence:

  1. How many robots were accepted by the customer or internal production owner, by site and quarter? 🟢
  2. What is the runtime / uptime / intervention-rate distribution under production conditions? 🟢
  3. What task output was achieved, and what baseline did it replace? 🟢/🟠 if baseline estimated.
  4. Did the same customer repeat, expand to another site, or sign a multi-period deployment? 🟢
  5. What is the customer ROI/payback, or what transparent inputs allow a payback estimate? 🟢/🟠
  6. What are robot gross margin, service/support cost, warranty reserve, and cash-conversion implications? 🟢
  7. Who carries safety, product-liability, cyber, downtime, maintenance and insurance risk? 🟢/🟡

If the answer is mostly “video, product page, roadmap, capacity target, customer logo, or TAM,” keep it S1-S4 depending evidence strength, but do not call it S5.

4. Signal vs noise

Signal

  • Primary-source operating metrics: accepted units, runtime, intervention, task count, yield, repeat order, revenue/margin, warranty reserve. 🟢
  • Customer-side confirmation: named site, production workflow, KPI, ROI/payback, safety/insurance requirements. 🟢/🟡
  • Filing-backed revenue/unit/margin disclosures, especially when segmented enough to isolate humanoid or embodied-intelligence contribution. 🟢
  • Standards / safety / insurability evidence that lets deployment scale beyond a controlled demo cell. 🟢/🟡

Noise unless upgraded

  • Viral videos without robot count, trial count, reset rules, teleoperation/autonomy boundary or failure cases. 🔴/🟠
  • Designed capacity treated as current production. 🔴 unsupported; Tesla itself warns installed capacity is not current production rate.
  • Customer logo without contract value, robot count, duration, task KPI, or economics. 🟠
  • Supplier exposure inferred from BOM math without named customer/order/revenue/margin. 🔴/🟠
  • Benchmark announcements without independent adoption, real-world validation, or deployment link. 🟠

5. Public-site draft section

The upgrade question: what turns robotics from S4 to S5?

The robotics story now has enough real evidence to track. That is different from saying the economics are solved.

S4 evidence tells us a project is real: a robot enters a customer site, a company publishes production yield, a filing discloses robot-related revenue, or a regulator/standards body starts defining deployment rules. S5 evidence tells us the system can repeat: customers accept robots, uptime and intervention rates hold up, repeat orders appear, ROI/payback is visible, gross margin and service burden are disclosed, and liability/insurance risk can be allocated.

That distinction is the core anti-hype rule. Tesla’s 1M/year and 10M/year Optimus line disclosures matter because they show manufacturing ambition, not because they prove output. Figure’s BMW runtime matters because it shows customer-site task evidence, not because it proves ROI. Unitree’s low prices matter because they expand access, not because they prove reliability or margin. UBTECH and 绿的谐波 filings matter because evidence is moving into financial statements, not because every robot-linked stock has become a beneficiary.

Footer: Evidence map only. No winner ranking. No trade recommendation. S4 is investable-research relevance; S5 requires repeatable economics.

6. Common misconceptions

  1. “S4 already means investable S5.”

    • Correction: S4 means material evidence worth tracking; S5 requires repeatable economics and risk absorption. 🟠

  2. “Capacity target means production.”

    • Correction: designed capacity is an infrastructure / intent signal. Actual output needs units, yield, utilization and acceptance. 🟢/🟠

  3. “Customer deployment means ROI.”

    • Correction: deployment KPI is stronger than demo, but ROI needs baseline, productivity, uptime, service cost and payback. 🟠

  4. “Filing-backed revenue solves the thesis.”

    • Correction: filing revenue is a major upgrade, but revenue quality, margin durability, warranty/service burden and customer repeatability still matter. 🟢/🟠

  5. “Safety/legal/insurance are afterthoughts.”

    • Correction: for physical robots around people and assets, risk allocation can determine whether deployments scale. 🟡/🟠

7. Think Deeper questions

  1. Which evidence will arrive first: accepted robot counts, intervention-rate distributions, repeat orders, or humanoid gross margin?
  2. If internal factory use comes before external customers, should S5 be measured through cost savings/payback rather than revenue?
  3. Does value migrate to OEMs, suppliers, model/data layer, compute platforms, integrators, insurers, or customers if hardware gross margin compresses?
  4. Which company will be first to disclose enough warranty/service/liability data for investors to trust scaled economics?
  5. Can benchmark progress become a leading indicator, or will only customer-side ROI matter?

8. Source list

  • Tesla Q1 2026 Update, Form 8-K Exhibit 99.1, filed 2026-04-22: https://www.sec.gov/Archives/edgar/data/1318605/000162828026026551/exhibit991.htm 🟢; captured in robotics-tesla-optimus-public-evidence-card-2026-06-16.md.
  • Figure, “F.02 Contributed to the Production of 30,000 Cars at BMW,” 2025-11-19: https://www.figure.ai/news/production-at-bmw 🟢; captured in robotics-why-now-catalyst-evidence-cards-2026-06-16.md.
  • Figure, “Ramping Figure 03 Production,” 2026-04-29: https://www.figure.ai/news/ramping-figure-03-production 🟢; captured in related Figure artifacts.
  • Unitree official R1 / G1 / H2 / H2 Plus / G1-D product pages, accessed/captured 2026-06-10 in prior artifacts. 🟢
  • UBTECH Robotics Corp Ltd Annual Report 2025, disclosed 2026-04-15; captured in robotics-oem-evidence-curves-beyond-tesla-figure-unitree-v1.md. 🟢
  • 绿的谐波 2025 annual report PDF: https://static.sse.com.cn/disclosure/listedinfo/announcement/c/new/2026-04-23/688017_20260423_EU45.pdf 🟢; captured in Leaderdrive / why-now artifacts.
  • RLWRLD / NVIDIA DexBench announcement, 2026-06-09; NVIDIA Isaac Lab-Arena blog, 2026-01-05 / updated 2026-02-03; NVIDIA Isaac-GR00T FAQ accessed 2026-06-16. 🟢/🟡; captured in robotics-dexterity-benchmark-gate-v1.md.
  • Directive (EU) 2024/2853; Regulation (EU) 2023/1230; OSHA Technical Manual robot safety chapter; IFR 2026 robotics trends; NIST AI RMF / robotics performance assessment. 🟢/🟡; captured in robotics-insurability-liability-gate-v1.md.
  • Charlie S4/S5 stage synthesis based on robotics research harness and above sources, 2026-06-16. 🟠

9. Public-safe flag

PUBLIC-safe as an evidence-upgrade checklist only. Do not include Hugo private portfolio data, position weights, purchase prices, trade rationale, tax context, private channel checks, paid-report excerpts, or rumors. Do not frame Tesla, Figure, Unitree, UBTECH, 绿的谐波, NVIDIA, insurers, suppliers, customers, or any public/private security as buy / sell / hold. This is research process guidance, not investment, legal, insurance, or tax advice.