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

Research library · updated 2026-06-14 · public

Robotics S5 Evidence Gap Tracker v1

Date: 2026-06-14 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe research artifact Visibility: PUBLIC, if used without private portfolio information Target use: morning review, /robotics/signals, slide speaker notes

Related files:

  • knowledge/robotics-research-harness.md
  • knowledge/robotics-tesla-optimus-public-evidence-pack-v2.md
  • knowledge/robotics-tesla-figure-unitree-public-evidence-ledger-v1.md
  • knowledge/robotics-figure-unitree-public-compression-guardrail-v1.md
  • knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md
  • knowledge/robotics-why-now-slide-3-catalyst-pack-v2.md

Source discipline:

  • 🟢 primary / official / filing
  • 🟡 credible secondary
  • 🟠 Charlie synthesis or estimate from primary evidence
  • 🔴 unsupported / guess

Public-safety:

  • No Hugo private portfolio data.
  • No buy / sell / hold language.
  • No trade recommendation.
  • No supplier-customer inference without primary confirmation.

0. One-line answer

截至 2026-06-14，robotics public research 的核心结论可以压缩成一句话：行业已经出现多个 S4 级别的可量化证据，但 S5 仍然缺一条完整经济性链条。

S4 已经可见：Tesla Optimus 有 filing-backed production-infrastructure / designed-capacity language；Figure 有 BMW deployment KPI 与 BotQ manufacturing KPI；Unitree 有 official price / platform-access anchors；绿的谐波有 filing-backed revenue / shipment / margin evidence。🟢/🟠

S5 仍缺：accepted units、paid deployment value、repeat orders、runtime / uptime / intervention rate、customer ROI/payback、vendor revenue / gross margin / service burden、humanoid-specific supplier revenue split。🟢/🟠 as of reviewed artifacts through 2026-06-14.

1. Why this artifact exists

前面已经分别做了 Tesla、Figure / Unitree、Leaderdrive、Why robotics now 的 source-backed packs。这个文件的增量不是再重复事实，而是做一个“morning review checklist”：

  1. 哪些证据已经足够强，可以公开讲？
  2. 哪些证据仍然不能升级成 S5？
  3. Codex / Warrior 包装时应该把下一个证据缺口放在哪里？

核心句：

S4 says “the evidence is now measurable.” S5 requires “the economics are repeatable and material.”

2. S4-to-S5 evidence ladder

Ladder stepWhat it asksCurrent best public anchorsCurrent statusSource grade / as-of
Demo / product existenceDoes the robot or module exist publicly?Tesla Optimus filings and AI page; Figure product posts; Unitree R1/G1/H2 pages; Leaderdrive product / filing product lines.Strong S3/S4 evidence.🟢 through reviewed files as of 2026-06-14
Production intent / manufacturing infrastructureIs there a path from prototype to repeatable build?Tesla Fremont / Texas designed-capacity language; Figure BotQ 350+ Figure 03, 1/day to 1/hour, >80% EOL FPY, 99.3% battery-line FPY, 9,000+ actuators.Tesla = S4 capacity-intent; Figure = S4 manufacturing KPI.🟢 Tesla filings filed 2025-04-22 to 2026-04-22; 🟢 Figure post 2026-04-29
Deployment KPIDoes a named customer site show measured work?Figure BMW: 11-month deployment, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles.S4. Strongest current public deployment KPI among Tesla / Figure / Unitree set.🟢 Figure post 2025-11-19; reviewed 2026-06-13
Cost / access curveIs experimentation becoming cheaper and easier?Unitree R1 AIR from US$4,900 / R1 from US$5,900; G1 from US$13.5K; H2 US$29,900; H2 Plus / G1-D developer and model workflow pages.S3/S4 cost-access / platform signal.🟢 Unitree pages captured 2026-06-10
Supplier revenue / shipment validationDoes component demand enter public financials?绿的谐波 2025 revenue RMB 570.714m +47.31%; net profit RMB 124.367m +121.42%; harmonic reducer sales 425,158 +72.48%; “工业及具身智能机器人零部件” revenue RMB 422.528m, GM 34.88%.S4 filing-backed supply-chain signal.🟢 2025 annual report disclosed 2026-04-23
Customer economicsDoes the customer confirm ROI/payback or repeat economics?Not disclosed for Tesla Optimus, Figure BMW/Catalyst, Unitree, or Leaderdrive humanoid customers in reviewed public artifacts.S5 gap.🟢 absence in reviewed primary sources / 🟠 synthesis as of 2026-06-14
Financial materialityDoes it show up in revenue, margin, backlog, cash flow, or segment disclosure?Leaderdrive has component-level financial evidence but not humanoid-only split; Tesla/Figure/Unitree do not disclose robot revenue / margin in reviewed public evidence.Partial S4 for supplier; S5 gap for OEM economics.🟢 filings / 🟠 synthesis

3. Company / layer scorecard

3.1 Tesla Optimus — S4 capacity ambition, not S5 economics

What is public-safe to say:

  • Tesla filings support a move from pilot-line / internal-use intent toward production-line installation, Gen 3 mass-production design, Fremont 1M/year designed line, and Texas 10M/year long-term designed line. 🟢 Tesla Form 8-K Exhibit 99.1 sequence filed 2025-04-22 to 2026-04-22; summarized in robotics-tesla-optimus-public-evidence-pack-v2.md.

What not to say:

  • Do not say 1M/year or 10M/year is current production. 🔴 unsupported.
  • Do not infer actual quarterly Optimus output, production yield, uptime, factory ROI, external paid customer, robot revenue, robot gross margin, or named suppliers. 🔴 unsupported as of reviewed filings.

Current signal grade:

  • S4 manufacturing-infrastructure / capacity-intent signal. 🟠 synthesis from 🟢 filings.

S5 upgrade checklist:

  1. Accepted Optimus units or quarterly output. 🟢
  2. Production yield / bottleneck / cost curve. 🟢
  3. Internal factory deployment metrics: runtime, uptime, task success, intervention rate, safety incidents. 🟢
  4. Factory productivity improvement or ROI/payback. 🟢/🟡
  5. External customer, pricing, contract value, repeat order. 🟢
  6. Robot revenue / gross margin / service cost. 🟢

3.2 Figure AI — strongest deployment KPI, still missing economics

What is public-safe to say:

  • Figure BMW deployment post discloses 11 months, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, and contribution to 30,000+ BMW X3 vehicles. 🟢 Figure official post dated 2025-11-19.
  • Figure BotQ post discloses 350+ Figure 03 delivered, production cadence from 1/day to 1/hour, >80% EOL first-pass yield, 99.3% battery-line first-pass yield, 500+ battery packs, and 9,000+ actuators. 🟢 Figure official post dated 2026-04-29.
  • Catalyst commercial agreement adds a second named deployment surface, starting at Reno, Nevada Distribution Logistics Center. 🟢 Figure official post dated 2026-05-26; 🟠 economics undisclosed.

What not to say:

  • Do not say BMW KPI proves customer ROI/payback. 🔴 not disclosed.
  • Do not infer robot count by customer, contract value, intervention rate, repeat-order economics, revenue, or gross margin. 🔴 unsupported in reviewed public posts.
  • Do not treat 350+ delivered Figure 03 as commercial sell-through without deployment utilization / revenue evidence. 🟢/🟠.

Current signal grade:

  • S4 deployment + manufacturing KPI signal. 🟠 synthesis from 🟢 official posts.

S5 upgrade checklist:

  1. BMW / Catalyst / another named customer confirms robot count by site. 🟢
  2. Paid contract value or recurring economics. 🟢
  3. Customer-confirmed ROI/payback. 🟢/🟡
  4. Uptime / intervention / safety metrics over long-duration production use. 🟢/🟡
  5. Repeat order or multi-site rollout. 🟢
  6. Figure revenue / gross margin / support burden tied to deployments. 🟢/🟠

3.3 Unitree / 宇树 — cost-access curve, not deployment economics

What is public-safe to say:

  • Unitree official pages list R1 AIR from US$4,900 / R1 from US$5,900, G1 from US$13.5K, and H2 US$29,900. 🟢 Unitree product pages captured 2026-06-10.
  • H2 Plus references NVIDIA Jetson T5000, FP4 2,070 TFLOPS, Isaac GR00T, Isaac TeleOp and Isaac Sim. 🟢 Unitree H2 Plus page captured 2026-06-10.
  • G1-D references data acquisition, labeling/review, model training and inference workflow. 🟢 Unitree G1-D page captured 2026-06-10.

What not to say:

  • Do not say low price proves industrial reliability, autonomy, margin quality, shipment scale, support cost, or customer ROI. 🔴 unsupported.
  • Do not say NVIDIA / Isaac stack references mean Unitree owns the AI/model economics. 🟢/🟠 alignment evidence only.

Current signal grade:

  • S3/S4 hardware cost-access / developer-platform signal. 🟠 synthesis from 🟢 official product pages.

S5 upgrade checklist:

  1. Humanoid shipments by product line and customer type. 🟢
  2. Named industrial deployment KPI: runtime, uptime, intervention rate, safety. 🟢/🟡
  3. Repeat orders / fleet expansion / channel sell-through. 🟢/🟡
  4. Warranty, maintenance and support burden. 🟢/🟠
  5. Product-line gross margin or paid software/service attachment. 🟢/🟠
  6. Evidence that Unitree captures software/data/deployment value, not only hardware ASP. 🟢/🟠

3.4 Leaderdrive / 绿的谐波 — S4 filing-backed supplier signal, not named-humanoid proof

What is public-safe to say:

  • 2025 revenue RMB 570.714m, +47.31%; net profit RMB 124.367m, +121.42%. 🟢 2025 annual report, disclosed 2026-04-23.
  • 2025 “工业及具身智能机器人零部件” revenue RMB 422.528m, +52.61%, gross margin 34.88%. 🟢 2025 annual report.
  • 2025 harmonic reducer sales 425,158 units, +72.48%; production 433,655 units, +72.27%; inventory 18,238 units, -6.34%. 🟢 2025 annual report.
  • 2026 Q1 revenue RMB 140.134m, +42.96%; net profit RMB 32.634m, +61.17%. 🟢 2026 Q1 report, disclosed 2026-04-23.
  • 2026-06-09 abnormal-trading announcement says 2026-06-04 to 2026-06-08 share price rose 36.00% versus SSE Composite -3.05% and automation-equipment index +2.13%, with no undisclosed major matter. 🟢 abnormal trading announcement.

What not to say:

  • Do not call RMB 422.528m pure humanoid revenue. 🔴 the line is “工业及具身智能机器人零部件,” not humanoid-only.
  • Do not imply Leaderdrive supplies Tesla, Figure, Unitree, AgiBot, UBTECH, or any named humanoid OEM without primary confirmation. 🔴 unsupported.
  • Do not treat share-price movement as order evidence. 🔴 contradicted by abnormal-trading caveat.

Current signal grade:

  • S4 filing-backed supply-chain validation; S5 still missing named customer / order / humanoid-specific economics. 🟠 synthesis from 🟢 filings.

S5 upgrade checklist:

  1. Named leading humanoid OEM design win with supply scope, volume, or contract period. 🟢
  2. Humanoid-specific revenue / backlog / order amount. 🟢
  3. Segment gross margin stability around current 34.88% or improvement while robot-related revenue grows above 30% YoY. 🟢/🟠 threshold based on current filing data.
  4. Integrated actuator / PRS / linear actuator revenue becomes material with attractive margin. 🟢
  5. OEM teardown or official BOM confirms production humanoid module attach. 🟢/🟡
  6. Downstream customer economics validate repeatable deployment, not only build-out. 🟢/🟡

4. Signal vs noise rules for morning review

Signals to keep

SignalWhy it mattersCurrent grade
Filing-backed capacity or production-line languageTrackable infrastructure signal, stronger than demo.S4 if quantified/site-specific 🟢/🟠
Named customer + quantified runtime / task KPIStrongest bridge from demo to deployment.S4 🟢
Manufacturing KPI with yield/cadence/outputShows process maturity, but not demand economics.S4 🟢
Published price anchors and toolchain pagesExpands experiment base and developer access.S3/S4 🟢/🟠
Filing-backed supplier revenue / shipment / marginStronger than concept narrative.S4 🟢

Noise to reject unless upgraded

NoiseCorrect treatment
Viral demo videos without runtime, task repetition, intervention or safety dataS1/S2; do not use as thesis evidence. 🔴/🟠
Designed capacity interpreted as achieved productionIncorrect compression. 🔴
Customer logo interpreted as ROI or recognized revenueNeeds economics, count, repeat order. 🟠
Low price interpreted as reliability or marginAccess signal only. 🟠
Supplier rumor interpreted as named customer exposureUnsupported without primary disclosure, filing, teardown or order evidence. 🔴
Share-price move interpreted as order signalNot operating evidence. 🔴

5. Slide / site-ready copy

S4 evidence is now measurable; S5 still needs economics

Robotics has moved beyond a pure demo cycle, but it has not yet completed the economics proof.

Tesla gives the strongest public capacity-ambition signal: filing-backed production-line and designed-capacity language around Fremont and Texas. Figure gives the strongest public deployment/manufacturing KPI signal: BMW runtime and task metrics plus BotQ yield/cadence/output metrics. Unitree gives the strongest public cost-access signal: sub-US$5k to US$30k humanoid price anchors and developer/model workflow pages. Leaderdrive gives a clean supplier case: filing-backed revenue, profit, shipment and robot-component gross-margin evidence.

The missing proof is the same across layers: accepted units, repeat paid deployments, low intervention, customer payback, vendor margin, service burden, and financial materiality. That is the bridge from S4 “measurable progress” to S5 “commercialization proof.”

Footer: Evidence map only. No winner ranking. No trade recommendation. S4 evidence does not equal S5 scaled commercial economics.

6. Common misconceptions

  1. Misconception: “If the evidence is measurable, commercialization is solved.”

    • Correction: measurable S4 evidence is progress, but S5 requires repeatable economics: customer payback, repeat order, low intervention, revenue, margin and service cost. 🟠

  2. Misconception: “The company with the largest capacity target leads deployment.”

    • Correction: capacity ambition and deployment KPI answer different questions. Tesla is stronger on capacity planning; Figure is stronger on public deployment KPI. 🟢/🟠

  3. Misconception: “Low-cost humanoids automatically create profit pools.”

    • Correction: low price can widen experiments while compressing hardware margins; value may migrate to model/data, integration, service, or customer workflow ownership. 🟠

  4. Misconception: “Supplier revenue acceleration proves named humanoid customer exposure.”

    • Correction: filing-backed supplier acceleration is meaningful, but named customer / humanoid-specific revenue / margin still need disclosure. 🟢/🟠

  5. Misconception: “Stock-price movement is a robotics signal.”

    • Correction: price movement is not order evidence; Leaderdrive’s 2026-06-09 abnormal-trading announcement is the clean cautionary example. 🟢

7. Think Deeper questions

  1. Which S5 disclosure would matter most first: robot count, intervention rate, customer ROI/payback, repeat order, vendor margin, or supplier humanoid-specific revenue?
  2. If Tesla proves internal factory productivity but not external robot sales, should that be treated as robot revenue proof or manufacturing-productivity proof?
  3. If Figure’s deployment KPI improves but remains private-company evidence, which public-market proxies actually capture the signal without overfitting supplier rumors?
  4. If Unitree lowers humanoid access cost faster than reliability improves, does the sector’s profit pool shift away from hardware bodies?
  5. If suppliers show revenue before OEMs show customer economics, how much of that is real demand versus inventory / pilot build-out / industrial-cycle recovery?

8. Source list

Primary / official sources cited through existing reviewed artifacts:

Internal synthesis cross-references:

  • knowledge/robotics-tesla-optimus-public-evidence-pack-v2.md 🟠
  • knowledge/robotics-tesla-figure-unitree-public-evidence-ledger-v1.md 🟠
  • knowledge/robotics-figure-unitree-public-compression-guardrail-v1.md 🟠
  • knowledge/robotics-leaderdrive-case-study-slide-evidence-v3.md 🟠
  • knowledge/robotics-why-now-slide-3-catalyst-pack-v2.md 🟠

9. Public-safe flag

Public-safe: yes, if used without Hugo private portfolio information and without trade language.

Exclude:

  • Hugo private portfolio data, position weights, purchase prices, tax context, or private rationale.
  • Buy / sell / hold language for TSLA, 688017.SH, suppliers, private humanoid companies, or any proxy.
  • Claims that designed capacity equals current production.
  • Claims that deployment KPI equals customer ROI.
  • Claims that low price equals reliability or margin.
  • Claims that supplier financial improvement confirms named humanoid customer exposure.
  • Paid-report excerpts or proprietary tables.