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

Research library · updated 2026-06-15 · public

Robotics Morning Review Next-Signal Matrix — 2026-06-15

Date: 2026-06-15 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe research artifact for morning review Visibility: PUBLIC Target use: morning review, slide speaker notes, /robotics/signals, /robotics/companies, /robotics/supply-chain

Related files:

  • knowledge/robotics-research-harness.md
  • knowledge/robotics-sharing-backlog.md
  • knowledge/robotics-codex-handoff.md
  • knowledge/robotics-tesla-optimus-s4-to-s5-checklist-v1.md
  • knowledge/robotics-figure-unitree-evidence-v1.md
  • knowledge/robotics-figure-unitree-warrior-brief-v1.md
  • knowledge/robotics-leaderdrive-s4-to-s5-evidence-debt-v1.md
  • knowledge/robotics-why-now-catalyst-evidence-card-v1.md

Public-safety: yes. No Hugo private portfolio data. No buy / sell / hold language. No external publishing action.

0. One-line answer

截至 2026-06-15，机器人研究的最高价值下一步不是再增加一个“谁赢”判断，而是建立四类证据曲线的 next-signal matrix：Tesla 代表 capacity-intent / internal factory path，Figure 代表 customer-site KPI / manufacturing cadence，Unitree 代表 low-cost access / developer platform，Leaderdrive / 绿的谐波代表 filing-backed supplier validation。四者都已经有 S3/S4 信号，但都缺 S5 scaled-commercial-economics proof。🟢 primary-source artifacts reviewed through 2026-06-15; 🟠 Charlie synthesis.

1. Why this artifact exists

早会最容易失真的地方，是把不同类型的 S4 信号混成同一种结论：

  1. Tesla 的 1M / 10M 是 designed / long-term capacity evidence，不是 current production。🟢 Tesla SEC filings, 2026-04-22.
  2. Figure 的 BMW KPI 是 customer-site work evidence，不是 full contract economics。🟢 Figure official BMW post, 2025-11-19.
  3. Unitree 的低价是 cost-access evidence，不是 reliability / margin / ROI proof。🟢 Unitree official product pages, checked 2026-06-10 in prior artifacts.
  4. 绿的谐波的收入 / 销量 / 毛利是 supplier filing evidence，不是 named humanoid OEM proof。🟢 2025 annual report and 2026 Q1 report, disclosed 2026-04-23.

This matrix turns them into a monitoring system: what is proven, what is missing, and what single future disclosure would upgrade the evidence fastest.

2. Four evidence curves at morning-review depth

CurveCurrent strongest public evidenceCurrent signalWhat it changesWhat it does not proveNext highest-value signal
Tesla Optimus — capacity-intent curveQ1 2026 Tesla filing says Fremont first-generation Optimus line is designed for 1M robots/year and Gigafactory Texas is being prepared for a second-generation line designed for long-term annual production capacity of 10M robots; Tesla capacity caveat says installed capacity is not current production rate. 🟢 as of 2026-04-22S4 manufacturing-infrastructure / capacity-intentOptimus is now a filing-backed manufacturing-scale project, not only a demo story.Actual output, accepted units, deployed robots, yield, runtime, intervention, ROI/payback, external paid customers, robot revenue, gross margin.Quarterly Optimus produced / accepted / deployed unit count, ideally with factory task KPI and intervention rate. 🟢 needed
Figure AI — customer-site KPI curveBMW deployment: 11-month deployment, 10-hour weekday shifts, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles; BotQ: 350+ Figure 03 robots delivered, cadence from 1/day to 1/hour, EOL FPY >80%, battery-line FPY 99.3%. 🟢 2025-11-19 / 2026-04-29 official postsS4 deployment + manufacturing KPIMoves humanoid evidence from demo into named customer workflow and production-process metrics.Contract value, robot count at BMW, customer-confirmed ROI/payback, uptime distribution, intervention rate, repeat order, vendor gross margin/service burden.Second or third named customer with same KPI set plus robot count / repeat order / ROI/payback. 🟢 needed
Unitree — cost-access curveR1 AIR from US$4,900 / R1 from US$5,900, G1 from US$13.5K, H2 US$29,900; H2 Plus and G1-D expose NVIDIA / Isaac / data-model workflow surfaces in prior product-page review. 🟢 official product pages checked 2026-06-10 in prior artifactsS3/S4 low-cost hardware / developer platformExpands experiment surface and pressures humanoid price expectations.Shipments by product line, industrial customer mix, reliability, uptime, warranty/support cost, gross margin, customer ROI, repeat orders.Official prospectus / filing-grade product revenue, shipment mix, gross margin and warranty/support cost by product category. 🟢 needed
Leaderdrive / 绿的谐波 — supplier filing curve2025 revenue RMB 570.714m (+47.31% YoY), net profit RMB 124.367m (+121.42% YoY), harmonic reducer sales 425,158 units (+72.48% YoY), “工业及具身智能机器人零部件” revenue RMB 422.528m (+52.61% YoY, GM 34.88%); 2026 Q1 revenue RMB 140.134m (+42.96% YoY), net profit RMB 32.634m (+61.17% YoY). 🟢 filings disclosed 2026-04-23S4 filing-backed supplier validationMoves supply-chain analysis from concept-stock narrative into revenue / volume / margin tracking.Named humanoid OEM, contract value, order volume, humanoid-only revenue/margin, BOM attach, single-robot value content, downstream customer ROI/payback.Named humanoid OEM design win or humanoid-specific revenue / margin / backlog disclosure. 🟢 needed

3. Priority order for the next evidence hunt

P0 — thesis-upgrading evidence

  1. Customer-confirmed deployment economics: ROI/payback, intervention rate, uptime, safety, service burden and repeat order for Figure / Agility-style customer workflows. 🟢 needed.
  2. Tesla actual Optimus output / accepted units / internal deployment KPI, not just line design capacity. 🟢 needed.
  3. Unitree prospectus-grade revenue / shipment / margin / warranty-support data by product line. 🟢 needed.
  4. Leaderdrive named humanoid OEM or humanoid-specific revenue / margin / backlog. 🟢 needed.

Why P0: these are the points that convert S4 evidence into S5 candidate evidence. Without them, more demos, product names, capacity screenshots and supplier speculation add noise more than signal. 🟠 synthesis.

P1 — evidence-quality strengthening

  1. Multi-site replication: same robot / workflow across more than one customer or factory. 🟢/🟡.
  2. Production quality: FPY, EOL yield, rework, bottleneck, uptime, component constraints, accepted deliveries. 🟢.
  3. Financial materiality: robot revenue, gross margin, service cost, capex attribution and cash conversion. 🟢.
  4. Supplier quality: inventory / receivables / operating cash flow moving with robot-related revenue, not against it. 🟢/🟠.

P2 — useful but not thesis-changing alone

  1. New demo videos without runtime / intervention / customer context. 🟠/🔴.
  2. New product pages without shipment / margin / reliability evidence. 🟢 for product existence; 🟠 for commercial implication.
  3. Funding / valuation rounds without deployment economics. 🟡/🟠.
  4. Stock-price moves without disclosed operating update. 🔴 as operating evidence.

4. Signal vs noise compression

Signal

SignalWhy it mattersCurrent best example
Named customer runtime + task countStronger than demo; can be tracked and falsified.Figure BMW: 1,250+ runtime hours, 90,000+ parts, 10-hour weekday shifts. 🟢
Production cadence / yieldShows manufacturing process, not just product ambition.Figure BotQ: 350+ Figure 03, 1/day to 1/hour, EOL FPY >80%. 🟢
Designed capacity with caveatUseful capacity-intent signal if not confused with output.Tesla Fremont 1M/year and Texas 10M/year designed/long-term capacity; installed capacity caveat. 🟢
Public low-price anchorsExpands experiment surface and changes cost expectations.Unitree R1 / G1 / H2 official pricing. 🟢
Filing-backed revenue / volume / marginStronger than concept mapping or supplier rumor.Leaderdrive 2025 revenue, reducer sales, robot-component revenue and GM. 🟢

Noise unless upgraded

NoiseCorrect interpretation
“Tesla 1M / 10M means production”Designed / long-term capacity ≠ current production rate. 🟢
“Figure BMW KPI proves full economics”Runtime / task count is S4; still missing ROI, intervention and contract economics. 🟢/🟠
“Unitree low price proves adoption”Price lowers experiment cost; it does not prove reliability, margin, support burden or customer ROI. 🟢/🟠
“Leaderdrive robot-related revenue equals humanoid revenue”Filing line is industrial + embodied-intelligence robot components, not humanoid-only split. 🟢
“Supplier/product fit proves customer”Needs named OEM, contract, teardown/BOM, filing revenue or customer confirmation. 🔴 without source.

5. What would change the morning-review answer

Upgrade the robotics field guide from “S3/S4-heavy evidence tracking” toward “S5 return-cycle proof” only if at least two curves cross their next gate:

  1. Tesla discloses actual Optimus output / deployed units plus factory productivity or intervention-rate data. 🟢 required.
  2. Figure or another OEM discloses repeat customer deployment with robot count, contract value and ROI/payback. 🟢 required.
  3. Unitree filing verifies high shipments, durable gross margin and manageable warranty/support cost for low-cost humanoids. 🟢 required.
  4. Leaderdrive or another supplier discloses named humanoid OEM / humanoid-specific revenue / margin / backlog. 🟢 required.
  5. A customer-side source, not only vendor, confirms robot payback and operational reliability across multiple sites. 🟢 strongest.

Downgrade if:

  1. 2026–2027 updates stay dominated by demos, roadmaps, product names and capacity without output / economics. 🟠.
  2. Customer pilots do not repeat or remain silent on intervention and ROI. 🟠.
  3. Low-cost hardware expands experiments but reliability / support cost blocks industrial deployment. 🟠.
  4. Supplier revenue grows while gross margin, cash conversion or receivables quality deteriorates. 🟢/🟠.
  5. Named supplier claims stay rumor-driven despite multiple reporting periods. 🔴.

6. Public slide / site draft

Suggested title

Robotics next signals: what would move S4 toward S5?

Suggested subtitle

Tesla, Figure, Unitree and Leaderdrive each show a different S4 evidence curve. None yet proves scaled commercial economics.

Four-card layout

  1. Tesla: capacity intent

    • Proven: SEC-filed Fremont 1M/year designed line and Texas 10M/year long-term line. 🟢
    • Missing: actual output, deployed units, yield, factory ROI, robot revenue. 🟢/🟠

  2. Figure: customer-site KPI

    • Proven: BMW runtime / parts / shift data and BotQ cadence / yield data. 🟢
    • Missing: robot count, contract value, customer ROI/payback, intervention rate, repeat order. 🟢/🟠

  3. Unitree: cost-access curve

    • Proven: low public price anchors and developer-platform surfaces. 🟢
    • Missing: shipments, reliability, support cost, margin, industrial ROI. 🟢/🟠

  4. Leaderdrive: supplier filings

    • Proven: revenue, reducer volume, robot-component revenue and GM are filing-backed. 🟢
    • Missing: named humanoid OEM, humanoid-only revenue/margin, contract/backlog. 🟢/🟠

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

7. Common misconceptions

  1. Misconception: “The best robotics evidence is one company’s demo.”

    • Correction: the best evidence is a chain: output -> deployment -> reliability -> customer economics -> revenue / margin. 🟠

  2. Misconception: “Capacity intent, customer KPI, low price and supplier revenue are interchangeable.”

    • Correction: they are different evidence curves and must not be averaged into one winner score. 🟠

  3. Misconception: “One S4 curve is enough for S5.”

    • Correction: S5 needs multiple gates together: repeat paid deployment, low intervention, ROI/payback, production output, and financial materiality. 🟠

  4. Misconception: “Public supplier exposure can be inferred from OEM momentum.”

    • Correction: supplier exposure requires named customer, order/revenue, BOM/teardown, or filing/customer confirmation. 🟠/🔴

8. Think Deeper questions

  1. Which curve crosses S5 first: Tesla internal factory economics, Figure repeat customer deployment, Unitree low-cost shipment economics, or Leaderdrive named supplier validation?
  2. If Tesla scales internally but never discloses external customers, should S5 be defined by robot revenue or factory productivity/payback?
  3. If Unitree commoditizes robot bodies, does value migrate to OEM scale, software/data, deployment service, or validated component suppliers?
  4. If suppliers show revenue before OEMs show deployment economics, should supplier filings be treated as leading indicator or potential inventory-cycle noise?
  5. What minimum disclosure package should a public robotics field guide require before calling a humanoid program commercially proven?

9. Source list

Primary sources / primary-source-derived artifacts 🟢:

10. Public-safe flag

Public-safe: yes, as an evidence matrix.

Exclude:

  • Hugo private portfolio data, position weights, entry logic or private rationale.
  • Buy / sell / hold language for TSLA, 688017.SH, or any public/private company.
  • Supplier rumors, private channel checks, paid-report excerpts, or unverified customer claims.
  • Claims that designed capacity, one customer KPI, low price, or supplier revenue alone proves S5 commercialization.