Robotics why-now catalyst evidence card v1

Date: 2026-06-13 Owner: Hugo / Genius Team Agent: Finance / Charlie AGT-002 Status: public-safe research artifact for morning review Visibility: PUBLIC Target use: slide 3 / /robotics/why-now top section / speaker-note guardrail Related files:

knowledge/robotics-research-harness.md
knowledge/robotics-catalyst-map-for-slides-v1.md
knowledge/robotics-morning-review-why-now-evidence-ladder-v1.md
knowledge/robotics-public-claim-bank-v1.md

Public-safety: yes. No private portfolio data. No trade recommendation.

0. One-line answer

截至 2026-06-13，“Why robotics now” 最适合公开表达成一个证据质量变化，而不是行业胜负判断：robotics 已经从 demo-heavy narrative 进入 S3/S4-heavy evidence tracking phase，因为 AI/model/data stack、低价硬件入口、客户现场 KPI、制造/产能 KPI、以及 filing-backed supplier / OEM revenue evidence 同时变得可量化；但仍缺 S5 scaled-commercial-economics chain。🟢/🟠 based on primary company filings/product pages/posts reviewed 2026-06-09 to 2026-06-13.

1. Core question

为什么是现在，而不是 2015/2021/2023？

更精确的问题不是“机器人 TAM 是否很大”，而是：

公开证据是否从 demo video 变成可跟踪 KPI？
KPI 是否足够接近客户经济性？
哪些证据只是 S3/S4，哪些才会升级到 S5？
如果 Codex 把它压成一页 slide，哪些 caveat 必须保留？

2. Morning-review catalyst card

Catalyst	As-of evidence anchor	Why it matters	Current signal	Missing before S5
1. AI / model / data stack becomes trackable	NVIDIA Isaac GR00T official developer page describes open robot foundation model / data pipeline / simulation / middleware / runtime / Jetson Thor stack; Physical Intelligence π0 / openpi / π0.5 / π0.7 and Skild AI official posts add model-layer evidence. Reviewed in internal why-now model files as of 2026-06-12. 🟢	机器人研究从单一硬件 demo 变成 model + data + simulation + deployment stack，可以持续跟踪工具链、benchmark、adoption 和 deployment effect。	S3/S4 platform-formation signal	Customer-site reduction in intervention rate, deployment time, failure rate, or labor cost; software/model revenue; multi-OEM paid adoption.
2. Hardware cost / access curve has public price anchors	Unitree official product pages: R1 AIR from US$4,900 / R1 from US$5,900; G1 from US$13.5K; H2 US$29,900; H2 Plus and G1-D add developer / model workflow surfaces. Accessed 2026-06-10. 🟢	价格锚降低实验成本，扩大 developer / researcher / pilot base，让机器人从少数实验室平台变成更宽的试错表面。	S3 cost-access / platform signal	Shipments by product line, customer mix, reliability, warranty/support cost, gross margin, repeat orders, industrial uptime.
3. Customer-site deployment KPI is appearing	Figure BMW official post dated 2025-11-19: 11-month Figure 02 deployment at BMW Spartanburg, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles. 🟢	这是从视频到工作量 KPI 的跃迁：时间、任务量、生产场景、客户现场都开始可量化。	S4 deployment KPI signal	Robot count, contract value, customer-confirmed ROI/payback, uptime, intervention rate, repeat order, service burden.
4. Manufacturing / capacity evidence is becoming numeric	Figure BotQ official post dated 2026-04-29: 350+ Figure 03 robots delivered, production cadence from 1/day to 1/hour, EOL FPY >80%, battery-line FPY 99.3%, 500+ battery packs, 9,000+ actuators. Tesla Q1 2026 SEC filing dated 2026-04-22: Fremont first-generation Optimus line designed for 1M robots/year; Texas second-generation line designed for long-term annual production capacity of 10M robots. 🟢	Humanoid company claims can now be tested against cadence, yield, output, line utilization, component supply and capacity caveats.	S4 manufacturing / capacity signal	Actual quarterly output, accepted units, utilization, yield over time, cost curve, field failure rate, delivery/customer acceptance, revenue/margin.
5. Filing-backed commercial / supplier evidence is no longer only rumor	UBTECH 2025 annual report disclosed 2026-04-15: revenue RMB 2.001bn (+53.3% YoY), full-size embodied humanoid revenue RMB 820.557m (+2,203.7% YoY), 1,079 full-size embodied humanoid units sold, over 6,000 annualized full-size humanoid capacity at end-2025. 绿的谐波 2025 annual report disclosed 2026-04-23: 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 with 34.88% gross margin. 🟢	上市公司 filings 把部分机器人链条从概念映射推进到收入、销量、毛利、产能可核查层。	S4 filing-backed validation signal	Humanoid-specific margin, named customer utilization, repeat orders, customer ROI/payback, service/warranty cost, named OEM supplier relationship, single-robot value content.

3. What changed vs old robotics cycles

Signal: research unit changed

旧周期更常见的是：

demo video；
TAM slide；
“即将量产”；
客户 logo；
未证实供应链传闻。

2025–2026 证据更强的地方在于：

runtime hours / task volume / shift schedule 开始出现。🟢 Figure BMW official post, 2025-11-19.
production cadence / yield / component count 开始出现。🟢 Figure BotQ official post, 2026-04-29.
designed capacity / named factory line / capacity caveats 开始进入 SEC filing。🟢 Tesla Q1 2026 Form 8-K Exhibit 99.1, filed 2026-04-22.
product price anchors 公开且可比较。🟢 Unitree official product pages, accessed 2026-06-10.
public filings 里出现 embodied humanoid revenue / robot-related component revenue / units / gross margin。🟢 UBTECH 2025 annual report; 绿的谐波 2025 annual report.

Noise: what still should not be over-read

这些证据仍不能推出：

humanoid commercialization solved。🔴 unsupported as of 2026-06-13.
Tesla 1M / 10M 是当前产量。🔴 Tesla language is designed/long-term capacity; actual output not disclosed.
Figure BMW KPI 已证明完整 economics。🔴 contract value, ROI, intervention rate and robot count not disclosed.
Unitree low price proves reliability or gross margin。🔴 shipments, margin, warranty/support cost not disclosed.
绿的谐波 robot-related revenue equals named humanoid OEM exposure。🔴 named humanoid customer / contract not disclosed in reviewed filings.

4. Stage classification

Current sector-stage label for public slide/site as of 2026-06-13:

S3/S4-heavy early deployment + supply-chain validation phase; not S5 scaled-commercial-economics / return cycle. 🟠 synthesis from primary evidence.

Why not S5 yet:

Paid deployment economics are incomplete: contract value, repeat order, customer ROI/payback often missing. 🟢/🟠
Reliability data is incomplete: uptime, intervention rate, failure rate, safety incidents rarely disclosed. 🟢/🟠
Production evidence is incomplete: designed capacity / cadence ≠ accepted customer output or profitable utilization. 🟢/🟠
Financial materiality is fragmented: filings are improving but humanoid-specific margin, service cost and cash conversion remain limited. 🟢/🟠
Supplier attribution remains risky: product fit / robot-related revenue ≠ named Tesla/Figure/Unitree/other OEM customer proof. 🟠

5. Codex compression guardrail

If Codex packages this into one slide, use this exact framing:

Title: Why robotics now: evidence type changed

Subtitle: Robotics is no longer only demo videos; five evidence streams are now measurable. But S4 evidence is not S5 economics.

Five cards:

AI/model/data stack
- Trackable stack: GR00T / open model tooling / simulation / data pipelines.
- Missing: customer-site intervention-rate and deployment-cost reduction.
Hardware cost/access
- Unitree public price anchors: R1 from US$5.9K; G1 from US$13.5K; H2 US$29.9K.
- Missing: shipments, reliability, margin, ROI.
Deployment KPI
- Figure BMW: 1,250+ runtime hours; 90,000+ parts; 10-hour shifts; 30,000+ X3 contribution.
- Missing: contract value, robot count, repeat order, ROI.
Manufacturing/capacity KPI
- Figure BotQ: 350+ robots; 1/day to 1/hour; >80% EOL FPY. Tesla: Fremont 1M/year designed line; Texas 10M/year long-term designed line.
- Missing: actual output, utilization, yield over time, economics.
Filing-backed validation
- UBTECH: RMB 820.557m full-size embodied humanoid revenue; 1,079 units sold. 绿的谐波: RMB 422.528m robot-parts revenue; 425,158 reducer units.
- Missing: customer ROI, humanoid-specific margin, named supplier/OEM proof.

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

6. What would change our mind

Upgrade the why-now thesis from “evidence tracking phase” toward “return-cycle proof” only if several of these appear together:

Multi-customer paid deployments with robot count, task KPI, contract value and repeat orders. 🟢
Customer-confirmed ROI/payback, not only vendor-disclosed operating metrics. 🟢
Low intervention rate / uptime / failure-rate disclosure over long shifts and multiple sites. 🟢
Actual production output and accepted deliveries, not only designed capacity or production intent. 🟢
Humanoid-specific revenue, gross margin, service/warranty cost and cash conversion from filings. 🟢
Supplier relationship confirmed by named OEM, filing-backed order/revenue, or identifiable production-part evidence. 🟢

Downgrade if:

2026–2027 newsflow remains mostly demos, new model names and customer logos without operating/economic KPIs. 🟠
Manufacturing output rises but deployed utilization, customer ROI and repeat orders do not follow. 🟠
Low-cost hardware expands experiments but warranty/support cost or quality constraints prevent industrial deployment. 🟠
Supplier revenue grows but margins compress materially or named humanoid exposure never appears. 🟠

7. Common misconceptions

Misconception 1: “Why robotics now = robots are now investable as a basket.”

Correction: why-now means the evidence base is finally measurable enough to track; it is not a basket trade recommendation. 🟠

Misconception 2: “Foundation models solve robotics commercialization.”

Correction: model progress matters only if it reduces intervention, setup time, failure rate or deployment cost in customer sites. 🟠

Misconception 3: “Low price proves Unitree has won.”

Correction: low price lowers experimentation cost; it does not prove reliability, gross margin, customer ROI or service economics. 🟢/🟠

Misconception 4: “Tesla designed capacity is production.”

Correction: Tesla’s capacity language is designed / long-term capacity, and reviewed filings do not disclose Optimus actual output, accepted units, external paid customers or unit economics. 🟢

Misconception 5: “Filing-backed supplier revenue proves named humanoid customer exposure.”

Correction: filings improve evidence quality, but named customer, contract value, humanoid-specific margin and value content are separate gates. 🟢/🟠

8. Think Deeper questions

Which evidence upgrades first in robotics: deployment KPI, customer ROI, production yield, or financial materiality?
If robot bodies become cheap, does value migrate to OEM hardware, model/data layer, deployment integration, maintenance, or customer workflow software?
Is Tesla’s internal factory path commercially equivalent to external customer deployment, or should it be evaluated as productivity/payback evidence first?
If Figure replicates BMW-style KPIs across Catalyst and a third customer, what exact missing metric still blocks S5?
If suppliers show robot-related revenue growth but no named customers, how much should public research infer without crossing into rumor?

9. Public site draft section

Why robotics now: the evidence unit changed

Robotics is not suddenly interesting because the TAM slide got bigger. It is interesting because the public evidence unit changed.

By 2025–2026, several evidence streams became measurable at the same time: model/data stacks, lower-cost hardware, real customer-site operating KPIs, manufacturing cadence/yield/capacity metrics, and filing-backed robot revenue or component revenue. That combination moves the field from pure demo watching into evidence tracking.

But the conclusion should stay disciplined. Current evidence is S3/S4-heavy: strong enough to justify a research system, not strong enough to declare scaled commercial economics solved. The next thesis-changing signals are not more viral videos; they are repeat paid deployments, customer ROI, low intervention rate, actual production output, durable gross margin and filing-backed financial materiality.

10. Source list

NVIDIA Isaac GR00T official developer page: https://developer.nvidia.com/isaac/gr00t 🟢 accessed / reviewed in internal robotics why-now files as of 2026-06-12.
Physical Intelligence official π0 / openpi / π0.5 / π0.7 posts: https://www.physicalintelligence.company/ 🟢 reviewed in robotics-why-now-ai-model-catalyst-v1.md, as of 2026-06-12.
Skild AI official company / research / funding / partnership posts: https://www.skild.ai/ 🟢 reviewed in robotics-foundation-model-layer-evidence-v1.md, as of 2026-06-12.
Unitree R1 product page: https://www.unitree.com/R1 🟢 accessed 2026-06-10; cited in robotics-unitree-deep-dive-v1.md.
Unitree G1 product page: https://www.unitree.com/g1/ 🟢 accessed 2026-06-10.
Unitree H2 product page: https://www.unitree.com/H2 🟢 accessed 2026-06-10.
Figure, “F.02 Contributed to the Production of 30,000 Cars at BMW”, 2025-11-19: https://www.figure.ai/news/production-at-bmw 🟢 checked 2026-06-10.
Figure, “Ramping Figure 03 Production”, 2026-04-29: https://www.figure.ai/news/ramping-figure-03-production 🟢 checked 2026-06-10.
Figure, “Figure Signs Agreement with Catalyst Brands to Scale Humanoid Operations”, 2026-05-26: https://www.figure.ai/news/figure-signs-agreement-with-catalyst-brands 🟢 checked 2026-06-10.
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 🟢 reviewed in robotics-tesla-optimus-filing-evidence-audit-v1.md.
UBTECH Robotics Corp Ltd Annual Report 2025, disclosed 2026-04-15 🟢 reviewed in robotics-ubtech-walker-evidence-v1.md.
绿的谐波 2025 annual report PDF, disclosed 2026-04-23: https://static.sse.com.cn/disclosure/listedinfo/announcement/c/new/2026-04-23/688017_20260423_EU45.pdf 🟢 reviewed in robotics-leaderdrive-case-study-slide-evidence-v3.md.

11. Public-safe flag

Public-safe: yes.

Exclude:

Hugo private portfolio weights or internal allocation rationale.
Any buy / sell / hold language.
Paid-report excerpts or private-channel claims.
Named supplier-customer inferences not supported by primary filings or official customer/OEM confirmation.
Claims that designed capacity, low price, customer KPI, or supplier revenue alone proves S5 commercialization.