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

Research library · updated 2026-06-10 · public

Robotics deployment evidence ladder v1

Date: 2026-06-10 Owner: Hugo / Genius Team Research owner: Finance / Charlie AGT-002 Status: public-safe research artifact Visibility: PUBLIC Related files:

  • knowledge/robotics-research-harness.md
  • knowledge/robotics-humanoid-evidence-scorecard-v1.md
  • knowledge/robotics-figure-ai-deep-dive-v1.md
  • knowledge/robotics-agility-digit-evidence-v1.md
  • knowledge/robotics-tesla-optimus-evidence-v2.md
  • knowledge/robotics-unitree-deep-dive-v1.md

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

0. One-line answer

截至 2026-06-10，humanoid robotics 最应该公开展示的不是“谁最强”，而是“deployment evidence ladder 到了哪一级”：Agility / Digit 与 Figure AI 已经有最接近真实客户部署的 S4 证据；Tesla Optimus 是更强的 production-infrastructure ambition 但缺 deployment KPI；Unitree / 宇树是更强的 cost / platform-access evidence 但缺客户运行 KPI。全行业仍缺 S5：客户确认的付费经济性、repeat orders、uptime / intervention rate、robot count、ROI/payback、revenue / gross margin。🟢/🟠

1. Core question

什么样的证据才说明 humanoid robotics 正在从 demo 走向 deployment？

公共分享里应避免把 demo、客户 logo、capacity ambition、低价、commercial agreement 混为一谈。更好的方法是按证据强度分层：

  1. Demo / official video.
  2. Productized robot / public specs / public price.
  3. Named customer pilot.
  4. Commercial agreement / RaaS / customer site deployment.
  5. Quantified operational KPI: runtime, task count, shifts, throughput, site count.
  6. Repeatability: repeat orders, multi-site expansion, same playbook across customers.
  7. Economics: contract value, ROI/payback, ASP, gross margin, maintenance cost, revenue line.

截至 2026-06-10，公开证据主要集中在第 3–5 层；第 6–7 层仍然缺口最大。🟠 Charlie framework judgment based on reviewed primary sources.

2. Deployment evidence ladder

Ladder levelWhat countsCurrent public examplesSignal gradeWhat is still missing
L1 DemoOfficial robot video / lab demoTesla / Figure / Unitree all have official product or demo materials. 🟢S1/S2 unless tied to KPIRuntime, task repetition, intervention rate, customer site.
L2 Productized hardwarePublic product page, specs, price, developer toolingUnitree R1 from US$4,900, G1 from US$13.5K, H2 US$29,900; H2 Plus references NVIDIA Jetson T5000 / Isaac GR00T / TeleOp / Sim; G1-D data/model tooling. 🟢 Unitree product pages, accessed 2026-06-10S3Customer deployment, runtime, reliability, margins.
L3 Named pilot / customer siteNamed customer or site, but limited economicsTesla internal factory-use language; Agility GXO proof-of-concept before RaaS; Figure BMW deployment start. 🟢S3/S4Robot count, contract value, customer-confirmed ROI.
L4 Commercial agreement / RaaSAgreement language stronger than logo-only partnershipAgility GXO multi-year RaaS agreement, 2024-06-27; Agility TMMC RaaS agreement, 2026-02-19; Figure Catalyst commercial agreement, 2026-05-26. 🟢S4 if named customer + use caseRevenue size, margins, duration economics, repeat order.
L5 Quantified operational KPITask volume, runtime, shift pattern, throughput, production contributionFigure BMW: 11-month deployment, 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to 30,000+ BMW X3 vehicles. 🟢 Figure, 2025-11-19. Agility Digit: 100,000+ totes moved at GXO Flowery Branch. 🟢 Agility, 2025-11-20Strong S4Uptime, intervention rate, robot count, failure distribution, customer ROI.
L6 Repeatability / scalingMulti-customer deployment with comparable KPI and repeat ordersAgility has multiple named agreements: GXO, Mercado Libre, TMMC, Schaeffler intent/investment. 🟢 2024-2026 Agility official releases; Figure has BMW + Catalyst. 🟢S4, not S5Same KPI across multiple customers, volume, repeat-order economics.
L7 Economics / financial materialityContract value, payback, gross margin, revenue line, unit economicsNot publicly disclosed in reviewed Tesla / Figure / Unitree / Agility materials as of 2026-06-10. 🟢/🟠S5 when disclosedThis is the main missing layer.

3. Company evidence mapping

3.1 Agility Robotics / Digit — strongest commercial-structure evidence

As-of: 2026-06-10.

Current classification: paid early deployment / commercial deployment in bounded logistics-manufacturing workflows; not scaled return-cycle proof. 🟠

Strongest evidence:

  • GXO and Agility signed a multi-year agreement to deploy Digit in GXO logistics operations after a late-2023 proof-of-concept pilot; Agility calls it the industry’s first formal commercial deployment and first RaaS deployment of humanoid robots. 🟢 Agility official release, 2024-06-27.
  • Digit moved over 100,000 totes at GXO’s Flowery Branch facility. 🟢 Agility official resource, 2025-11-20.
  • Mercado Libre and Agility announced a commercial agreement to deploy Digit robots, beginning in San Antonio, Texas. 🟢 Agility official release, 2025-12-10.
  • Toyota Motor Manufacturing Canada signed a Robots-as-a-Service agreement after a successful pilot. 🟢 Agility official release, 2026-02-19.

Why it matters:

  • Agility has the cleanest public mix of named customers + commercial/RaaS language + quantified workflow milestone. 🟢/🟠
  • The workflow is still bounded: tote movement and logistics/manufacturing tasks, not broad general-purpose autonomy. 🟠

Missing before S5:

  • Deployed robot count, contract value, RaaS pricing, revenue contribution, gross margin, uptime, intervention/failure rate, customer-confirmed ROI/payback, and repeat-order volume. 🟠

3.2 Figure AI — strongest quantified deployment KPI + manufacturing KPI

As-of: 2026-06-10.

Current classification: early commercial deployment / production ramp; not scaled commercial economics. 🟠

Strongest evidence:

  • Figure disclosed an 11-month Figure 02 deployment at BMW Group Plant Spartanburg. 🟢 Figure official post, 2025-11-19.
  • BMW deployment metrics: 10-hour shifts Monday-Friday, 90,000+ parts loaded, 1,250+ runtime hours, contribution to production of 30,000+ BMW X3 vehicles, estimated 1.2M+ robot steps / 200+ miles. 🟢 Figure official post, 2025-11-19.
  • Figure announced a commercial agreement with Catalyst Brands to deploy humanoids into Catalyst’s distribution/logistics network, starting at Reno, Nevada. 🟢 Figure official post, 2026-05-26.
  • BotQ delivered 350+ Figure 03 robots, increased cadence from 1/day to 1/hour, achieved EOL first-pass yield >80%, battery-line first-pass yield 99.3%, shipped 500+ battery packs, and produced 9,000+ actuators across 10+ SKUs. 🟢 Figure official post, 2026-04-29.

Why it matters:

  • Figure gives the clearest public “runtime + task count + manufacturing cadence/yield” packet among reviewed humanoid OEMs. 🟢
  • BotQ manufacturing metrics make the evidence stronger than a one-off BMW demo, but still do not prove sell-through economics. 🟢/🟠

Missing before S5:

  • BMW / Catalyst contract value, robot count by customer, customer-confirmed ROI/payback, repeat-order terms, uptime/intervention rate, field failure distribution, gross margin. 🟠

3.3 Tesla Optimus — strongest capacity ambition, weakest public deployment KPI

As-of: 2026-06-10.

Current classification: pilot + production-line installation / infrastructure preparation; not scaled commercial deployment. 🟠

Strongest evidence:

  • Tesla Q1 2026 update says the Fremont first-generation Optimus line is designed for 1M robots/year. 🟢 Tesla Q1 2026 Form 8-K Exhibit 99.1, filed 2026-04-22.
  • Tesla Q1 2026 update says Gigafactory Texas is being prepared for a second-generation Optimus line designed for long-term annual capacity of 10M robots. 🟢 Tesla Q1 2026 Form 8-K Exhibit 99.1.
  • Tesla Q4 2025 update says Gen 3 Optimus is its first design meant for mass production and preparations are underway for the first production line, with start of production planned before end-2026. 🟢 Tesla Q4 2025 Form 8-K Exhibit 99.1, filed 2026-01-28.

Why it matters:

  • Tesla is the biggest capacity-ambition case and the most important public-market anchor for humanoid manufacturing scale. 🟢/🟠
  • However, designed capacity is not actual output, and Tesla’s reviewed filings do not disclose Optimus operating KPI. 🟢/🟠

Missing before S5:

  • Actual Optimus production output, line yield, utilization, cost curve, ASP, robot revenue, external paid deployment, factory task KPI, uptime, intervention rate, safety incidents, payback. 🟠

3.4 Unitree / 宇树 — strongest cost/access evidence, not deployment proof

As-of: 2026-06-10.

Current classification: productized low-cost hardware platform / developer ecosystem; deployment economics not yet validated in reviewed official pages. 🟠

Strongest evidence:

  • Unitree R1 official page lists price from US$4,900. 🟢 Unitree R1 page, accessed 2026-06-10.
  • Unitree G1 official page lists price from US$13.5K. 🟢 Unitree G1 page, accessed 2026-06-10.
  • Unitree H2 official page lists H2 at US$29,900, tax and shipping excluded. 🟢 Unitree H2 page, accessed 2026-06-10.
  • Unitree H2 Plus references NVIDIA Jetson T5000 onboard compute, FP4 2,070 TFLOPS, NVIDIA Isaac GR00T / TeleOp / Sim workflows. 🟢 Unitree H2 Plus page, accessed 2026-06-10.
  • Unitree G1-D describes data acquisition, processing, labeling, review, data asset management, model training, and inference tools. 🟢 Unitree G1-D page, accessed 2026-06-10.

Why it matters:

  • Unitree makes humanoid hardware look more like an accessible developer / experimentation surface than a rare lab asset. 🟢/🟠
  • This can accelerate ecosystem learning, but low price is not a proxy for deployment reliability or value capture. 🟠

Missing before S5:

  • Humanoid unit shipments, named industrial customer runtime, task KPI, uptime, intervention rate, repeat orders, gross margin, warranty/maintenance cost, humanoid-specific revenue. 🟠

4. Signal vs noise

Signal

  • Named commercial/RaaS agreement with customer and task context. 🟢
  • Runtime hours, shift schedule, task count, throughput, production contribution. 🟢
  • Manufacturing cadence and first-pass yield, if tied to actual delivered robots. 🟢
  • Multi-customer replication with comparable KPIs. 🟢/🟠
  • Customer-confirmed ROI/payback and repeat orders. 🟢 S5 candidate.

Noise unless upgraded

  • Viral demo without reset/intervention/runtime disclosure. 🔴/🟠
  • “Commercial agreement” without robot count, site count, revenue, or task KPI. 🟢/🟠
  • Customer logo with no deployment status. 🟢/🟠
  • Designed capacity without actual output/yield/utilization. 🟢/🟠
  • Low hardware price treated as winner proof without reliability/margin/service data. 🟠

5. What would change our mind

Upgrade the sector closer to S5 if:

  1. A named customer independently confirms robot count, deployment duration, uptime, intervention rate, task success, ROI/payback, and repeat order. 🟢
  2. A humanoid OEM discloses robot revenue/backlog, ASP, gross margin, service/support cost, and production yield. 🟢
  3. The same deployment playbook works across 3+ customers with comparable KPI and falling deployment cost. 🟢/🟠
  4. Public filings show robotics revenue becoming financially material rather than narrative optionality. 🟢

Downgrade if:

  1. Deployments remain narrow showcase workflows with no repeat-order economics through 2026. 🟠
  2. RaaS/service burden makes gross margin unattractive despite customer interest. 🟠
  3. Capacity and production announcements do not convert into actual output and customer use. 🟠
  4. Low-cost hardware expands demos but not paid industrial usage. 🟠

6. Common misconceptions

Misconception 1: “Commercial agreement = commercialization solved.”

Correction: commercial/RaaS language is stronger than a logo-only partnership, but S5 requires contract economics, reliability, repeatability, and margin evidence. 🟠

Misconception 2: “100,000 totes or 90,000 parts proves general-purpose humanoids.”

Correction: these are strong bounded-workflow deployment KPIs. They do not by themselves prove broad task generalization, multi-customer economics, or customer ROI. 🟢/🟠

Misconception 3: “Tesla’s 1M / 10M designed capacity is equivalent to production scale.”

Correction: designed capacity is an infrastructure ambition; actual output, yield, utilization, and task economics remain undisclosed in reviewed filings. 🟢/🟠

Misconception 4: “Low-cost robots automatically win the market.”

Correction: low price expands access, but value may migrate to software, data, deployment integration, maintenance, or customer-side process redesign. 🟠

7. Think Deeper

  1. If humanoid deployment starts in narrow logistics/manufacturing workflows, is “general-purpose” a destination or a marketing shortcut?
  2. Which S5 signal matters more first: customer-confirmed ROI/payback or OEM-disclosed robot revenue/gross margin?
  3. If RaaS is the early deployment model, does value accrue to robot OEMs, fleet operators, integration software, or customers?
  4. Does low-cost hardware compress OEM margins before deployment economics are proven?
  5. Should public robotics research rank companies, or rank evidence types: deployment, manufacturing, cost/access, AI/data, economics?

8. Public site / slide draft

From demo to deployment: the ladder that matters

The humanoid robotics question is no longer just “can the robot move?” A better question is: how far up the deployment evidence ladder has each company climbed?

As of 2026-06-10, Agility and Figure have the strongest public deployment evidence. Agility has named commercial/RaaS agreements and a quantified GXO milestone: Digit moved 100,000+ totes. Figure has BMW deployment metrics: 1,250+ runtime hours, 90,000+ parts loaded, 10-hour shifts Monday-Friday, and contribution to 30,000+ BMW X3 vehicles. These are not S5 proof, but they are materially stronger than demo videos.

Tesla and Unitree show different evidence types. Tesla has the largest disclosed capacity ambition: a Fremont line designed for 1M robots/year and a Texas line designed for long-term 10M robots/year. Unitree has the clearest cost/access signal: R1 from US$4,900, G1 from US$13.5K, and H2 at US$29,900. Both are important, but neither replaces deployment economics.

The decisive next signals are not more videos. They are repeat orders, robot count, uptime, intervention rate, customer payback, production yield, revenue, and gross margin.

9. Source list

Agility Robotics:

Figure AI:

Tesla:

Unitree / 宇树:

10. Public-safe flag

Public-safe: yes.

Exclusions:

  • No Hugo private portfolio weights.
  • No paid-report excerpts.
  • No direct trade recommendation.
  • Do not present the ladder as a company ranking or stock ranking; it is an evidence-quality framework.