Robotics OEM evidence curves beyond Tesla / Figure / Unitree v1

Date: 2026-06-15 Research owner: Finance / Charlie AGT-002 Status: public-safe research artifact Visibility: PUBLIC Output target: company-map section / slide appendix / site module Public-safety flag: no trade recommendation; no Hugo portfolio data; no private channel checks.

One-line answer

The next useful company-map upgrade is not another winner ranking. It is a four-lane evidence map: UBTECH shows the clearest filing-backed humanoid revenue/unit disclosure; Agility shows bounded logistics RaaS/commercial workflow proof; Apptronik shows a partner-rich manufacturing and AI-stack scaling path; 1X shows the hardest home-humanoid preorder/capacity/data-loop experiment. All four are S4 or S3/S4 signals, not S5 scaled-commercial economics, because public sources still lack repeat-order ROI, uptime/intervention, customer economics, and durable humanoid gross-margin proof. 🟢/🟠 as of 2026-06-15

Core question

After Tesla / Figure / Unitree, which additional humanoid OEM evidence curves are worth tracking, and what distinct S4-to-S5 bottleneck does each test?

Why this artifact exists

Existing robotics research already has deep public-safe artifacts for:

• Tesla Optimus: capacity-intent / manufacturing infrastructure, not commercialization proof.
• Figure AI: strongest disclosed deployment/manufacturing KPI among reviewed private humanoids, still missing S5 economics.
• Unitree: cost-access and developer-platform curve, still missing industrial deployment economics.
• Leaderdrive / 绿的谐波: supplier filing signal, not named humanoid customer proof.

This note fills the company-map gap for Agility, Apptronik, 1X, and UBTECH so Codex can package a broader OEM landscape without flattening different evidence types into a horse race.

Evidence map

Stage classification

Current company-map read:

1. UBTECH: S4 because annual-report disclosure moves beyond demo into filing-backed revenue and unit evidence. Not S5 because the filing still lacks repeat-order ROI, humanoid-specific margin, service burden, and customer-side utilization. 🟢/🟠 as of 2026-04-15 annual report.
2. Agility: S4 because commercial/RaaS language plus a 100,000+ tote milestone is stronger than a customer-logo announcement. Not S5 because economics, fleet count, uptime, and intervention are undisclosed. 🟢/🟠 as of 2026-02-19 latest reviewed Agility release.
3. Apptronik: S4 because customer, manufacturing, AI-stack, and funding signals align. Not S5 because evidence is still mostly partner/configuration proof rather than measured production or customer economics. 🟢/🟠 as of 2026-02-11 funding release.
4. 1X: S3/S4 because preorder/capacity/home-deployment intent is measurable, but shipment and real-home operating evidence remain pending. 🟢/🟠 as of 2026-06-04 World Model Lab post and 2026-04-30 NEO Factory post.

Overall: the non-mainline OEM set strengthens the thesis that robotics evidence has moved from viral-demo watching into measurable S3/S4 tracking. It does not yet prove a broad S5 humanoid return cycle. 🟠 Charlie synthesis, 2026-06-15.

Signal vs noise

Signal

• Filing-backed humanoid revenue / units: UBTECH RMB 820.557m humanoid revenue and 1,079 units in 2025. 🟢
• Commercial workflow with throughput milestone: Agility / GXO 100,000+ totes and RaaS language. 🟢
• Manufacturing partner loop: Apptronik / Jabil building Apollo and using factory environment for validation tasks before customer deployment. 🟢
• Home deployment architecture: 1X preorder/capacity plus Expert Mode, which is important because it admits autonomy gaps while creating a potential data/service loop. 🟢/🟠

Noise / do-not-overread

• Do not rank these four as simple winners. They test different bottlenecks: filing scale, RaaS workflow, partner manufacturing, and home data loop. 🟠
• Do not treat annualized capacity as current output. UBTECH >6,000 capacity and 1X 10,000/year capacity are capacity anchors, not actual production rates. 🟢/🟠
• Do not treat RaaS/commercial language as full economics. Agility still lacks public robot count, RaaS pricing, customer ROI, and gross margin. 🟢/🟠
• Do not treat partnerships/funding as product-market fit. Apptronik’s nearly US$1bn capital and Google/NVIDIA/Jabil/Mercedes ecosystem lower execution-risk questions but raise the bar for commercial proof. 🟢/🟠
• Do not treat Expert Mode as either failure or proof. For 1X, the key metric is intervention rate and cost trend over time, not the label “autonomous” vs “teleoperated.” 🟢/🟠

S5 upgrade checklist by company

Public company-map copy draft

Beyond Tesla / Figure / Unitree: four additional evidence curves

The broader humanoid OEM map should not be compressed into one leaderboard. UBTECH, Agility, Apptronik, and 1X each answer a different question.

UBTECH asks whether a listed humanoid OEM can turn robots into reportable revenue: its 2025 annual report disclosed RMB 820.557m of full-size embodied humanoid products/services revenue and 1,079 units sold. Agility asks whether humanoids can work in bounded logistics workflows: GXO moved into a multi-year RaaS agreement and Digit later crossed 100,000+ totes at one facility. Apptronik asks whether a partner-rich stack — Mercedes, GXO, Jabil, NVIDIA, Google DeepMind, and nearly US$1bn capital — can convert into measured production and deployment. 1X asks the hardest home question: can a consumer humanoid use preorder demand, a 10,000/year factory, Expert Mode, and a data loop to reach real-home usefulness?

The public conclusion is disciplined: these are four S3/S4 evidence curves, not S5 proof. The next gate is not another demo video; it is robot count, uptime, intervention rate, customer ROI/payback, repeat orders, revenue quality, gross margin, and service burden.

Common misconceptions

1. “UBTECH has revenue, so humanoid commercialization is solved.”

   • Correction: filing-backed revenue is strong S4 evidence, but S5 needs repeatability, customer economics, humanoid-specific margin, and service burden. 🟢/🟠

2. “Agility’s 100,000 totes proves general-purpose humanoids.”

   • Correction: it proves meaningful throughput in a bounded logistics workflow; it does not prove broad autonomy or full customer economics. 🟢/🟠

3. “Apptronik has Google/NVIDIA/Jabil/Mercedes, therefore it wins.”

   • Correction: partner stack is a strong configuration signal; actual deployment economics still need measured output and customer ROI. 🟢/🟠

4. “1X Expert Mode means the product is not important.”

   • Correction: Expert Mode reveals autonomy debt, but may also be a practical home-deployment/data-loop strategy. The decisive metric is intervention-rate decline versus service cost. 🟢/🟠

Think Deeper questions

1. Which evidence curve is most transferable across companies: filing-backed revenue, throughput KPI, manufacturing partner validation, or real-home data-loop evidence?
2. If early humanoid commercialization is RaaS-heavy, where does value accrue: robot OEM, fleet operator, integration/service layer, software/data layer, or customer productivity capture?
3. Does the home route produce more generalizable data than warehouse/factory routes, or does it impose too much safety/privacy/support complexity before economics work?
4. Should public-company robotics research prioritize UBTECH-style filing disclosure even if private-company deployment KPI may be technically stronger?

Packaging recommendation for Codex / Warrior

Status suggestion: SYNTHESIS_CANDIDATE, PUBLIC, Output: site or slide.

Suggested title:

• “Beyond Tesla / Figure / Unitree: four evidence curves, four missing S5 gates”

Suggested layout:

1. Four cards: UBTECH filing scale / Agility RaaS workflow / Apptronik partner-manufacturing stack / 1X home data loop.
2. One S5 gate strip: robot count, uptime, intervention, ROI/payback, repeat order, revenue quality, gross margin, service burden.
3. Footer: “Evidence map only. No winner ranking. No trade recommendation. Capacity, revenue, RaaS, partnerships, and preorders are S3/S4 signals until economics are repeatable.”

Source list

• UBTECH Robotics Corp Ltd Annual Report 2025, disclosed 2026-04-15; captured in robotics-ubtech-walker-evidence-v1.md. 🟢
• UBTECH official Walker S / S1 / S2 product pages, accessed 2026-06-11 in prior artifact. 🟢
• Agility Robotics, “GXO Signs Industry-First Multi-Year Agreement with Agility Robotics,” 2024-06-27; captured in robotics-agility-digit-evidence-v1.md. 🟢
• Agility Robotics, “Digit Moves Over 100,000 Totes in Commercial Deployment,” 2025-11-20; captured in robotics-agility-digit-evidence-v1.md. 🟢
• Agility Robotics, Mercado Libre commercial agreement 2025-12-10 and Toyota Motor Manufacturing Canada RaaS agreement 2026-02-19; captured in robotics-agility-digit-evidence-v1.md. 🟢
• Apptronik Apollo product page, Mercedes-Benz agreement, GXO proof-of-concept, NVIDIA collaboration, Google DeepMind partnership, Jabil collaboration, and 2026 Series A-X release; captured in robotics-apptronik-apollo-evidence-v1.md. 🟢
• 1X NEO product page, manufacturing page, NEO Factory post dated 2026-04-30, NVIDIA collaboration post dated 2025-03-18, World Model Lab post dated 2026-06-04; captured in robotics-1x-neo-evidence-v1.md. 🟢
• Charlie cross-company evidence-curve synthesis and S4/S5 stage classification, 2026-06-15. 🟠

Public-safety exclusions

Do not include:

• Hugo private portfolio information, watchlist weights, trade rationale, private channel checks, paid-report excerpts, tax/legal advice, or buy/sell/hold language.
• Any claim that UBTECH / Agility / Apptronik / 1X has proven S5 scaled-commercial economics.
• Any claim that annualized capacity equals current production.
• Any winner ranking unless future sources disclose comparable robot count, economics, and customer ROI metrics.