China OEM evidence normalization for robotics company map

Date: 2026-06-18 Owner: Hugo / Genius Team Research owner: Finance / Charlie AGT-002 Visibility: PUBLIC Status: SYNTHESIS_CANDIDATE Public-safety flag: public-source only; no Hugo portfolio data; no trade recommendation; no private channel checks.

One-line answer

For the public robotics company map, AgiBot / 智元, Fourier / 傅利叶, Galbot / 银河通用, and EngineAI / 众擎 should not be collapsed into a single “China humanoid winners” bucket. As of the reviewed evidence through 2026-06-12 and this synthesis dated 2026-06-18, they are four different S3/S4 evidence curves: data/platform breadth, rehab-institutional transfer, productized wheeled manipulation, and production-intent speed. None has public S5 scaled-commercial-economics proof yet. 🟠 Charlie synthesis based on existing source packets.

Why this artifact exists

The main public company page already has strong P0 evidence cards for Tesla, Figure, Unitree, UBTECH and selected US OEMs. The remaining packaging risk is that China private OEMs get shown as a broad list of logos or demos. This file turns the existing v1 evidence packets into a compact, source-graded comparison that Codex can use for /robotics/companies without overclaiming.

Source packets used:

knowledge/robotics-agibot-evidence-v1.md — 🟢 official AgiBot pages accessed 2026-06-12; 🟠 Charlie interpretation.
knowledge/robotics-fourier-evidence-v1.md — 🟢 official Fourier pages accessed 2026-06-12; 🟠 Charlie interpretation.
knowledge/robotics-galbot-evidence-v1.md — 🟢 official Galbot site / SPA bundle accessed 2026-06-12; 🟠 Charlie extraction and interpretation.
knowledge/robotics-engineai-evidence-v1.md — 🟢 official EngineAI pages accessed 2026-06-12; 🟠 Charlie interpretation.

Public company-map table

Company	Evidence curve	Source-backed anchors	Current signal	Missing before S5	Safe public sentence
AgiBot / 智元	Broad embodied-AI platform + data-service layer	Product families include Yuanzheng A-series, Lingxi X-series, Genie G-series, D-series, OmniHand, C5; DaaS page claims hundreds of robots, multiple real interaction environments and 7*24h collection. 🟢 official pages accessed 2026-06-12	S3/S4 product breadth + data-infrastructure signal	Paid deployments, customer robot count, contract value, uptime/intervention, robot revenue, gross margin, repeat orders, proof that DaaS improves autonomy/deployment cost.	“AgiBot’s public evidence is strongest as product-matrix + data-infrastructure ambition, not yet customer economics.”
Fourier / 傅利叶	Rehab robotics incumbent attempting humanoid transfer	Founded 2015; product lineup serves >2,000 institutions across 40 countries/regions; GR-1/GR-2/GR-3 roadmap; named application exploration with CCB Shanghai Branch and SAIC-GM. 🟢 official pages accessed 2026-06-12	S4 institutional base for rehab; S3/S4 humanoid transfer watch	Humanoid-specific deliveries, humanoid revenue/margin, runtime, uptime, intervention rate, customer ROI/payback, repeat orders.	“Fourier’s edge is not a viral demo; it is whether a rehab/institutional base can convert into humanoid economics.”
Galbot / 银河通用	Productized wheeled dual-arm / heavy-duty manipulation platform	Official G1 specs: 173 cm, 85 kg, 190 cm arm span, 0–240 cm reach, 10h runtime, 5 kg payload; S1: up to 50 kg payload, 8h operation, hot-swap battery, 320 kg total weight. 🟢 official site / SPA bundle accessed 2026-06-12	S3/S4 productized-platform watch	Named paid customers, deployed robot count, contract value, uptime/intervention, production output, revenue, gross margin, repeat-order data.	“Galbot is productized enough for the company map, but specs are still not deployment economics.”
EngineAI / 众擎	Fast product cadence + T800 production-intent signal	Founded 2023-10; PM01/SE01/T800/SA01/JS01/S2 product matrix; official 2026-05-22 news headline says first batch of T800 humanoids rolled off production line and company is moving toward 10,000-unit scale delivery. 🟢 official pages accessed 2026-06-12	S3/S4 productized-platform + production-intent watch	Delivered/accepted units, customer names, contract value, production yield, warranty/service cost, revenue, gross margin, runtime/intervention, repeat orders.	“EngineAI’s strongest public signal is production intent; do not read it as 10,000 shipped units.”

Signal ranking by evidence type, not winner ranking

Evidence type	Stronger current example	Why it matters	Guardrail
Data / platform infrastructure	AgiBot	DaaS and ecosystem claims map directly to the embodied-AI data bottleneck. 🟢/🟠 as of 2026-06-12	Data collection is not a data moat unless it improves model performance and lowers deployment cost.
Institutional customer access	Fourier	>2,000 institutions across 40 countries/regions gives a real robotics go-to-market base. 🟢 as of 2026-06-12	Rehab customer base is not automatically humanoid revenue or humanoid ROI.
Bounded manipulation hardware	Galbot	G1/S1 specs create a clearer warehouse/retail/logistics workflow hypothesis than generic humanoid demos. 🟢/🟠 as of 2026-06-12	Product specs do not prove utilization, uptime, or payback.
Production-intent speed	EngineAI	T800 first-batch roll-off / toward-10,000-unit delivery headline is stronger than pure demo flow. 🟢/🟠 as of 2026-06-12	“Toward 10,000-unit scale delivery” is not 10,000 accepted deliveries, revenue, or margin.

Stage classification

Current label: S3/S4 cluster, not S5.

Why S3/S4:

Each company has at least one official public evidence channel with product, company, or application claims. 🟢 existing source packets, accessed 2026-06-12.
Several have stronger-than-demo anchors: AgiBot DaaS / product matrix, Fourier >2,000-institution base and named application exploration, Galbot G1/S1 product specs, EngineAI T800 production-intent headline. 🟢 existing source packets, accessed 2026-06-12.
These signals are useful for company-map coverage and monitoring, but they remain mostly productization, channel, institutional-base, or production-intent evidence. 🟠 Charlie synthesis, 2026-06-18.

Why not S5:

None of the four evidence packets provides filing-level or customer-confirmed humanoid revenue, gross margin, contract value, repeat order, robot count by customer, uptime, intervention rate, service burden, or customer ROI/payback. 🟢 negative evidence from reviewed official pages as of 2026-06-12; 🟠 synthesis dated 2026-06-18.
Product breadth, product specs, open ecosystems, institutional footprint, and production-intent headlines can be S4 monitoring signals, but they do not prove repeatable commercial economics. 🟠 Charlie framework, 2026-06-18.

Signal vs noise rules for Codex packaging

Signal:

Official product specs with dated access and source grade. 🟢
Named counterparties or customer/application exploration, if wording is preserved and economics are not inferred. 🟢/🟠
Production-line or first-batch language, if separated from accepted deliveries and revenue. 🟢/🟠
Institution count or geographic footprint, if tied to the correct business line. 🟢

Noise / overclaim risk:

Logo lists without contract terms.
“World first,” “mass-produced,” “scale delivery,” or “commercialized” wording without unit, revenue, margin or customer proof.
Viral walking/stunt videos as commercialization evidence.
Treating product specs as uptime, intervention-rate, or customer ROI proof.
Treating China OEM velocity as an investable conclusion without value-capture and economics evidence.

What would change our mind

Upgrade the cluster toward S5 only when public sources show several of:

Customer-accepted robot count by company/site/task. 🟢/🟡 required.
Contract value, recognized robot revenue, backlog, or recurring service/software economics. 🟢 required for financials; 🟡 if credible customer documents.
Runtime, uptime, intervention rate, failure rate, maintenance burden, safety incidents, and warranty/service cost. 🟢/🟡 required.
Customer ROI/payback and repeat-order or multi-site expansion. 🟢/🟡 required.
Humanoid-specific gross margin or contribution margin after deployment/service burden. 🟢 required.

Downgrade if:

2026–2027 updates remain product/spec/demo-heavy without customer-site KPI.
“Scale delivery” language does not translate into accepted units and revenue.
Institutional or channel claims fail to produce repeat customer economics.
Service/warranty burden or teleoperation labor makes apparent hardware progress economically weak.

Common misconceptions

“中国 OEM 产品上新快，所以已经接近 S5。” Correction: product cadence is useful S3/S4 evidence; S5 requires delivered units, uptime, intervention rate, economics, and repeat order.
“康复/机构客户基础可以直接等同 humanoid 商业化。” Correction: Fourier’s rehab base matters, but transfer into humanoid deployment must be measured separately.
“量产意图 headline 等于实际出货。” Correction: EngineAI’s T800 headline should be tracked as production-intent signal, not as accepted deliveries or revenue.
“产品参数越强，客户 ROI 越好。” Correction: specs must become utilization, reliability, support cost, and payback before they matter economically.

Think Deeper questions

Which China OEM evidence converts first into customer economics: data services, rehab/institutional access, bounded logistics manipulation, or production scale?
If private China OEMs move from S4 to S5, does value accrue more to OEM brands, data/model layers, deployment integrators, or filing-backed suppliers?
What is the right public dashboard metric: accepted units, productive hours, intervention rate, repeat orders, revenue/margin, or service burden?

Public-site draft module

China OEMs: fast productization, different evidence debts

The China OEM cluster should be read as four evidence curves, not a winner list. AgiBot shows the broadest product/data-platform ambition; Fourier has the clearest rehab/institutional base; Galbot has productized wheeled manipulation specs; EngineAI has the most explicit production-intent headline around T800. These are all useful S3/S4 signals for a company map.

The common S5 gap is still the same: accepted units, customer ROI, uptime, intervention rate, contract economics, repeat orders, revenue, gross margin, and service burden. Public readers should track these companies because the evidence has moved beyond simple demos, but should not treat product cadence or production-intent language as proof that humanoid economics are solved.

Footer caveat: Evidence map only. No winner ranking. No trade recommendation. Productization ≠ deployment economics; production intent ≠ accepted deliveries; institutional base ≠ humanoid ROI.

Source list

AgiBot official English / Chinese site, company profile, product pages, DaaS page, and news pages, accessed 2026-06-12 via robotics-agibot-evidence-v1.md. 🟢 Primary for claims; 🟠 for commercial implication.
Fourier overview, GRx, GR-1, GR-2, GR-3, FDH-6, Rehab, and business pages, accessed 2026-06-12 via robotics-fourier-evidence-v1.md. 🟢 Primary for claims; 🟠 for humanoid-transfer implication.
Galbot official website and official SPA front-end bundle, accessed 2026-06-12 via robotics-galbot-evidence-v1.md. 🟢 Primary for official site text; 🟠 for extraction and commercial implication.
EngineAI official homepage, about, development-history, news, PM01, SE01, T800, SA01, JS01 and S2 pages, accessed 2026-06-12 via robotics-engineai-evidence-v1.md. 🟢 Primary for claims; 🟠 for production/commercial implication.

Public-safety exclusions

Do not include Hugo private portfolio data, position weights, trade rationale, private channel checks, paid-report excerpts, tax/legal context, or rumors.
Do not frame AgiBot, Fourier, Galbot, EngineAI, Unitree, UBTECH, suppliers, or any related public/private company as buy / sell / hold.
Do not imply any of the four has proven S5 scaled-commercial economics.
Do not present company promotional claims as independently verified customer economics.