2026-06-18 👁️ — views

Humanoid Robot Deployment Tracker — Where Robots Are Actually Working Today

Unit counts, tasks, and productivity data on every humanoid robot doing real commercial work in mid-2026 — versus the ones still in press-release mode.

Article 31 in the Physical AI Benchmark Series — The Deployment Reality Check

The humanoid robotics industry operates on two timescales simultaneously: the press-release timescale, measured in dramatic announcements and polished demo videos, and the deployment timescale, measured in units on factory floors doing actual work. This article tracks the second timescale only. The question is not what companies have announced or promised for 2028. The question is: as of mid-2026, which humanoid robots are doing real work, where, how many, and how productively?

This is the humanoid equivalent of Waymo’s ride count — a ground-level audit of actual deployment versus stated ambition. All unit counts are estimates based on company announcements, third-party reporting, and analyst research. Neither the robots’ manufacturers nor their customers publish comprehensive deployment statistics. Where figures are estimated, they are labeled as such.

Section 1 — Active Deployment Table

The table below covers every humanoid robot platform with credible evidence of commercial or pre-commercial factory deployment as of mid-2026. “Credible evidence” means a named customer, third-party reporting, or disclosed unit count — not a demo video or a press release alone.

Reading the table: Only three platforms have cleared the bar from research demo to genuine commercial-adjacent deployment: Tesla Optimus (internal factory use), Figure 02 (external Tier-1 OEM customer), and Agility Digit (external Fortune-10 customer). The rest are research pilots or pre-commercial. Unitree occupies a distinct category — volume B2B sales in China with credible unit counts, but productivity data is limited and the customer base is less publicly verifiable.

Section 2 — The Announcement vs. Deployment Gap

The humanoid sector has a large and systematic gap between press-release activity and actual deployment. This gap is not unique to robotics — it characterizes every capital-intensive emerging technology — but it is wider here than in autonomous vehicles, because humanoid robots require solving both hardware reliability and AI task generalization simultaneously.

Announcement-only red flags — signals that a “deployment” is not yet real:

• No third-party customer named — or the only deployment is the manufacturer’s own factory
• Video demos only, with no evidence of continuous multi-shift operation
• No productivity or uptime metrics disclosed, even as estimates
• Vague timelines — “commercially available in 2025” with no update or delivery confirmation
• Absence of expansion announcements — a pilot that doesn’t expand after 12 months is failing

Real deployment signals — signals that a deployment is substantive:

• A named Tier-1 customer (BMW, Amazon, Hyundai) whose reputation creates accountability
• A disclosed unit count, even as an estimate range
• Third-party reporting from journalists or analysts at the deployment site
• Repeat orders or a publicly announced Phase 2 expansion
• Uptime or productivity metrics, even partial ones

Applying this filter to mid-2026: Tesla Optimus (internal), Figure 02 at BMW, and Agility Digit at Amazon are the only platforms that pass a strict deployment test. Atlas at Hyundai is a research pilot with no production intent disclosed. Unitree passes on volume but not on independent customer verification. 1X and Sanctuary are genuine early-stage pilots, not deployments.

The gap matters because investment theses, supply chain decisions, and government policy are increasingly shaped by the press-release timescale. This article uses the deployment timescale instead.

Section 3 — Tesla Optimus: The Ramp That Matters Most

Optimus is the only humanoid platform with a credible path to six-figure annual unit production by 2027–2028. No other manufacturer has disclosed production targets at Tesla’s stated scale, and no other manufacturer has Tesla’s vertical integration across AI training, hardware manufacturing, and factory deployment.

Tesla Optimus mid-2026 status (all figures estimated):

The internal deployment is strategically rational: Giga Texas provides a controlled, high-value environment where Optimus failures cost Tesla but not an external customer. The robots collect training data while doing real work. Each task cycle generates the labeled manipulation data that trains the next model version. This is Tesla’s data flywheel strategy applied to embodied AI — the same logic that drove FSD’s improvement from supervised miles.

The key challenge is not production volume. Tesla has demonstrated it can manufacture complex hardware at scale. The challenge is task generalization: going from 50–500 units doing 3–5 tasks in a controlled factory to 50,000 units doing 20+ tasks across diverse environments is a qualitative leap, not a linear scaling problem. Optimus needs to demonstrate that its AI can generalize to new tasks without requiring months of training per task — and that generalization has not yet been publicly demonstrated at scale.

If Optimus achieves external commercial sales in H2 2026 or early 2027 at a price point below $200,000, it would be the single most significant commercial milestone in humanoid robotics history.

Section 4 — Why Figure + BMW Is the Most Significant External Deployment

Among all external (non-internal) deployments, Figure 02’s pilot at BMW’s Spartanburg, South Carolina plant is strategically the most important — not because of its current scale, but because of what it proves if it succeeds.

Why the BMW deployment is the key proof point:

1. Named Tier-1 OEM customer — BMW’s reputational stake in production quality makes this a genuine accountability signal. BMW would not allow a dysfunctional robot in a live body shop.

2. Automotive body shop conditions — one of the harshest industrial environments for any robot: sparks, heavy metal parts, high precision tolerances, safety-critical welds. If Figure 02 works here, it works almost anywhere in manufacturing.

3. Continuous operation footage — Figure has released video evidence of the robot in sustained operation, not a scripted demo. Third-party observers have reported on the deployment.

4. OpenAI VLA policy stack — Figure’s AI runs on Vision-Language-Action models built with OpenAI. The BMW deployment is the largest real-world test of VLA-based manipulation in an adversarial industrial environment. Its results will shape how the broader AI industry views VLA for physical tasks.

5. Expansion signal — Figure has stated intent to scale beyond the initial pilot, moving from the body shop toward full assembly line integration. A Phase 2 announcement with unit counts above 100 would confirm this is a production program, not a showcase.

The benchmark threshold: If Figure 02 achieves 80%+ of human worker productivity at BMW sustained over 6 months of continuous operation, it becomes the strongest proof point for commercial humanoid robotics outside Tesla. At that level, the economics become compelling for automotive OEMs facing labor cost pressure — and Figure’s order pipeline would likely reflect it.

Section 5 — Deployment Milestones to Watch in H2 2026

The next six months will resolve several of the key open questions in humanoid robotics deployment. The milestones below are the ones with the highest informational value — each one either validates or invalidates a major hypothesis about the commercial readiness of the sector.

What each milestone tells us:

• Optimus commercial sale: The existence of a paying external customer at any price point validates the demand side of the market. The price point and task scope tell us how far task generalization has actually progressed.

• Figure BMW Phase 2: If Figure scales to 100+ units in the body shop, it validates VLA-based manipulation at production scale. If the pilot stalls or shrinks, it reveals a generalization ceiling.

• Digit Amazon scale-up: Amazon’s procurement decisions are driven by unit economics, not narrative. If Digit order volumes increase, the cost per unit of work is competitive with labor. If volumes plateau, the economics are not yet there.

• Unitree Western launch: Unitree’s Chinese deployment numbers suggest the company has solved hardware reliability at B2B scale. The question is whether its AI stack is competitive with Figure’s and Tesla’s VLA approaches — or whether cheaper hardware with narrower AI is a viable market segment.

Section 6 — About This Series

This is article 31 in the Physical AI Benchmark Series. This series has covered the ramp index, the humanoid race, unit economics, global competition, HD mapping, fleet operations, software and OTA, insurance and liability, consumer demand, partnerships, competitive moats, Cybercab versus Model Y, safety data, Waymo Gen 6, Optimus manufacturing, scorecard snapshots, the 2030 forecast scenarios, the investor framework, Waymo’s city expansion pipeline, Tesla’s state approval map, AV weather and climate constraints, the talent war, the forward-looking regulatory calendar, robotaxi fare pricing analysis, and the AV data flywheel comparison.

This article adds the humanoid deployment tracker: a ground-level audit of which robots are actually working, where, and how productively — as distinct from what has been announced, demonstrated, or promised. The deployment gap between announcement and factory floor is the most important metric in humanoid robotics right now. As of mid-2026, only three platforms have meaningfully crossed it. The next six months will determine whether that number grows to five or ten — or stays at three.

Sources

• Tesla Optimus production updates — Tesla AI ↗
• Figure AI BMW deployment — Figure AI blog ↗
• Agility Robotics Amazon partnership — Agility Robotics ↗
• Boston Dynamics Atlas commercial — Boston Dynamics ↗

Tags