Humanoid robots at work: 5 real deployments happening right now

The word “deployment” gets used loosely in the robotics industry. A robot operating in a controlled research environment, or performing a staged demonstration at a trade show, is often described as “deployed.” This article uses a stricter definition: a humanoid robot is deployed when it is performing a repeatable, revenue-relevant task in a live operational environment, with human workers present, on a schedule that did not exist for the cameras.

By that definition, 2026 is the first year that humanoid deployment is genuinely newsworthy. Counterpoint Research estimates 16,000 humanoid units were deployed globally in 2025, with projections for cumulative installations exceeding 100,000 by 2027. The deployments below are the ones with the clearest, best-documented evidence — specific robots, specific tasks, specific facilities, specific results.

None of them is a solved problem. All of them are real.

The Five Deployments at a Glance

Sources: BMW Group, Tesla SEC filings, Agility Robotics, AGIBOT / Interesting Engineering, Accenture. All data as of May 2026.

DEPLOYMENT 01  ·  BMW + FIGURE AI — SPARTANBURG, SOUTH CAROLINA

The First 30,000 Cars

The most carefully documented humanoid deployment in automotive history began in a BMW body shop in Spartanburg, South Carolina, in 2025. The robot was Figure 02, developed by Figure AI, working a single dedicated station: the precise removal and positioning of sheet metal parts for the welding process. The task was chosen deliberately — high physical demand, consistent geometry, high repeatability.

The results, disclosed by BMW in February 2026 when the company announced its European expansion, were striking:

• 90,000+  Sheet metal components handled  — over 10 months
• 1,250  Operating hours logged  — 10-hour shifts, Mon–Fri
• 1.2 million  Steps covered  — across the factory floor
• 30,000+  BMW X3s produced during pilot  — supported by Figure 02

BMW’s own characterisation of the task was telling: “particularly demanding in terms of speed and accuracy while also being physically exhausting.” That framing — ergonomically difficult for humans, repetitive enough for a robot — is exactly the profile where first-generation humanoids have the strongest commercial case.

The Spartanburg pilot led directly to BMW establishing a “Center of Competence for Physical AI in Production” and expanding the programme to BMW Plant Leipzig in Germany — the first humanoid deployment in a European automotive facility — where the AEON robot from Hexagon Robotics began test operations in December 2025, with a full pilot planned for summer 2026.

“Physical AI can deliver measurable added value under real-world conditions.” — BMW Group press statement, February 2026

DEPLOYMENT 02  ·  TESLA OPTIMUS — GIGAFACTORY TEXAS

1,000 Robots. One Factory.

In January 2026, Tesla confirmed that over 1,000 units of its Optimus Gen 3 humanoid robot were actively deployed across its global manufacturing footprint — primarily at Gigafactory Texas, with additional units at Fremont. The deployment represents the largest single humanoid robot rollout in manufacturing history.

The current task portfolio is deliberately constrained: sorting 4680 battery cells and handling logistics kits. These are not the most complex tasks in Tesla’s factory. They were chosen because they represent the category of work — repetitive, physically consistent, high-volume — where the cost-benefit case for humanoid deployment is clearest while the technology matures.

The training methodology is worth understanding. Tesla is collecting training data by recording video of human workers performing tasks at its Fremont factory and Austin Gigafactory. According to Tesla’s SEC filing, a dedicated Optimus manufacturing factory at Fremont (converting the Model S/X lines) is being prepared, with a first-generation line designed for 1 million robots per year and a second-generation line at Giga Texas targeting 10 million units annually. Full Gen 3 (V3) production is expected to begin in late July/August 2026.

The cost trajectory matters as much as the deployment numbers. Manufacturing costs have declined 40% between 2023 and 2024 according to Goldman Sachs data cited in Deloitte’s 2026 Tech Trends report. Bank of America projects unit costs below $17,000 by 2030. At that price point, the ROI calculus for humanoid deployment changes substantially.

“The era of the humanoid laborer has transitioned from a Silicon Valley fever dream into a hard-coded reality on the factory floor.” — Financial Content, January 2026

DEPLOYMENT 03  ·  AGILITY ROBOTICS DIGIT — GXO LOGISTICS / SPANX, GEORGIA

The First Commercial Contract

Agility Robotics’ Digit is the humanoid robot with the most clearly commercial deployment story in the US market. The path from pilot to revenue ran through a GXO Logistics-operated warehouse in Georgia managing fulfilment for Spanx, where Digit was deployed to handle a single, physically demanding task: transferring empty totes from autonomous mobile robots to conveyor belts.

The task selection was deliberately unglamorous. Tote transfer is repetitive, ergonomically hard on human workers (constant bending, reaching, lifting), and sufficiently consistent in geometry that a humanoid robot can perform it reliably without needing to handle novel objects. This is exactly the profile that early-generation humanoid deployments need — and the GXO/Spanx pilot demonstrated it works in a live commercial setting.

• $10–12/hr  Operational cost of Digit  — vs $30/hr for human labour in equivalent role
• 98%  Task success rate at Amazon Sumner facility  — after 18 months of testing
• 35 lbs (16 kg)  Payload capacity  — Digit v4, commercially deployed
• 10,000+ units/year  Factory production capacity  — Agility’s Oregon facility

Amazon’s involvement is strategically significant. Amazon’s Industrial Innovation Fund participated in Agility’s $150 million Series B funding round, and Amazon has tested Digit at its Sumner, Washington facility — achieving a 98% task success rate after 18 months of testing, at an operational cost of $10–12/hour compared to $30/hour for human workers performing equivalent tasks. Specific details of Amazon’s ongoing deployment remain under NDA, but the investment and testing history signals strong institutional confidence.

The multi-year commercial contract with GXO — the world’s largest pure-play contract logistics provider — marks the transition that matters most: from pilot to revenue. Agility’s Digit review notes the robot currently operates at a 2:1 uptime ratio (two Digits working while one charges), which is the honest operational reality of current battery constraints — and the type of detail that does not appear in press releases but matters significantly in deployment planning.

DEPLOYMENT 04  ·  AGIBOT G2 — LONGCHEER TECHNOLOGY, SHANGHAI

The Electronics Factory Benchmark

China’s most significant humanoid deployment milestone of 2026 is not about automotive. AgiBot’s G2 robots are deployed at Longcheer Technology’s tablet manufacturing facility in Shanghai — integrated directly into multimedia integrated testing (MMIT) stations on active consumer electronics production lines. This is arguably the most technically demanding deployment environment on this list.

Electronics manufacturing requires sub-millimetre precision, consistent force control, high throughput, and the ability to handle components that vary in size and weight within the same production run. The G2 performs precision loading and unloading — picking up tablets, placing them into testing fixtures with millimetre-level accuracy, and sorting finished or defective units.

• Up to 310 units/hr  Throughput  — reported by AgiBot
• 19–20 seconds/task  Cycle time  — per device at MMIT station
• 99.9%  Task success rate  — on live production line
• 36 hours  Integration time  — from delivery to production-ready
• 140 hours  Continuous operation accumulated  — as of April 2026 deployment report

The 36-hour integration time is the figure most worth examining. It suggests that AgiBot has made meaningful progress on deployment friction — one of the most underappreciated barriers to humanoid adoption in manufacturing. Traditional industrial robot deployments require weeks or months of integration work; 36 hours for a live production line is a materially different proposition.

AgiBot’s scale context matters here. The company shipped 5,168 humanoid robots in 2025 — more than Tesla, Figure, and Agility combined, according to Omdia — and rolled out its 10,000th unit on March 30, 2026. The Longcheer deployment is part of a planned expansion to 100 units at that facility by Q3 2026, with further rollouts into automotive, semiconductor, and energy sectors. For Western buyers, AgiBot’s deployments are Chinese-first; the company’s international footprint in Europe, Japan, and North America is growing but its production track record is currently anchored in China.

“Embodied AI is no longer experimental. It is a practical, production-ready capability that can operate reliably in real industrial environments and deliver measurable economic value.” — Maoqing Yao, SVP AgiBot

DEPLOYMENT 05  ·  ACCENTURE + SAP + VODAFONE — DUISBURG, GERMANY

The Enterprise Integration Test

The most recently announced deployment on this list is also the most instructive for enterprise buyers. In April 2026, Accenture, SAP, and Vodafone Procure & Connect presented a joint pilot at Hannover Messe 2026 — the world’s largest industrial technology exhibition — showing a humanoid robot operating at Vodafone’s logistics warehouse in Duisburg, Germany.

The robot is powered by Accenture’s “Robot Brain” solution and trained in a digital twin of the warehouse environment, built on Accenture’s Physical AI Orchestrator using NVIDIA Omniverse libraries. What makes this deployment distinctive is not the hardware — it is the enterprise system integration.

The robot does not simply move boxes. It identifies operational inefficiencies, detects safety risks, and reports findings directly into SAP’s warehouse management system in real time. Specific scenarios tested in the pilot:

• Detection of misplaced or damaged products, with automatic SAP logging
• Assessment of pallet stacking and weight distribution against compliance standards
• Identification of unused storage space and optimisation recommendations
• Hazard detection — obstacles in aisles, misaligned pallets — reported to operational dashboards

The SAP integration layer — powered by Joule, SAP’s AI execution interface for embodied AI — is what elevates this from a robotics pilot to a business process automation deployment. As Dr. Lukasz Ostrowski, head of Embodied AI & Robotics at SAP explained: “By grounding actions in trusted SAP data, we can automate health and safety incident reporting and real-time inventory validation to protect workers and strengthen compliance through consistent, auditable workflows.”

For enterprise buyers evaluating humanoid deployment in 2026, this pilot answers the question that BMW, Tesla, and AgiBot do not: what does the robot connect to? Operational insight that lives only on the robot is interesting. Operational insight that flows into the ERP system that runs the business is actionable.

What These Five Deployments Tell You

Reading across all five cases, four consistent patterns emerge.

Task selection is the most important deployment decision. Every deployment on this list was anchored to a task that is ergonomically difficult for humans, geometrically consistent, and high-volume. Tote transfer. Sheet metal positioning. Battery cell sorting. Tablet loading. These are not the hardest tasks in their respective facilities — they are the ones with the best profile for first-generation humanoid reliability. The deployments that are struggling (not documented here but well-documented in the industry) are the ones where the task selection was too ambitious for the current technology generation.

Integration friction is the hidden deployment cost. AgiBot’s 36-hour integration stands out precisely because most humanoid deployments require significantly longer. The RoboticsTomorrow analysis of humanoid deployment notes that current humanoid robots resemble AMRs and AGVs 15 to 20 years ago — capable of performing the task, but dependent on immature supply chains, hands-on vendor support, and components not yet optimised for the deployment environment.

Enterprise system integration will determine scale. The Accenture/SAP/Vodafone deployment illustrates a principle that applies across all five cases: robots that report into business systems are fundamentally more valuable than robots that operate in isolation. ISG’s 2026 Intelligent Robotics report makes this explicit — the competitive advantage is in orchestration, not hardware.

The West and China are running different experiments at different speeds. AgiBot’s 10,000-unit production milestone and BYD’s planned 20,000-unit fleet stand in sharp contrast to the carefully staged, single-facility pilots in the US and Europe. China accounts for over 80% of global humanoid deployments by volume. The deployment pace is not comparable. What is comparable is the direction: in automotive, logistics, and electronics manufacturing, humanoid robots are performing real work on both sides of that divide.

The Bottom Line

The deployments on this list are early, constrained, and frequently dependent on vendor engineering support. They are also real. The gap between “robots performing staged demonstrations” and “robots clocking ten-hour shifts in body shops” has closed — quietly, without a single product launch event. What is being tested now, in these five environments, is not whether humanoid robots can perform industrial tasks. That question has been answered. What is being tested is whether they can perform them reliably enough, cheaply enough, and integrably enough to justify deployment at scale.

The next 18 months will answer that question.

Key Sources

• BMW Group — First Humanoid Robot Introduced in Plant Leipzig (Feb 2026)
• BMW Group Press — Humanoid Robots in Production, Germany (Feb 2026)
• Tesla SEC Filing (8-K) — Optimus Factory and Production Plans (2026)
• Agility Robotics — Broadens Relationship with Amazon (Official)
• Scaling Deep — How Agility Robotics Convinced Amazon to Deploy Digit
• Accenture Newsroom — Humanoid Robotics Pilot, Hannover Messe 2026 (Apr 2026)
• Interesting Engineering — AgiBot G2 Live Production Line (Apr 2026)
• Gizmochina — AgiBot 10,000 Humanoid Robots Milestone (Apr 2026)
• RoboticsTomorrow — What It Takes to Deploy Humanoid Robots (Apr 2026)
• ISG — Advances in Physical AI Reshape Robotics (May 2026)
• KuCoin / Counterpoint Research — Humanoid Robots 2026 Market Data
• IIoT World — Physical AI Deployment ROI: BMW’s 30,000-Car Proof