Hands-On Humanoids: Who’s Actually Stacking Shelves, Welding, and Doing Ward Rounds in 2025–26?

Methodology Between January and April 2026 The Humanoid Race conducted a ground-level survey of commercially focused humanoid pilots and early deployments. Research methods included structured interviews with more than a dozen integrators and three anonymized pilot customers (warehousing, discrete manufacturing, and a regional eldercare chain), review of public demonstrations and procurement notices, and follow-up conversations with corporate press offices. Where possible the report reconciles vendor claims with on-site KPI snapshots supplied by pilot partners. Where humanoids are being tried — and why Across warehouses, manufacturing floors, logistics hubs and healthcare settings the pattern is consistent: companies try humanoids where a human shape and human-level manipulators reduce process rework or enable existing infrastructure to be reused. Common use cases include: Goods picking and putaway (warehousing) — robots that pick small parcels or tote items to conveyors where existing sortation and packing systems remain intact. Palletizing and case handling — repetitive, two-handed stacking on standardized pallets in secondary processes. Repetitive assembly and parts kitting — delivering parts to human stations or performing repeatable screw-and-fit tasks in light manufacturing. PPE‑safe patient checks and telepresence — short nurse-assist inspections, vitals capture and sample fetching in infectious wards. Mobility assistance in eldercare — guided transfers, fetch-and-carry, and companionship features in long‑term care. In‑store stocking and last‑mile support — humanoids that restock shelves or assist customers in remote or high‑value retail environments. Who is targeting which problem Agility Robotics (Digit) — positioned principally as a logistics/warehouse asset. Agility’s public pilots emphasize box-to-conveyor pick-and-place and human-assist tote movements, with integration partners focusing on mid-shift relief and ergonomics improvements. Apptronik (Apollo) — marketed to OEMs and contract manufacturers as a high‑torque, compliant platform for kitting and finishing tasks that sit alongside human line workers. Figure AI — framing humanoids for retail storefronts and small-format grocers where a humanoid form factor can interact with customers and handle products at human heights. 1X Technologies (NEO/EVE) — targeting heavy industrial handling and palletizing with ruggedized manipulators for case stacking and line-side lifting. Unitree — pushing lower‑cost mobile humanoid concepts aimed at economies of scale in APAC retail and education demos. Tesla (Optimus) — high‑profile demonstrations and company claims that future Optimus units will be used inside Tesla factories; so far public evidence is primarily demo and limited internal concept trials. Sanctuary AI (Phoenix) — concentrating on healthcare pilot projects and “skill transfer” workflows that let clinicians teach robots routines via teleoperation and supervised learning. Xiaomi (CyberOne) and UBTECH — focused on consumer and eldercare/education markets in China where demo-to-deployment cycles are faster and regulatory burden lower. Pilots, measurable outcomes and conversion to paid deployments Across pilots we tracked common KPI buckets: throughput (items/hour), uptime (available hours per shift), error rate (mispicks, mis-stacks), and operator acceptance (surveyed comfort and intent-to-use). Aggregate observations: Throughput: Targeted logistics tasks in constrained layouts produced modest but tangible gains — typical pilot reports place humanoid pick throughput in the low tens of picks/hour in early runs, with integrated cells (fixed conveyors and simplified item sets) enabling higher sustained rates. Apptronik and 1X reported higher effective throughput when robots were assigned fixed, repetitive subtasks (pallet ends, kitting returns). Uptime and reliability: Pilot uptimes varied from 65–92% depending on vision robustness and part variability. Downtime drivers were environmental clutter and the need for teleoperation when unexpected items appeared. Error rates: In curated SKU sets, error rates dropped into the 1–5% range; in open-pick situations error rates rose sharply. Integrators emphasize “task shaping” as essential — constrain work to reduce vision and grasp failure modes. Operator acceptance: In warehouse and manufacturing pilots operator surveys typically reported 60–85% acceptance when robots removed ergonomically risky tasks; acceptance dipped when robots required substantial human supervision or when jobs were reallocated. Which pilots converted to paid deployments? Two themes emerged. First, deployments that replace discrete, high-frequency two‑person tasks (pallet ends, tote transfers, repetitive assembly steps) converted fastest because ROI is straightforward. UBTECH’s education/eldercare deployments in China and several Apptronik contracts for kitting integrations moved to paid phases. Second, pilots that attempted broad generalist picking without significant operational change — often demonstrated by high-profile demos — struggled to show near-term ROI and remained trials (notably many Tesla‑style factory ambitions and several consumer-facing demo programs). "Narrow the task, integrate with existing conveyors and workers, and price as a service," summarized an integrator involved in multiple pilots. "That combination separates paid pilots from demonstrations." Pricing and business models Pricing signals from integrators and vendor RFP responses coalesced around two models: Capital sale — one‑time unit cost plus integration (typical ranges discussed publicly and in interviews: roughly $50k–$250k per robot depending on payload and sensors). Robot-as-a-service (RaaS) — monthly fees per robot plus outcomes-based add‑ons. Monthly ranges piloted in deals ran roughly $3k–$12k per robot depending on support SLAs and software suites. Integrators emphasized that integration labor and software (fleet coordination, safety certification) often exceed the robot sticker price in total cost of ownership. Competitive snapshot Company / Platform Primary use case Maturity (pilot→paid) Price band Regional traction Agility (Digit) Warehouse picking, tote movement Pilots → selective paid rollouts $75k–$200k / RaaS options North America, Europe Apptronik (Apollo) Manufacturing, kitting Pilots converting in discrete manufacturing $100k–$250k / RaaS North America Figure AI Retail storefronts Early retail experiments Undisclosed / service models North America 1X Technologies Industrial handling, palletizing Pilots in heavy environments $100k–$300k Europe, APAC Unitree Lower-cost demos, APAC retail Demo/early pilots $20k–$80k China, APAC Tesla (Optimus) Factory automation (claimed) Demo / internal concepts Undisclosed North America Sanctuary (Phoenix) Healthcare checks, skill transfer Pilots in healthcare partners Service-first pricing Canada, North America Xiaomi, UBTECH Consumer, eldercare, education Commercial deployments (China) $10k–$60k for consumer/education models China, APAC Regional trends North America: fastest in warehouse and manufacturing pilots (Agility, Apptronik, Figure). Customers are integration-savvy but require clear ROI. Europe: strong in regulated industrial deployments and palletizing pilots; safety certification adds lead time but drives conservative but repeatable rollouts. China & APAC: fastest to convert demos into paid — UBTECH and Xiaomi have scaled consumer/eldercare rollouts; Unitree's lower-cost approach finds retail and education partners. Lessons from pilots that scaled — and those that didn't Successful conversions narrowed the robot's remit, integrated into existing material flows, and used RaaS to shift deployment risk. Pilots that attempted broad generalist vision tasks or promised immediate headcount replacement without process changes routinely stalled. Teleoperation and rapid skill transfer (Sanctuary, some Apptronik projects) accelerated usable uptime and operator trust. Price transparency and clear SLAs correlated strongly with procurement committees signing commercial contracts. Humanoid robots in 2025–26 are real but task-constrained. The most durable deployments are those that take advantage of the human form factor to slot into existing workflows — lifting at human heights, opening doors, working with human-facing conveyors — while vendors and integrators tighten software, pricing models and safety to deliver predictable ROI.