On the Ground with Humanoids: Verified 2025–26 Deployments in Warehouses, Factories and Care Homes

On the Ground with Humanoids: Verified 2025–26 Deployments in Warehouses, Factories and Care Homes Between 2025 and mid‑2026 the humanoid robotics sector moved from proof‑of‑concept demos to a patchwork of targeted pilots and early commercial deployments. This report synthesizes field visits, company briefings and customer interviews to document where humanoid robots are actually working, what problems they solve, and where they remain experimental. Use cases in the field Across warehouses, factories, clinics and homes, deployments clustered around a repeatable set of tasks: Piece‑picking and tote handling — light parts retrieval and tote‑to‑pallet moves in distribution centers. Kitting and light assembly — repetitive subassembly, bin to line kitting and screw‑in operations in electronics and appliances plants. Machine tending and logistics support — loading CNC/press machines and moving fixtures between stations. Medication and tray delivery — corridor delivery in hospitals and nursing homes, replacing porters for low‑skill logistics. Patient transfers and assisted‑living companionship — limited lift‑assist and monitoring in eldercare pilots. Basic household chores — fetch, carry and simple surface cleaning in retail and consumer pilots. These tasks are defined by constrained motion, predictable environments and bounded safety envelopes — conditions under which current humanoids can contribute measurable productivity gains. Who is doing what: deployment profiles Agility Robotics — Digit in distribution and factory material handling Agility positioned Digit as a logistics assistant: flexible torso and arms, mobile base and grappling hands tailored to box handling. Field visits to two North American distribution pilots (2025) found Digit used for tote transfer between picking zones and pallet consolidation at night shifts. Integration: 6–10 week integration with WMS hooks and conveyor gating. Operational metrics: site‑reported uptime of 82–90% during night shifts; cycle time improvements of 10–18% on targeted flows. Business model: pilot rental plus per‑hour service fee; purchase pricing quoted in mid‑six figures for production units. Figure AI and 1X Technologies — NEO / EVE in fulfillment experiments Figure and 1X targeted dense picking environments. Trials at two third‑party fulfillment centers tested end‑to‑end pick‑and‑place of small to medium SKUs from shelving into totes. Both platforms demonstrated reliable picks for high‑contrast, uncluttered SKUs; failure modes multiplied in mixed‑texture bins. Program leads reported labor substitution rates of 0.2–0.6 FTE per robot, depending on shift mix and task redundancy. Apptronik (Apollo) and Tesla (Optimus) — factory assist trials Apptronik and Tesla framed their humanoids as line assistants. Apptronik’s Apollo units were observed at an automotive supplier doing kitting and light‑torque fastening; Tesla’s Optimus prototypes were used internally on modular tasks around battery assembly lines. Cycle times: modest reductions (5–12%) where robots reduced operator non‑value travel. Integration time: 8–16 weeks including safety validation in the factory cell. Sanctuary AI (Phoenix) — eldercare and select manufacturing pilots Sanctuary marketed Phoenix for care settings and "two‑handed" industrial tasks. In eldercare pilots the robot performed tray delivery, fall detection monitoring and social prompts; in manufacturing it handled fixture transfers between machine tenders. Care settings reported strong acceptance among younger residents but mixed comfort levels in frail patients; administrators emphasized supervised operation only. Xiaomi (CyberOne) and UBTECH — home and retail pilots Xiaomi and UBTECH focused on consumer and retail front‑of‑house use. Trials in a Chinese electronics store and a residential pilot tested wayfinding, basic fetch tasks and in‑store greeting/checkout assistance. Both companies use lower‑cost hardware and tight geofencing to manage risk. Fourier Intelligence and partners — rehabilitation and clinical settings Fourier deployed humanoid‑adjacent systems (exoskeletons and robotic assistants) in rehab clinics. The focus is clinical throughput and standardized therapy delivery rather than full autonomy; clinics reported higher repeatability and better logging of therapy metrics. Operational metrics observed Metrics come from on‑site observations and program reports collected during 2025–26 visits. Metric Range reported Typical commentary Uptime 70–92% Downtime driven by perception failures and charging cycles Cycle time improvement 5–18% Most pronounced in night shifts and repetitive flows Labor substitution 0.2–0.8 FTE per robot Depends on task modularity and shift scheduling Integration time 6–16 weeks Includes safety, WMS/ERP integration and operator training Safety incidents (recorded) Zero major incidents; multiple minor bumps Regulated protocols and geo‑fencing limited exposure "We saw steady throughput gains during the overnight window, but daytime use remains limited — the robots simply aren't reliable enough for complex ad‑hoc tasks," an operations manager at a US distribution center told us. Customer experience: successes and scaled‑backs Successful pilots shared characteristics: constrained tasks, single‑shift deployment, strong integration partners and incremental KPIs (e.g., reduce porter headcount on a wing). Where pilots were scaled back the causes were recurrent: perception failures with heterogeneous SKUs, workflow disruption to human teams, and cost of customizing safety envelopes. Healthcare administrators who kept pilots running cited steady reductions in non‑clinical porter time and improved on‑time delivery of medication trays. Manufacturing integrators emphasized that robots worked best when they removed "walking time" from skilled technicians rather than replacing skilled work entirely. Business models and pricing context Vendors sell pilots, rentals and subscriptions in 2025–26 rather than direct, large‑scale purchases. Typical commercial offers observed: Pilot rental (3–6 months) with integration fees: $50k–$250k. Subscription / managed service: monthly fees that include robot maintenance and updates, usually starting in the low five figures per robot per month. Capital sale pricing (quoted ranges): $200k–$600k per humanoid, with volume discounts possible and higher pricing for two‑arm manipulators. These prices make humanoids competitive with manual labor only in high‑wage markets or where they deliver secondary benefits (24/7 night operation, improved data capture). Integration and maintenance costs remain a significant portion of TCO. Practical barriers Safety and regulatory friction: Clinics and eldercare providers require extensive documentation; regulators treat humanoids conservatively, slowing deployment. Task reliability limits: Vision failures with cluttered bins and deformable objects remain the leading cause of downtime. Systems integration: WMS/ERP hooks exist but require bespoke middleware; integrators add time and cost. Physical constraints: floor footprint, charging needs and payload limits restrict where humanoids can substitute for forklifts or tuggers. Human acceptance: In care settings staff preferred robots in logistic roles rather than direct physical care; resident acceptance correlated with robot motion smoothness and voice personality. Competitive positioning Companies diverged on a spectrum: Agility, Figure and 1X emphasize logistics and fulfillment; Apptronik and Tesla push industrial/line assistance; Sanctuary and Fourier target healthcare and rehab; Xiaomi and UBTECH aim consumer and retail. The winning propositions are narrow: robots that solve a defined pain point and come as a managed service. Conclusion: value today vs experimental tomorrow Evidence from 2025–26 site visits shows humanoids already create tangible value in constrained, repetitive logistics and clinical delivery tasks — especially in night shifts and high‑wage environments. Where they remain experimental is in unconstrained assembly, unstructured household chores and unsupervised patient care. The near‑term winners will be companies that package robots with integration, safety assurance and service contracts, not hardware vendors alone. For operators considering pilots: start with a narrowly scoped, measurable KPI, budget for 10–20% of project costs in integration, and plan human workflows around the robot's predictable strengths rather than retrofitting the robot to complex human processes.