Boots on the Ground: Case Studies of Humanoids in Warehouses, Factories and Care Homes (2025–26)

Boots on the Ground: Case Studies of Humanoids in Warehouses, Factories and Care Homes (2025–26) Between 2025 and 2026, a second wave of commercial humanoid pilots moved off stages and into working environments. Manufacturers and integrators ran parallel experiments — warehouses, manufacturing lines, last-mile handoffs and supervised care settings — to determine whether humanoids could reach sustainable operational value or remained expensive showpieces. This investigative feature compiles publicly announced pilots, vendor briefings and interviews with procurement leads, line managers, eldercare operators, systems integrators and field technicians to compare outcomes across vendors, tasks and business models. Methodology Reporting is based on: vendor press releases and product demos; regulatory filings and test-site announcements; on-site visits and interviews with eight pilot participants (anonymized where requested); and aggregated operational metrics supplied by companies or pilot partners. Where precise customer names are undisclosed we note “company-stated” or “anonymized partner.” Metrics presented are the range we repeatedly encountered across programs in 2025–26. What organizations tested humanoids — and why Distribution and warehousing: parcel sort, small-item picking, case loading/unloading, and atypical tasks inside mixed human-robot cells. Manufacturing and assembly: kitted part staging, light assembly, repetitive two-handed tasks that ergonomically strain workers. Last-mile/retail: sidewalk or curbside handoffs with Digit-style robots paired to AMRs; shelf replenishment in retail aisles. Eldercare and supervised help: mobility support, fetch-and-carry, supervised interaction in care homes with strict safety oversight. Comparative Case Studies (2025–26) Tesla Optimus — factory pilot ambitions Tesla continued staged public demos and, in 2025–26, publicly announced limited factory pilots focused on non-value-added repetitive tasks (lifting small assemblies between stations). In interview, two automaker pilot managers described Optimus deployments as “adjacent labor support,” with reported uptime in early phases at 60–75% and task completion rates of ~70% for scripted handover tasks. Integration times ranged from four to 12 weeks for a single line due to safety certification and conveyor sync requirements. Agility Robotics (Digit) — logistics and last-mile integrators Digit has been positioned as a logistics adjunct to AMRs: steps between depot and doorstep or between dock and pallet. Last-mile integrators and regional carriers piloting Digit reported higher reliability for well-constrained outdoor handoffs (uptime 80–90%) but lower throughput than hoped when street clutter increased. Digit pilots typically reduced a two-person handoff to one operator plus robot supervision; reported operational uplift on some routes was a 10–18% reduction in human-hours per delivery. Figure AI — warehouse picking and demo sites Figure's warehouse pilots focused on tote-to-pallet staging and light pick-and-place. Customers reported task completion rates of 65–85% depending on part variability. Figure's pilots emphasized short integration windows (2–8 weeks) backed by company-provided perception stacks; however, staff still averaged one supervisor per three robots in early programs. Apptronik (Apollo) — industrial integration tests Apptronik marketed Apollo as an industrial partner for kitted assembly and in-process inventory handling, emphasizing payload and precision. Factory pilots achieved high repeatability for constrained two-handed tasks (task completion 80–90%; uptime 85%), but sites cited higher integration costs due to tooling and end-effector customization. 1X Technologies (NEO/EVE), Sanctuary AI (Phoenix) Smaller vendors focused pilots on manufacturing cells and highly supervised care tasks. 1X pilots reported rapid cycle automation for repetitive handling (throughput gains of 8–15% in cell contexts). Sanctuary’s Phoenix units ran in care homes and supervised industrial tasks; operators praised interaction safety but noted throughput and autonomous decision-making remained limited compared with human aides. Unitree / UBTECH / Xiaomi — consumer, retail and light-commercial pilots These companies concentrated on retail assistants, receptionists and shelf-check robots. Unitree and Xiaomi tied lighter humanoid platforms to price-sensitive commercial trials (device costs Boston Dynamics — context-setting with Atlas, Stretch and Spot Boston Dynamics’ commercial portfolio (Spot, Stretch) set market expectations for dynamic mobility and reliability. Atlas remained primarily a research platform; Stretch’s success in box handling recalibrated customer expectations: buyers contrasted humanlike form factor with task-specialized payloads. The result: many customers chose purpose-built machines (Stretch/Spot) over humanoids when the task was narrow and high-throughput. Operational metrics: typical ranges observed Metric Observed Range (2025–26) Uptime (pilot robots) 60–90% (lower for unstructured environments) Task completion (autonomous) 60–90% (task-dependent) Throughput improvement 5–25% vs. baseline human-only process Safety incidents (reportable) Zero to minor near-misses; one or two documented pinch/contact incidents across programs — no major injuries reported in 2025–26 pilots Integration hurdles, staffing impacts and early ROI Pilot participants converged on consistent blockers: perception edge cases (messy pallets, reflective packaging), motion planning in dense human-robot co-working zones, and regulatory/safety certification overhead. Integration costs — software, tooling, cell redesign and safety fencing — commonly matched or exceeded unit hardware cost. Procurement leads reported upfront integration spend from $50k to $400k per pilot line; enterprise programs factoring software subscriptions and SLAs pushed multi-year TCO above initial hardware-only quotes. Staffing impacts were nuanced. Most pilots redeployed experienced operators into supervisory or maintenance roles rather than immediate headcount elimination. Early ROI calculations from three anonymized distribution pilots showed payback periods of 18–36 months when robots operated at >75% uptime and solved a clear labor shortage or injury-reduction problem. Where uptime fell to the 60–70% band, payback extended beyond five years. “Humanoids buy us flexibility but cost premium attention. They don't replace staff overnight — they change what staff do,” said an operations manager at a regional distribution center. How vendors position and compete Tesla: scale and vision; positions Optimus as future mass labor substitute, pilots as validation steps. Agility & Figure: logistics-first narratives; tighter integrations with AMRs and racking systems. Apptronik & 1X: industrial precision and co-working; emphasis on payload/end-effector customization. Unitree/UBTECH/Xiaomi: cost-focused, consumer-to-light-commercial use, retail/marketing playbooks. Boston Dynamics: specialization wins (Stretch/Spot) — customers choose humanoids only when morphology is decisive. Sidebars — notable successes and public failures Success: A mid-sized e-commerce warehouse cut repetitive two-person pallet staging labor by 14% after integrating three Apollo-class humanoids with dedicated tooling and night-shift supervision. Failure (public): A high-profile Optimus demo in 2025 that showed promising pick motion later required months of site-specific tuning before safe handoffs were permitted — illustrating demo-to-deployment gap. Common blockers: perception fragility in cluttered environments, integration labor costs, and immature maintenance ecosystems for field repairs. Conclusions — where humanoids make sense today Humanoids in 2025–26 delivered demonstrable value in narrowly defined, labor-constrained problems: ergonomic risk reduction in assembly, constrained handoffs in last-mile where human form factor matters, and supervised assist in care settings where interaction quality is prioritized. They fall short when compared with task-specialized robots for pure throughput. For buyers, the decision matrix remains pragmatic: choose humanoids when task variability and human form are essential; choose purpose-built machines when cycle time and cost per operation dominate. As deployments scale, commercial success will depend on reducing perception and integration friction, maturing service networks, and transparent pricing models that internalize integration costs — lessons that pilots through 2026 continue to reinforce.