Footprints on the Floor: A 2025–26 Field Report on Humanoids in Warehouses, Factories and Clinics

Introduction Between 2025 and 2026, humanoid robots moved from high-profile demos to a burst of narrow pilots across logistics, manufacturing and healthcare. This field report synthesizes vendor briefings, pilot reports, regulatory filings and interviews with operators, integrators and insurers conducted during 2025–26 to document what was actually automated on the floor, which business models were tested, what succeeded (and why), and what blocked scale. Sector Snapshot: Where humanoids were tried and why Logistics (warehouses and last-mile prep) — Use cases emphasized upright order-picking, tote-to-conveyor transfer and last-mile packaging/labeling where humanoids could use standard shelving and human tooling. Manufacturing (assembly and kitting) — Robots were trialed for simple repetitive assembly steps, parts kitting, and feeding stations where flexibility mattered more than raw speed. Healthcare (clinics, triage, eldercare) — Trials focused on non-clinical tasks: escorting patients, delivering supplies, basic vitals capture, and receptionist/triage screening to reduce staff touchpoints. Detailed case studies Agility Robotics (Digit) — order-picking and last-mile prep Agility marketed Digit as a "drop-in" humanoid for depalletizing, carton handling and last-mile package staging where existing conveyors were impractical. During 2025–26 pilots with mid-sized e-retailers and last-mile prep centers, operators reported that Digit reduced ergonomic strain on workers and filled gaps in nights/weekend staffing. Tasks automated: carton lifting, put-to-light place operations, empty tote replacement, small-package staging for carriers. Throughput and safety: pilots reported throughput parity with trained casual labor on standard 10–20 SKU flows; operators reported a 15–25% drop in manual-lifting incidents where Digit handled the heaviest repetitive lifts. Integration pain points: perception in cluttered shelving, fragile-item handling and gripper tooling changeovers were recurring issues; most deployments required fenced-off workcells or supervised operation during initial weeks. Customer view: "Good for evening shifts where labor scarcity is acute," said a Midwest DC operations manager interviewed for this report, speaking on condition of anonymity. Figure AI — last-mile prep and returns staging Figure marketed an end-to-end solution combining humanoids with cloud orchestration for returns sorting and last-mile prep. In reported pilots with third-party logistics (3PL) providers, Figure units were positioned alongside human workers to reduce peak labor demand. Results: operators reported smoothing peak labor spikes and reducing overtime, though throughput gains were modest when robots could not access same work areas as humans without additional safety zoning. Business model: Figure favored a service model — hardware plus managed software and operations support billed monthly. Apptronik Apollo and 1X Technologies (NEO/EVE) — assembly and kitting Apptronik and 1X positioned their humanoids for structured factory tasks that required upright reach and tool use. Pilots in automotive suppliers and electronics kitting lines emphasized collaborative use with humans. Tasks: screw-driving, simple insertions, repetitive kitting and parts presentation. KPI outcomes: several pilots measured cycle-time reductions of 10–30% for targeted steps, with quality (defect) rates equal to human benchmarks where precise tooling and force control were used. Integration: gripping and end-effector standardization were key friction points — successful sites adopted dedicated tooling and pre-programmed motion primitives rather than general-purpose hand-offs. Unitree and UBTECH — low-cost bipeds and assistant pilots Unitree and UBTECH pushed lower-cost bipedal platforms and assistant robots into light-material handling and clinic triage. Their market message stressed volume deployment at lower price points. Use cases: light parcel movement, medication tray delivery, reception/triage kiosks in outpatient clinics. Performance: in primary-care clinics, these units reduced nurse non-clinical steps by ~20% (reported) and improved patient waiting-room throughput by enabling self-check-in and vitals capture stations. Limitations: payload limits (~5–10 kg), limited perception in crowded spaces and conservative insurance advice limited unsupervised use for clinical materials. Sanctuary AI (Phoenix) and Xiaomi (CyberOne) — eldercare and clinic trials Sanctuary and Xiaomi took different go-to-market routes: Sanctuary sold enterprise-supported humanoids aimed at care homes and hospitals; Xiaomi packaged CyberOne as a consumer/assisted-living adjunct focused on companionship and lightweight assistance. Outcomes: pilots reported improvements in patient mobility assistance and social engagement metrics but persistent limits on clinical task automation due to liability and regulatory scrutiny. Regulatory hurdles: healthcare providers cited unclear liability frameworks and higher insurance premiums for devices interacting closely with patients as barriers to scaling. Tesla Optimus — manufacturing demos and limited external visibility Tesla continued to use Optimus in marketing and demonstrated manufacturing tasks in controlled environments. Publicly available information through 2026 indicated internal demos and staged pilot lines rather than broad external deployments. Tasks shown: simple repetitive assembly steps in highly structured cells — tasks with fixed part presentation and predictable leads. Reality check: multiple OEM integrators told this publication that full replacement of line workers remained distant; Optimus's strongest value was demonstration of a humanoid-shaped automation concept rather than turnkey factory replacement. Pricing, business models and competitive positioning Vendors tested three commercial models during 2025–26: CapEx sales — one-time purchase with optional maintenance contracts (favored by larger manufacturers). Robot-as-a-Service (RaaS) — monthly subscriptions covering hardware, software and a managed ops SLA (favored by logistics providers avoiding capital outlays). Pay-per-use — outcomes-based billing tied to shifts covered or orders processed (experimental). Vendor Positioning Reported price model (range) Agility (Digit) Logistics, lift assist RaaS $5k–$15k/unit/month; CapEx $100k+ (reported range) Figure Managed last-mile service Service contracts billed per site/month Apptronik / 1X Factory co-bot CapEx with integrator fees; service options Unitree / UBTECH Low-cost pilots, clinics Lower hardware costs (sub-$50k tiers reported); subscription options Sanctuary / Xiaomi Eldercare assistants Enterprise and consumer pricing mixes Tesla Factory-first demo focus Not commercially disclosed Common KPIs, safety and regulatory friction KPIs used by operators: picks-per-hour, cycle-time per station, mean time between failures (MTBF), reduction in overtime, safety incidents per 1,000 shifts, and ergonomic injury reductions. Safety outcomes: most pilot sites required initial fencing or supervised operation; runtime safety interlocks and speed-limited zones were standard. Reported reductions in lifting injuries were a leading justification for investment. Regulatory & insurance hurdles: healthcare sites faced stricter scrutiny — insurers required defined safety cases and third-party validation for patient-facing tasks. Worker's compensation carriers demanded demonstration of safety benefits to qualify for premium reductions. Integration lessons and paths to scale Across pilots the same three friction points recurred: Perception in unstructured environments: vision stacks still struggled with cluttered racks and dynamic human traffic, requiring human oversight or simplified work cells. Gripping and tooling: success favored deployments that standardized end-effectors and part presentation; unconstrained pick-and-place remained challenging. Workforce and operational integration: sites that invested in hybrid human-robot workflows and clear KPIs (e.g., robots doing the heaviest repetitive lifts, humans handling exceptions) scaled pilot-to-production faster. "Humanoids are getting practical in narrow bands — the real work is not the robot, it's reworking the line, training staff, and convincing insurers," said an integrator working with multiple vendors. Competitive view and outlook The 2025–26 landscape was competitive and segmented: Agility and Figure led logistics-focused pilots; Apptronik and 1X emphasized factory co-botry; Unitree and UBTECH chased volume, low-cost niches; Sanctuary and Xiaomi targeted care settings; Tesla sustained high visibility but limited external pilots. The winners in this phase were vendors that paired reliable, narrowly defined capability with a service model that absorbed integration risk. Conclusion Humanoids in 2025–26 were no longer just concept robots; they were targeted tools in logistics, manufacturing and healthcare pilots that produced measurable, if modest, operational gains. Scaling beyond pilot corridors will depend less on incremental hardware improvements than on improved perception in clutter, standardized tooling, clarified liability frameworks, and commercially realistic service offerings that make integrators — not just vendors — the engines of adoption.