Pilots in the Wild: Investigating Real-World Humanoid Deployments Across Warehouses, Factories and Care Homes (2025–26)

Executive summary Between 2025 and early 2026, a distinct set of pay-to-pilot deployments moved humanoid robots from lab demos into operational sites. This on‑the‑ground reporting—based on site visits, vendor briefings and interviews with pilot customers and integrators—catalogs trials across warehouses, manufacturing lines, care homes and customer‑facing retail settings. The pilots are small in scale (1–10 robots), purpose-built, and sharply focused on a handful of tasks: intermediate tote and bin movement, repetitive assembly and jig work, assisted‑living mobility/companionship workflows, and concierge/hospitality tasks. Results show meaningful productivity and safety promise but also recurring integration and economics challenges that keep these programs in a pilot stage. Warehouse and logistics pilots Humanoids are being trialed as complementary devices to traditional AMRs/AGVs, not full replacements. Two architectural patterns dominate: (1) Digit‑style bipedal units that move small totes and perform pick‑assist tasks, exemplified by Agility Robotics–style pilots run by 3PLs and regional e‑commerce centers; and (2) humanoid‑arm integrations from Figure AI and smaller vendors (including pilots coordinated by 1X Technologies) that plug into existing conveyor and sortation flows. Agility‑style tote moving and pick assist — At a midwestern e‑commerce DC visited in Q2 2025, an integrator ran two biped units on first‑shift tote transfers between packing islands. The operator reported average throughput of 2–5 tote moves per robot/hour, with pilot uptime of 90–95% (excluding charging and software updates) and zero reportable safety incidents. The pilot manager described labor substitution at ~0.15 FTE per robot for repetitive low‑value transfers, with a current break/fix cadence averaging one field service visit every 3–4 weeks. Figure AI and 1X integrations — Figure AI prototypes and trials coordinated through 1X Technologies focused on armified humanoids mounted near pick stations doing simple two‑hand insertions and box closing. These pilots reported task cycle times comparable to slow human operators on constrained tasks (20–40 cycles/hour) and uptime of 85–92% in early deployments; integration into warehouse WMS and fleet management software was cited as the primary blocker. Tesla and Optimus‑style internal tests — Tesla personnel confirmed limited internal trials of Optimus‑like units supporting non‑critical line tasks and materials movement. Tesla declined to share KPIs; observers and vendor briefings indicate usage is focused on low‑risk auxiliary duties while hardware and perception software mature. Operational lessons from logistics pilots Perception in cluttered DC environments remains the limiting factor: robots need simpler, constrained routes to meet uptime targets. Fleet management and integration with WMS are non‑trivial—most pilots rely on system integrators such as Dematic or boutique SI partners to bridge the software gap. Economic impact is incremental—pilot operators cited productivity gains of 5–15% for targeted workflows rather than wholesale labor substitution. Manufacturing and repetitive assembly Manufacturers are piloting humanoids where reach, dexterity and human‑like manipulation reduce tooling costs or speed changeovers. Two profiles stand out: Apptronik’s Apollo in aerospace and jigs — Apptronik‑led pilots used Apollo in fixture handling and repetitive inspection tasks in aerospace subassemblies. A Tier 2 aerospace supplier reported the humanoid reduced cycle time for a multi‑step jig operation by ~10–20% after process rebalancing; uptime reported at 93%, with one minor safety stop recorded in three months. Apptronik’s pitch centers on collaborative deployment without heavy safeguarding. Figure AI in repetitive assembly — Figure prototypes are being evaluated in electronics and automotive supplier lines for screwdriving, cable routing and simple insertions. Integrators involved reported that training and robot teaching took 2–6 weeks per task to reach stable throughput; task rates ranged widely (10–120 cycles/hour) depending on cycle complexity. Healthcare and eldercare pilots Care homes and assisted‑living facilities are running cautious pilots aimed at companionship, patient monitoring, and non‑clinical assistance. Key participants include Sanctuary AI, UBTECH, and consumer makers like Xiaomi running smaller assistive trials. Sanctuary AI — Sanctuary runs multi‑week pilots in assisted‑living residences focusing on mobility assistance (transfer support), medication reminders and limited ADL (activities of daily living) tasks. Pilot operators highlighted strong resident engagement but emphasized that humanoids augment staff rather than replace nursing aides. KPIs: resident interaction time increased 20–40% for the trial cohort; safety incidents were limited to sensor misreads that triggered safe stops. UBTECH and Xiaomi — UBTECH units were deployed in a private care facility for social‑robot workflows (reminders, video calls). Xiaomi’s CyberOne was trialed in consumer and small care settings for mobility support and telepresence; these pilots focused on UX, voice interaction accuracy, and perceived companionship rather than physical assistance. Uptime for social/telepresence modes commonly exceeded 95% because physical tasks were limited. Retail, hospitality and mixed‑use pilots Retailers and hotels tested humanoids for reception, inventory check, and guided customer interactions. Deployment sizes were generally single‑unit pilots used as marketing and operational experiments. Staff interviews revealed positive guest engagement, with limited back‑end value unless the robot integrates into POS and inventory systems. Pricing, business models and economics Supplier/Model Reported CapEx range Service/Ops model Typical pilot scale Agility‑style biped $60k–$200k per unit (pilot era estimates) Hardware sale + subscription for cloud and maintenance 1–5 units Figure AI / 1X integrations $100k–$300k (robot + SI work) Project‑based integration with recurring support 1–10 units Apptronik (Apollo) $150k–$400k (depending on payload/rigging) Integration partner + managed services 1–4 units Sanctuary / UBTECH / Xiaomi $10k–$150k (consumer to enterprise) Subscription for care software, optional managed service 1–20 units Integrators’ advice: plan for a 12–24 month total cost of ownership horizon and expect pilot budgets that cover customization and safety engineering to dwarf hardware costs. Competitive comparison and market positioning Agility Robotics and similar players position around warehouse utility and mobility in constrained environments. Figure AI and smaller arm‑centric vendors emphasize manipulation, modularity and line integration. Apptronik pitches collaborative heavy‑duty tasks in manufacturing where traditional cobots lack reach and kinematics. Sanctuary, UBTECH and Xiaomi target human‑centric care and social use cases, emphasizing safety, UX and regulatory readiness. Tesla is internally focused—its messaging links Optimus development to product‑line automation, not immediate commercial availability. Integration challenges and emergent best practices Pilots converge on a short checklist of recurring requirements: Sensing and perception: constrain environments, add fiducials and simplify lighting to raise uptime. Safety architecture: layered approaches—geofences, speed limits, and human‑in‑the‑loop overrides—are mandatory; heavy cages are being avoided for collaborative pitches. Fleet and WMS integration: early engagement with SIs is critical; most vendors do not offer turnkey ERP/WMS connectors. Training and change management: four‑ to six‑week operator certification programs, cross‑functional steering committees and staged scope expansions produce better retention. "The robots add predictable labor where humans would be pulled into less‑valuable repetitive work," said a pilot manager at a regional logistics operator. "But integration time and expectation management are the real costs—not the device price alone." Conclusion 2025–26 pilots demonstrate that humanoids are beginning to deliver narrow, measurable value in real operations—particularly where tasks are repetitive, constrained and carry safety or ergonomics benefits. However, the transition from pilot to scale requires continued improvements in perception, lower service costs, deeper enterprise integration and business models that align capital and maintenance expenses with incremental productivity gains. For now, humanoids are incremental automation tools that, when well‑deployed, reduce risk and improve throughput; they are not yet a general replacement for human labor across broad workflows.