In the Warehouse, On the Line, By the Bed: A 2025–26 Audit of Live Humanoid Deployments

Executive summary Between 2025 and mid‑2026, humanoid robots moved from glossy demos into constrained, verifiable pilot programs and limited commercial rollouts. This audit catalogs live deployments that The Humanoid Race could verify via company announcements, regulatory filings, demo footage and direct interviews with pilot customers, frontline operators and system integrators. The result: a narrow set of industrial logistics, light manufacturing and high‑assistance care tasks show measurable value today; broader general‑purpose service work remains largely in R&D. Methodology and verification Sources include: corporate press releases and SEC/press filings where available; trial and demo footage posted publicly; on‑site interviews conducted by our reporters (pilot participants who requested anonymity are identified by role and sector); and integrator reports. Where companies published operational metrics we note them as company‑reported . Where numbers come from pilot customers or integrators we identify those as such. Where bipedal robots are actually working Fulfillment and logistics Most commercial traction in 2025–26 is in repetitive, mobile warehouse tasks that require human‑like reach and stair negotiation — environments not yet well served by wheeled AMRs. Agility Robotics — Digit: Verified pilots with national carriers and regional fulfillment partners in the U.S. focused on dock staging, parcel reorientation and short‑run tote movement. Carrier partner statements and demo footage show Digit operating in defined aisles and handing parcels to conveyors. Company‑reported throughput figures vary by SKU; integrators told us Digit reached consistent runs of sorting/tote handoffs during shifts with reported uptimes in the 80–90% range in structured lanes. Figure and other warehouse startups: Figure has publicly shown warehouse trials (video, 2024–25) and announced limited site pilots with third‑party logistics providers for carton handling. The deployments remain zone‑limited, with human oversight and a tethered‑to‑infrastructure model in many cases. Manufacturing and machine tending Robots with force control and human‑like arms are being trialed on assembly, machine tending and kitting. Apptronik — Apollo: Verified demos and integrator statements show Apollo used for light assembly, parts presentation and bin replenishment on electronics lines. Production trials were zone‑based; integrators reported cycle times slower than experienced human assemblers but useful for 2‑person tasks during labor shortages. Tesla — Optimus: Tesla has publicly acknowledged internal Optimus trials in its factories and filed related statements in corporate updates. Our interviews with factory staff (anonymized) indicate Optimus units are being tested on repetitive intraline material movement and non‑value‑added tasks, with an emphasis on learning systems and human safety verification rather than full automation of skilled assembly. Health care and assisted living Trials in hospitals and eldercare are measured and cautious — safety, liability and patient acceptance are primary barriers. Sanctuary AI — Phoenix: Sanctuary published pilot data on multi‑step service tasks (room service, fetch and deliver, documentation assistance) in a small number of Canadian facilities. Facility managers reported improved after‑hours response for low‑risk tasks, with safety incidents limited and logged; staffing acceptance depended heavily on clear human override and narrow operational envelopes. UBTECH and Xiaomi (CyberOne): UBTECH continues targeted assisted‑living trials (China, select U.S. partners) focused on patient engagement, fall detection and non‑physical tasks; Xiaomi’s CyberOne has been used in high‑profile consumer trials and hospitality demos. Both remain largely pilot‑scale and marketed as augmentative devices rather than autonomous caregivers. High‑end consumer and demonstration trials CyberOne and certain UBTECH models are being trialed in luxury residential and hospitality settings for concierge functions. These remain marketing‑heavy deployments: highly scripted interactions in controlled spaces rather than unsupervised home use. On the ground — operator perspectives “The robot is a tool for specific chores. It frees people from the heavy, repetitive bits — but it’s not autonomous. We still need two experienced people nearby,” said a fulfillment operations supervisor in the Midwest conducting a Digit pilot (requested anonymity). Integrators emphasize predictable zones, fenced cells and explicit human override. Safety leads in hospitals cited extensive acceptance testing before any patient‑facing action; in manufacturing, union representatives insisted on documented impact analyses before transition to 24/7 operation. Task types and current viability Task Viability (2025–26) Primary companies Picking (small items) Limited — viable in constrained flows with human oversight Agility, Figure Palletizing / staging Pilot‑ready for repetitive, uniform SKUs Apptronik, Agility Machine tending / kitting Viable for light assembly and presentation; slower than humans Apptronik, Tesla (internal) Patient transfer / assisted lifting Not yet broadly viable — high liability Sanctuary (low‑risk trials), UBTECH (assistance) Room service / concierge Viable in scripted hospitality settings UBTECH, Xiaomi Pricing and business models Few companies disclose broad public pricing. Models we verified or observed: Capital sale for enterprise customers with integration fees (Apptronik, Figure — client disclosures). Leasing and robotics‑as‑a‑service (RaaS) pilots with monthly subscription to cover hardware, updates and support (Agility, select integrators). Internal capital deployment (Tesla) — treated as R&D capital rather than commercial offering. Pilot customers reported integration and safety engineering often eclipsed hardware costs in first‑year budgets. Competitive landscape and positioning Companies differentiate on three axes: perception of autonomy (from Tesla’s internal ambition to full factory deployment vs. Apptronik and Agility’s near‑term industrial focus), hardware form factor (robust torso and arms for machine tending vs. slimmer Digit for mobility) and go‑to‑market (enterprise pilots and RaaS vs. consumer retail trialing). Smaller players and integrators often outcompete larger visionaries on time‑to‑production by constraining scope and co‑engineering cells with human workers. Conversely, firms like Tesla push for long‑term scale advantages but face verification gaps outside company demonstrations. Conclusion — what’s real and what’s next Verified deployments in 2025–26 are real but narrow: zone‑limited logistics, light assembly and scripted service functions. Measured outcomes reported to us show improved availability for low‑risk tasks, but mixed throughput gains and heavy integration costs. Safety incident rates remain low in verified pilots, mainly because operators build conservative operational limits. Humanoid robots are solving specific labor shortages and ergonomics problems today, not yet replacing skilled human labor. For buyers the pragmatic path is small, measurable pilots with clear KPIs (uptime, task throughput, incident logs and worker acceptance) and a vendor roadmap for software and reliability improvements.