On the Floor: A 2025–26 Audit of Real-World Humanoid Deployments
Executive summary In 2025–26 a growing, but still nascent, set of humanoid robot pilots moved from controlled demos into grounded, repeatable work at warehouses, logistics hubs, light‑assembly lines,...
Published by The Humanoid Race
Archived source unverified — no source link is available. The claims, figures and quotations have not been checked against an available source. Retained for historical context, not presented as newly verified reporting.

Executive summary In 2025–26 a growing, but still nascent, set of humanoid robot pilots moved from controlled demos into grounded, repeatable work at warehouses, logistics hubs, light‑assembly lines, clinics and assisted‑living facilities. Vendors including Agility Robotics , Figure AI , Apptronik , Tesla , 1X Technologies , Sanctuary AI , Xiaomi and UBTECH led headline projects. Integrators and third‑party auditors increasingly role‑checked vendor claims; the picture that emerges is mixed: robots accomplish narrow, repetitive tasks and reduce discrete human exertion, but they fall short of widescale replacement due to safety, floor‑integration and cost barriers. Where humanoids actually worked in 2025–26 Pilots clustered into six use cases. Below we list industries, typical tasks and representative vendors. Parcel sorting and last‑mile micro‑fulfillment (warehousing & logistics) — Agility Robotics’ Digit was the most visible humanoid in this sector, trialed for case/piece manipulation and conveyor handoffs in distribution centers. Trials emphasized bag/case deconsolidation and short‑walk pick-and-place between conveyors. Pallet handling and tote transfers (logistics hubs) — Hybrid deployments combined humanoids with fixed conveyors and collaborative robots (cobots) to move totes between palletizing stations; vendors used humanoids for intermediate handoff and human‑style grasping where bin geometry varied. Light assembly and pick‑and‑place (manufacturing) — Figure AI and Apptronik focused on repetitive assembly subtasks in electronics and automotive subassemblies, often in factory assistance roles where humans retained complex dexterity operations. Patient handling and clinical support (clinics and hospitals) — Sanctuary AI and 1X Technologies ran early trials for non‑clinical support: room checks, supply runs, and limited patient transfers (with strict protocols and human supervision). Eldercare and assisted living — Trials with 1X EVE and Sanctuary Phoenix emphasized room checks, medication reminders, and assisted sit‑to‑stand help where safety sensors and human handoff remained mandatory. Consumer and companion experiments — Xiaomi’s CyberOne and UBTECH platforms appeared in retail and consumer‑experience deployments as companion/concierge robots, emphasizing social interaction over heavy labor. How companies position their robots Vendors make two distinct sales narratives: Productivity augmentation: Agility, Figure and Apptronik frame humanoids as “workforce extenders” in repetitive, ergonomically risky tasks—aiming to improve safety and throughput without replacing skilled workers. Generalist future promise: Tesla and Sanctuary foreground multi‑domain potential, using factory testbeds and healthcare pilots as demonstrations of a generalist platform that will scale with software improvements. Marketing language emphasizes anthropomorphic affordances (stairs, human‑grade doorways, handoff with people), while integrators stress the current reality: most pilots are constrained to simplified, instrumented workflows and substantial human oversight. Pilot outcomes and KPIs — the real numbers (aggregated) Across documented pilots in 2025–26, the following ranges summarize outcomes reported by vendors, site customers and independent integrators. Where possible numbers are drawn from public vendor dashboards, customer briefings and third‑party pilot audits; ranges reflect variation across sites. Metric Typical range reported Notes / source types Availability / uptime 70–90% Lower bound during early field configuration; higher with 24/7 ops in instrumented cells (vendor & integrator reports) Cycle time vs human baseline 1.2–3× human time Humanoids slower on dexterous picks but consistent; some tasks reached parity after retooling workflow (customer reporting) Human handoffs per shift 4–20 per robot per 8‑hr shift Frequency depends on task complexity and safety policy; many pilots required human interruption for error recovery Safety incidents Near‑misses common; few recordable injuries Incidents often tied to edge‑case interactions; independent auditors pressured vendors to soft‑limit speed in mixed traffic Measured ROI Multi‑year payback (3–6+ yrs) in most pilots ROI improved when robots reduced overtime/turnover costs or avoided ergonomic injuries; capital intensity high Representative pilot vignettes Agility Robotics (Digit) — parcel handling: Multiple US distribution centers ran 6–12 week pilots where Digit performed conveyor-to-staging transfers and short‑walk picks. Sites reported improved ergonomics on repetitive lifts, but throughput gains depended on conveyor redesign and simplified grasp targets; independent integrators noted a 20–40% reduction in human lift tasks but a 10–30% throughput penalty until workflows were adapted. Figure AI / Apptronik — factory assistance: Trials at electronics subassembly cells focused on screw‑insertion and board transfer. Companies emphasized fast‑iteration control stacks; plant managers said humanoids reduced night‑shift hiring needs but required protected work cells and human intervention for exception handling. Tesla (Optimus) — internal testbeds: Tesla continued to run prototype Optimus units in select Fremont/Nevada facilities for repetitive non‑critical tasks and demonstrations. Outcomes remained internal; one qualifying integrator briefed The Humanoid Race that Tesla emphasized rapid software cycles over incremental hardware sales. Sanctuary AI & 1X — healthcare/eldercare trials: Trials focused on low‑risk interactions: deliveries, room checks, mobility assistance under a human’s direct supervision. Nursing staff acceptance hinged on clear failure modes and easy override; unions and regulators demanded recorded safety logs and human presence during patient transfers. Pricing, business models and integrator economics Pricing remained opaque but followed three models: Direct sale / capex: High upfront cost (vendor quotes reported in industry ranges of $50k–$200k+ per unit for advanced humanoids, depending on sensing and tooling). Robot‑as‑a‑Service (RaaS): Monthly leasing or service contracts commonly reported at $3k–$15k per robot per month, usually plus integration and site engineering fees. Outcome‑based contracts: A few integrators tied fees to throughput or downtime SLAs, but vendors were generally cautious about performance guarantees given environmental variability. Competitive comparisons and pairing with legged platforms Competition split between task‑focused humanoids (Agility, Apptronik/Figure) and broad‑vision platforms (Tesla, Sanctuary). Key differentiators: Robustness vs cost: Agility emphasized incremental deployability in warehouses; Figure and Apptronik marketed high‑torque actuators and factory‑grade durability. Tesla and Sanctuary argued for software scalability but faced skepticism about field reliability. Sensing and integration: Players with mature software stacks and integrator partnerships delivered shorter pilot cycles. Vendors that paired humanoids with legged or mobile bases (some pilots paired Unitree or Boston Dynamics quadrupeds for rough terrain support or payload shuttle duties) used the hybrid approach to bypass humanoid mobility limits. "Customers are buying predictable performance more than generality. Humanoids win where human‑like form solves the task without expensive reengineering of the site," said a third‑party integrator who runs multiple pilots. Barriers to scale Safety certification and regulation: Healthcare and eldercare pilots faced the strictest regulatory oversight; most patient‑interaction trials required documented safety cases and human oversight—limiting autonomous deployment. Floor and workflow integration: Many facilities had to rewire conveyors, add markers or deploy fenced cells; the integration cost often exceeded robot acquisition expense in pilot budgets. Labor relations: Unions pressed for clear redeployment plans; several pilots included joint labor‑management committees for monitoring. Robustness and exception handling: Real environments generate edge cases—packing variations, spilled liquids, clogs—that still require human judgment and engineered mitigations. Conclusion: incremental adoption, pragmatic expectations Across 2025–26 the industry moved from spectacle to repetition: humanoids are no longer just stage demos, but they are not yet generalist workforce drop‑ins. Measured gains are specific—ergonomic lift reduction, consistent repetitive handling, and 24/7 availability in instrumented cells—while costs, safety requirements and integration complexity throttle rapid scale. Expect the next 24 months to be defined by tighter integrator playbooks, clearer RaaS offerings, and sectoral pockets (logistics micro‑fulfillment, light assembly, supervised eldercare tasks) where humanoids begin to deliver predictable ROI.