On the Ground with Bipeds: A 2025–26 audit of real-world humanoid pilots and deployments

Executive snapshot In 2025–26 humanoid robots have moved from concept videos into limited, real‑world pilots — predominantly narrow tasks inside logistics and light manufacturing, with early healthcare and hospitality trials and consumer demos. What separates hype from reality is clear: a scattering of contracted trials and short, closely supervised deployments; no mass commercial rollouts. Vendors position products either as drop‑in replacements for specific human tasks (picking, kitting, machine tending) or as “flexible arms-on-legs” for environments where wheeled robots struggle. This report audits the live pilots, the firms running them, what each pilot is actually measuring, and what procurement teams should require today. Logistics — piece‑picking and palletizing Use case: narrow-bin piece picking, put‑wall kitting, and semi‑structured palletizing in e‑commerce and last‑mile hubs. Who’s running pilots Agility Robotics (Digit) — focused on goods‑to‑person adjuncts and piece‑picking trials in distribution centers; multiple trial announcements since 2023 matured into year‑long contracted pilots in 2025 with 3PLs and regional e‑commerce operators. Figure and other narrow‑task humanoid startups — several small teams are trialing constrained picking lanes where items are presented in fixed bins. Integrators — systems integrators continue to select humanoids for pilot corridors where reconfigurability and human‑like manipulation reduce integration cost. How vendors position them: Agility emphasizes “humanoid where wheels can’t go” — stepping over gaps, manipulating shelves, and using human tools without extensive rewiring of the site. Figure and similar teams market pick‑rate parity in narrow SKU sets and simplified human‑robot handoffs. Pilot outcomes & KPIs: public metrics are sparse. Typical pilot targets are: Throughput targets of 20–60 picks/hour for constrained, single‑SKU lines; Pick accuracy targets above 99% for e‑commerce order accuracy; Uptime aims in the 80–95% window during supervised shifts; Safety incidents: expected to be zero‑tolerance; real pilots report low‑severity incidents only during integration trials. Barriers: trained perception for varied SKUs, grasping reliability, and integration with warehouse management systems. Procurement criteria center on demonstrable pick‑rates on customer SKUs and a clear path to certifications for industrial floors. Manufacturing — machine tending, kitting, collaborative tasks Use case: collaborative machine tending, line kitting, and small‑parts assembly tasks where a biped’s reach, balance and ability to use human tooling are advantageous. Who’s running pilots Apptronik (Apollo) — marketed as a collaborative humanoid for light industrial contexts; Apptronik has run factory floor demonstrations and limited collaborative trials with integrators focused on kitting and sequential part feeds. Boston Dynamics — continues R&D demonstrations showing robust balancing and payload handling; commercial production remains focused on non‑biped assets, with humanoid work largely research‑to‑demo. Tesla (Optimus) — publicly stated internal beta testing inside Tesla facilities for material handling and simple repetitive tasks, positioning Optimus as an internal productivity multiplier rather than an immediate product. Pilot outcomes & KPIs: manufacturers track cycle‑time, changeover reduction, and collaborative safety metrics (force‑limiting interactions, emergency stop performance). Reported early benefits are reduced ergonomics injuries and rapid redeployment between stations; demonstrable ROI typically projected over multiple years. Barriers: industrial‑grade safety certification, ruggedness in shift cycles, and the economics of replacing inexpensive cobots or humans on highly repetitive tasks. Healthcare — patient handling and monitoring pilots Use case: assisted patient transfer, vitals monitoring, and logistics within hospitals (linen, supplies). Who’s running pilots: small pilots with regional hospitals have involved commercial humanoid platforms adapted for patient‑safe transfer and contactless vitals recording. Vendors highlight reduced staff strain and triage assistance; hospitals emphasize strict safety reviews and pilot controls. Pilot Outcomes & KPIs: hospitals require zero adverse patient events, accuracy in vitals within clinical tolerances, and demonstrable improvements in staff time saved. Progress is incremental: most hospitals confine humanoids to non‑critical support roles during trials. Hospitality, concierge and home experiments Use case: front‑of‑house concierge trials, reception assistance, and controlled home demos. Xiaomi and UBTECH run staged consumer demos and retail concierge pilots; these are marketed toward early adopters and lifestyle uses rather than heavy duty commercial service. Hotel chains and large venues have run short concierge pilots focusing on information delivery and package handling on controlled routes. KPIs: guest acceptance, successful task completion rate, and uptime during lobby hours. Business models skew toward capex leasing or service contracts for on‑site support. Commercial models, pricing and integration Pricing context remains variable. Vendors offer: Robot sale + integration services for large customers; Robot‑as‑a‑service (RaaS) where monthly fees include maintenance and software updates; Pay‑per‑performance pilots with milestone payments tied to KPIs. Typical market expectation: initial pilots cost hundreds of thousands of dollars including integration; scaling to dozens of units requires clear task constraints and long‑term contracting to approach ROI positivity. Many vendors emphasize a RaaS approach to lower entry cost and assume integration risk. Comparative snapshot of pilot maturity Use case Leading vendors in pilots Pilot maturity (2025–26) Typical KPIs disclosed Logistics — picking/palletizing Agility Robotics, Figure, niche startups Limited production pilots in constrained lanes Picks/hour, accuracy, uptime, safety incidents Manufacturing — kitting/machine tending Apptronik, Tesla (internal), integrators Collaborative line trials; internal beta tests Cycle time, downtime, safety performance Healthcare Specialist integrators & vendor pilots Controlled clinical pilots for support tasks Patient safety incidents, accuracy of monitoring, staff time saved Hospitality & Home Xiaomi, UBTECH, hospitality integrators Demo pilots, short‑term hotel trials Guest interaction success, reliability Composite insight from customer and integrator briefings: "For now the value case is real only when the task is constrained and the integration scope is tightly defined. Safety, consistent perception and maintenance costs determine whether to scale." Viability timeline and procurement guidance Industry assessment: Warehousing (logistics) — 2–5 years for broader adoption in constrained zones; pilots today validate use in picking aisles with controlled SKUs. Manufacturing — 3–6 years for steady adoption in collaborative, light‑duty roles where ergonomics benefit is clear. Healthcare — 4–8 years; regulatory and patient‑safety hurdles slow progress. Home — consumer humanoids remain experimental for the near term; wide adoption unlikely within five years without major cost and capability breakthroughs. Takeaways for procurement teams What to pilot: start with constrained, high‑ergonomics or safety‑sensitive tasks (e.g., repetitive pick lanes, kitting stations, nonclinical patient handling). Require vendor‑run proof‑of‑value pilots that include integration scope and fallbacks. Required metrics: Task throughput (picks/hour or cycles/hour) on your SKU set or part family; End‑to‑end accuracy and quality metrics (order accuracy, assembly error rate); Availability/Uptime during staffed shifts and Mean Time To Repair (MTTR); Safety incidents and near‑miss logging, with independent validation of fail‑safe behavior; Total cost of ownership over a 3–5 year window including maintenance and software licensing. Which vendors today: a small set — Agility Robotics and Apptronik among them — are offering production‑grade pilots for constrained industrial tasks; others (Boston Dynamics, Tesla, Xiaomi, UBTECH) remain closer to R&D, internal pilots or consumer demo stages. Integration partners and 3PLs will be decisive in turning pilots into repeatable deployments. Bottom line: humanoid robots in 2025–26 are leaving the lab for narrowly defined, closely supported commercial pilots. Procurement teams should pilot selectively, insist on hard operational KPIs, and treat early humanoid trials as long‑horizon automation investments rather than immediate labour replacements.