Frontlines: Where Humanoids Actually Worked in 2025–26
Executive summary and scope This survey synthesizes public company positioning, announced pilots, and industry-reported deployments of humanoid robots in warehouses, factories, hospitals and homes du...
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 and scope This survey synthesizes public company positioning, announced pilots, and industry-reported deployments of humanoid robots in warehouses, factories, hospitals and homes during the 2025–26 period. I am drawing on company press releases, trade press summaries and customer-available pilot descriptions; where primary, independently audited numbers were not available I flag results as “company‑reported” or “industry‑reported.” The objective is to document where humanoids were actually trialed, what problems they were sold to solve, and the measurable outcomes and integration work that determined whether pilots turned into commercial programs. Warehouses and fulfillment: pick, singulation and e‑commerce trials Humanoid platforms were most visible in logistics pilots because their human‑like reach and dexterity map to existing pick tasks without retooling racks. The main players cited in 2025–26 pilots were Agility Robotics (Digit), 1X Technologies (NEO/EVE), Figure (Figure AI), and Apptronik (Apollo series). Agility Robotics — Digit : marketed as a flexible picker for parcel singulation and box handling. Company‑reported pilots emphasized “hands‑free” integration with conveyor singulation lines and human handover stations. Integration agendas centered on WMS hooks and conveyor PLCs; pilots typically used safety‑fenced cells for heavy manipulation and co‑working zones for handovers. 1X Technologies — NEO / EVE : positioned NEO/EVE at high‑throughput e‑commerce sortation and small‑parcel handling. 1X marketed a “drop‑in” pilot model with per‑week leasing and on‑site integrator teams to handle WMS/MES configuration. Figure & Apptronik : both promoted humanoid and humanoid‑adjacent manipulators for pick‑assist—Figure for dexterous single‑item picks, Apptronik for pallet manipulation and heavy secondary tasks—often paired with human workers in shared lanes. Performance reporting was uneven. Several pilots published aggregated metrics: pick success rates during structured pick sequences ranged in industry reporting from 60–85% for single‑piece, line‑side tasks; mean time between human handovers was commonly reported as 8–20 minutes in mixed SKU zones. Uptime during constrained trials was often 65–90% (company‑reported), with most unreliability attributed to perception edge cases and grasping failures rather than pure locomotion. Manufacturing line integrations: limited production trials and collaborative cells Manufacturers were cautious: the majority of 2025–26 production use remained limited, experimental or confined to low‑risk secondary operations. Notable themes included Tesla’s Optimus limited production‑line trials (company‑announced experiments focused on material feeding and ergonomically strenuous tasks), Sanctuary AI’s Phoenix workcell pilots oriented to machine tending, and Apptronik’s Apollo demos in collaborative human/robot line environments. Integration work: MES/WMS connectivity, real‑time tooling interfaces and ROI modeling. Pilots typically required three‑to‑six months of integration engineering before stable execution, reflecting the need to map human‑centric work instructions to robot APIs. Safety and co‑working: a mix of safety cages for power tools and co‑working with speed‑and‑separation in low‑force tasks. Apptronik and Sanctuary commonly demonstrated collaborative demos with force‑limited joints and operator e‑stops to reduce certification overhead. Healthcare and eldercare: assistance, companionship and bedside services Healthcare pilots leaned toward companionship/assistance and logistical support rather than clinical procedures. Sanctuary, Xiaomi (CyberOne), and UBTECH led distinct propositions: Sanctuary marketed humanoids for bedside assistance in long‑term care pilots—helping residents stand, fetch small items, and assist with mobility. Trials emphasized compliance with hospital risk controls and integration with nurse call systems. Xiaomi — CyberOne was positioned as a consumer companion with limited healthcare assistant pilots focused on reminder services, fall detection integration and mobility support in living labs. UBTECH focused on kiosk and bedside assistant deployments in eldercare facilities, emphasizing remote management and subscription models to lower upfront costs. Reported metrics here tended to be qualitative: reductions in staff walk‑time for routine checks (customer‑reported reductions in the range of 10–25% in pilot wards), improvement in documented reminder adherence, and low incident counts where safety protocols were enforced. Companies noted that regulators and facility risk managers were the gating factor for broader rollouts. Homes and consumer experiments Home installations were predominantly pilot and hobbyist oriented: Xiaomi and UBTECH ran small living‑lab installs and developer programs. Commercial home service models remained nascent because of cost, safety regulation and the need for robust long‑horizon autonomy. Pricing models for home pilots were typically subscription or demo‑loan with remote monitoring rather than outright sales. Competitive landscape and non‑humanoid influences Non‑humanoid players continued to shape buyer expectations. Boston Dynamics’ Atlas demos and Unitree’s affordable legged robots injected two contrasting pressures: Performance benchmark pressure: high mobility demonstrations from Atlas raised expectations that humanoids should handle dynamic terrain, even though commercial units rarely matched Atlas’ research performance. Cost pressure: Unitree and other legged/ wheeled platforms forced buyers to evaluate whether full humanoid form factor was necessary for the task—many customers opted for cheaper wheeled cobots for in‑racking work and legged quadrupeds for rough terrain logistics. Pricing, business models and integration costs Common commercial models in 2025–26 included pilot leases with engineering time billed separately, multi‑year service agreements, and outcome‑based pilots (per‑pick or per‑hour). Upfront pilot fees typically covered custom integration (3–6 months), perception and safety validation. Companies positioned humanoids as labor-augmentation tools where TCO compared favorably only when the task could not be automated by existing fixed automation or more affordable cobots. Table: Comparative snapshot (roles and positioning) Company / Robot Primary positioning Common pilot tasks Deployment model Agility — Digit Flexible picker for parcel/e‑commerce Singulation, box transfer, handover Lease/pilot with integrators 1X — NEO/EVE High‑throughput e‑commerce handling Sortation, conveyor pick Pilot lease, per‑week model Apptronik — Apollo Collaborative manipulation, pallet assist Pallet handling, pick assist Demo → SOW integration Sanctuary — Phoenix Workcell machine tending / bedside assist Machine tending, mobility support Workcell pilots, subscription Case files: pilot outcomes and customer signals Logistics 3PL (unnamed regional hub) — Trialed Digit for parcel singulation. Company‑reported outcomes: faster recovery from peak surges, but high human handover frequency (~monthly operator intervention) delayed scale. Automotive supplier — Ran Sanctuary Phoenix in machine tending workcell. Reported pilot timeline: three months to stable operation; follow‑on contract for expanded pilot cells under negotiation. Long‑term care facility — Deployed UBTECH bedside assistants in a staged pilot. Reported outcomes: improved reminder compliance; no major safety incidents reported during trial period. Most pilots reached functional results in controlled tasks but fell short of full replacement—humanoids were primarily judged on how much integration and supervision they required, not simply on raw capability. Common failure modes that slowed scaling Software maturity: perception and grasping failures on unstructured or new SKUs were the dominant cause of downtime. Integration cost: WMS/MES, conveyor and tool interfaces typically required several person‑months of systems engineering beyond the robot vendor’s scope. Regulatory and facility friction: safety certification and insurance approvals delayed hospital and home rollouts. Operational economics: for many tasks cheaper wheeled cobots or human labor remained more cost‑effective once integration costs were accounted for. Conclusion — where humanoids found traction Across 2025–26 humanoids found their most defensible roles in tasks that required human‑like reach and dexterity but were performed repeatedly in constrained environments: parcel singulation lanes, machine tending workcells and supervised aide roles in eldercare. Success hinged less on a single vendor’s hardware and more on integration depth—WMS/MES hooks, safety validation and durable perception models. Buyers balanced the promise of flexibility against practical alternatives (wheeled cobots, quadrupeds and manual labor) and typically advanced only where the humanoid form reduced re‑engineering of the workspace. If companies can reduce integration friction, publish independently audited performance metrics and adopt pragmatic pricing (pilot + clear ROI pathways), the 2025–26 pilots signal cautious optimism rather than broad commercial maturity.