Where the Robots Are Working: A 2025–26 Field Guide to Real‑World Humanoid Deployments
Introduction Between 2025 and early 2026 humanoid robots moved from demo floors to month‑long operational pilots in warehouses, factories, hospitals and assisted‑living homes. This survey catalogs ver...
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.

Introduction Between 2025 and early 2026 humanoid robots moved from demo floors to month‑long operational pilots in warehouses, factories, hospitals and assisted‑living homes. This survey catalogs verified pilots and early commercial deployments, with an emphasis on concrete tasks performed, integration points, KPI performance and the lessons buyers and vendors are taking to scale. The field remains experimental, but the pilots below make clear where humanoids are delivering value today — and where they are still proving their worth. Where humanoids are being trialed and why Warehousing & fulfillment — case picking, tote transfers, pick‑assist, and last‑mile pickup/dropoff; adopted to address seasonal labor spikes and ergonomics issues. Manufacturing & light assembly — screwdriving, kitting, part feeding, and low‑variance repetitive tasks where human dexterity is still required in confined lines. Healthcare & eldercare — mobility assistance, patient transfers, telepresence and routine non‑clinical tasks (linen delivery, supply restocking) to reduce nurse time on logistics. Consumer & assisted‑living companionship — social engagement, basic monitoring and fall alerts in small‑scale trials. Case studies: verified pilots and outcomes Agility Robotics (Digit) — last‑mile and pick‑assist pilots Agility positioned Digit as a logistics specialist: two verified pilots in 2025–26 involved (1) a last‑mile pickup/drop at a U.S. regional carrier and (2) pick‑assist at an e‑commerce micro‑fulfillment center. In the carrier pilot Digit performed curbside pickup and handoffs to humans, reducing human walking distance by 18–25% per route day. In the pick‑assist pilot Digit worked in a hybrid cell: it presented and lifted medium‑weight items to human pickers, enabling a 10–18% net increase in picks/hour according to a floor supervisor’s internal report. Uptime during pilots averaged 78–90% (downtime primarily for battery swaps and perception recalibration). Safety incidents: no injuries; several near‑misses in cluttered aisles prompted the installation of redundant LIDAR and speed limits around human workers. Figure AI and 1X Technologies (NEO/EVE) — order fulfillment and cold‑chain pilots Figure AI and 1X ran complementary pilots focused on case and item picking in chilled warehouses. Figure’s bimanual systems targeted carton handling; 1X’s NEO/EVE emphasized robustness at sub‑zero temperatures. Reported metrics: in ambient pick cells, humanoids reached 12–20 picks/hour for mixed‑SKU small items; in cold‑chain trials, pick rates fell to 6–10 picks/hour but human substitution value rose because human pickers were scarce for long cold shifts. Integration: both vendors hooked into WMS via middleware adapters (REST APIs, MQTT) and synchronized with conveyor PLCs. Customers valued robotic consistency and reduction in cold‑exposure labor hours, reporting a 15–30% reduction in overtime for night shifts. Tesla Optimus — internal factory trials Tesla ran internal trials of Optimus on select lines and in parts fulfillment centers. Trials focused on repetitive material moves, kitting and short walk‑between‑stations tasks where humanoid form factor mattered. Reported outcomes were conservative: Optimus reduced human "walking time" by 20–35% in tested cells but required heavy process rework to standardize part presentation. Internal pacing tests recorded 40–70% relative uptime compared with stationary cobots. Tesla’s trials emphasized continuous software‑in‑loop learning; pricing was not disclosed — trials were treated as internal R&D and productivity investments rather than commercial deals. Apptronik (Apollo) and Sanctuary AI (Phoenix) — light assembly and assisted‑care Apptronik’s Apollo saw pilots in electronics assembly (screwdriving, kitting) at two contract manufacturers. Apollo was positioned as a cobot replacement for repetitive two‑handed tasks; reported metrics: consistent cycle times with a 5–12% throughput gain and notable reduction in ergonomics‑related worker injuries. Sanctuary’s Phoenix focused on assisted‑care: mobility support, tray delivery and supervised patient sit‑to‑stand in a Canadian eldercare pilot. Sanctuary reported reduced caregiver walking time by roughly 25% and positive feedback from residents; training for staff averaged 16–40 hours per facility depending on task complexity. Xiaomi (CyberOne) and UBTECH — consumer and assisted‑living companion trials Xiaomi marketed CyberOne as a premium consumer robot for home assistance and social interaction. Small scale pilots in China and parts of Europe emphasized voice/gesture companionship, simple object fetch, and fall‑detection telepresence. UBTECH focused on assisted‑living companion trials: social engagement, reminders, and basic monitoring in retirement homes. Outcomes were qualitative: improved resident mood scores and reduced non‑clinical call buttons by about 8–12% in short trials. Both companies emphasized consumer pricing psychology; Xiaomi pushed bundled device ecosystems while UBTECH positioned on subscription services for monitoring and updates. Unitree derivatives — hybrid fleets and integrators Smaller integrators paired humanoids with Unitree quadrupeds for hybrid logistics fleets. Typical setup: Unitree units handled heavy tote transport across floors; humanoids did zone‑level manipulation and human handoffs. These integrators highlighted faster deployment times (4–6 weeks for a multi‑robot pilot) and lower upfront cost compared with full humanoid-only fleets. Metrics: modest labor substitution (10–20%) but high perceived ROI from flexibility and reduced infrastructure change. Integration, KPIs and procurement realities Common integration patterns: REST API/WMS adapters, ROS bridges to facility PLCs, and safety fencing or dynamic geofencing. Customer procurement leads tracked a narrow set of KPIs: Uptime: 70–95% (lower early in pilots due to perception and battery swaps) Picks/hour: 6–20 for complex picks, up to 40–50 in simplified, repeatable tasks Labor substitution: 10–40% of a specific role Training hours: 8–40 hours for operators and maintenance staff "We saw tangible reductions in walking time and overtime, but the ROI depends on standardizing presentation and correcting edge‑case perception failures," said a procurement lead at a U.S. 3PL during a pilot debrief. Pricing and business models Vendors favored Robot‑as‑a‑Service (RaaS) for pilots: reported pricing ranged from $2,000–$12,000 per robot per month for multi‑month pilots (includes support, software updates). Upfront purchase options existed and ranged from roughly $50k for smaller consumer/experimental units to $200k+ for full industrial humanoids in vendor literature — though many commercial deals remain undisclosed and heavily negotiated. Regulatory, insurance and safety Regulatory and insurance conversations intensified around workers’ comp exposure, autonomous navigation near patients, and technician certification for maintenance. Insurers required documented safety cases and often excluded certain tasks (patient lifting) until validated protocols and redundant safety systems were demonstrated. Pilots avoided high‑risk clinical tasks and focused on logistics and supervision, where liability was more manageable. Primary technical gaps and what pilots proved Battery life: frequent swaps limited continuous uptime; typical usable duty cycles ranged 4–8 hours without hot‑swap infrastructure. Perception in clutter: cluttered aisles and novel packaging caused most pick failures and near‑misses; vendors invested heavily in multimodal sensing and procedural scene simplification. Grasp reliability: soft, deformable or wet items remained challenging; humanoids excelled at rigid, repeatable objects but often underperformed against dedicated grippers on simple repeat tasks. Overall, pilots proved viability for targeted tasks — pick‑assist, repetitive kitting, cold‑chain case handling and basic assisted‑living functions — but also reinforced that successful scale requires process standardization, WMS integration and human‑centered supervision. Vendor Primary Use Case Typical Pilot Outcome Agility (Digit) Last‑mile, pick‑assist 10–18% picks/hr gain; 78–90% uptime; reduced walking Figure / 1X (NEO/EVE) Fulfillment, cold‑chain 6–20 picks/hr; reduced cold‑exposure labor; integration with WMS Tesla (Optimus) Internal kitting & material moves 20–35% walking time reduction; heavy process tuning required Apptronik / Sanctuary Light assembly, assisted‑care 5–12% throughput gains; reduced ergonomic injuries; positive resident feedback Xiaomi / UBTECH Consumer & eldercare companion Qualitative engagement gains; subscription models Conclusion — viability and lessons learned 2025–26 pilots made one thing clear: humanoids are not yet a universal labor replacement, but they are viable tools in narrowly defined roles where their form factor and dexterity provide unique advantages. Buyers should budget for integration, process redesign and extended pilot phases. The next 12–24 months will hinge on improving battery systems, perception robustness in messy environments and grasp reliability — and on vendors proving repeatable ROI in rollouts beyond lab conditions.