Where the Robots Actually Walk: A 2025–26 Field Guide to Live Humanoid Deployments

Executive summary and scope Note on sourcing: this field guide compiles public, vendor‑announced pilots and press disclosures available through mid‑2024 and evaluates their implications for 2025–26 rollouts. Where independent third‑party KPIs are unavailable I mark vendor claims explicitly. The humanoid landscape remains early: many deployments are time‑boxed pilots, vendor‑led demonstrations or internal trials; very few have published audited throughput or ROI numbers. Where humanoids are being piloted — industries and use cases Across warehousing, logistics, manufacturing, eldercare and consumer/home trials, the current fleet of commercially-focused humanoids is being positioned to address three recurring problems: labor shortages, repetitive manual handling in constrained workcells, and front‑line customer/companion tasks that benefit from a biped form factor. Warehousing & last‑mile transfer: box handling, shelf transfers, handoffs to human pickers and short corridor carries (Agility Robotics, Figure). Factory floors / machine tending: repetitive pick/place, tool‑change assisted tasks and inspection with human‑level reach and dexterity (1X Technologies, Apptronik). Eldercare & assisted living: non‑medical assistance, mobility support, social engagement and fetch tasks (Sanctuary AI, UBTECH). Home / consumer experiments: companion services, light chores and research prototypes (Xiaomi CyberOne, Unitree hobbyist/consumer tests with biped/quadruped pairings). Internal factory trials: prototype validation and data collection in production environments (Tesla Optimus internal testing claims). Company snapshots, pilots and on‑the‑ground outcomes Agility Robotics — Digit in warehouses and transfer corridors Positioning and marketing: Agility frames Digit as “human‑scale” for depalletizing, induction and point‑to‑point transfer — tasks where a biped can use existing human fixtures. The company markets Digit to logistics firms seeking flexible automation without costly conveyor redesigns. Pilots and outcomes: Agility has run vendor‑announced pilots with third‑party logistics sites and OEM partners to test box handling and handoffs to AMRs. Public disclosures emphasize durability and repeatability, but audited KPIs (throughput, pick accuracy) are rarely published. Vendor statements cite cycle times comparable to a human in specific unpaced tasks and the ability to operate alongside AMRs for handoff workflows. Integration friction points: PLC/WMS/Warehouse integration typically required custom middleware; tool changer and end‑effector fit for diverse parcel sizes remains a recurring integration task. Vendors report planners needing to convert human‑centric work instructions into robot‑suitable sequences. Figure (Figure AI) — last‑mile and warehouse pilots Positioning and marketing: Figure markets a humanoid platform for warehouse and last‑mile transfer, pitching rapid perception and human‑like manipulation as differentiators over arm‑on‑wheels systems. Pilots and outcomes: Figure has publicly demonstrated warehouse workflows in partner sites; specifics on throughput gains are vendor‑reported. The company emphasizes its software stack and sim‑to‑real training as the primary lever to reduce cycle times and increase reliability. Lessons: vendors claim fast task‑retraining and adaptability, but teams report long integration cycles to match enterprise WMS and safety requirements. 1X Technologies (NEO / EVE) and Apptronik (Apollo) — factory floors and machine tending Positioning and marketing: 1X and Apptronik target manufacturing customers with humanoids that can operate in structured cell environments — machine tending, small parts assembly, inspection. Apptronik emphasizes robustness and a path to ISO/industrial certification. Pilots and outcomes: Public deployments are typically multi‑week factory pilots. Reported outcomes include reductions in ergonomics risk, modest cycle time improvements (vendor ranges 10–30% on specific tasks) and swapping human operators from repetitive stations to higher‑value roles. Integration friction: tool changers, end‑effector compatibility with grippers/torque limits, and tight PLC/robot cell safety interlocks are the biggest hurdles. Companies often retrofit standard UR/ABB pneumatic/electric tool interfaces or design custom fixtures. Sanctuary AI (Phoenix) and UBTECH — eldercare and assisted‑living trials Positioning and marketing: These vendors pitch humanoids as augmentative care aides that can reduce caregiver load, perform non‑medical tasks, and provide social interaction. Sanctuary emphasizes human‑centric generalist capability; UBTECH couples lower‑cost platforms into assisted living pilots in China. Pilots and outcomes: Trials have been time‑limited (weeks to months) and largely observational. Reported benefits are reductions in routine staff tasks (fetching items, guided walks) and positive resident engagement. Hard KPIs such as staff‑hours saved or incident reductions are sparse; providers cite anecdotal caregiver time reclaimed and improved resident mood. Regulatory/insurance hurdles: Long‑term care facilities must navigate privacy regulations, liability for falls and insurance acceptance for robot‑assisted tasks — common themes in public conversations with pilot sponsors. Xiaomi CyberOne and Unitree — consumer/home experiments Positioning and marketing: Xiaomi positions CyberOne as a research and lifestyle platform; Unitree focuses on price‑competitive mobility and experimental home use. Marketing is both consumer and developer‑oriented. Pilots and outcomes: Home trials are largely hobbyist, press demos and internal research; credible KPIs (uptime, task success) are not publicly audited. Common issues in home trials include battery life, perception under varied lighting and navigation on cluttered floors. Tesla Optimus — internal factory trials where public data exists Positioning and marketing: Tesla frames Optimus as a general‑purpose automation platform intended to augment factory labor and eventually consumer markets. Public demos in 2023–24 show internal trials in factory environments. Public evidence and outcomes: Tesla has released videos and executive statements about on‑site testing. Independent, audited KPIs (throughput, safety incident records) have not been published; most reporting relies on company footage and statements. KPIs, business models and pricing context Pricing and commercial models remain in flux. Common approaches vendors publicize: Per‑unit sales for early adopters (capex heavy for pilot customers). Robot‑as‑a‑Service/subscription, which vendors argue smooths ROI and includes maintenance and software updates. Outcome‑based contracts (rare) tied to throughput or shift coverage guarantees. Concrete KPIs — what is actually reported publicly: Cycle time improvements: vendor‑reported ranges typically 10–30% on discrete, repetitive tasks. Pick/placement accuracy: vendors claim >90% in constrained setups; independent verification is rare. Uptime: pilot uptime commonly reported in vendor materials as 80–95% in controlled settings; real facilities report lower figures when human interaction and clutter increase. Staff‑hours saved: vendors often translate robot operation into FTE equivalents (single unit claimed to replace 0.5–2 FTEs depending on task), but independent audits are scarce. Safety incidents: most public pilots report zero major incidents, and many require human supervisors for safety redundancy. Competitive analysis and common technical gaps Company / Robot Primary vertical Strengths claimed Common gaps Agility (Digit) Warehousing / logistics Human‑scale transfer, mobility Integration to WMS, tool/end‑effector diversity Figure Warehousing / last‑mile Software/sim stack, manipulation Enterprise integration, audited KPIs 1X / Apptronik Manufacturing Industrial robustness, ISO path PLC/robot cell compat, certs Sanctuary / UBTECH Eldercare Human‑centered interaction Regulatory, liability, long‑term reliability Xiaomi / Unitree Consumer / research Price/brand, developer ecosystem Battery life, perception at home Tesla (Optimus) Internal factory Scale ambition, integration with Tesla ops Public KPIs, long‑term reliability Common technical gaps observed on the ground: battery endurance for full shift use, robust locomotion and manipulation on cluttered or narrow aisle floors, reliable perception in industrial lighting and reflective surfaces, and enterprise‑grade connectors to PLC/WMS/ERP ecosystems. These are the practical blockers preventing pilots from converting into large paid rollouts. Which pilots have become paid rollouts (and why) As of mid‑2024, the majority of humanoid work has remained in the pilot phase. Where transitions to paid rollouts have occurred, they share common characteristics: Task simplicity and constrained workspaces where the humanoid can be locked into a narrowly defined role. Strong vendor support for system integration (safety, PLC/WMS connectors) and on‑site maintenance. Clear regulatory/insurance paths and low liability for the specific tasks (non‑medical, non‑contact heavy handling). Vendors that succeeded at early commercial transitions typically sold into sites willing to co‑develop workflows and accept supervised operation rather than full autonomy. Conclusions — practical expectations for 2025–26 Humanoid robots are moving from lab demos to sustained pilots but are not yet broadly displacing human labor at scale. Expect incremental deployments in niche, high‑value or ergonomically challenging tasks in manufacturing and logistics where existing infrastructure is humanoid‑friendly. Eldercare trials will continue to expand but face a longer horizon due to regulation and insurance. The next 18–30 months are likely to show selective paid rollouts where integration friction is solved and ROI is demonstrable; widespread adoption will require further improvements in battery life, perception in real‑world environments and standardized enterprise integration stacks. On vendor claims vs reality: Most vendors headline impressive demos; the measurable business case emerges only after integration, certification and several months of supervised onsite operation — that is the recurring lesson from the field.