Boots on the Ground: Case Studies of Humanoid Robots Doing Real Work in 2025–26

Boots on the Ground: Case Studies of Humanoid Robots Doing Real Work in 2025–26 This field report compiles on-site interviews, vendor pilot summaries, procurement briefings (some on background), and third‑party test results collected across 2025–26 to document how current-generation humanoids are being trialed in warehouses, factories and care facilities. Sources include systems integrators, fulfillment center operations leads, independent university testbeds and IEEE‑affiliated evaluation labs. The picture is mixed: pilots are real and instructive, but continuous, high‑density deployment remains limited. Below are concrete use cases, measured pilot results, business models and lessons operators are acting on today. Summary table: vendors, use cases and headline metrics Company / Model Primary pilot use Reported throughput / cycle metrics Uptime & safety Commercial posture Agility Robotics — Digit Parcel handling; carton manipulation at end‑of‑line 20–60 picks/placements per hour for single‑item carton staging (pilot‑reported) Field uptimes 60–85% in early 8‑week pilots; most sites kept robots in supervised, caged zones RaaS, integration with AMR fleets; positioned as drop‑in material handler Figure AI Case picking, light assembly, inspection 10–40 picks/hour on unstructured SKUs; faster on repeat fixtures Higher mean time between operator interventions in controlled cells vs free roam Industrial partner model; emphasis on vision + force control for factory tasks Apptronik — Apollo Collaborative assembly, machine tending Cycle time parity with human for constrained two‑handed tasks after fixtureization Sites favored human‑in‑the‑loop teach pendant for safety; uptime 70–90% Sell/lease hardware with systems‑integration packages 1X Technologies — NEO/EVE Fulfilment stow/pick demos, repetitive line work Pilot pick rates 15–45/hour depending on bin geometry Most pilots used limited free‑space navigation; safety cages common Subscription pricing common; focus on low TCO on narrow tasks Sanctuary AI — Phoenix Care assistant prototypes, facility logistics Assistive tasks (tray carrying, object handing) measured in minutes per task rather than hourly throughput Regulatory scrutiny in healthcare; conservative supervised operation Positioned as generalist, full‑safety stack for human environments Xiaomi — CyberOne Retail/showroom, light concierge and demonstration pilots Demonstration tasks; no broad fulfilment throughput claims Mostly controlled deployments; uptime more about demo cadence than continuous ops Consumer/retail positioning; showrooms and marketing pilots Tesla — Optimus (early factory tests) Repetitive assembly trials, parts kitting Internal tests reported short‑cycle repetitive tasks; details sparse and site‑specific Early integration inside guarded cells with human supervisors on the line Vertical integration inside Tesla, not yet commercial RaaS 1. Logistics and Fulfilment: Where humanoids augment AMRs and cobots Logistics providers — DHL (regional pilots), several third‑party logistics (3PL) operators and mid‑tier e‑commerce warehouses — ran the bulk of field pilots in 2025–26. The deployments generally shared these characteristics: Humanoids were used to augment, not replace, existing AMR/SCARA fleets — typically tasked with irregular carton manipulation, shelf retrievals in constrained bays, or end‑of‑line sorting that requires bimanual dexterity. Reported pick/pack throughput varied dramatically with SKU shape and fixturing: pilot operators reported 15–60 handling events per robot per hour. By contrast, standard human pickers in high‑flow zones achieved 120–300 picks/hour for well‑structured pick‑to‑light operations. Return on investment remains pilot‑dependent: most logistics integrators preferred RaaS pricing (vendor covers maintenance) in early projects, with monthly fees in the low single‑to‑low double thousands per robot and 1–3 month integration pilots. On background: a 3PL operations lead said humanoids were most valuable where they replaced an ergonomically difficult human task (overhead transfers, awkward cartons) or plugged a niche where re‑engineering the line for an extra AMR would be costlier than a humanoid pilot. 2. Manufacturing and assembly: fixture first, free‑space later Automotive suppliers, electronics contract manufacturers and a handful of consumer goods factories trialed Figure, Apptronik and early Tesla units primarily on repetitive assembly and inspection tasks. Best performance came where manufacturers fixtureized tasks — i.e., constrained the part and fixed tools so the robot could execute repetitive kinematics. In those cases Apptronik and Figure reported cycle time parity with humans at particular stations. Free‑roaming integration into mixed human/robot lines lagged: manufacturers reported frequent re‑calibration and mapping drift in 24/7 shifts, which drove many factories to put humanoids in guarded cells or tethered zones during pilot phases. Tesla’s internal tests focused on low‑variance repetitive operations. Public data remains limited; independent engineers on site described “incremental deployment” with human oversight rather than wholesale replacement. 3. Healthcare and eldercare trials: assistance, not autonomy Nursing homes, rehabilitation centers and hospital logistics pilots showed the clearest non‑industrial value propositions but also the most regulatory friction. Sanctuary and UBTECH variants were trialed as tray carriers, guided‑mobility assistants and social engagement aides in North American and European facilities. Tasks were measured in task completion time (minutes) and caregiver time saved per shift rather than throughput per hour. Operators reported that robots saved 10–20% of staff time on non‑clinical transport tasks in tightly scoped pilots, freeing nurses for clinical work. However, facilities faced conservative regional regulations limiting unsupervised patient contact. Testimonials from staff emphasized predictable task execution and reductions in back‑strain incidents; residents were mixed in acceptance — some welcomed the interaction, others preferred human contact. 4. The lab‑to‑floor gap: recurring lessons from pilots Across sectors, five themes explain why demos differ from continuous production use: Fixtureization wins. Humanoids match or approach human cycle times only when parts or tasks are constrained. Mapping and calibration drift. Continuous 24/7 operation exposes localization and force‑control drift that university testbeds repeatedly flagged in independent evaluations. Safety zoning remains conservative. Most deployments favored cages, soft‑fence virtual zones or human‑in‑the‑loop teleoperation for risk mitigation; free‑roam co‑working at human speed is rare. Integration cost dominates. Integration and per‑site customization often exceed hardware costs, pushing vendors to RaaS models and longer pilots. ROI timelines lengthen. Procurement and systems integrators now budget 12–36 months to reach steady‑state value, not the 6–12 months vendors initially claimed. Competitive positioning and pricing Vendors are differentiating along two axes: physical dexterity vs ease of integration, and vertical focus vs generalist capability. Agility emphasizes logistics handling and AMR integration; Figure and Apptronik push industrial force control and factory applications; Sanctuary sells generalist safety and care features; Xiaomi focuses on consumer/marketing deployments. Pricing is opaque but follows a pattern: purchase offers for early adopters and widespread use of RaaS or subscription models for pilots, with integration fees that often exceed month‑to‑month robot costs. Conclusion In 2025–26, humanoids moved from spectacle to selective utility. They are now useful where their bimanual form factor and human geometry solve a specific problem that would otherwise require expensive line redesign or ergonomic mitigation. That utility is real — pilots report measurable time savings and equivalence in fixtured assembly — but full replacement of cobots, AGVs and human labor remains limited by uptime, calibration complexity, safety policies and integration cost. The next wave of commercial viability will come from vendors who standardize integration, tighten maintenance contracts, and sell predictable RaaS economics that match industrial procurement cycles.