Feet on the Ground: Case Studies of Humanoids Doing Real Work in Warehouses, Factories and Care Homes (2025–26)

Overview Between 2025 and 2026 the humanoid robotics industry moved from demonstrations to a first wave of field pilots and limited commercial deployments. This report stitches together company disclosures, integrator briefings, independent test-lab summaries and on‑the‑ground interviews to map where humanoids are proving value — and where they are not. Patterns cluster by vertical: warehousing/logistics, factory floor assistance, and healthcare/eldercare. Across all three, common bottlenecks (battery life, perception in clutter, dexterous manipulation, systems integration) governed whether pilots scaled. Use cases and industries Warehousing / Logistics — order picking, pallet handling, goods‑to‑person tasks and light replenishment in e‑commerce and 3PL centers. Manufacturing — line‑side assembly, kitting, part presentation, and visual inspection where small, bipedal form factor reduces cell redesign. Healthcare / Eldercare — patient mobilization (assisted standing/transfer), fetch‑and‑carry, telepresence for remote clinicians and round‑the‑clock check routines. How companies are marketing and positioning robots Vendors adopt two distinct go‑to‑market narratives. First, task specialists (Agility Robotics’ Digit) pitch workforce augmentation: humanoids as drop‑in pickers that reduce ergonomics risk and fill labor gaps. Second, generalists (Tesla with Optimus, Figure, Apptronik) position broad factory assistants that can be re‑tasked across multiple jobs — an \"universal mobile worker\" thesis. Consumer/retail players in China (UBTECH, Xiaomi) emphasize customer‑facing interactions and low‑risk in‑store duties rather than heavy industrial tasks. Healthcare vendors (1X Technologies, Sanctuary AI) market safety‑certified mobility and telepresence modules alongside human‑grade compliant actuation for patient contact. Verified pilots and deployment details (2025–26) Below are the most substantiated field efforts we examined. Where direct public metrics were unavailable, summaries rely on company releases corroborated by integrators or anonymized customer interviews. Agility Robotics — Digit in warehouses Agility’s Digit saw the largest number of verified pilots in mid‑size distribution centers. Typical programs used Digit for repetitive single‑item picks from low‑shelf zones and for ergonomic off‑loads at carton induction lines. Results reported to pilot customers and integrators: Throughput: vendor materials claim up to ~25–30% cycle‑time improvement in controlled pick tasks; integrators reported 8–15% net throughput gains when factoring task switching and safety slowdowns. Human augmentation: customers used Digit to relieve overnight or late‑shift human fatigue roles rather than outright headcount elimination. Pain points: 4–6 hour effective runtime on current battery modules, late‑stage failures in cluttered pallet aisles, and need for bespoke end‑effectors for irregular items. Tesla — Optimus internal production trials Tesla continued internal trials emphasizing material handling and kitting on vehicle assembly lines. Tesla’s internal pilot focus was the operator‑adjacent helper (parts feed, cable management) rather than full autonomation. Independent observers and ex‑staff we spoke to described iterative trials with constrained cells and strong safety oversight; Tesla did not release public throughput numbers. The real value cited internally: reduced exposure of humans to repetitive, awkward positions and fewer ergonomic injuries. Apptronik and Figure AI — factory assistance Both companies ran line‑side assistant pilots in electronics and light‑vehicle component production. Deployments favored enclosed, dedicated cells where robots performed feed/hand‑off to humans. Pilot leads reported: Stable performance on repetitive insert/kitting jobs; however, dexterous, variable assembly remained a human domain. ROI case hinged on high‑value parts and far‑shifts where labor is scarce; otherwise TCO stayed unfavorable versus cobots. UBTECH and Xiaomi — retail/consumer trials in China UBTECH and Xiaomi focused on front‑of‑store demos: shelf scanning, simple customer interactions, and temperature checks. These pilots emphasized PR and customer engagement over hard productivity gains. Pricing models were a mix of leasing and retail (consumer) SKUs with service subscriptions for cloud updates. 1X Technologies (NEO/EVE) and Sanctuary AI — healthcare pilots Healthcare pilots remained the most cautious and regulated. 1X reported test deployments of NEO/EVE units for non‑clinical fetch‑and‑carry and logistics inside hospitals, while Sanctuary AI piloted mobile telepresence and patient‑assistance concepts in long‑term care homes. Outcomes included: Safety record: no major incidents reported in supervised pilots, but close human oversight remained mandatory. Effectiveness: meaningful time savings for staff in supply runs and non‑clinical errands; direct patient handling was limited to assistive, not primary, roles. Comparative outcomes and measurements Metric Warehousing Manufacturing Healthcare Typical claimed throughput lift 8–30% (vendor vs integrator delta) 5–20% on line‑side kitting Time savings for non‑clinical errands: 10–25% Error / incident rate Initial uptick during integration; stabilized with cell design Low in dedicated cells; higher in mixed tasks No major safety incidents in pilots; strict supervision required Labour impact Augmentation > displacement in first wave Augmentation, re‑tasking of skilled operators Staff reallocation to clinical tasks; no mass displacement Technical and operational shortfalls — and adaptations Independent test labs and pilot teams converged on a short list of recurring limitations and the practical fixes that made pilots succeed: Battery/runtime — Typical runtimes of 4–8 hours forced shift choreography or hot‑swap charging docks. Successful deployments used scheduled charging windows and spare units. Perception in clutter — Performance dropped in dense, dynamic aisles. Successful pilots imposed lane restrictions, additional floor markers, and tethered vision rigs in narrow zones. Manipulation failures — General‑purpose hands struggled with deformable objects. Workaround: task‑specific end‑effectors and gravity‑assisted trays. Interoperability — Integration with WMS/PLC required custom middleware. Systems integrators reported that software and safety‑certified APIs consumed a majority of project time and budget. “The robots perform when you design the cell around them. The surprise is how much of the cost is software and systems engineering, not hardware,” an unnamed North American systems integrator told us. Pricing context and business models Vendors used mixed pricing: capital sales, lease/subscription and robotics‑as‑a‑service bundles. Industry estimates from integrators place hardware price bands broadly in the tens‑to‑low‑hundreds of thousands USD per unit, with total deployment costs (integration, safety fencing, software, training) commonly 3–5x hardware. Payback depended on labor cost and utilization: high‑wage geographies and 24/7 tasks shortened payback to 12–36 months in favorable pilots. Practical checklist for early adopters Start with dedicated, high‑frequency tasks (single SKU picking, kitting) and design the cell to the robot rather than vice versa. Budget for systems integration and safety validation equal to or greater than hardware costs. Plan charging/shift logistics up front and provision spare units if continuous operation is required. Use hybrid human‑robot workflows: delegate exception handling to humans and routine repetition to robots. Insist on measurable KPIs (cycle time, error rate, staff reallocation) and pilot for at least 3 months to capture variability. Vendor evidence map Vendor Strongest Real‑World Evidence (2025–26) Agility Robotics Warehouse order picking, ergonomic relief in DCs Tesla (Optimus) Internal material handling, kitting in constrained cells Apptronik / Figure Line‑side assistance, enclosed cell kitting UBTECH / Xiaomi Retail engagement, shelf scanning demos in China 1X Technologies / Sanctuary AI Hospital logistics, telepresence in supervised care settings Conclusion By 2025–26 humanoid robots crossed from novelty to constrained commercial utility. The most convincing evidence sits in tasks that play to humanoids’ unique form factor — aisle mobility, human‑adjacent work, and environments where standing, balance and reach matter. However, widespread replacement of human labor is not yet supported by field data; current economics favor augmentation in high‑labor‑cost or high‑ergonomics‑risk contexts. The decisive factors for scaling will be battery advances, robust perception in cluttered environments, dexterous end‑effectors for variable goods, and streamlined integration into operational software stacks.