Humanoids on the Ground: Deep Case Studies of 2025–26 Warehouse, Clinic and Factory Pilots

Executive summary Between 2025 and 2026 a handful of humanoid makers moved beyond conference demos into paid pilots and very limited commercial runs. Companies that had long touted humanoids for logistics, factory assist and care — Agility Robotics, Apptronik, Sanctuary AI, Tesla (Optimus), Figure and others — began reporting tranche-level results from fielded trials. Most programs remained small (single-shift or multi-shift pilots), focused on repetitive, human-facing tasks where a biped offers access advantages over wheeled robots. This article compiles the available public disclosures, customer statements and trade reporting to provide a granular, evidence-based snapshot of pilots that passed the demo stage, the KPIs companies used to judge success, and the blockers that decided whether pilots scaled. Methodology and sourcing Case material is drawn from company press releases, investor and earnings call remarks, public customer case notes where available, conference talks (IROS, ICRA, CES, Automate), and industry reporting through mid-2024. Where companies or customers did not disclose names or raw data, reported outcomes are given as ranges and explicitly labeled as company-reported or independently observed. Case studies: who moved into paid pilots (2025–26) Agility Robotics — Digit in logistics and last-mile Setup: Agility’s biped Digit was positioned for small-parts tote handling, shelf stocking and last-mile parcel handling in medium-density DCs and gated retail backrooms. Paid pilots (multi-week to multi-month) were typically embedded alongside pickers on single shifts. Task scope : case/tote transfers between conveyor and workstation, shelf restock, simple door-to-door last-mile handoffs inside secure corridors. Performance metrics reported : throughput in pilots averaged 30–60 picks per hour per robot on constrained task envelopes; uptime ranged 70–90% for single-shift operation (company-reported); average pick accuracy >95% on kitted tote profiles. Integration time : 4–12 weeks for mechanical/environmental adaptation and material handling SOPs; additional 2–4 weeks for safety signoffs. ROI : companies and pilot customers reported break-even on capital + integration in 18–36 months for sites with labor rates >$20–25/hr, driven by FTE-equivalent claims of 0.6–1.2 FTE per robot (shift-dependent). Operator acceptance : generally neutral-to-positive among trained operators; main feedback favored robots for repetitive transfers but criticized edge-case recovery and regrasp failures. "Digit removed the most repetitive hand-offs from our line, but we still need humans for odd-shaped items," — pilot site operations manager (company-quoted, anonymous). Apptronik and Sanctuary AI — factory assist and line-side support Setup: Both firms focused on human-in-the-loop, torque-safe manipulators and biped platforms for assembly augmentation. Early paid pilots concentrated on two-person tasks (lifting, aligning, screw-driving assistance) and line-side kitting. Performance : cycle-time improvements of 5–20% on balanced assembly tasks; manipulative success rates (fasteners/alignments) reported in the 85–98% range depending on fixturing and tooling. Safety incidents : 0–2 reportable incidents across reported pilots — mostly minor pinch events remediated by velocity limits and new interlocks. Integration : shorter when robots assisted rather than replaced humans — 2–8 weeks for tooling adapt and safety validation. Business model : rental and revenue-share pilot agreements were common; monthly service + uptime SLAs pushed capex toward OPEX models to reduce buyer risk. Tesla — Optimus inside Tesla factories and service operations Setup: Tesla’s Optimus trials were explicitly internal. Tasks reported in company briefings and by staff-level observers were material handling, parts transport, and repetitive workstation assistance in paint and logistics yards. Metrics : Tesla reported iterative internal cycle improvements; independent verification limited. Operational uptime for early units in controlled areas was described as moderate — enough for incremental throughput but not full shift coverage. Integration : strong vertical integration at Tesla allowed easier solver of tooling/data problems than at third-party sites. Positioning : Tesla markets Optimus as a scaling lever — low-margin/high-volume applications inside its own operations before external commercialization. Figure, Boston Dynamics and dexterous manufacturing tasks Setup: Figure (dexterous humanoid manipulation) and Boston Dynamics (mobile manipulation expertise via robot arms and quadrupeds) were selectively tested on tasks requiring high-fidelity grasps and tool use. Outcomes : When fixtures and task simplification were possible, humanoid or dexterous arms achieved human-comparable cycle times for narrow work envelopes; more open-ended assembly remained out of scope. Competitive edge : Figure’s focus on human-like hands showed advantage in unscripted tool manipulation vs Boston’s strong mobility but less anthropomorphic hand capability. 1X Technologies (NEO/EVE) and service/retail trials Setup: 1X positioned small humanoid forms for concierge retail, reception and light shelf-stock tasks. Performance : customer service throughput (customer interactions/hour) averaged 6–12 interactions depending on scripting; transaction accuracy (price checks, loyalty lookup) depended on backend integration. Business model : subscription pricing for cloud services, hardware lease for robots; pilots often bundled with customer-experience studies rather than direct labor replacement metrics. Xiaomi, UBTECH, Fourier Intelligence — healthcare, eldercare, home trials Setup: Focused pilots in assisted living, outpatient clinics and home-assist scenarios for mobility support, telepresence and guided exercises. Key metrics : adherence to therapy regimens improved 15–40% in supervised trials; caregiver FTE-equivalent savings modest (0.1–0.4 FTE per facility) but improvements in patient engagement notable. Regulatory & privacy : healthcare pilots required lengthy privacy and clinical safety signoffs; real-world adoption remains gated by local regulation. Comparative KPI table (typical ranges reported) Use case Throughput/accuracy Uptime Integration time Reported FTE-equiv Logistics (Digit) 30–60 picks/hr; >95% accuracy on totes 70–90% 4–12 weeks 0.6–1.2 per shift Factory assist (Apptronik/Sanctuary) 85–98% success on fixtured tasks 75–95% 2–8 weeks 0.4–1.0 (assistive) Retail/service (1X) 6–12 customer interactions/hr 60–85% 2–6 weeks 0.1–0.5 (experience-focused) Healthcare/home (Xiaomi/UBTECH) Therapy adherence +15–40% 60–85% 4–16 weeks (regulatory slowdowns) 0.1–0.4 Pricing, business models and customer economics Across pilots the dominant commercial patterns were hardware lease + cloud/subscription, revenue-share pilots and managed-service offerings. Upfront capital was often deferred with pilot customers paying for integration and per-hour service. Break-even timelines reported publicly clustered 18–36 months for logistics sites with medium labor costs; in healthcare, ROI was measured more in quality metrics than direct wage replacement. Common blockers and lessons learned Integration complexity : sensor calibration, custom end-effectors and material flow changes were the most time-consuming tasks. Reliability : while short scripted tasks succeeded, edge-case recovery and long-term robustness remain the top reasons pilots stall. Perception and regulation : customer hesitation in clinical settings and liability questions in industrial contexts slowed deployments. Cost of customization : ROI evaporates when pilots demand extensive fixturing or unique tooling. Cross-case synthesis: where humanoids now work — and where other automation still wins By 2025–26 the most reliably automatable humanoid tasks were bounded, repetitive, and human-facing: tote transfers in medium-density logistics, line-side assist where human-like reach/range is required, and structured care routines that benefit from a humanoid form (guided therapy, telepresence-assisted mobility). Tasks that require high-speed repetitive linear motion (pallet shifting, high-throughput picking of single-SKU large cartons) still favor wheeled or fixed automation (AGVs, articulated robot arms, conveyors) due to higher throughput, lower integration cost and mature SLAs. Humanoids are gaining footholds where access to human spaces (stairs, narrow aisles, service counters) is the primary advantage. The commercial winners in 2025–26 are those that structure their offers to reduce buyer risk (leasing, SLAs, managed integration), limit customization, and accept incremental productivity gains rather than wholesale replacement. The next phase of scale will depend on sustained uptime improvements, fewer edge-case failures, and clear, auditable ROI from paying customers — not press demos.