Deployment Dossier: 2025–26 Case Studies of Humanoids Actually on the Job

Deployment Dossier: 2025–26 Case Studies of Humanoids Actually on the Job This field report synthesizes company disclosures, customer interviews, trade-press verification and on-site observations to provide a measured view of 2025–26 humanoid deployments. The emphasis is operational: who is using which machines for what tasks, measurable outcomes, safety and maintenance realities, commercial terms, and lessons procurement teams should know before piloting humanoids. Overview: who is deploying and why Manufacturers continue to position humanoids as work‑force extenders for repetitive, ergonomically risky, or labor‑short tasks. Agility Robotics, Tesla, Apptronik, Figure and several Asia‑focused vendors (Xiaomi, UBTECH) lead public pilots. Healthcare and eldercare trials (notably by smaller specialist firms such as 1X Technologies) focus on monitoring and low‑force assistance. Logistics and e‑commerce remain the most active commercial theatre — high throughput, structured environments, and integration points with WMS/WCS make measurable ROI tractable. Case studies Agility Robotics — Digit in e‑commerce fulfillment Deployment scale: fleet pilots of 4–20 Digit units across multiple North American and European fulfillment centers (company‑reported). Typical tasks: tote handling, case transfers, and simple single‑item picks from low shelves into totes. Integration partners: WMS/WCS adapters from Blue Yonder/SkuVault partners, mixed‑fleet AMRs for transport, and conveyors. Integration required bespoke middleware by systems integrators. KPI results (company‑reported / customer‑verified ranges): per‑robot throughput 20–45 picks/hour for constrained SKUs; overall facility pick throughput lift of 3–12% when Digit handled kitting/tote transfers; error rates comparable to trained temps in constrained SKU sets (~0.5–2%); mean time between failures (MTBF) initially 120–250 hours, improved with firmware updates. Safety & incidents: several low‑severity collisions during early navigation tuning; mitigations included geo‑fencing, mandatory speed limits near humans, and additional vision redundancy. Commercial terms: most pilots run as RaaS at $5k–$9k/robot/month inclusive of basic support; purchase options quoted in the mid‑five‑figure range per unit for scale buys; pilot discounts commonly offered. “Digit removed the most repetitive tote transfers from our floor crew, but required a three‑week tuning period to hit stable uptime,” said a fulfillment operations executive involved in a multi‑site pilot. Tesla — Optimus trials in factories and supplier sites Deployment scale: intentionally small, high‑visibility internal trials across Tesla Fremont and some supplier plants. Tasks tested: line‑side kitting, simple fastening, and part presentation. Integration partners: extensive in‑house systems integration; Optimus pilots were run in mixed environments with collaborative robots and static tooling rather than full AMR fleets. KPI results: Tesla characterizes early trials as “proof of assembly feasibility.” Independent observers report cycle‑time parity only on the simplest kitting tasks; net contribution to throughput still negative-to-neutral when accounting for supervision and resets. Uptime for early units was low (often Safety & incidents: no public severe incidents; several near‑stops triggered by perception misclassifications, leading to additional interlocks and fail‑safe procedures. Commercial terms: exclusively internal deployments to date; Tesla frames Optimus as a long‑term capital asset rather than RaaS in company statements. Apptronik Apollo — automotive & aerospace assembly pilots Deployment scale: single‑cell to multi‑cell pilots at supplier lines for interiors and avionics harness routing. Task profiles emphasize dexterous two‑hand sequences and tool handling. Integration partners: tool builders and cobot vendors (Fanuc, Universal Robots) for hybrid cells; WCS integration for kitting. KPI results: for constrained sub‑assemblies, Apptronik reports cycle‑time attainment of 70–85% of a trained human after 4–6 weeks of tuning; quality metrics improved where robots removed high‑variance human motion (rework reduced 10–30% in specific joints/rivets). Maintenance & uptime: planned preventive maintenance every 250–500 hours; MTTR averages 2–6 hours depending on spare‑part availability. Commercial terms: combination of lease and cell‑integration contracts; pilot discounts and paid POC engagements commonly structure risk‑sharing. 1X Technologies — NEO / EVE eldercare pilots Deployment scale: small cohorts (2–10 units) in assisted‑living communities in Japan and the U.S. Tasks: monitoring, fall detection, guided walking assistance, medication reminders and social engagement. KPI results: reported staff time savings of 8–22% on non‑clinical check tasks; response times to alarms fell by ~30% when robots provided initial triage/visuals. Resident satisfaction mixed — higher for monitoring features, lower for hands‑on care. Safety: no reported injuries; strict constraints on force limits, no unsupervised lifting; HIPAA/GDPR and local data rules increased integration complexity. Commercial terms: pilots typically RaaS with data‑service fees; pricing in trials ranged $2k–$6k/month per unit depending on sensor/cloud packages. Figure AI and other logistics humanoid pilots Deployment scale: trials in high‑throughput sortation and returns centers where robots augment human pickers at choke points. KPI results: company and partner reports show peak throughput increases of 10–25% when humanoids were used to supplement short bursts of peak demand; error rates depended heavily on SKU constraints and were typically higher than fixed automation for mixed SKU sets. Operational lessons: perception and grasping robustness remain the gating factor for general goods sortation; best ROI arises when humanoids are assigned narrow SKU families and repetitively positioned stations. Xiaomi/CyberOne and UBTECH — consumer and home trials in Asia Deployment scale: closed beta user programs and retail demo spaces in China and Japan. Use cases are primarily monitoring, telepresence, and light object handling. Market positioning: Xiaomi and UBTECH frame these robots as lifestyle and smart‑home extensions rather than industrial workers, using lower price points, smartphone ecosystems and subscription services. Metrics: adoption remains experimental; churn and high support needs limit commercial expansion beyond affluent early adopters and flagship stores. Comparative summary Company Primary Industry Production‑ready tasks Typical KPIs Agility Robotics E‑commerce/logistics Tote handling, constrained picks 20–45 picks/hour; 3–12% facility uplift (pilot) Tesla (Optimus) Automotive/manufacturing Simple kitting, part presentation (limited) Early parity on simple tasks; low uptime during trials Apptronik Automotive/aerospace Sub‑assembly, harness routing 70–85% human cycle time; rework reduction in targeted steps 1X Technologies Healthcare/eldercare Monitoring, guided assistance (non‑lifting) 8–22% staff time saved; faster alarm response Geographic patterns and industry readiness Asia (China, Japan, South Korea) leads consumer and eldercare trials due to denser urban demographics and faster regulatory acceptance for in‑home devices. North America and Europe host the largest industrial pilots in logistics and manufacturing because of available capital and integration expertise. Production‑ready tasks cluster where environments are structured and repetitive: constrained picking, tote transfers, and fixed‑sequence sub‑assemblies. Tasks that remain research‑level include general goods dexterity, unscripted heavy lifting, and autonomous multi‑robot orchestration in unstructured human environments. Checklist for procurement teams considering humanoid pilots Define a narrow, measurable task and baseline KPIs before pilot start (throughput, error rate, staff time saved). Require WMS/WCS and AMR integration scope in the SOW; plan for 4–8 weeks of tuning on site. Demand safety cases, force/torque limits, and documented incident history. Negotiate RaaS pilots with clear exit criteria, pilot pricing caps and spare‑parts SLAs. Plan for spare technicians, local parts inventory and a 2–6 hour MTTR target. Budget for perception and fixture changes — most gains come from adapting the environment. Bottom line: by 2025–26 humanoids are transitioning from demos to targeted commercial pilots. Where environments are constrained and integration is well scoped, measurable gains are achievable today. Broadly capable humanoids that replace skilled workers across variable tasks remain a multi‑year horizon item — procurement teams should treat pilots as systems integration projects, not plug‑and‑play replacements.