Walking the Floor: Inside 2025–26’s Real-World Humanoid Deployments

Executive snapshot Through 2025–26 humanoid robots moved from staged demos to limited, monitored workfloors. Logistics and light-manufacturing pilots dominate deployments; healthcare and consumer trials are maturing but remain small and supervised. This field report synthesizes company disclosures, pilot summaries shared with The Humanoid Race, and public materials to map who is actually walking the floor—and what they reliably do. Where humanoids are being piloted and why Warehouses & logistics: picking, tote handling, sortation and parts transport where free-form legged mobility and human-scale manipulation are advantages. Manufacturing lines: kitting, light assembly and parts handoff that require human reach and dexterity but are repetitive and ergonomically challenging for workers. Hospitals & eldercare: caregiving assistance, longitudinal presence, and low-contact logistics (linen, meal trays) under nursing supervision. Homes and R&D: smart-home integrations, companion trials, and mobility research labs testing hybrid human–robot workflows. How companies position their robots Figure AI: marketed to logistics operators as "warehouse-first humanoids"—focus on integration into existing conveyor and sortation systems with a SaaS + robot subscription model. Agility Robotics (Digit): pitched as a bolt-on partner for existing robotic fleets—mobile, bipedal pick-and-place in mixed human environments. Tesla (Optimus): positioned internally as a factory worker replacement for repetitive, ergonomically taxing tasks; public narrative emphasizes scale and future cost-per-unit advantages. Apptronik & 1X Technologies: focused on manufacturing kitting/light-assembly with payload-capable manipulators and deterministic motion suited to automotive/electronics. Sanctuary AI (Phoenix): selling long-duration caregiving trials that stress safety, explainability, and human-like interaction as keys for eldercare adoption. UBTECH & Xiaomi (CyberOne): targeting consumer/home markets—voice, smart-home integration and social interaction over heavy manipulation. Boston Dynamics & Unitree: marketed mainly through research/adjunct solutions—hybrid setups (Atlas research work; Spot + human teleop) or low-cost adjunct roles for proof-of-concept. Case studies: pilots, metrics and business context Figure AI & Agility Robotics — logistics picking and sortation Both companies reported multi-site pilot programs in 2025 involving e-commerce 3PLs and parcel sortation centers. The pilots shared common themes: Tasks: mixed-SKU tote picking, case-to-conveyor handoffs, replenishment of pick faces. Uptime: company-reported operational uptime ranged 72–88% during supervised day shifts; downtime largely due to perception edge-cases and battery swaps. Throughput: pilot throughput equated to 20–45 picks/hour per humanoid depending on SKU complexity—roughly 30–60% of an experienced human picker on diversified SKUs, but competitive on repetitive, large-item picks where reach matters. Integration effort: 4–12 months including custom gripper development, WMS integration, and safety validation; systems integrators (several regional integrators and the logistics partners' automation teams) were core to deployments. Safety incidents: no major injuries reported in public pilots; most incidents were navigation stalls and one fall per several thousand hours in early trials. “Digit shortened ergonomic tasks and freed seasoned pickers for exception handling—time-to-value in our pilot was marginally under two years on conservative ROI models,” an anonymized logistics operations director told The Humanoid Race. Tesla Optimus — factory trials at Gigafactories Tesla’s public and internal statements have shown Optimus prototypes performing simple transport and positioning tasks. In 2025–26 several limited factory trials focused on: Tasks: parts transport between stations, simple assembly assist (holding, aligning), and repetitive placement that reduced worker bending and carrying. Metrics: Tesla reports tasks completed on fixed cycles with near-robotic repeatability; external verification is limited. Uptime in factory trials reported above 85% in scheduled runs but required on-floor supervision and frequent human intervention for exception handling. Business model: currently internal deployment cost-offset vs. headcount; Tesla frames Optimus as long-term scale play rather than immediate resale product. Apptronik Apollo & 1X Technologies (NEO/EVE) — automotive and electronics kitting Apptronik and 1X positioned their humanoids for predictable, conveyor-fed kitting and light-assembly tasks: Throughput: 40–70 kit placements per hour in validated cell conditions; better than general-purpose humanoids where repeatability and fixed fixtures are exploited. Integration: preferred model uses cobot safety cages, fixed part feeders, and PLC integration—deployment times reported 3–9 months. ROI: clients in electronics reported 18–36 month payback when replacing highly repetitive, ergonomically risky roles. Sanctuary AI (Phoenix) — eldercare longitudinal trials Sanctuary’s Phoenix has been trialed in supervised long-duration shifts in multiple eldercare facilities. Focus areas: Use cases: supervised social interaction, patient monitoring, non-contact delivery (meals, medication trays) and mobility-assist reminders. Uptime & safety: system uptime in trials averaged 78% across 24/7 monitoring schedules; safety emphasis and human override were mandatory. Reported incidents were limited to navigation bumps and false alarms triggered by motion sensors. Outcomes: facilities reported improved resident engagement metrics and nursing time reallocation, but no facility claimed full headcount substitution—humanoids were adjunct, not replacements. UBTECH & Xiaomi (CyberOne) — consumer/home trials UBTECH’s Walkers and Xiaomi’s CyberOne prototypes appeared in curated home trials and smart-home integration pilots: Focus: voice-driven assistance, remote monitoring, and smart-home orchestration rather than heavy lifting. Consumer metrics: trial households reported novelty value high but utility limited to scheduled reminders and telepresence; uptime depends on charging behaviors and Wi‑Fi robustness. Boston Dynamics & Unitree — research and adjunct roles Boston Dynamics’ Atlas remains research-focused; Spot has been widely deployed in inspections, and hybrid projects use Spot or Atlas in tandem with human operators for mobility-challenging tasks. Unitree’s low-cost platforms are used as adjunct systems for proof-of-concept and non-safety-critical roles where budget constrains adoption. Pricing, business models and competitive positioning Company/Model Primary model 2025–26 pricing context Figure AI RaaS + integration Subscription and integration fees; pilots reported monthly fees in the low five-figures (company-reported ranges) Agility (Digit) Sale or lease + SI Capex-equivalent pricing for enterprise customers; market estimates place enterprise humanoids in the mid-six-figure range or RaaS alternative Tesla (Optimus) Internal first, future sale No public list price; Tesla projects long-term cost parity assumptions via scale Apptronik / 1X Sale + SI Target pricing optimised for industrial customers; ROI claims focused on 18–36 month payback Sanctuary (Phoenix) Pilot agreements + service Facility pilots under multi-year service contracts; pricing structured to share risk with providers Competitive takeaways Specialized depth wins: purpose-tuned systems (Apptronik in kitting, Figure for tote handling) outcompete generalist humanoids in integration speed and throughput. Systems integrators are critical: few customers adopted humanoids without SI partners; integration effort, safety validation and WMS/PLC connections make SI costs a majority of early-deployment spend. Price and business model matter: RaaS lowers adoption friction; outright purchases plus SI are realistic only for large-scale, risk-tolerant customers. Lessons learned — what humanoids can and cannot do reliably in 2025–26 Can do reliably: repetitive kitting/lift-assist in constrained cells, supervised tote handling in warehouses, non-contact delivery in care facilities, and social/telepresence roles in homes when mobility is simple. Struggle with: high-mix small-part picking that requires rapid dexterous regrasping; extended unstructured interactions without human oversight; rapid reconfiguration without SI support. Safety & trust: humanoids require conservative speed/sensing envelopes on shared floors, lowering throughput vs. humans or fixed-arm robots but improving safety. Who’s adopting first and why Logistics & 3PLs: first adopters because tasks are repetitive, facilities have modular layouts, and labor variability creates demand for automation. Electronics & automotive manufacturing: adopt for kitting and ergonomics-driven substitution where predictable motion and high repeatability deliver ROI. Healthcare/eldercare: adopt slowly with heavy regulation and human-touch requirements; early pilots focus on augmentation and presence rather than replacement. In 2025–26 humanoids are economically and operationally viable in narrowly defined roles. The firms that win will combine robust hardware, targeted task specialization, SI partnerships, and service models that share deployment risk. For the next 24 months expect incremental increases in uptime and throughput as perception and battery systems improve, but not the wholesale replacement of human labor across complex, high-mix tasks.