Legs on the Line: Inside 2025–26 Commercial Humanoid Deployments

Executive summary Through 2025–26 the commercial humanoid landscape moved from research demos to targeted, field-driven pilots. The most consequential deployments have concentrated on warehouse & logistics, light manufacturing/line support, healthcare & eldercare trials, and retail/guest-service demos. Companies are positioning humanoids not as wholesale replacements for humans but as flexible, mobile manipulators that bridge gaps between fixed automation and humans. This report catalogs the leading pilots, evaluates task fit and early outcomes, and provides practical guidance for operations leaders selecting vendors. Where the legs are used: industry map and tasks Use cases that have reached real-world pilot scale fall into five buckets: Warehousing & logistics — case picking, conveyor-side work, order-staging and last-mile handoffs. Manufacturing line support — parts kitting, light assembly, machine tending and cobot collaboration. Healthcare & eldercare — non-clinical assistance, payload carriage, guided mobility and companion services in care facilities. Retail & guest services — reception, product demos, and guided customer interactions in controlled environments. Consumer/home trials — early demonstrations and limited home pilot programs focused on perception, voice interaction and fetch tasks. Task fit — what humanoids can realistically do in 2025–26 Picking: best for large, rigid items in constrained pick zones; two-handed grasp remains nascent. Conveyor-side work: strong fit where mobility and simple pick/place are needed between stations. Parts kitting: valuable when human dexterity is still required but travel time reduction matters. Patient assistance: effective for non-clinical tasks (carrying trays, fetching supplies, guided walking with supervision). Reception/retail: effective for novelty, signage, and guided navigation; full customer-service replacement not yet demonstrated. Vendor snapshots, pilots and early outcomes The following summaries synthesize vendor briefings, public disclosures and operator reports observed in field deployments through 2025–26. Agility Robotics — Digit (warehousing & logistics) Agility’s Digit has been visible in multiple order-fulfillment and distribution pilots. Operators report Digit’s mobility and two-arm form factor excel when tasks require walking between static stations (conveyor pick/pack and pallet staging). Reported pilot outcomes vary: throughput uplifts commonly reported in the mid-single digits to low double digits (5–20%) depending on site layout; downtime is dominated by perception and grasp failures when inventory diversity increases. Deployment model: direct pilots with in-house automation teams or through integrator partners. Pricing: pilots typically structured with pilot fees and transition to RaaS or purchase options; vendors decline to publish uniform pricing but proposals observed in the field range from pilot fees (~$50k–$200k) to RaaS leases ($5k–$25k/month per unit) depending on service scope. Figure and 1X Technologies (NEO/EVE) — order-fulfillment & last-mile Figure AI and 1X have emphasized full-stack autonomy for order-fulfillment zones and last-mile handoffs. Early pilots place their humanoids at the conveyor or pack-station boundary where robots can standardize handoffs into automated sorters or delivery tethers. Reports show reduced manual touches per order and lower ergonomic strain for workers; however, integration cost with existing WMS and sorter systems remains the gating factor. Apptronik — Apollo (cobot proofs on assembly lines) Apptronik’s Apollo has been trialed as a co-bot assisting human assemblers with repetitive insertions, screw-driving and payload handing. Lines that adopted Apollo reported more consistent takt times and a reduction in single-operator fatigue tasks. Apptronik markets Apollo as a collaborative bridge between industrial robots and human workers — emphasizing safety, control APIs, and factory-floor certifications. Sanctuary AI — Phoenix (care-facility assistance) Sanctuary’s Phoenix has been trialed in eldercare and assisted-living settings where companies positioned the humanoid to perform routine, non-clinical support (deliveries to rooms, guided walking with a staff attendant, and cognitive-engagement exercises). Facilities that trialed Phoenix reported increased staff availability for higher-skill caregiving, though liability and clinical-regulatory review prolonged deployments. Operators emphasize the need for rigorous safety interlocks and supervised operation models. Xiaomi CyberOne, UBTECH — consumer/home and retail demos Xiaomi and UBTECH have focused on consumer and retail demonstrations: in-store product showcases, concierge demos, and limited home trials emphasizing human-like interaction. These are largely marketing- and experience-driven pilots; both vendors pitch humanoids as consumer-facing brand ambassadors rather than industrial workhorses. Unitree and hardware-first builders — low-cost pilots Several smaller hardware-first firms and research spinouts have run low-cost, targeted pilots in warehouses and campus environments. Their value proposition: lower capital cost and focused functionality (e.g., walking to transfer boxes). Outcomes are mixed — lower price points accelerate experimentation but often require more in-house integration work. Tesla Optimus and Boston Dynamics’ Atlas — factory pilots and benchmarks Tesla’s Optimus has been showcased internally on assembly lines and in limited public demos. Tesla positions Optimus as a future volume robot that will reduce labor cost for repetitive tasks, but public data on sustained, external commercial pilots remains limited. Boston Dynamics’ Atlas remains the industry technical benchmark — a research platform that informs perception, balance and dynamic locomotion but is not marketed for commercial deployments. Early pilots consistently report one-to-three operator hours freed per robot per shift in targeted tasks, but uptime and error rates vary widely with environment complexity. Pilot metrics, uptime and operator feedback Across pilots the broad patterns are: Throughput improvement: site-reported gains range from 5% (complex picks) to 30% (well-structured conveyor tasks). Error rates: pick and place failures range from low single digits to >10% in high-mix SKUs; perception and grasp are the main failure modes. Uptime: mature line-support pilots report 80–95% uptime for scheduled shifts; more experimental warehouse pilots see 60–80% uptime due to tuning and retraining needs. Operator feedback: positive on ergonomic relief and reduced walking time; negative on maintenance burden, calibration time, and edge-case failures. Pricing, contract structures and deployment models Three contract archetypes have emerged: Pilot fee + integration: fixed pilot engagement (often $50k–$200k) covering customization and evaluation. RaaS/lease: monthly subscriptions covering hardware, software updates and service ($3k–$25k/month depending on SLA and onsite support). Perpetual sale: outright purchase for integrators and enterprises with internal maintenance (wide price dispersion; vendor-disclosed list prices remain uncommon). Integrator-led rollouts are common where vendors lack scale in systems integration; enterprises with automation centers of excellence sometimes run in-house pilots before scaling. Competitive analysis — who fits which use case Vendor Best-fit industries Strengths Limitations Agility Robotics (Digit) Warehousing & logistics, conveyor work Purpose-built for warehouse gait, two-arm manipulation Perception/grasp limits in high-mix environments Figure / 1X (NEO/EVE) Order-fulfillment, last-mile handoffs Full-stack autonomy, sorter integration focus Integration with legacy WMS is costly Apptronik (Apollo) Light manufacturing, assembly line co-bot Cobot-safety, factory certifications Lower mobility compared with warehouse specialists Sanctuary AI (Phoenix) Healthcare & eldercare trials Human-in-the-loop assistance focus Regulatory and liability hurdles; supervised operation required Xiaomi, UBTECH Retail/demo, consumer trials Customer-facing interaction, product ecosystems Not optimized for industrial durability Tesla Optimus Factory automation (future) Ambitious cost-targeting, vertical integration Limited external commercial data; maturity still emerging Boston Dynamics (Atlas) Research & benchmarking Leading dynamic locomotion Not a commercial work platform Practical guidance for operations leaders Start with high-repeatability, constrained tasks (conveyor-side, kitting islands) where perception demands are limited. Budget for integration — expect SI or internal robotics teams to consume substantial effort for WMS/ERP coupling. Negotiate SLAs that tie payments to demonstrable throughput or uptime milestones; include clear escalation paths for maintenance. Plan human supervision models: early deployments benefit from a single technician supervising several robots rather than fully autonomous operation. Run 6–12 month pilots to capture lifecycle metrics (mean time between failures, mean time to repair, operator acceptance) before scaling. Conclusion — which vendors look best suited, by use case By 2025–26 the field shows specialization. Agility and Figure/1X are the best current fits for warehouse and conveyor-adjacent logistics tasks; Apptronik is strongest where factory-cobotics safety and certification matter; Sanctuary leads early trials in care settings where supervised assistance is the objective; Xiaomi and UBTECH succeed at retail and consumer-facing pilots. Tesla and Boston Dynamics play different roles — Tesla as a potential future-volume supplier and Boston Dynamics as the technical benchmark. Humanoid robots are commercially meaningful today only in narrowly scoped deployments. For operations leaders, disciplined pilot scoping, realistic KPIs, and candid assessment of integration burden are the most reliable paths to extracting value from legs on the line.