On the Ground: Where Humanoid Robots Are Doing Real Work in 2025–26

Executive summary Humanoid robots moved from lab demos to real-world pilots in 2025–26, but their value is highly task- and environment-dependent. Across warehouses, manufacturing floors, clinics and care homes, vendors have started multi-month pilots and limited deployments that show clear value in a narrow set of operations — repetitive, two‑handed material handling in structured environments and inspection/interaction tasks that benefit from a human-form factor. At the same time, uptime, throughput and integration costs frequently fall short of vendor marketing claims. This on-the-ground review catalogues confirmed pilots and public deployments, compares platforms, and summarizes verified performance, integration friction and likely ROI timelines. Where humanoids are actually being used Warehousing & fulfillment: picking, packing and tote-to-pallet material handling in structured aisles or conveyor-fed workstations. Manufacturing: light assembly, parts kitting, inspection, and simple tool use in testbeds and pilot lines. Healthcare & assisted living: patient transfer assistance, social interaction/companion functions, and pharmacy restocking trials. Last-mile & support: point-to-point delivery and site inspection using lower-cost biped or non-humanoid platforms as complementary devices. Company profiles, verified pilots and results Agility Robotics — Digit (logistics & factory pilots) Agility’s Digit is the most visible workhorse in logistics trials. Publicly announced pilots and on-the-ground vendor briefings indicate deployments with regional 3PLs and factory integrators focusing on box handling, conveyor interfacing and feeder tasks. Operators quoted in public materials describe Digit as useful for night-shift pick-and-place and conveyor-side handoffs where tasks are repeatable and fixtures are standardized. Performance overview reported in vendor materials and pilot summaries: Uptime during trials: often cited in the 70–85% range once on-site systems were mature; initial integration weeks see frequent downtime. Cycle time: 20–40% slower than experienced human operators on unstructured picks; parity reached in fixtured, repetitive tasks. Error rates: mis-grasps and failed handoffs reduced to low single digits after vision retraining for specific SKUs. Figure AI — fulfillment and packing trials Figure positions its humanoid around high-speed fulfillment and claims rapid cycle capability. Public demos and trade-show pilots focused on box-to-tote and packing workflows. Independent integrators report that Figure’s strength is in dense, conveyor-fed packing stations where the robot works within tightly controlled kinematic envelopes. 1X Technologies — NEO/EVE in healthcare & assisted living pilots 1X has emphasized patient-facing use cases. Trials publicized by the company and partner organizations include mobility assistance and companion functions in assisted-living facilities. Early results: companion engagement metrics are promising for resident satisfaction, but physical patient handling remains highly constrained by safety protocols and requires human supervision. Sanctuary AI — Phoenix in manufacturing testbeds Sanctuary’s Phoenix has been used in controlled manufacturing testbeds for light assembly and inspection. The company markets Phoenix as a generalist that can be trained to perform a range of human tasks. Testbed results show good repeatability for short-cycle assembly steps, but integration with tooling and fixturing is a significant cost driver. Apptronik — Apollo demos in manufacturing and space-adjacent roles Apptronik markets Apollo both as a factory-assistant and as a candidate for space-rated humanoid development. Field demonstrations with aerospace suppliers and manufacturing partners have focused on payload handling, simple assembly and cable routing tasks. Apollo’s modular joints and industrial connectors make it attractive for suppliers that need a predictable payload envelope. Unitree — lower-cost platforms for last-mile & site support Although Unitree’s machines are primarily quadruped, providers and fleet operators have used them as low-cost alternatives for last-mile support, inspection and complementary material movement in facility perimeter tasks. Unitree’s lower price point positions it as a pragmatic step for operators not yet ready to invest in full humanoids. Xiaomi (CyberOne) and UBTECH — consumer & eldercare trials Xiaomi and UBTECH are experimenting with consumer-facing humanoids and eldercare trials in China. These pilots skew toward social interaction, remote monitoring and simple assistance (bringing objects, reminders). Companies emphasize user experience and safety rather than heavy-duty material handling. Integration, operations and verified KPIs Across pilots, three consistent integration challenges emerged: Safety and certification: meeting local workplace safety standards and obtaining clearance for patient-handling tasks adds months to deployment timelines. Perception & vision in clutter: pickups in unstructured bins or crowded shelves produce the majority of operational errors. Fleet orchestration & software: integrating humanoids into WMS/ERP and AMR fleets requires custom middleware and often a 6–12 month integration sprint. Typical KPI ranges observed (from vendor materials, pilot summaries and operator reporting): Operational uptime: 65–90% after stabilization (lower during early integration). Throughput vs human: 60–110% depending on task structure (best parity in repetitive, fixture-based tasks). Error/mispick rate: 1–8% in matured pilots; higher in novel SKUs or crowded shelves. Pricing, business models and ROI Most humanoid companies sell robots as part of an integrated solution rather than as simple hardware units. Common business models： Hardware + integration contracts (one-time integration + annual maintenance). Robot-as-a-Service (RaaS) for pilots, with monthly fees covering hardware, software and managed services. Publicly disclosed pricing is rare. Broad industry context: lower-cost quadrupeds and simple mobile platforms can be procured for low five-figure sums, while full humanoid deployments — when sold as integrated solutions — are commonly structured around six-figure total engagement costs for initial pilots (hardware, systems integration, safety validation). Typical ROI windows vendors present range from 18–48 months, with shorter payback when robots enable night shifts or reduce agency labor premiums. Competitive comparison (summary) Company / Robot Primary pilots Deployment stage Pricing disclosure Noted strengths Agility Robotics / Digit Logistics, factory handoffs Multi-site pilots Not publicly disclosed Proven balance and box handling Figure AI Fulfillment/packing Demo & pilot stations Not publicly disclosed High-cycle packing scenarios 1X Technologies / NEO & EVE Assisted living, clinics Early pilots Not publicly disclosed Human-facing interaction Sanctuary AI / Phoenix Manufacturing testbeds Testbed & pilot lines Not publicly disclosed Generalist skill training Apptronik / Apollo Manufacturing, space demos Demonstrations & pilots Not publicly disclosed Industrial-grade joints Unitree Last-mile support, inspection Commercial low-cost deployments Some platforms priced mid-thousands Low-cost, pragmatic entry Xiaomi / CyberOne, UBTECH Consumer, eldercare trials Consumer pilot programs Not publicly disclosed User experience, social features Short case studies: ROI timelines & failure modes Case A — Warehouse night-shift packing pilot Situation: A regional e‑commerce 3PL ran a 90‑day pilot swapping a single packing station to a humanoid-equipped, conveyor-fed cell. Outcome: throughput reached parity with daytime human performance after 8 weeks; uptime averaged 78% over the pilot. ROI: projected payback 30–36 months when factoring in premium night-shift labor reduction. Failure modes: intermittent vision failures on mixed-sized polybags and unexpected tote orientations. Case B — Assisted living companion pilot Situation: An assisted-living facility trialed a humanoid companion for resident engagement and medication reminders. Outcome: engagement metrics and satisfaction improved; physical assistance tasks were restricted by safety protocols, requiring human backup. ROI: qualitative benefits cited for retention and brand differentiation; quantitative ROI unclear. Failure modes: battery life and network connectivity affecting reliability. Regulatory and labor implications Regulatory frameworks lag deployments: safety certification for patient handling and human-robot workcells adds months and cost. On labor, pilot sites report reassigning workers from repetitive tasks to supervision, quality and exception handling — but unions and operators remain wary of aggressive headcount reductions. Short-term labor economics favor humanoids where labor is expensive or dangerous, or where human workers are in shortage. Conclusion — what humanoids add value to today In 2025–26, humanoid robots add measurable value in structured, repeatable tasks that leverage a human form factor: conveyor handoffs, fixture-based light assembly and social/monitoring roles in care settings. They are not yet universal drop-in replacements for complex unstructured picking or heavy patient transfers. Expect continued improvements, but also persistent integration costs and safety validation timelines that shape where, when and how operators adopt humanoids. Note: This report is based on public company disclosures, on‑the‑ground vendor demonstrations, pilot summaries and industry reporting collected by The Humanoid Race. Where independent third‑party verification was unavailable, claims are described as company statements or pilot reports.