Humanoid Robot Fleet Management: Is Your Operation Ready to Scale to 10+ Units?
Physical AI Journal Research Team — Independent analysis by Sekason Research Limited, United Kingdom
Evidence as of 16 October 2026
Scope & Disclaimer
This report, published by the Physical AI Journal and operated by Sekason Research Limited, United Kingdom, provides market and adoption intelligence only. It is not legal, financial, investment, engineering or safety-certification advice or opinion, and it makes no statement on the safety, engineering soundness or technical conformity of any robot or deployment. Naming a company or platform is not an endorsement. Company-claimed figures are labelled and have not been independently verified by Sekason Research Limited. Do not rely on this report as the sole basis for a procurement, investment or safety decision. Full terms are in the Disclaimer.
1. Executive Intelligence Synthesis
Direct answer: Public evidence does not yet support a standard operating model for humanoid robot fleet management at 10+ units, so readiness is a set of gates the operator must prove in its own pilot. This report found no confirmed operating fleet of 10+ humanoids at a single named customer in public sources (NOT FOUND in public sources). Gartner predicts that fewer than 20 companies will reach production stage in supply chain and manufacturing by 2028 (analyst estimate, Gartner; 21 January 2026; a prediction, not an observed fact).
The decision this report supports is whether to expand a humanoid pilot to 10+ units, extend the pilot, move to Robots-as-a-Service (RaaS), wait, or use conventional automation. The public record cannot settle it: the largest named-site count found is 7 units of the Digit robot at Toyota Motor Manufacturing Canada, and that figure comes from trade press, not from Agility Robotics or Toyota (reported by trade press, unverified). The operator therefore decides on five measurements taken in its own pilot, and Sections 4 and 8 set out how. The 10+ threshold is this report's decision point, not an industry benchmark. The 10+ figures that do exist are plans, orders, capacity or aggregate sales. No supervisor ratio, intervention rate, uptime percentage, spares or service-level data was found in public sources for any commercial humanoid fleet, so the operator must measure these itself and write them into contracts.
Five signals carry the verdict.
1. Scale is claimed, not documented. Agility reports more than $300 million of Digit v5 orders that are subject to contractual milestones (company-claimed). UBTech reported 921 full-size humanoids sold in the first half of 2026, with no per-site split (company-claimed). Hyundai plans Atlas deployment at its Georgia plant from 2028 (company-claimed). Treat each as a market signal, and ask any vendor for the count operating at a named site.
2. The operating model is unmeasured in public. Agility names uptime, throughput and mean time between incidents (MTBI) as KPIs of its Arc platform and publishes no values (company-claimed). Figure AI lists zero interventions per shift as a requirement on its BMW page, not as a result (company-claimed). Require measured values from the pilot, because no public benchmark can substitute for them.
3. Economics depend on productive hours, not purchase price. Published payback figures are models with undisclosed assumptions. IDTechEx models about 6 months at high utilisation and about 15 months at medium utilisation (analyst estimate, IDTechEx). McKinsey describes early-pilot payback as "often exceeding two years" (analyst estimate, McKinsey). Build the cost per productive hour from the operator's own inputs before weighing any payback claim.
4. Pilots extend in single digits. GXO describes a "small fleet" of Digit units without stating a count (company-claimed). Trade press reports 3 pilot units and 7 commercial units at Toyota Canada (reported by trade press, unverified). BMW announced 1 Figure 03 unit (company-claimed). Plan the next stage as an extension of a single-digit pilot, and fix its measurement plan before the units arrive.
5. Fleet scale shifts labour into maintenance. Amazon says its Shreveport site needs 30% more employees in reliability, maintenance and engineering roles (company-claimed; AMR analogue, not humanoid data). It is the closest public indicator that fleet scale moves labour into maintenance, and it concerns autonomous mobile robots, not humanoids. Budget maintenance and engineering headcount as a fleet cost line.
Evidence grading follows the Journal's Research Methodology, and every claim is handled under its Editorial Standards. Each figure below carries one of four grades, shown with its inline label in the key that follows. Further analysis is available from the Physical AI Journal.
Grade | What it means | Inline label used |
Independently verified | Supported by an accessible primary or independent record | (verified, [source]) |
Customer-confirmed | Disclosed by the customer, with scope recorded; none found for outcomes at 10+ units | (verified, [source]) |
Vendor-reported | Attributed to the manufacturer | (company-claimed); (company-claimed, relayed) |
Modelled or projected | Forecast or Journal illustration with assumptions stated | (analyst estimate, [firm]); (editorial estimate — illustrative assumption, not observed data) |
Other labels used: (reported by trade press, unverified); (AMR analogue, not humanoid data); (disputed); (NOT FOUND in public sources).
2. Platform / Market Landscape
Direct answer: Shipments of humanoid robots are measurable; industrial fleet deployment is not. Omdia counted about 13,000 global humanoid shipments in 2025 and IDC about 18,000 (analyst estimates, differing totals). Interact Analysis states that real-world application was about 10% of 2025 production, with academic research and entertainment dominating (analyst estimate, Interact Analysis). No verified count of humanoids operating in industrial fleets was found in public sources, so shipped units should not be read as productively deployed units.
2.1 Market structure and who supplies units
Volume leadership sits with Chinese vendors. Omdia's 2025 count attributes 5,168 units to AgiBot, 4,200 to Unitree and 1,000 to UBTech (analyst estimate, Omdia). Unitree's prospectus, as reported by state media, gives more than 5,500 humanoids shipped in 2025 (verified, SSE / China Daily, as reported from the prospectus; relayed), which conflicts with Omdia's figure (disputed). The methodology behind Omdia's count was not available. AgiBot also reports about 15,000 embodied-intelligence robots rolled out as of June 2026, a total that includes wheeled and quadruped robots and not only humanoids (company-claimed, relayed). The Western suppliers named in deployments, Agility, Figure AI, Boston Dynamics, Apptronik and Tesla, publish no verified count of units operating at a named customer site (NOT FOUND in public sources). The Journal's Quarterly Humanoid Platform Index records specifications, pricing and deployment counts by platform and is the reference for supplier-level detail.
2.2 The 10+ numbers that exist
Every 10+ figure found falls into one of four types: target, order, capacity or aggregate sale. None is an operating fleet.
Figure | Type | Source and date | Grade |
Hyundai plan for 25,000 Atlas units across Hyundai and Kia plants "over the next few years" | Target | The Robot Report, 21 Sep 2026 | Vendor-reported (company-claimed) |
Hyundai US factory able to make up to 30,000 robots a year by 2028 | Capacity | The Robot Report, 21 Sep 2026 | Vendor-reported (company-claimed) |
Agility RoboFab designed for up to 10,000 Digits a year; actual output not retrieved | Capacity | Form 425 Ex. 99.2, Jun 2026 | Vendor-reported (company-claimed) |
Agility Digit v5 orders above $300 million, subject to contractual milestones | Order | Form 425 Ex. 99.1, 24 Jun 2026 | Vendor-reported (company-claimed) |
UBTech 921 full-size units sold in H1 2026, no per-site split | Aggregate sale | HKEXnews, 28 Aug 2026 | Vendor-reported (company-claimed) |
Capacity is not output, an order is not a delivery and a sale is not a deployment. Agility's investor deck footnote states that the orders are not current-period revenue. UBTech's reported full-size revenue split puts RMB569 million under custom solutions for other industries and RMB21.247 million under education (company-claimed, relayed); the automaker share cannot be determined from it.
2.3 Shipments and analyst framing
Estimates of shipped units differ by firm, period and method, and none measures units at work.
Firm | Period | Estimate | Grade |
Omdia | 2025 | About 13,000 (AgiBot 5,168; Unitree 4,200; UBTech 1,000) | Modelled or projected (analyst estimate, Omdia) |
IDC | 2025 | About 18,000, up about 508% year on year | Modelled or projected (analyst estimate, IDC) |
Smart Analytics Global | H1 2026 | About 19,100 (AgiBot about 8,400; Unitree about 5,900); methodology not read | Modelled or projected (analyst estimate, Smart Analytics Global; relayed) |
Interact Analysis | Annual | Below 100,000 units today; real-world application about 10% of 2025 production | Modelled or projected (analyst estimate, Interact Analysis) |
Goldman Sachs reportedly projects 75,000 units in 2026 (analyst estimate, Goldman Sachs; reported secondhand). That figure compares with 2025 shipments of 13,000 to 18,000 and an H1 2026 estimate of about 19,100; no full-year 2026 actuals exist to test it. Market volume is a supplier-health signal; it does not measure how many units run productive shifts.
3. Deployment Evidence
Direct answer: Outcome evidence for humanoid robots in industrial operations exists only as vendor-reported cumulative throughput without a fleet-size denominator. Agility reports more than 100,000 totes moved at GXO (company-claimed; fleet size and start date not disclosed) and more than 65,000 operating hours across 9 customer facilities (company-claimed; the investor deck also states 8). No customer-confirmed or independently verified productivity, uptime or intervention data was found in public sources.
3.1 The evidence ladder
No deployment found reaches Production at 10+ units. Every deployment in this report is placed at one of four stages, and each stage has a test that the evidence must pass.
Stage | Test the evidence must pass | Cases in this report |
Announcement | A named party states an intent, plan or agreement | Figure 03 at BMW (1 unit arriving) (company-claimed); Hyundai Atlas plant deployment; Catalyst Brands; Schaeffler |
Pilot | Units operate on site on a defined task for a defined period, with an outcome reported | Figure 02 at BMW (robot count not stated) |
Commercial Deployment | A paid contract is in place and units operate under it, with scope confirmed | Toyota Canada (RaaS); GXO (multi-year RaaS) |
Production | Units run as an ongoing operation at 10+ with reported availability, supervision and cost data | None found in public sources |
The evidence ladder diagram shows the four stages from Announcement to Production and the test each stage must pass before a deployment moves up.
3.2 What outcome data exists and who reported it
Three outcome claims exist, and every one is vendor-authored. The Journal's Deployment Outcome Tracker follows such claims after the press release; the Humanoid Deployment Evidence Report 2026 holds the wider evidence base, and this report does not duplicate either.
Reported outcome | Reporter and date | What it cannot tell you |
More than 100,000 totes and about 98% accuracy (undefined) at GXO Flowery Branch | Agility, reported 24 Nov 2025 and 16 Sep 2026 (company-claimed) | Number of Digits, start date and period; so no rate, uptime or labour productivity |
More than 65,000 operating hours across 9 customer facilities | Agility, 24 Jun 2026 (company-claimed) | Fleet size, customer split and whether all 9 are active; the deck also states 8 |
30,000+ BMW X3 vehicles, 90,000+ parts loaded, 1,250+ runtime hours, 10-hour shifts Monday to Friday | Figure AI, 19 Nov 2025 (company-claimed) | Robot count, which is not stated; per-robot productivity |
Figure's page also lists an 84-second cycle, a 37-second load, above 99% placement and zero interventions per shift (company-claimed). These are requirements or targets on the page, not achieved results. Fortune reported on 6 April 2025 that a BMW spokesperson said only 1 Figure robot worked at the Spartanburg plant at a time, outside production hours (verified, Fortune, as reported). No GXO-origin outcome evidence for Digit was found in public sources.
3.3 The metrics no public source provides
Seven operating metrics decide whether a fleet scales, and none has a public value. Each was searched for up to 16 October 2026, and a metric without a public value cannot be benchmarked against a vendor's claim.
Metric | Status |
Supervisor-to-robot ratio | NOT FOUND in public sources |
Remote-operator ratio | NOT FOUND in public sources |
Intervention rate | NOT FOUND in public sources |
MTBI value | NOT FOUND in public sources; Arc names MTBI as a KPI without a value (company-claimed) |
Uptime percentage | NOT FOUND in public sources |
Spares and service-level terms | NOT FOUND in public sources |
Charging infrastructure per robot | NOT FOUND for any deployment; a 2024 Agility CEO target of 2 robots per charger, planned 4:1 and possibly 10:1, is stale (company-claimed) |
The practical consequence is that these seven metrics must be requested from the vendor as logged pilot data. Absence from public sources does not show that vendors lack the data; it shows that a buyer cannot compare vendors without asking for it.
4. Economics & ROI Analysis
Direct answer: Cost per productive hour equals total fleet operating cost divided by total productive robot-hours, where productive hours are completed work and exclude charging, faults, recovery, waiting and scheduled maintenance. Scale improves the result only if supervision, recovery and integration costs grow more slowly than productive hours. No public data fixes those inputs, and the payback models found, such as IDTechEx's about 6 to 15 months, do not state them (analyst estimate, IDTechEx), so the operator supplies them.
4.1 Cost per productive hour: definition, formula and cost stack
Productive hours, not powered-on hours, form the denominator of cost per productive hour, the metric used throughout this report.
Cost per productive hour = Total fleet operating cost ÷ Total productive robot-hours
The numerator is the sum of seven cost rows. The denominator counts only hours of completed work. Charging, faults, recovery, waiting and scheduled maintenance are excluded, so a robot that is powered on but idle adds cost without adding productive hours.
Cost row | Anchor in public sources | Grade | Input the reader replaces |
Acquisition, lease or RaaS fee | Digit RaaS illustration: $8,500 per month, about $100k per year to the customer, customer payback 1.1 years (company-claimed / management model — not verified market pricing). Digit v4 bill of materials about $125k, a cost and not a price (company-claimed). Manufacturing cost $30,000 to $150,000 per unit (analyst estimate, McKinsey). Unitree list prices including $4,900 (R1 Air) and $29,900 (H2) (company-claimed, relayed; not an industrial-fleet price) | Vendor-reported; Modelled or projected | Quoted fee per unit, term, escalators |
Integration and site preparation | NOT FOUND in public sources | None | Integrator quote, WMS and MES connection scope, site changes |
Supervision and remote intervention labour, counted separately | NOT FOUND in public sources | None | Supervision hours per robot-hour, loaded hourly rate |
Software and fleet platform | Illustration above describes the $8,500 as SaaS pricing per Digit (company-claimed / management model — not verified market pricing) | Vendor-reported | Fee, API and integration scope |
Maintenance and spares | Repair cost "up to $1,000" per repair, including shipping and technician labour (analyst estimate, McKinsey); spares and service-level pricing NOT FOUND in public sources | Modelled or projected | Spares kit, repair turnaround, technician hours |
Energy and charging | NOT FOUND in public sources | None | Energy per robot-hour, chargers installed |
Downtime and recovery | NOT FOUND in public sources | None | Recovery time per incident, substitution policy |
Supervision labour has its own row so that it cannot be absorbed into a lease or service fee. No volume-discount, service-contract or spares pricing was found for any vendor. The Journal's Sector ROI Model Library applies this method by sector; the method is set out here so the operator can rerun it with its own inputs.
4.2 Pricing and payback anchors
Published payback figures span from about 6 months to more than 2 years, and the spread is a reason to model, not a range to rely on.
Anchor | Value | Assumptions stated | Grade |
IDTechEx payback model | About 6 months at high utilisation (2026); about 15 months at medium; operating cost below US$5 per hour in high-utilisation industrial scenarios around 2030 | Discount rate, lifespan, wages, maintenance, integration and remote-assist costs not stated; IDTechEx says it "should not be interpreted as guaranteed profitability" | Modelled or projected (analyst estimate, IDTechEx) |
McKinsey, 30 Jun 2025 | Early-pilot payback "often exceeding two years" | Not stated | Modelled or projected (analyst estimate, McKinsey) |
Agility CEO, reported 17 Jul 2024 | "under 2 year ROI vs a human at a fully loaded $30 an hour" | A 2024 target; stale | Vendor-reported (company-claimed) |
Agility investor deck, Jun 2026 | Customer payback 1.1 years under RaaS | Unaudited illustration (company-claimed / management model — not verified market pricing) | Vendor-reported |
Average selling price | IDTechEx about US$114,700 (2024) to about US$37,000 (2030); Morgan Stanley about $200,000 today (unverified against the original) | Not stated | Modelled or projected (analyst estimate) |
4.3 Fleet-size stress test
The stress test is break-even analysis, and it takes its thresholds from the operator's cost base.
Let B be the operator's benchmark labour cost per hour,
w the loaded hourly cost of supervision labour,
s supervision hours per robot-hour,
c the non-supervision operating cost per scheduled robot-hour,
F the fixed fleet cost per scheduled hour (integration, fleet platform, site preparation),
N the fleet size and a productive availability, the share of scheduled hours spent on completed work. This is the Journal's editorial model, not observed data.
Cost per productive hour = (c + F ÷ N + s × w) ÷ a
Break-even availability a* = (c + F ÷ N + s × w) ÷ B
Break-even supervision s* = (B × a − c − F ÷ N) ÷ w
Availability a = 1 − charging share − waiting share − maintenance share − (k × t ÷ H)
Here k is interventions per robot-shift, t recovery hours per intervention and H shift hours. The grid below sets the five fleet sizes against the four drivers. The Journal supplies no observed values. Each driver is entered in low, base and high columns by the reader, and every cell is a reader input or an editorial estimate — illustrative assumption, not observed data.
Fleet size | Fixed cost per robot (F ÷ N) | Drivers (a) supervision, (b) interventions, (c) availability, (d) recovery: where low, base and high values come from | What the break-even answers |
1 unit | F carried by one robot | Measured in the pilot | Does the single-robot case clear B at all? |
5 units | F ÷ 5 | Measured in the pilot | Does fixed cost per robot fall faster than supervision rises? |
10 units | F ÷ 10 | Extrapolated from the pilot; confirm in an extended pilot | Is 10+ justified at the measured drivers? |
20 units | F ÷ 20 | Extrapolated; no public analogue found | Which driver breaks first as coordination load grows? |
50 units | F ÷ 50 | Extrapolated; no public analogue found | Does recovery time or supervision set the ceiling? |
The break-even chart plots cost per productive hour against fleet size from 1 to 50 units for the operator's low, base and high driver inputs, and marks where each curve crosses the operator's benchmark labour cost.
The scaling risk is that added units raise coordination and supervision cost faster than they raise throughput. Dividing F across more robots lowers cost only while s and the recovery term in a hold steady. The only ratio-like statement found is a hypothetical:
Sankaet Pathak, CEO of Foundation, said that under pure teleoperation
"one robot running three shifts can imply three human operators"
(company-claimed, relayed by Sacra, 9 March 2025; an opinion, not a measurement).
4.4 Buy, lease, RaaS or wait
Each model moves different risks to the buyer, and the contract term that protects the buyer differs by model.
Model | Transfers to the buyer | Contract term that protects the buyer |
Buy | Capital, maintenance, upgrade risk, exit cost | Acceptance tests tied to measured availability; spares and response-time terms |
Lease | Fixed term cost, maintenance as contracted | Early exit and upgrade clauses; replacement-unit commitments |
RaaS | Recurring fee; least capital | Fee linked to logged productive hours or committed KPIs; data-export rights |
Wait | Delay cost; no learning from a live pilot | Re-test date tied to a vendor availability milestone |
Timing is an input. Agility describes Digit 5 as still in development, with early access in the first half of 2027 and general availability by the end of 2027 (company-claimed); features and specifications may change. An operator choosing between RaaS now and waiting should price the cost of waiting for that general availability date against the value of learning from a live pilot.
5. Named Deployment Case Studies
Direct answer: Four named cases exist and none reaches a documented 10+ operating fleet: BMW with Figure AI (Figure 02 Pilot; Figure 03 Announcement), Toyota Canada with Agility (Commercial Deployment), GXO with Agility (Commercial Deployment) and Hyundai with Boston Dynamics (Announcement). Only the BMW Figure 02 Pilot has vendor-reported outcome data (company-claimed), and it carries no robot-count denominator, so per-robot productivity cannot be derived from it.
The four cases span the Pilot and Announcement stages, and the largest site count reported for any of them is 7 units (reported by trade press, unverified); each case below closes with the question an operator should ask.
Case | Stage | Units | Outcome | Grade | Source date |
BMW with Figure | Pilot (Figure 02); Announcement (Figure 03) | Figure 02 not stated; Figure 03 1 arriving (company-claimed) | 30,000+ X3, 90,000+ parts, 1,250+ runtime hours (company-claimed) | Vendor-reported | 19 Nov 2025; 30 Jun 2026 |
Toyota Canada with Agility | Commercial Deployment | Not disclosed; trade press 3 then 7 (unverified) | None | Vendor-reported | 19 Feb 2026 |
GXO with Agility | Commercial Deployment | "Small fleet", count not specified | 100,000+ totes, about 98% accuracy (undefined) (company-claimed) | Vendor-reported | 24 Nov 2025 |
Hyundai with Boston Dynamics | Announcement | Atlas count at centre not found | None | Vendor-reported plans | 21 Sep 2026 |
BMW Plant Spartanburg, South Carolina, with Figure AI
Figure 02 ran a Pilot of about 11 months (company-claimed, relayed) loading sheet metal to a welding fixture, with completion announced on 19 November 2025. Figure reports 30,000+ BMW X3 vehicles, 90,000+ parts loaded and 1,250+ runtime hours on 10-hour shifts Monday to Friday (company-claimed). The page does not state how many robots ran, so no total converts to per-robot productivity, and the cycle, placement and intervention figures on it are targets, not results. The Fortune report in Section 3.2 puts the count at 1 Figure robot at a time (verified, Fortune, as reported). BMW's own results statement was not found in public sources. Figure 03 is an Announcement: 1 unit arriving at Hall 52 for a sequencing use case on 30 June 2026 (company-claimed), with BMW's Ulrich Wieland saying BMW looks forward to deploying it, a statement of intent with no metric. The reason for the change of task was not stated. The case shows a Pilot followed by a single-unit Announcement, with no published fleet decision. The operator's question: how many robots, shifts and interventions sit behind the 1,250 hours?
Toyota Motor Manufacturing Canada, Woodstock, Ontario, with Agility
Agility's release, dated 19 February 2026 from its web address, announces a RaaS agreement after a pilot. The task is reported as swapping an empty tote for a full tote on an automated tugger (reported by trade press, unverified). The release states no unit count, financial terms or start date (company-claimed). Trade press differs: Automotive Manufacturing Solutions reports 3 pilot units and 7 in the commercial deployment from early April 2026, while Global News reports 7 allocated, beginning with 3; neither names a source (reported by trade press, unverified). Toyota's own unit statement and the operating status after April 2026 were not found in public sources. The stage is Commercial Deployment, announced, with no outcome. If 7 is correct, the largest named site is still below 10. The operator's question: what unit count is contractually committed, and which acceptance test releases the next tranche?
GXO, Flowery Branch, Georgia, with Agility
The Pilot began on 8 December 2023 at a SPANX facility, with Digit moving totes from AMRs to conveyors (company-claimed, relayed). A multi-year RaaS agreement followed in June 2024, described as a "small fleet" with the count not specified, and Agility's Arc as the management layer (company-claimed). Agility reports more than 100,000 totes and about 98% accuracy, which is undefined (company-claimed). Fleet size, start date and period are not disclosed, so the totals cannot be turned into a rate, uptime or labour productivity. No GXO-origin outcome evidence was found, and GXO's Q2 2026 call summary of 8 August 2026 contained no humanoid content. Conversion from pilot to contract is confirmed as announced; scale-up beyond a small fleet is not evidenced. The operator's question: what does the 98% measure, and what interventions sit behind it?
Hyundai Motor Group Metaplant America, near Savannah, Georgia, with Boston Dynamics
The Robotics Metaplant Application Center opened on 21 September 2026, where Atlas is being trained on logistics and sequencing (company-claimed, relayed). It is a training centre, not a production deployment, and the Atlas count there was not found. Hyundai's plans are Atlas deployment at the Georgia plant from 2028 for parts sequencing, expansion to assembly by 2030, 25,000 units across Hyundai and Kia plants and capacity up to 30,000 robots a year by 2028 (company-claimed). The Hyundai Motor branch of the Korean Metal Workers' Union said in January 2026 that no robots should enter the production floor without a labour-management agreement (verified, Korea JoongAng Daily; a reported union statement, relayed; sector automotive manufacturing, South Korea). A labour agreement is therefore a gating item for any plan at this size. The operator's question: which labour agreements must exist before units reach the floor?
Announcement-only programmes
Schaeffler stated in November 2024 an intent to deploy a significant number of Digit units across its 100 plants by 2030, a target with no confirmed unit count (company-claimed). Mercedes-Benz and Apptronik were reported on 18 March 2025 to involve a handful of Apollo units (verified, AOL / Reuters-syndicated report, as reported), and Jabil and Apptronik announced a collaboration in February 2025 without quantities. Mercado Libre is named as a Digit customer without detail. Catalyst Brands signed with Figure on 26 May 2026 in Reno, Nevada, for sequencing, with counts and terms not disclosed (company-claimed). None has outcome data. The Journal's Deployment Outcome Tracker follows each after the announcement.
6. Friction, Risk & Unresolved Issues
Direct answer: The main risk in scaling humanoid robots beyond a pilot is unmeasured operating load, because no public source reports supervision, intervention or uptime for a commercial humanoid fleet. Two sourced barriers add to it. McKinsey (30 June 2025) found "no clear playbooks for scaling pilots" (analyst estimate, McKinsey), and Gartner (21 January 2026) lists dexterity limits, integration with existing systems and battery life (analyst estimate, Gartner).
6.1 Supervision and teleoperation dependence
Arc names uptime, throughput and MTBI as KPIs without publishing values (company-claimed). Figure's zero interventions per shift is a target, and the 98% tote accuracy at GXO is not an intervention rate (company-claimed). The ratio-like Foundation hypothetical appears in Section 4.3. A related signal comes from a consumer product: MIT Technology Review reported on 23 February 2026 that 1X's $20,000 Neo can be piloted by a remote operator when stuck, with customer consent sought first (verified, MIT Technology Review; consumer analogue, not an industrial fleet). Whether Tesla's Optimus demonstrations were operator-assisted is disputed: Electrek inferred assistance in a demonstration reported on 7 December 2025, and Tesla's CEO, Elon Musk, has denied it (disputed; no independent audit exists). Require remote-assist hours per shift as a reported contract metric.
6.2 Maintenance, battery and integration
Cost and uptime drivers come mostly from non-vendor sources. McKinsey says the best humanoids typically operate for 2 to 4 hours, that repairs may cost up to $1,000, and that bespoke parts, a lack of accepted standards and planetary roller screw shortages complicate maintenance (analyst estimate, McKinsey). Gartner adds integration with existing systems, high upfront and upkeep cost, and a tendency for most deployments through 2028 to stay in tightly controlled environments (analyst estimate, Gartner). Vendor battery claims differ: Digit 5 at 90 minutes with a 9-minute charge (company-claimed). Integration is described only at API level; no vendor documents PLC-level or named WMS connectors publicly (NOT FOUND in public sources). The Amazon Shreveport indicator in Section 1 also applies here (company-claimed; AMR analogue, not humanoid data).
6.3 Contingent orders and the claims gap
Announced volume has run ahead of confirmed volume. Agility's more than $300 million of orders are contingent on milestones, and its deck cites both 9 and 8 deployments (company-claimed). Backlog cancellation rights were not retrieved. Figure's "fleet" statements sit against BMW's one-robot account (Section 3.2). Reports of Tesla unit goals rest on anonymous sources (unconfirmed), and Musk said on the Q1 call of 22 April 2026 that he did not know the production rate for the year (verified, relayed transcript). Abdil Tunca, Senior Principal Analyst at Gartner, stated in the Gartner press release of 21 January 2026: "CSCOs must carefully evaluate readiness and avoid overcommitting resources". The statement supports tying contract milestones to measured readiness, not to announced volume.
6.4 Labour and workforce signals
Three labour signals exist in the research base, each with a stated sector, timeframe and confidence tier, and none is a displacement forecast for the reader's operation.
• Morgan Stanley (via Fortune, 19 August 2025): sector logistics and physical retail; savings "likely take many years to achieve" and adoption could take years "if not decades", through attrition first; scenario analysis, low-to-moderate confidence (analyst estimate, Morgan Stanley).
• McKinsey (30 June 2025): productivity gains "without necessarily eliminating jobs"; general-purpose robots, no timeframe given; Journal confidence tier low (analyst estimate, McKinsey).
• Hyundai union (January 2026): automotive manufacturing, South Korea; a stated position, not a measured labour effect (verified, Korea JoongAng Daily; relayed).
No measured productivity gain against human labour from a customer-verified humanoid deployment was found, and no union action outside Hyundai was found, in public sources.
Risk | Evidence | Grade | Mitigation contract term |
Unmeasured supervision and intervention load | No public values; Arc KPIs unvalued | NOT FOUND in public sources | Logged interventions and remote-assist hours per shift |
Contingent orders and vendor claims | More than $300 million subject to milestones | Vendor-reported | Delivery-linked payments; cancellation rights reviewed |
Battery and charging cycle | 90-minute battery, 9-minute charge; McKinsey 2 to 4 hours | Vendor-reported; Modelled or projected | Charging share of cycle measured over a defined window |
Integration scope | API level only | NOT FOUND in public sources | Named connector scope in the statement of work |
Workforce agreement | Hyundai union position | Verified as reported, relayed (Korea JoongAng Daily) | Labour agreement as a condition precedent |
Repair and spares | Up to $1,000 per repair; no service-level pricing | Modelled or projected | Spares and response-time terms |
7. Competitive / Platform Comparison
Direct answer: Humanoid fleet orchestration software justifies its cost only if the contract commits to measured results, because it is a vendor claim with named KPIs and no published values. Arc and Orbit are single-vendor platforms (company-claimed). VDA 5050 and ISO 21423, the interoperability efforts, are oriented to AMRs, and ISO 21423 was not finalised as of 15 September 2026. No humanoid-specific fleet interoperability standard was found in public sources.
7.1 Fleet software
Two fleet-software platforms are documented for humanoids, Arc and Orbit, and neither has published performance values.
Vendor and product | Claim | KPI values | Grade |
Agility Arc, launched 11 Mar 2024 | Facility mapping, workflow definition, operations, troubleshooting; tracks uptime, throughput, MTBI and robot status; APIs for WMS, WES and MES | NOT FOUND in public sources | Vendor-reported (company-claimed) |
Boston Dynamics Orbit | Connects Atlas to MES, WMS and other industrial systems; a task learned by one Atlas can be replicated across the fleet | NOT FOUND in public sources | Vendor-reported (company-claimed) |
Figure, Apptronik, Tesla, UBTech, Unitree, AgiBot | Fleet software | NOT FOUND in public sources | None |
Agility's CEO, Peggy Johnson, said on 24 June 2024 that the human job is to manage the fleet of robots (company-claimed). A pilot by Accenture, Vodafone Procure & Connect and SAP in Duisburg on 22 April 2026 had robots receive inspection tasks through SAP Extended Warehouse Management and write findings back; the manufacturer is unnamed and no robot count or result was disclosed (company-claimed).
7.2 Lock-in, interoperability and standards status
No ratified humanoid fleet standard was found, so portability between vendors is a contract matter. Require data-export and API terms. The table reports status only.
Standard or initiative | Status per sources |
VDA 5050 v3.0.0 | Adopted by VDMA and VDA on 17 Feb 2026 per a participant; AMR-focused; humanoids not mentioned |
MassRobotics AMR Interoperability Standard | v1.0 published May 2021; v2.0 planned per an Oct 2022 article; 2026 status not confirmed |
ISO 21423 | Scope: data exchange between industrial mobile robots and fleet managers; not finalised as of 15 Sep 2026; InOrbit's OpenRobOps claims the first reference implementation (company-claimed) |
ISO 25785-1 | Working draft as of early 2026; Oct 2026 status not confirmed |
IEEE Humanoid Study Group | Landscape analysis and roadmap, focused on safety and performance; no fleet-interoperability gap identified in the Sep 2025 pathway study |
No humanoid vendor was found publicly committing to VDA 5050 or MassRobotics, and no humanoid adapter documentation for Open-RMF was found.
7.3 Platform comparison matrix
Specifications below are vendor statements. The independently verified column is empty for every platform because no independent test report was found. The Journal's Quarterly Humanoid Platform Index carries the fuller specification record.
Platform | Payload | Runtime and swap | Fleet software | Per-site units | Independently verified |
Digit 5 | 50 lb (company-claimed) | 90-minute battery, 9-minute charge (company-claimed) | Arc | NOT FOUND | NOT FOUND |
Atlas | Up to 50 kg (company-claimed) | Runtime NOT FOUND; self-swap (company-claimed) | Orbit | NOT FOUND | NOT FOUND |
Walker S2 | NOT FOUND | About 3-minute swap (company-claimed) | NOT FOUND | NOT FOUND | NOT FOUND |
Figure 03 | NOT FOUND | NOT FOUND | NOT FOUND | 1 announced at BMW (company-claimed) | NOT FOUND |
Apollo | NOT FOUND | NOT FOUND | NOT FOUND | A handful at Mercedes-Benz (Mar 2025) (verified, AOL / Reuters-syndicated report, as reported) | NOT FOUND |
Optimus | NOT FOUND | NOT FOUND | NOT FOUND | NOT FOUND | NOT FOUND |
Vendor spec sheets were not retrieved, so most cells remain NOT FOUND in public sources.
8. Strategic Recommendations (Operations Leader)
Direct answer: Expand only when the operator's own pilot data show cost per productive hour at or below its benchmark, at the measured intervention and availability levels, across repeated shifts. Hold when metrics are unproven or cost is above benchmark. Use RaaS to limit capital while evidence matures, and choose conventional automation when task variability does not need humanoid flexibility. Thresholds are buyer-set: no public benchmark exists for any gate metric (NOT FOUND in public sources).
8.1 The 10-Robot Readiness Gate scorecard
Eight gates decide readiness. Uptime has four meanings: power-on (the robot is energised), available (ready to accept work), successful task execution (the task is completed to specification) and productive output (completed work that counts toward cost per productive hour). Productive availability means the fourth.
Gate metric | Definition and basis | Evidence the buyer must see | Public benchmark status | Threshold |
Task success rate | Tasks completed to specification ÷ attempted | Logged tasks over a defined window | GXO's about 98% is undefined (company-claimed) | Set from the Section 4 break-even at the buyer's benchmark |
Interventions per robot-shift | Human interventions ÷ robot-shifts | Intervention log from the vendor | Figure's zero is a target only (company-claimed) | Set from the Section 4 break-even at the buyer's benchmark |
Productive availability | Productive hours ÷ scheduled hours | Time-stamped state logs | NOT FOUND in public sources | Set from the Section 4 break-even at the buyer's benchmark |
MTBI | Mean time between incidents | Incident log | Arc names MTBI, no value (company-claimed) | Set from the Section 4 break-even at the buyer's benchmark |
Recovery time and substitution | Time to resume work; can another unit take the task | Recovery log; substitution test | NOT FOUND in public sources | Set from the Section 4 break-even at the buyer's benchmark |
Supervision hours per robot-hour | Supervisor and remote-assist hours ÷ robot-hours | Staffing records | NOT FOUND in public sources | Set from the Section 4 break-even at the buyer's benchmark |
Charging and swap share of cycle | Charging and swap time ÷ scheduled time | Charger and swap logs | Digit 5 90-minute battery, 9-minute charge; Walker S2 about 3-minute swap; Atlas self-swap (company-claimed) | Set from the Section 4 break-even at the buyer's benchmark |
Repeatability across shifts | Gate results across repeated shifts | Multi-shift data | NOT FOUND in public sources | Set from the Section 4 break-even at the buyer's benchmark |
8.2 Expand, hold, RaaS or replace
Gate results, cost per productive hour against the benchmark, task variability, vendor contract terms and capital position set the outcome. The expand, hold or replace decision tree maps those inputs to one of five outcomes: expand, extend the pilot, move to RaaS, wait, or use conventional automation.
1. Expand: gates pass at the buyer's thresholds and cost per productive hour is at or below the benchmark across repeated shifts.
2. Extend the pilot: gates are partly measured, or results are not repeatable across shifts.
3. Move to RaaS: the case holds but capital or upgrade risk is the constraint.
4. Wait: the platform is not yet available; Digit 5 early access is the first half of 2027 (company-claimed).
5. Conventional automation: task variability is low, or cost per productive hour stays above the benchmark.
Caleb Thomson, Senior Director Analyst at Gartner, stated in the Gartner press release of 21 January 2026:
"we expect polyfunctional robots to be the superior solution".
8.3 Contract metrics to require
Six metrics belong in the contract, each with a stated definition and a log the buyer can audit.
Metric | Contract definition | Evidence |
Uptime | One of the four meanings in Section 8.1, named in the contract | Time-stamped logs |
MTBI | Mean time between incidents, with incident defined | Incident log |
Recovery time | Time from fault to resumed work | Recovery log |
Response time | Vendor time to respond and to restore | Service tickets |
Parts availability | Spares on site or delivery time | Spares ledger |
Task completion | Tasks completed to specification | Task log |
The Operations Leader can now make this decision: approve units beyond the pilot only when the vendor has supplied logged data for the five measurements (task success, interventions per robot-shift, productive availability, recovery and substitution, and supervision hours per robot-hour), the Section 4 break-even shows cost per productive hour at or below the company's own benchmark across repeated shifts, and the contract carries the six metrics above with data-export rights and delivery-linked payment. If any of the three is missing, extend the pilot, move to RaaS or use conventional automation. Stop expansion if the measured cost per productive hour exceeds the benchmark at the measured levels, or if the vendor will not commit to logged metrics.
9. Executive FAQ
Direct answer: The first questions to settle before scaling to 10 humanoid robots are five measurements: task success, interventions per robot-shift, productive availability, recovery and substitution, and supervision hours per robot-hour. All five come from the operator's own pilot, because no public benchmark exists for any of them (NOT FOUND in public sources). Section 8 turns each into a pass, hold or stop test.
What should I verify before scaling a humanoid robot pilot to 10 or more units?
Five measurements: task success rate, interventions per robot-shift, productive availability, recovery and substitution, and supervision hours per robot-hour. No public benchmark exists for any of them, so each needs logged data from the operator's pilot. Figure's zero interventions per shift is a target, not a result (company-claimed).
How many people do I need to supervise a humanoid robot fleet?
No public ratio exists. Agility names uptime and MTBI as Arc KPIs without values (company-claimed), and the only ratio-like statement is a Foundation CEO hypothetical of three operators per robot under pure teleoperation (company-claimed; an opinion, not a measurement). Measure supervision hours per robot-hour in the pilot.
How do I calculate the cost per productive hour of a humanoid robot?
Divide total fleet operating cost by total productive robot-hours, excluding charging, faults, recovery, waiting and scheduled maintenance. Build the seven-row cost stack in Section 4.1. A rate of $8,500 per month per Digit is a management model, not a market price (company-claimed / management model — not verified market pricing).
What does humanoid robot uptime mean, and what should I require from a vendor?
Uptime has four meanings: power-on, available, successful task execution and productive output. Digit 5's 90-minute battery and 9-minute charge are vendor claims (company-claimed). Require time-stamped logs of all four over a defined window, with the contract naming which one governs payment.
When does fleet orchestration software justify its cost, and should I buy a single-vendor or mixed fleet?
Only when the contract commits to KPIs with values. Arc and Orbit are vendor-claimed with no published values (company-claimed), and VDA 5050 and ISO 21423 are AMR-oriented, with ISO 21423 not finalised as of 15 September 2026. No humanoid-specific interoperability standard was found, so mixed fleets rest on contract terms.
When should I expand, pause, lease or choose conventional automation instead of humanoid robots?
Expand when gate results and cost per productive hour clear the buyer's benchmark across repeated shifts. Pause when metrics are unproven, use RaaS to limit capital, and choose conventional automation when task variability is low. Gartner expects polyfunctional robots to be the superior solution (analyst estimate, Gartner; 21 January 2026).
Research limitations:
This report uses public sources only and reflects evidence as of 16 October 2026. Vendor figures are labelled company-claimed and were not independently verified. Several figures reached the research through secondary or relayed sources. Agility’s SEC Form S-4/A was only partly read. Vendor specification sheets were not retrieved. “Not found” means not found in public sources, not that the data does not exist. This report gives no engineering, safety or conformity opinion; see the Disclaimer.
References & Strategic Sources
The sources below are those cited in this report, classified as in the research record.
Figure AI. F.02 Contributed to the Production of 30,000 Cars at BMW. Date: 19 Nov 2025. URL: https://www.figure.ai/news/production-at-bmw. Classification: Tier 1; COMPANY-CLAIMED.
Figure AI. F.03 Arrives at BMW. Date: 30 Jun 2026. URL: https://figure.ai/news/f-03-at-bmw. Classification: Tier 1; COMPANY-CLAIMED.
Figure AI. Figure Signs Agreement with Catalyst Brands to Scale Humanoid Operations. Date: 26 May 2026. URL: https://figure.ai/news/figure-signs-agreement-with-catalyst-brands. Classification: Tier 1; COMPANY-CLAIMED.
Fortune (Jason Del Rey). Figure AI / BMW partnership details, reality vs exaggeration. Date: 6 Apr 2025. URL: https://fortune.com/2025/04/06/figure-ai-bmw-humanoid-robot-partnership-details-reality-exaggeration. Classification: Tier 4; VERIFIED (that it was reported).
The Robot Report. BMW Group deploys Figure 03 humanoid after tests with previous version. Date: 29 Jun 2026. URL: https://www.therobotreport.com/bmw-group-deploys-figure-03-humanoid-after-tests-previous-version/. Classification: Tier 4; VERIFIED (reporting); BMW statement relayed.
Repairer Driven News. Humanoid robots complete 11-month project at BMW plant. Date: 25 Nov 2025. URL: https://www.repairerdrivennews.com/2025/11/25/humanoid-robots-complete-11-month-project-at-bmw-plant/. Classification: Tier 4; relays COMPANY-CLAIMED.
Agility Robotics via Business Wire. Commercial Agreement with Toyota Motor Manufacturing Canada. Date: 19 Feb 2026 (from URL). URL: https://secure.businesswire.com/news/home/20260219029433/en/Agility-Robotics-Announces-Commercial-Agreement-with-Toyota-Motor-Manufacturing-Canada. Classification: Tier 1; COMPANY-CLAIMED.
Automotive Manufacturing Solutions. Toyota Canada signs humanoid robot deal with Agility. Date: Feb 2026. URL: https://www.automotivemanufacturingsolutions.com/automation/toyota-canada-signs-humanoid-robot-deal-with-agility/2611551. Classification: Tier 4; COMPANY-CLAIMED (relayed).
Global News. Digit, Toyota's humanoid robot. Date: 21 Feb 2026. URL: https://globalnews.ca/news/11677383/digit-toyotas-humanoid-robot/amp/. Classification: Tier 4; relayed, unsourced counts.
Supply Chain 24/7. Agility Robotics Digit 5 humanoid warehouse work. Date: 16 Sep 2026. URL: https://www.supplychain247.com/article/agility-robotics-digit-5-humanoid-warehouse-work/cloud. Classification: Tier 4; COMPANY-CLAIMED (relayed); Tunca comment relayed.
Agility / Churchill Capital Corp XI. Form 425 Ex. 99.1 release. Date: 24 Jun 2026. URL: https://www.sec.gov/Archives/edgar/data/0002074973/000121390026071290/ea029548401ex99-1.htm. Classification: Tier 1; COMPANY-CLAIMED.
Churchill Capital Corp XI. Form 425 Ex. 99.2 investor presentation. Date: June 2026. URL: https://www.sec.gov/Archives/edgar/data/2074973/000121390026071287/ea029548401ex99-2.htm. Classification: Tier 1; COMPANY-CLAIMED (unaudited, self-contradicting).
Agility Robotics. Agility Arc launch release. Date: 11 Mar 2024. URL: https://agilityrobotics.com/news/agility-robotics-brings-operational-visibility-to-deployment-of-digit-fleets-with-the-launch-of-agility-arc. Classification: Tier 1; COMPANY-CLAIMED.
Agility Robotics. Digit product page. Date: undated. URL: https://www.agilityrobotics.com/products/digit. Classification: Tier 1; COMPANY-CLAIMED.
Agility Robotics. Humanoid Robots Entering the Workforce (Peggy Johnson interview). Date: 24 Jun 2024. URL: https://agilityrobotics.com/content/humanoid-robots-entering-the-workforce. Classification: Tier 1; COMPANY-CLAIMED.
Agility Robotics. GXO signs industry-first multi-year agreement. Date: 27 Jun 2024. URL: https://agilityrobotics.com/content/gxo-signs-industry-first-multi-year-agreement-with-agility-robotics. Classification: Tier 1; COMPANY-CLAIMED.
Robotics 24/7. GXO begins pilot of Agility Digit. Date: 8 Dec 2023. URL: https://www.robotics247.com/article/gxo_begins_pilot_of_agility_robotics_digit_robot/gxo_logistics. Classification: Tier 4; COMPANY-CLAIMED (relayed).
Robotics 24/7. Agility Robotics, GXO sign multi-year humanoid deployment agreement. Date: 28 Jun 2024. URL: https://www.robotics247.com/article/agility_robotics_gxo_sign_multi_year_humanoid_deployment_agreement/gxo_logistics. Classification: Tier 4; COMPANY-CLAIMED (relayed).
Assembly Magazine. Humanoid robot Digit moves 100,000 totes. Date: 24 Nov 2025. URL: https://www.assemblymag.com/articles/99671-humanoid-robot-digit-moves-100-000-totes-at-distribution-center. Classification: Tier 4; COMPANY-CLAIMED (Agility).
MarketBeat. GXO Q2 2026 earnings call highlights. Date: 8 Aug 2026. URL: https://www.marketbeat.com/instant-alerts/gxo-logistics-q2-earnings-call-highlights-2026-08-08/. Classification: Unranked; secondary summary.
The Robot Report. Schaeffler plans global use of Agility Digit. Date: 13 Nov 2024. URL: https://www.therobotreport.com/schaeffler-plans-global-use-agility-robotics-digit-humanoid/. Classification: Tier 4; COMPANY-CLAIMED.
EVwire. Tesla Q1 2026 earnings call transcript. Date: 22 Apr 2026. URL: https://evwire.com/p/tesla-tsla-q1-2026-earnings-call-full-transcript. Classification: Unranked; VERIFIED against transcript (relayed).
TechSpot. Tesla halts mass production of Optimus citing... (The Information report). Date: 8 Oct 2025. URL: https://www.techspot.com/news/109781-tesla-temporarily-halts-mass-production-optimus-robots-citing.html. Classification: Tier 4; report exists, figures UNCONFIRMED.
The Robot Report. Boston Dynamics opens Metaplant Application Center to train Atlas. Date: 21 Sep 2026. URL: https://www.therobotreport.com/boston-dynamics-opens-metaplant-application-center-train-atlas-humanoid-robots/. Classification: Tier 4; COMPANY-CLAIMED (relayed).
Korea JoongAng Daily. Hyundai union moves to block humanoid robots on production lines. Date: 22 Jan 2026. URL: https://www.koreajoongangdaily.com/business/hyundai-union-moves-to-block-humanoid-robots-on-production-lines/12040140. Classification: Tier 4; VERIFIED (reported union statement, relayed).
Boston Dynamics. Boston Dynamics unveils new Atlas robot (blog). Date: 5 Jan 2026. URL: https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/. Classification: Tier 1; COMPANY-CLAIMED (text not read; basis is Robotics 24/7).
Robotics 24/7. CES 2026: Boston Dynamics unveils new Atlas. Date: 7 Jan 2026. URL: https://www.robotics247.com/article/ces-2026-boston-dynamics-unveils-new-atlas-humanoid-robot/boston_dynamics. Classification: Tier 4; COMPANY-CLAIMED (relayed).
UBTECH Robotics Corp Ltd (HKEXnews). Interim Results Announcement, six months ended 30 Jun 2026. Date: 28 Aug 2026. URL: https://www1.hkexnews.hk/listedco/listconews/sehk/2026/0828/2026082801691.pdf. Classification: Tier 1; issuer-reported; interpretation COMPANY-CLAIMED.
Gasgoo. UBTech reports 1.27 billion yuan in first-half revenue. Date: 1 Sep 2026. URL: https://autonews.gasgoo.com/articles/news/ubtech-reports-127-billion-yuan-in-first-half-revenue-sells-921-full-size-humanoid-robots-2094660109459607553. Classification: Tier 4; relayed.
Robotics 24/7. UBTech Walker S2 autonomous battery swap. Date: 30 Jul 2025. URL: https://www.robotics247.com/article/ubtech_robotics_walker_s2_humanoid_robot_offers_autonomous_battery_swap_capabilities. Classification: Tier 4; COMPANY-CLAIMED.
AOL / Reuters-syndicated. Mercedes-Benz takes stake in robotics. Date: 18 Mar 2025. URL: https://www.aol.com/news/mercedes-benz-takes-stake-robotics-150148224.html. Classification: Tier 4; VERIFIED (as reported, exec quoted).
Apptronik / Jabil via Nasdaq. Apptronik and Jabil collaborate. Date: 25 Feb 2025. URL: https://www.nasdaq.com/press-release/apptronik-and-jabil-collaborate-scale-production-apollo-humanoid-robots-and-deploy. Classification: Tier 1; COMPANY-CLAIMED.
Gartner. Gartner Predicts Fewer Than 20 Companies Will Scale Humanoid Robots... by 2028. Date: 21 Jan 2026. URL: https://www.gartner.com/en/newsroom/press-releases/2026-01-21-gartner-predicts-fewer-than-20-companies-will-scale-humanoid-robots-for-manufacturing-and-supply-chain-to-production-stage-by-2028. Classification: Analyst (unranked, treated as Tier 3); VERIFIED as published forecast.
McKinsey. Will embodied AI create robotic coworkers?. Date: 30 Jun 2025. URL: https://www.mckinsey.com/industries/industrials-and-electronics/our-insights/will-embodied-ai-create-robotic-coworkers. Classification: Analyst (unranked, treated as Tier 3); VERIFIED as published assessment.
Interact Analysis (Prezly). Humanoid robot revenue to reach $15bn by 2035. Date: 21 May 2026. URL: https://interact-analysis.prezly.com/humanoid-robot-revenue-to-reach-15bn-by-2035. Classification: Tier 3; ANALYST ESTIMATE.
Goldman Sachs (via Humanoid.guide). 2035 shipments and market value revision. Date: 15 Sep 2026. URL: https://humanoid.guide/?p=18115. Classification: Tier 3; ANALYST ESTIMATE; SECONDARY report.
Morgan Stanley (via Humanoid.guide). Humanoid market and ASP forecasts. Date: 29 Apr 2025. URL: https://humanoid.guide/?p=3205. Classification: Tier 3; ANALYST ESTIMATE; unverified against original.
Morgan Stanley (via Fortune). $920bn savings from AI, agentic and embodied. Date: 19 Aug 2025. URL: https://fortune.com/2025/08/19/morgan-stanley-920-billion-sp-500-savings-ai-agentic-robots-jobs/. Classification: Tier 3 via Tier 4; ANALYST ESTIMATE.
IDTechEx (EMSNow). Humanoid robots 2026-2036 payback and ASP. Date: 15 May 2026. URL: https://www.emsnow.com/?p=54508. Classification: Analyst (unranked); ANALYST ESTIMATE.
Omdia (TechNode). AgiBot leads global humanoid shipments in 2025. Date: 9 Jan 2026. URL: https://technode.com/2026/01/09/chinas-agibot-leads-global-humanoid-robot-shipments-in-2025-omdia-says/. Classification: Analyst (unranked); ANALYST ESTIMATE.
IDC (Jiemian). 2025 humanoid shipments. Date: undated. URL: https://en.jiemian.com/article/13925066.html. Classification: Analyst (unranked); ANALYST ESTIMATE.
Humanoid.guide (SCMP / Smart Analytics Global). AgiBot overtakes Unitree in humanoid shipments. Date: 11 Aug 2026. URL: https://humanoid.guide/agibot-overtakes-unitree-in-humanoid-shipments-before-ipos/. Classification: Analyst estimate, relayed via aggregator.
SSE / China Daily. Unitree IPO news release. Date: 11 Aug 2026. URL: https://english.sse.com.cn/news/newsrelease/voice/c/c_20260811_10828578.shtml. Classification: Tier 4; VERIFIED as reported from prospectus (relayed).
Gasgoo. AgiBot launches Hong Kong IPO process. Date: 28 Jul 2026. URL: https://autonews.gasgoo.com/articles/icv/agibot-launches-hong-kong-ipo-process-2082120389458100225. Classification: Tier 4; COMPANY-CLAIMED (relayed).
eWeek. How much does a humanoid robot cost. Date: 19 Aug 2026 (mod. 12 Sep 2026). URL: https://www.eweek.com/news/how-much-does-a-humanoid-robot-cost/. Classification: Tier 4; COMPANY-CLAIMED (relayed).
The Robot Report. Here's what it could cost to hire a Digit humanoid. Date: 17 Jul 2024. URL: https://www.therobotreport.com/heres-what-it-could-cost-to-hire-a-digit-humanoid/. Classification: Tier 4; COMPANY-CLAIMED (stale).
Accenture newsroom. Accenture, Vodafone Procure & Connect and SAP pilot humanoid robotics. Date: 22 Apr 2026. URL: https://newsroom.accenture.com/news/2026/accenture-vodafone-procure-connect-and-sap-pilot-humanoid-robotics-in-warehouse-operations. Classification: Tier 1; COMPANY-CLAIMED.
Idealworks. VDA 5050 3.0.0 release. Date: Mar 2026. URL: https://idealworks.com/en/news/vda5050-300-release-march-2026. Classification: Tier 1 (participant); VERIFIED event, described by a participant.
MassRobotics. Interoperability standard demonstration. Date: 24 Oct 2022. URL: https://www.massrobotics.org/massrobotics-partners-demonstrate-interoperability-standard-at-amr-and-logistics-week/. Classification: Tier 4; dated.
The Robot Report. InOrbit OpenRobOps / ISO 21423 reference implementation. Date: 15 Sep 2026. URL: https://www.therobotreport.com/inorbit-ai-releases-openrobops-iso-21423-reference-implementation/. Classification: Tier 4; COMPANY-CLAIMED.
InOrbit. Open-RMF adapter. Date: undated. URL: https://www.inorbit.ai/blog/inorbit-rmf-adapter. Classification: Tier 1 (vendor); COMPANY-CLAIMED.
IEEE RAS. Humanoid Study Group page. Date: modified 17 Dec 2025. URL: https://www.ieee-ras.org/industry-activities/standards/active-projects/study-group-humanoid-robots/. Classification: Tier 2; VERIFIED (status).
IEEE (via The Robot Report). A Pathway Study for Future Humanoid Standards. Date: Sep 2025. URL: https://www.therobotreport.com/wp-content/uploads/2025/09/IEEE-Humanoid-Report-of-Future-Standards-Development.pdf. Classification: Tier 2; VERIFIED (published report).
i-scoop. ISO 25785-1 explained. Date: early 2026. URL: https://www.i-scoop.eu/iso-25785-1-explained-and-what-it-means-for-humanoid-robot-safety/. Classification: Unranked; status as of early 2026.
Amazon. Amazon's one millionth robot and DeepFleet. Date: 1 Jul 2025. URL: https://www.aboutamazon.com/news/operations/amazon-million-robots-ai-foundation-model. Classification: Tier 1 (ANALOGUE); COMPANY-CLAIMED.
Sacra. Foundation CEO interview. Date: 9 Mar 2025. URL: https://sacra.com/chat/h/93dc70de-7775-4bc2-b477-ce4cd741da03/. Classification: Unranked; COMPANY-CLAIMED opinion.
MIT Technology Review. The human work behind humanoid robots is being hidden. Date: 23 Feb 2026. URL: https://technologyreview.com/2026/02/23/1133508/the-human-work-behind-humanoid-robots-is-being-hidden. Classification: Tier 4; VERIFIED.
Electrek. Tesla Optimus takes suspicious tumble in new demo. Date: 7 Dec 2025. URL: https://electrek.co/2025/12/07/tesla-optimus-robot-takes-suspicious-tumble-in-new-demo/. Classification: Tier 4; DISPUTED (Electrek inference vs Musk denial)
© 2026 Sekason Research Limited, United Kingdom (Company No. 14339910). Published by Physical AI Journal. All rights reserved. No part of this report may be reproduced or redistributed without written permission, except for brief quotation with attribution.
