Industrial Deployment
The installed base robot learning is trying to enter, and the jam-rate arithmetic that decides whether a 99 percent cell ships.
- Last reviewed
- Reading time
- 14 min
- Citations
- 23
IFR's World Robotics 2025 reports a global operational stock of 4,663,698 industrial robots in 2024; annual installations exceeded 500,000 in each year from 2021 through 2024, with 542,076 installed in 2024 Müller 2025. Those machines are the field robot learning is trying to enter, and almost nothing about how they got there resembles the pipeline a paper describes. In 2024, China had 2,027,190 industrial robots in operation and Japan had 450,530. China had about 4.5 times the stock of Japan, the second-ranked country, and received 295,045 new installations that year, reported as 54 percent of the global total. IFR's May 5, 2026 release repeats the rounded stock and installation-share figures from World Robotics 2025 Müller 2025 International Federation of Robotics 2026. By customer industry in 2024, electrical/electronics accounted for 128,899 installations (reported as 24 percent), automotive for 126,088, and metal and machinery for 88,777 (16 percent). The IFR summary gives automotive 23 percent on page 13 but 24 percent on page 16; its chart and automotive discussion both report 126,088 units. These are customer-industry categories, not application families. IFR says the customer industry was unspecified for 14 percent of installations Müller 2025. The humanoid programmes this site follows count deployments in pilots and hundreds of hours: Unitree claims roughly 5,500 humanoid units shipped across 2025 Noreika 2026, a count Omdia puts nearer 4,200 while ranking AgiBot first RoboZaps 2026. The industrial market installs a hundred times either figure every year. This page is that field.
The economics you can drag
Use the sliders to compare two cases in this authored model. Keep robot cost at 80,000 USD, integration at 2.5 times, cycle time at 6 seconds, uptime at 95 percent and hourly wage at 25 USD. With 15-second jam clearing, lowering per-pick success from 99.9 to 99 percent changes calculated payback from 11.56 to 11.82 months: an increase of 0.26 months, or 2.2 percent. Repeat the comparison with jam clearing set to 300 seconds: calculated payback rises from 12.11 to 17.30 months, and modeled output falls from 542.9 to 380.0 picks per elapsed hour, a 30 percent decrease. Both paybacks remain within the model's chosen 24-month horizon. These are calculations from assumed inputs, not measured deployments; the payback readout rounds to one decimal place.
Cell economics calculator
Assumption: $80k is a chosen example, not a sourced arm-price quote. EVST gives complete-cell budgets, not fixed list prices.
Assumption: 2.5x is chosen within EVST’s 2-3x complete-cell guidance, not a measured cell.
Assumption: a paced piece-picking cycle; drag it to match any quoted cell.
Assumption: availability net of maintenance and faults; 95% is a working figure, not a vendor claim.
Assumption: the policy headline number; the demonstration teaching step drops this to 99.
Assumption: seconds of human attention per failed pick; this dial is the whole argument.
Assumption: fully loaded picker wage; set it to your own facility number.
Cost per pick
0.008 USD over 5 yr
Payback
11.6 months
Verdict
Pays back inside 24 months
Cell cost $200,000; 569 modeled picks per elapsed hour and 415,062 per 730-hour month; jam rate 0.10%. All seven defaults and slider ranges are authored assumptions; EVST’s guide supplies context, not exact inputs; none is a measured deployment result.
Where each elapsed hour goes
One elapsed hour is 3600 seconds; the three shares sum to it at every control setting.
This calculator is an authored worked example, not a measured deployment or financial forecast. Its 80,000 USD robot cost is not a sourced arm-price quote; the 20,000 to 250,000 USD range and 5,000 USD step are chosen controls. All seven defaults, ranges and steps are authored assumptions: a 2.5 integration multiple, 6-second cycle, 95 percent uptime, 99.9 percent per-pick success, 15-second jam clearing and 25 USD hourly wage accompany that robot cost. The model also chooses 730 hours per month, a 60-month amortization period and a chosen 24-month verdict horizon. Throughput is modeled picks per elapsed hour, including downtime. Payback divides cell capital by modeled monthly labour value; it does not add damage or throughput savings.
EVST supplies vendor context, not those exact inputs: its guide describes complete-cell budgets, explicitly does not publish fixed list prices, and places the robot at a third to half of total cost EVST Engineering Team 2026. Its payback guidance is 12 to 24 months for multi-shift cells, extending toward 24 to 36 months for single-shift or lower-throughput operations EVST Engineering Team 2026. The calculator's 2.5 multiple and 24-month threshold are selections within that context, not quoted measurements.
What is actually automated at scale
Strip the installed base down by task and one honest sentence covers most of it: repetitive motion in a fixed cell, with little or no perception worth the name. OSHA’s Technical Manual lists uses of industrial robot systems including arc and resistance welding, painting and spraying, machine-tool loading and unloading, assembly, materials handling and packaging Administration 2026. A spot-welding cell in a body shop repeats one trajectory against one fixture thousands of times a shift; the "perception" is the fixture guaranteeing the part is exactly where the program expects it. In A3's North American order data for 2025, collaborative robots accounted for 7,212 units valued at $241 million: 19.6 percent of the 36,766 robots ordered and 10.7 percent of the $2.25 billion in order value. A3 began reporting collaborative robots as a distinct category in Q1 2025 Automation 2026.
The genuinely learned frontier inside this base is logistics, and it is worth naming its subsystems because they are the machines a manipulation policy actually competes with. Automated storage and retrieval puts high-bay racking and shuttles where aisles and forklifts used to be. Autonomous mobile robot fleets move cases and totes under a traffic manager. Goods-to-person stations bring the shelf to a stationary human, which is the arrangement that sets the productivity bar any automated picker must beat, and piece picking from mixed bins is the task at that bar's edge. In its October 18, 2023 Sequoia and Digit announcement, Amazon reported over 750,000 robots working collaboratively with employees. Sequoia was operating at one Houston fulfillment center; testing Digit for tote recycling was planned Dresser 2023. The company’s fleet overview, published October 9, 2024 and updated June 4, 2026, reports more than one million robots deployed across its operations network since 2012. It describes Sequoia inventory handling, Sparrow piece picking, Vulcan picking and stowing, and Proteus cart transport, while separately describing next-generation Proteus as a lab pilot awaiting deployment Greenawalt 2024. In the 2023 announcement, Amazon said Sequoia could identify and store received inventory up to 75 percent faster than its then-current process, and reduce the time to process an order through a fulfillment center by up to 25 percent Dresser 2023. Amazon’s Vulcan article, published May 7, 2025 and updated June 4, 2026, describes force-feedback picking and stowing of approximately 75 percent of the types of items stored at its fulfillment centers, at speeds comparable to front-line employees. It describes high and low inventory-pod work at Spokane and Hamburg, handoff to employees for items the system cannot move, and further European and US deployment as planned. The coverage figure describes item types, not pick success or independently validated reliability Davies 2025. These are company-reported figures for different tasks and reporting scopes. The cumulative deployment total is not a simultaneous active-fleet count.
Symbotic's FY2025 Form 10-K distinguishes contracted work from revenue. It reports approximately $22.5 billion of backlog as of September 27, 2025 and $2.246922 billion of total revenue for the fiscal year ended that day. Backlog means unperformed obligations under existing contracts, not revenue already earned. The filing says the Walmart agreement, expanded in May 2022, covers implementation across all 42 regional distribution centres; that is contractual scope, not a count of operating sites Inc. 2025. For the year ended September 27, 2025, Symbotic reports 50 systems in deployment and 48 operational systems under software maintenance and support contracts. It then expected approximately 12 percent of backlog to be recognized as revenue in fiscal 2026. That is a revenue-recognition forecast, not an order-to-running duration; remaining performance obligation estimates can change with terminations, contract scope and other adjustments Inc. 2025. The unwind reads the same way: Kroger closed three of the eight automated sheds it built with Ocado, said it was monitoring the remaining five, and pays Ocado around £190 million in compensation for the privilege Hawkins 2025. Scale can go both ways. Ocado's own automation division, Ocado Intelligent Automation, sells the same grid-robot warehouses outward from grocery into healthcare, third-party logistics and industrial parts Group 2026.
The integrator's multiple
A robot arm is a minority of what a robot cell costs. Integrator guidance for palletising cells quotes complete systems at two to three times the arm's price, with the robot body itself a third to half of total cell cost EVST Engineering Team 2026.
EVST's July 15, 2026 commercial palletising-cell guide lists application-specific end-of-arm tooling; perimeter fencing, light curtains or area scanners, interlocked gates and safety-rated PLCs; mechanical and electrical installation, PLC/HMI integration with upstream conveyors or wrapping equipment, and on-site commissioning and testing; plus pattern programming, software licences and any vision-guided pick logic for mixed loads. The surrounding-line work depends on what is already installed
EVST Engineering Team 2026.
OSHA's Technical Manual, discussing ANSI/RIA R15.06-2012, says the systems integrator is responsible for completing and documenting the robot application's risk assessment before commissioning and providing the results to the employer. Manufacturers or employers may also act as integrators. The employer retains responsibility for a safe workplace, and having a risk assessment alone does not establish that the application protects workers
Administration 2026.
The practical consequence for a policy: buy a robot and deploy a policy is wrong by an order of magnitude as a model of what shipping costs, and the buyer who signs the cheque is buying a line that must hit a rate, not a demo that must impress once. In its April 16, 2026 investor outlook, Bessemer predicts that near-term value will accrue to full-stack, vertically integrated players rather than pure-play foundation-model companies. The authors say deployment requires domain-specific data collection, target-environment fine-tuning, hardware integration and operational infrastructure. They also stress expensive data collection and foundation models that are not yet general enough to work out of the box Levine 2026.
The metrics that decide a deployment
Five quantities decide whether a cell ships, and none of them is a benchmark score. Cycle time is the time required to produce a part or complete a process, measured for that operation Institute n.d.. Takt time is available production time divided by customer demand over the same period, expressed as time per unit. Its purpose is to match production with demand Institute n.d.. NASA distinguishes inherent availability, which uses mean time between failures and mean time to repair, from operational availability. The inherent measure excludes administrative and logistics delays and preventive maintenance; the operational measure includes corrective and preventive maintenance, administrative delays and logistics support time. Here, uptime means time in an operable state, not necessarily time spent producing NASA 1994. The calculator above reports capital cost per modeled pick: robot price times the integration multiple, divided by its modeled lifetime pick count over a fixed amortization period. It does not include running costs. This is a local worked-example calculation, not a measured deployment result.
The payback period is total system capital cost divided by monthly labour, reduced-damage and throughput value. EVST's guide estimates 12 to 24 months for multi-shift palletising cells and says single-shift or lower-throughput operations typically stretch closer to 24 to 36 months EVST Engineering Team 2026.
The jam rate is the lesson
Here is a separate authored arithmetic example, not a measured intervention rate: assume a 99 percent per-pick success rate and that every failed attempt needs clearing. At ten seconds per clearing, the expected overhead under those assumptions is (1 − 0.99) × 10 = 0.1 seconds per attempted pick. This calculation alone does not establish profitability. Ken Goldberg proposes combining model-based engineering with model-free learning so robots can perform useful work, collect real-world data, and use those data to improve performance and learn adjacent skills. This is a proposed way to bootstrap data collection, not a claim that engineering removes the need for learning. He also expects model-free AI eventually to enable fully general-purpose robots Goldberg 2025. The demo-to-production gap this site tracks at The Reliability Gap is reframed by this: what looks like an arithmetic problem in per-step success is very often an operations-design problem in intervention cost, and it is why the intervention rate and the cost of clearing each intervention, not the success rate alone, decide whether a cell ships. Morgan Stanley's 2026 note on the humanoid industry's "PR problem" is the same observation about the hype cycle from the capital side Wilkins 2026, and Figure's eight-hour autonomous sorting shift, impressive as it is, was one task at one site, on the company's own livestream, with no published success rate Belmonte 2026. The deployment dashboard at The Reliability Gap collects the site's selected deployment examples; this page deliberately does not repeat it.
Brownfield versus greenfield
The same technology succeeds or fails on where it is installed. A greenfield site is designed around the automation: floor loading, ceiling height, power, network, and traffic lanes exist because the system needed them. A brownfield deployment must fit the automation into a building that already runs. OSHA's Technical Manual describes safeguards for non-collaborative robot applications during automatic operation: guards such as fences and barriers, interlocked guards, and presence-sensing devices. It also describes reduced-speed manual mode with an enabling device for workers teaching inside the safeguarded space, plus lockout/tagout procedures and training. Its maintenance guidance calls for control of hazardous energy under 29 CFR 1910.147 or 29 CFR 1910.333 Administration 2026. Separately, Symbotic, describing its own systems in its FY2025 Form 10-K, says it can install them in phases while the existing warehouse continues to operate; this is a company-reported capability, not a guarantee for every retrofit Inc. 2025. The safety-and-assurance obligations of deployment have their own page at Safety and Assurance.
The labour evidence, disagreement intact
What robots do to jobs is a live empirical fight, and this page takes no side. Daron Acemoglu and Pascual Restrepo, in the Journal of Political Economy, find that one additional robot per thousand workers reduces the US employment-to-population ratio by about 0.2 percentage points and wages by 0.42 percent, with the losses concentrated in commuting zones exposed to industrial automation Acemoglu 2020. The MIT Task Force on the Work of the Future, co-chaired by David Autor and David Mindell with Elisabeth Reynolds as executive director, reported in 2020 that it found no compelling evidence of technological advances driving a jobless future. It describes automation displacing human labour from some tasks while creating new work, with the jobs available and the skills they demand shaped by economic incentives, policy choices and institutional forces Autor 2020. Both positions are named and both are serious; the field has not converged. What neither side disputes is the scale mismatch in the current cycle: the industrial installed base grew over decades, and the humanoid fleet being pitched as the next one is, so far, a rounding error against it, with even the volume leader's revenue growth decelerating from 332 to 68 percent and its Q1 2026 adjusted net profit down 52.55 percent year on year Ramsey 2026.
The boundary with the rest of this site is deliberate. This page is the operating field: the installed base, the integrator's invoice, and the arithmetic of keeping a cell running. The Reliability Gap is where the demo-to-production argument lives, and the market map is the capital cycle that finances all of it.
See also
- Hardware Taxonomy
Arms, humanoids, hands, sensors, and compute: a buyer's guide from SO-101 to Jetson Thor.
- Teleoperation Rigs
ALOHA, GELLO, UMI, and VR teleop: cost, data quality, throughput, and the embodiment gap.
- The Data Bottleneck
Robot-hours versus LLM tokens: the log-log reality of embodied data and teleop-farm economics.
Linked from
- Hardware Taxonomy
Arms, humanoids, hands, sensors, and compute: a buyer's guide from SO-101 to Jetson Thor.
- The Robot Learning Stack
Data capture, schemas, training, simulation, evaluation, serving and robot integration as one reproducible system rather than a model checkpoint.
- The Bear Case
Why this could be another robotics winter, and the milestones that would prove it wrong.
References
Christopher Müller, IFR Statistical Department, VDMA Services GmbH, Frankfurt am Main, Germany, 2025.
https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf
International Federation of Robotics, IFR press release, 2026-05-05.
https://ifr.org/ifr-press-releases/news/china-makes-ai-powered-robots-core-of-national-strategy
Association for Advancing Automation, A3, 2026-02-06.
https://www.automate.org/robotics/news/robot-orders-grow-6-6-in-2025-as-general-industries-drive-broader-automation-adoption
Symbotic Inc., U.S. Securities and Exchange Commission, 2025.
https://www.sec.gov/Archives/edgar/data/1837240/000183724025000278/sym-20250927.htm
Scott Dresser, Amazon company press, published 2023-10-18.
https://www.aboutamazon.com/news/operations/amazon-introduces-new-robotics-solutions
Tyler Greenawalt, Amazon company press, published 2024-10-09; updated 2026-06-04.
https://www.aboutamazon.com/news/operations/amazon-robotics-robots-fulfillment-center
Alex Davies, Amazon company press, published 2025-05-07; updated 2026-06-04.
https://www.aboutamazon.com/news/operations/amazon-vulcan-robot-pick-stow-touch
Daron Acemoglu, Pascual Restrepo, Journal of Political Economy 128(6), 2188-2244, 2020.
https://doi.org/10.1086/705716
David Autor, David Mindell, Elisabeth Reynolds, MIT Task Force on the Work of the Future, MIT, final report of the Task Force, 2020.
https://ipc.mit.edu/research/work-of-the-future/
Lean Enterprise Institute, Lean Lexicon, n.d.; accessed 2026-09-22.
https://www.lean.org/lexicon-terms/takt-time/
Lean Enterprise Institute, Lean Lexicon, n.d.; accessed 2026-09-22.
https://www.lean.org/lexicon-terms/cycle-time/
NASA, NASA Lessons Learned Information System, Lesson 841, 1994-12-01; submitting organization: jsc.
https://llis.nasa.gov/lesson/841
EVST Engineering Team, EVST (EVS TECH CO., LTD), 2026-07-15.
https://www.evsint.com/palletizing-robot-cost-roi-price-payback-2026/
Emily Hawkins, This is Money, 2025-11-18.
https://www.thisismoney.co.uk/money/markets/article-15303311/Warehouse-closures-crush-Ocado-shares-US-partner-shuts-three-sites-devastating-blow-UK-firm.html
Occupational Safety and Health Administration, U.S. Department of Labor, as of 2026-08-22.
https://www.osha.gov/otm/section-4-safety-hazards/chapter-4
Ocado Group, Ocado Group, as of 2026-08-22.
https://ocadointelligentautomation.com/
Alius Noreika, 2026.
https://www.technology.org/2026/07/18/humanoid-robots-in-2026-what-is-actually-deployed/
Mireya Ramsey, TechTimes, 2026.
https://www.techtimes.com/articles/320197/20260711/robot-boom-meets-earnings-reality-unitree-profits-halved-optimus-not-sale.htm
Joseph Wilkins, CNBC, 2026.
https://www.cnbc.com/2026/07/29/morgan-stanley-humanoid-robots-pr-problem.html
Kyle Belmonte, 2026.
https://www.techtimes.com/articles/316632/20260514/figure-ais-helix-02-robots
Ken Goldberg, Science Robotics, 2025.
https://doi.org/10.1126/scirobotics.aea7390
Jeremy Levine, Talia Goldberg, Janelle Teng Wade, Alexandra Sukin, Bhavik Nagda, Jason Scheller, Christine Deakers, 2026.
https://www.bvp.com/atlas/bessemer-predicts-robotics-and-physical-ai
Spot a factual error or missing qualification? Report a content correction.