The future of robotics and automation: where it stands in 2026

Where industrial robotics stands in 2026

The International Federation of Robotics (IFR) counts an industrial robot by the ISO 8373:2021 definition: an automatically controlled manipulator that can be reprogrammed for different tasks, moves in three or more axes, and is fixed in place or mounted on a mobile platform. Welding arms, painting robots, pick-and-place arms and cobots all count; software bots do not.

World Robotics 2026, published on September 24, 2026, reports on 2025:

• Installations: more than 600,000 new industrial robots, up 11% on 2024. The IFR's installations chart puts the figure at 603,000, after three flat years at 541,000 to 553,000, and against 254,000 in 2015.
• Operational stock: a record 5 million robots in service, up 9%, and more than double the stock of seven years earlier.
• Forecast: installations rising 9% to 655,000 in 2026 and reaching 806,000 in 2029.

The largest markets in 2025:

The customer base is changing too. The IFR's press conference presentation says the centre of gravity has shifted from the automotive industry to general industry, where demand quintupled over a decade. The IFR expects the growth to continue for three reasons: manufacturers relocating production into high-wage economies and countries short of workers, demographic change, and advances in AI, machine vision and sensing that widen what robots can do while easier programming and integration cut the cost of deploying them.

Which countries use the most robots

Installations measure how much a country buys; robot density measures how automated its factories are. Density is the number of operational industrial robots per 10,000 manufacturing employees, which puts large and small economies on one scale. The latest figures, published by the IFR in April 2026, are for 2024:

China shows why both measures matter. It operates the world's largest stock, around 2 million industrial robots, about 4.5 times Japan's, yet ranks 22nd on density because its manufacturing workforce is so large. Korea's lead comes from its electronics and automotive industries, the two largest customer industries for industrial robots.

Collaborative robots: sharing the work cell

A traditional industrial robot works behind fences or light curtains, and people enter its cell only when it is stopped. A collaborative robot, or cobot, is designed to share space with people. The IFR describes the range of designs: at one end, conventional robots whose workspace sensors slow them down or stop them when someone walks in; at the other, cobots built with light materials, rounded contours, padding, sensor skins and sensors in the base or joints that keep force and speed under set thresholds if they touch someone.

Cobots made up 11% of industrial robot installations in 2025, according to the World Robotics 2026 presentation. The IFR's cobot statistics count robots designed for collaborative use under ISO 10218-1, the safety standard for the robot itself.

Safety, though, belongs to the application, not the arm alone. ISO 10218-2:2025, published in February 2025 to replace the 2011 edition, covers how robots are integrated into cells with their tools and surrounding machines, and ISO says it complements ISO/TS 15066, which addresses collaborative robot applications. A cobot swinging a sharp part at full speed is not collaborative just because the arm is. The IFR points to the practical gain: people can enter the workspace without first shutting down and securing the cell, a procedure it says can cost thousands of dollars per minute of machine downtime.

Mobile robots in warehouses and logistics

Outside the factory, the IFR counts service robots. In 2025, 250,000 service robots for professional use were sold, including autonomous mobile robots (AMRs), up 24%, and almost every other one was built for transport and logistics, according to the World Robotics 2026 presentation.

Two kinds of mobile robot do most of this work. An automated guided vehicle (AGV) follows routes fixed by markers, tape or wires in the floor, which makes it predictable but slow to change when the layout does. An AMR builds a map with its own sensors and plans its path, so it can go around an obstacle, a pallet or a person. Amazon describes Proteus, its first fully autonomous mobile robot, as able to move carts around employees in open, unrestricted areas of its sites, while its Hercules robots lift up to 1,250 pounds of inventory.

Scale changes the problem from moving one robot to managing traffic. On June 30, 2025, Amazon said it had deployed its millionth robot, across more than 300 facilities, and introduced DeepFleet, a generative AI foundation model trained on its own inventory-movement data that coordinates how the robots move through a building. Amazon says it improves the fleet's travel efficiency by 10%.

Buying is not the only model. The IFR's presentation notes that robots-as-a-service (RaaS), where the customer pays a recurring fee for robots, software and support, lowers adoption barriers considerably. So was the GXO humanoid contract described below.

Humanoid robots: what is announced and what is deployed

Humanoid robots attract most of the attention and a small share of the robots. The IFR counts nearly 7,000 full-size humanoids (taller than 140 cm) sold worldwide in 2025, and has identified 174 humanoid manufacturers, more than half of them in China. Its assessment in the World Robotics 2026 presentation is that humanoids are still moving from laboratories to pilots, with three obstacles: safety and functional reliability, a lack of training data, and a lack of an economically viable business case. The IFR's 2026 trends report sets the bar: to compete with traditional automation, humanoids have to match its cycle times, energy use and maintenance costs.

The case for the form is that factories, warehouses and homes are built for human bodies: stairs, shelves at arm height, carts, doors. The record so far, as of September 2026:

The details show what "deployed" means today. GXO's agreement with Agility, signed on June 27, 2024 after a pilot in late 2023, was presented as the industry's first formal commercial deployment of humanoids and was a RaaS contract: Digit moved totes from cobots onto conveyors at a SPANX facility. By Agility's September 2026 count, Digit had reached 100,000 totes at GXO's Flowery Branch site near Atlanta at about 98% accuracy while on task, and Agility's new Digit 5 adds an independent safety controller so it can work near people without barriers. Figure's BMW report of November 2025 lists its targets (an 84-second cycle, more than 99% of cycles placed correctly, zero interventions per shift) without reporting results against them, and names the forearm as the robot's top hardware failure point.

Figure's Figure 03 update of June 30, 2026 calls the new work a demonstration. Boston Dynamics describes its Hyundai site as a testbed and training centre, with the 25,000-unit rollout planned over the next few years. 1X's order page says plainly that, for tasks NEO does not know, a 1X expert can supervise it remotely at scheduled times.

The pattern is consistent: the humanoids in paid work do one well-defined materials-handling task, such as moving totes or loading parts. General-purpose work remains a plan.

Robot foundation models: from research to paid work

The IFR links the humanoid wave to advances in generative AI such as vision-language-action (VLA) models, part of a broader shift to robot foundation models. Physical Intelligence introduced π0 in October 2024 as a prototype general-purpose robot foundation model: one model, trained on data from many robots and tasks, that can be prompted with a plain instruction or fine-tuned for a job, instead of a robot programmed for the same weld in the same spot. Google DeepMind's Gemini Robotics 2, announced on July 30, 2026, extends its vision-language-action model to whole humanoid bodies. How these models see, learn from demonstrations and plan is covered in our guide to AI in robotics.

Access is still early. As of September 2026, the Gemini Robotics 2 VLA and on-device models are available to early-access partners, with the ER 2 reasoning model in Google AI Studio, and DeepMind itself reports that multi-finger dexterous manipulation remains challenging and its robots need to move faster.

The first paid uses look different from the demos. In February 2026 Physical Intelligence published reports from two partners running its models in live work. Ultra, which packs e-commerce orders in customer warehouses, showed π0.6 packing orders for a full shift at 96.4% autonomy; when the model stumbles, Ultra's human-in-the-loop intervention system steps in, and those interventions generate data that improves the next model. Weave, which folds laundry in a San Francisco laundromat, reported that adding its own data to the model's pretraining cut missed grasps by 42% and interventions by its remote specialists by 50%.

The other early use is perception rather than motion. Since April 8, 2026, Boston Dynamics' inspection software for its Spot robot has used Google's Gemini Robotics ER 1.6 reasoning model for all customers enrolled in AIVI-Learning, reading analog gauges and sight-glass levels and checking 5S compliance. The model does not move the robot; it interprets what the robot's cameras capture.

Software automation: RPA and AI agents in the office

The office has its own "robots", and they have nothing to do with hardware. The IFR's statistics explicitly exclude software bots and robotic process automation (RPA). In Microsoft Power Automate, for example, desktop flows are the RPA feature: they automate rule-based tasks on a workstation, including legacy applications such as terminal emulators, by working with interface elements, images or screen coordinates. That is how RPA reaches systems that have no API, and also why a changed screen can break it.

AI agents are the newer layer. Anthropic draws the line in its guide to building effective agents: in a workflow, language models and tools run along code paths fixed in advance; in an agent, the model directs its own process and decides which tools to use. The label is now stretched. In June 2025 Gartner predicted that more than 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear business value or inadequate risk controls, and warned of "agent washing": assistants, RPA and chatbots rebranded as agents. Gartner estimated that only about 130 of the thousands of agentic AI vendors are real.

Most sound designs mix them, the way a warehouse mixes conveyors, AMRs and people. A document arrives; a model reads it and turns it into structured fields; a rule checks those fields; cases the rule cannot pass go to a person's review queue; and only then does an API call, or an RPA step where there is no API, write the result into the business system.

That is the pattern our AI and automation service builds: workflows, assistants and agents that use a model only where a step needs judgment, with a rule you signed off, or a named person, approving anything that changes data.

What research says about robots, AI and jobs

The most cited evidence on industrial robots comes from Daron Acemoglu and Pascual Restrepo. In Robots and Jobs: Evidence from US Labor Markets, published in the Journal of Political Economy in 2020, they studied the rise in robot use across US local labour markets from 1990 to 2007. One more robot per thousand workers reduced the employment-to-population ratio by about 0.18 to 0.34 percentage points and wages by 0.25% to 0.5%, an effect they separate from imports from China and Mexico, offshoring, the decline of routine jobs and other IT investment.

Germany tells a different story. In The Adjustment of Labor Markets to Robots (Journal of the European Economic Association, 2021), Wolfgang Dauth, Sebastian Findeisen, Jens Suedekum and Nicole Woessner found that robot exposure displaced manufacturing jobs, but new jobs in services fully offset the losses. The adjustment fell mostly on young people entering the labour force; incumbent workers kept more stable employment, often by taking on new tasks in the same plant, and workers in complementary roles such as managers and technical scientists benefited.

Read together, the studies say the effect depends on the country, the period and who is starting out, not that robots either destroy or create jobs on balance everywhere. Employers describe the shift in their own terms: Amazon says the advanced robotics at its next-generation fulfillment center in Shreveport, Louisiana require 30% more employees in reliability, maintenance and engineering roles.

Office automation shows a similar early pattern. The August 2026 revision of "Canaries in the Coal Mine?" by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, based on ADP payroll data through June 2026, finds no evidence of economy-wide job displacement. Employment of workers aged 22 to 25 in AI-exposed occupations, however, stands 19% below where it would be had it kept pace with less-exposed peers, mainly through reduced hiring, and the declines concentrate where AI substitutes for tasks rather than complementing them. The authors present these as early descriptive indicators, not causal estimates. For the full evidence on office jobs, see the impact of AI on white-collar jobs, and for AI's wider effects, how AI is changing society.

What to expect through 2030, and how to weigh the claims

The forecasts point to steady growth, not a break. The IFR expects 806,000 installations in 2029, and its presentation shows Asia growing 11% in 2026 while Europe stays flat and the Americas grow 2%. Hyundai's plan puts Atlas on component assembly by 2030. Gartner expects at least 15% of day-to-day work decisions to be made autonomously through agentic AI by 2028, up from none in 2024.

Announcements will keep outrunning those numbers, so it helps to know what kind of claim you are reading:

For a technical read of a robot video (was it autonomous, sped up, in a scene it had seen before), our AI in robotics guide includes a checklist. For more on these subjects, see the AI topic page.