Factories installed more than 600,000 industrial robots in 2025, an increase of 11% on the previous year, and more than five million are now operating worldwide. The figures come from the International Federation of Robotics (IFR) and its World Robotics 2026 report. The global operational stock grew by 9%. This is how industrial robot installations stand at the end of the year, along with several other developments that bear on where the field is heading.
China widens its lead as Europe slips
China's factories installed 354,000 robots, which is 59% of the global total. Chinese manufacturers also supplied 55% of their domestic market.
The United States overtook Japan to become the second-largest market, with almost 38,500 installations. India installed nearly 10,500 robots, up 15%. Europe moved the other way. Installations fell by 8% in Germany, 11% in Italy, 8% in France and 15% in Spain.
IFR forecasts that worldwide installations will reach 655,000 units in 2026 and 806,000 in 2029.
Toyota has given investors a separate indication of scale. The company said that modernizing factories run by Toyota, its group businesses and major suppliers could require around 400,000 robots. The estimate covers replacement machines, new industrial robots, automated logistics systems and possible future humanoid deployments. Toyota said the programme could cost ¥1 trillion, or approximately $6.4 billion, annually from 2028. It has not made a firm spending commitment.
On the product side, Universal Robots launched Gen 7 at the IMTS trade show in Chicago. It is a redesigned collaborative robot platform built around greater computing power, tool-flange connectivity, sensor integration and support for AI-assisted applications.
The people training robots in Indian homes
In Rajasthan, India, women are being paid to record household activities from cameras worn on their heads. The footage shows cooking, cleaning and folding clothes from the worker's point of view.
The recordings can be divided into individual actions and used to teach robots how people handle objects in real homes. Household settings contain clutter, variation and improvised behaviour that are hard to reproduce in a laboratory, which makes this data valuable.
An investigation into the work estimates that between 20,000 and 30,000 people may be doing it in India. That figure is an extrapolation, not an official count. The investigation also does not establish which robot companies ultimately receive the featured workers' recordings.
Autonomous weapons talks move to November
Three years of discussions by the United Nations Group of Governmental Experts ended in September with growing support for formal negotiations on new international rules. Human Rights Watch says 76 states now back negotiations on legally binding measures covering autonomous weapons systems.
The group's final report was weakened in negotiation. References to ethical concerns and to the design and development of the weapons were removed or reduced after objections from states including Russia and the United States.
The issue now goes to the Convention on Certain Conventional Weapons Review Conference in November, where states may decide whether to establish a negotiating mandate. The Stop Killer Robots campaign argues that further discussion is no longer enough and that states should begin negotiating prohibitions and regulations.
Computer scientist Stuart Russell also joined the debate in a short UC Berkeley video. He called for a ban on small autonomous anti-personnel weapons and warned that machines able to select and attack people could become cheap, scalable weapons of mass destruction. His position is narrower than a call to prohibit every military use of AI. It targets systems that can independently identify human targets and apply lethal force at scale.
Research: fairness in security robots, and a hand that walks
A study in the International Journal of Human–Computer Interaction examined whether perceived fairness can reduce conflict between people and security robots. Xin Ye and Lionel Robert looked at two forms. Distributive fairness concerns whether a robot's decision or its consequences appear fair. Informational fairness concerns whether the robot gives an adequate explanation for the decision. Both reduced conflict and improved acceptance. The results come from controlled human–robot interaction research, not from robots exercising authority in operational policing.
Separately, researchers at ETH Zurich have taught an anthropomorphic robotic hand to walk on its fingertips. The self-contained hand uses reinforcement-learning policies to crawl, steer and recover after falling. It can support its own weight while pressing successive keyboard commands, and it can push an object towards targets using overhead visual feedback. The work is an eight-page arXiv preprint, and separate policies handle different tasks.
What the numbers do and do not show
The installation data shows a market that is growing and concentrating. China's 59% share is larger than the combined share of every other country, and the decline in four major European markets sits against double-digit growth in India. IFR's forecast of 806,000 units in 2029 would be about a third above the 2025 level of roughly 600,000, a trajectory that depends on demand holding up in the countries now contracting.
Toyota's figure is worth reading carefully. An estimate of 400,000 robots across one company's supply base is equal to about two-thirds of everything installed worldwide in 2025. Toyota presented it as a need, not an order, with no firm spending commitment and with humanoids listed only as a possibility. It shows how large a single manufacturing ecosystem's appetite could be, but it is not yet demand.
The home-recording story points to a quieter constraint. Manipulation research needs real-world data, and that data is being produced by paid human labour at a scale no one has verified. The 20,000 to 30,000 estimate is an extrapolation, and the destination of the footage is unknown. Until those gaps are closed, claims about how much household data is feeding robot learning cannot be tested.
The regulatory picture is similarly unsettled. Support from 76 states is a clear signal, but the weakened final report shows that the largest military powers can still dilute text. November is the first real test of whether a negotiating mandate follows.
The two research items are early. The fairness study supports a design principle, that security robots should explain themselves, but it was run in controlled settings. The fingertip-walking hand is a preprint with task-specific policies, so it does not demonstrate a single controller that moves, balances and manipulates. Both are worth tracking, and neither has yet been shown to work outside the lab.