Robotics Enters Its AI Era as Service Robots, Factory Automation and Embodied Intelligence Accelerate

SteelGate

Sep 30

10 min read

Updated: Oct 2

STEELGATE · AI · TECHNOLOGY

Professional service robot shipments jumped 24% in 2025 as logistics, medical systems and increasingly capable AI-driven machines moved deeper into commercial use. But the biggest story may be what is happening underneath the hardware: robotics is beginning to inherit the rapid advances of modern artificial intelligence.

The global robotics market entered 2026 with two trends unfolding at the same time.

The first is scale.

Factories, warehouses, hospitals and commercial facilities are deploying more robots than ever before. Global shipments of professional service robots increased 24% in 2025 to nearly 250,000 units, according to new data from the International Federation of Robotics.

The second is intelligence.

A new generation of artificial-intelligence systems is beginning to change how robots perceive their surroundings, interpret human instructions, plan tasks and adapt when the physical world does not behave exactly as expected.

Together, those developments are pushing robotics beyond the traditional model of machines programmed to repeat narrowly defined movements.

The result is an industry increasingly divided into two layers: a large, established automation market already producing measurable economic value, and an emerging field of AI-powered machines attempting to become more flexible, autonomous and general-purpose.

That distinction matters.

The robots generating most commercial deployments today are still not humanoids. They are warehouse vehicles, industrial arms, cleaning machines, surgical systems and other machines engineered around specific jobs.

But the software being developed for humanoids and general-purpose robotics could eventually make many of those machines substantially more capable.

Professional Service Robots Approach 250,000 Annual Shipments

The International Federation of Robotics reported on September 30 that global shipments of professional service robots rose 24% in 2025 to almost 250,000 units.

Transportation and logistics remained the industry's largest professional application by a wide margin.

Approximately 117,500 transportation and logistics robots were sold during the year, an increase of 21%, representing roughly 47% of the professional service robot market.

These machines typically move goods through warehouses and factories, transport materials between workstations, replenish inventory or support delivery operations.

Their success illustrates an important feature of robotics economics: robots tend to scale fastest where the environment and task can be controlled.

A warehouse aisle is considerably easier for a machine to understand than an unpredictable household. A robot transporting a pallet between predefined locations faces fewer variables than a humanoid being asked to clean an unfamiliar kitchen.

That is one reason logistics has become one of the clearest commercial proving grounds for autonomy.

The business model is changing as well.

IFR reports that direct sales remain common, but Robots-as-a-Service, or RaaS, continues to expand. In the United States, IFR says RaaS has already become the dominant business model for logistics service robots.

Instead of purchasing an entire fleet upfront, customers can effectively treat robotic capacity more like an operating expense.

That can reduce the initial financial barrier to automation while allowing robotics companies to generate recurring revenue through deployment, software, maintenance and fleet-management services.

In economic terms, the shift resembles what cloud computing did to enterprise computing infrastructure: turning expensive equipment that once required large upfront investment into capacity that can increasingly be purchased as a service.

The Larger Automation Market Is Already Much Bigger

Service robotics is growing quickly, but it sits beside an industrial robotics market operating on a much larger installed base.

IFR's World Robotics 2026 report shows that factories worldwide installed more than 600,000 industrial robots during 2025, an 11% increase from the previous year.

The number of industrial robots operating globally reached approximately 5 million, up 9% in a single year.

That installed base has more than doubled in seven years.

China remains the center of this expansion.

Factories in China installed approximately 354,000 industrial robots in 2025, up 20% from the previous year and representing 59% of worldwide installations. Chinese robot manufacturers themselves supplied approximately 195,000 units domestically, giving local manufacturers a 55% share of their home market.

The United States moved into second place globally, installing nearly 38,500 industrial robots in 2025, a 12% increase. Japan installed 36,219 units, while South Korea installed roughly 30,000.

Those numbers illustrate why discussions about robotics should not focus exclusively on futuristic humanoids.

Automation is already occurring at industrial scale.

What AI may change is the range of jobs those machines can perform.

From Programmed Automation to Adaptive Automation

Traditional industrial robots are extraordinarily capable when their environment is predictable.

A robotic arm can weld the same location on thousands of vehicle bodies with exceptional precision. A packaging robot can repeat the same pick-and-place operation millions of times.

But change the object, move the workstation or introduce an unexpected obstacle and traditional automation can require reprogramming.

Modern AI is attacking that limitation.

Computer vision allows robots to recognize objects and environments.

Machine learning can help systems optimize movement.

Natural-language models allow operators to communicate instructions without writing conventional robotic code.

Multimodal AI can combine video, images, language, sensor readings and spatial information.

And increasingly, vision-language-action models attempt to translate perception and language directly into physical movement.

IFR itself identifies artificial intelligence, machine vision and sensing as important drivers expanding the range of robot applications and reducing programming and integration requirements.

This represents a potentially significant shift in the economics of automation.

Historically, buying a robot was only part of the cost.

Companies also needed engineers, integrators and programmers to build the surrounding system and teach the robot precisely what to do.

If AI reduces that integration burden, automation could become practical for smaller manufacturers and lower-volume production environments where traditional robotics previously made less economic sense.

Google, NVIDIA and the Rise of 'Physical AI'

Some of the most important robotics development is now happening inside companies traditionally associated with artificial intelligence and computing rather than industrial machinery.

Google DeepMind has developed a series of Gemini Robotics models designed specifically to connect multimodal artificial intelligence with physical machines.

Its newest Gemini Robotics 2 system combines vision-language-action models with what Google describes as embodied reasoning.

Instead of merely recognizing what is visible through a robot's cameras, the system is designed to understand a physical task, plan multiple steps and translate those instructions into movements.

Google says Gemini Robotics 2 can control systems ranging from robotic arms to full humanoid bodies and can handle whole-body movement, dexterous manipulation and multi-robot collaboration.

The company's Robotics ER 2 model functions as a higher-level planner. It can interpret instructions, understand physical surroundings, break work into steps and coordinate execution with lower-level robot-control models.

This distinction between a reasoning layer and a movement layer could become important.

One AI system effectively decides what should happen next.

Another determines how the robot's motors should physically accomplish it.

NVIDIA is pursuing a related strategy through its Isaac robotics platform and GR00T project.

NVIDIA describes Isaac GR00T as a general-purpose humanoid robotics platform combining robot foundation models, training data, simulation environments, accelerated computing software and Jetson hardware for local robotic inference and control.

That architecture reveals another major piece of the emerging robotics industry: simulation.

Training physical robots exclusively in the physical world is expensive, slow and potentially dangerous.

Simulation allows developers to generate environments, failures, object configurations and movement examples digitally before transferring those learned behaviors to physical machines.

The long-term objective is essentially to create a robotics development pipeline that looks increasingly like modern AI training:

data → simulation → model training → robotic deployment → real-world feedback → additional training.

If that feedback loop becomes efficient, robotic capabilities could improve much faster than systems that must be individually programmed and tested.

Medical Robotics Shows What Mature Commercialization Can Look Like

Healthcare offers another view of what happens when robotics successfully crosses from experimental technology into established infrastructure.

IFR reported that approximately 8,400 surgical robots were sold globally in 2025, representing growth of 57%.

One of the clearest examples is Intuitive Surgical's da Vinci ecosystem.

Intuitive reported that approximately 3.15 million da Vinci procedures were performed globally during 2025, an 18% increase from approximately 2.68 million in 2024.

The company placed 1,721 da Vinci systems during the year and ended 2025 with an installed base of approximately 11,106 systems worldwide, up 12%.

The significance extends beyond unit sales.

Medical robotics increasingly operates as an ecosystem involving hardware, instruments, software, procedure volumes, training and recurring usage.

That model offers an important lesson for the broader robotics industry.

The value of robotics may ultimately come less from selling machines individually and more from the recurring economic activity built around each deployed machine.

Humanoids Remain Small — but Strategically Important

Against these much larger established markets, humanoid robotics remains relatively small.

IFR estimates that approximately 7,000 full-size humanoid robots taller than 140 centimeters were sold globally in 2025 for professional and commercial applications outside research and entertainment.

That is a fraction of the hundreds of thousands of professional and industrial robots deployed during the same period.

The gap provides a useful reality check.

Humanoids attract disproportionate attention because their potential application range is enormous. But commercially, the category remains much earlier in its development.

IFR notes that many current humanoid deployments remain specialized and still rely on human teleoperation.

Significant obstacles remain around safety standards, reliability, training, maintenance expenses, manipulation and the basic economic case for using a humanoid instead of a simpler machine.

That final question may become one of the most important in robotics:

Does the task actually require a human-shaped machine?

For moving pallets, an autonomous mobile platform may be cheaper and more efficient.

For welding vehicles, a stationary industrial arm may be faster.

For cleaning floors, wheels and rotating brushes generally make more sense than legs and hands.

Humanoids become more compelling when businesses want automation inside environments already designed for human workers—stairs, shelves, tools, doors, factory stations and equipment engineered around human reach and movement.

That is why manufacturing and logistics are becoming important testing environments for humanoid systems.

Where AGI Fits — and Where It Does Not

The relationship between robotics and artificial general intelligence requires careful distinction.

There is currently no broadly accepted demonstration that AGI has been achieved.

Today's most advanced robotic systems are better described as examples of embodied AI or physical AI: artificial-intelligence models designed to perceive, reason about and act within the physical world.

But robotics could become one of the most important testing grounds for increasingly general AI systems.

Digital AI operates in relatively forgiving environments.

A language model generating an incorrect sentence can generate another sentence.

A physical robot making an incorrect movement could damage equipment, drop an object or injure a person.

That changes the standard required for reliability.

A genuinely general-purpose robot would need to combine several abilities simultaneously:

• visual perception • spatial reasoning • natural-language understanding • memory • task planning • motor contro l• manipulation • navigation • error detection • adaptation • understanding of human intent • safety awareness

Modern multimodal models are beginning to combine several of those capabilities.

Google DeepMind explicitly describes its Gemini Robotics work as part of a longer-term path toward general-purpose physical intelligence and ultimately AGI in the physical world.

That remains the company's stated research direction rather than evidence that such AGI exists today.

The near-term importance is more practical.

AI does not need to reach AGI for robotics to become dramatically more useful.

A warehouse robot that can understand ten times as many environments without manual programming has economic value.

A factory robot that can learn a new manipulation task from a small number of demonstrations has economic value.

A medical system that helps clinicians perform procedures more consistently has economic value.

A mobile robot capable of navigating an unfamiliar facility without a precisely mapped route has economic value.

The commercial threshold is therefore much lower than the AGI threshold.

The Hard Problems Are Becoming Physical

AI progress does not eliminate the difficult engineering problems that distinguish robotics from software.

Batteries still determine how long mobile systems can operate.

Motors and actuators determine speed, strength and precision.

Hands remain exceptionally difficult to reproduce mechanically.

Sensors must function reliably in changing light, dust, vibration and temperature.

Robots operating around people require extensive physical safety systems.

And every machine eventually needs maintenance.

Economics may be an even stronger constraint.

A robot can be technologically impressive and still fail commercially if its purchase price, maintenance cost, energy requirements, downtime and supervision expenses exceed the value of the labor or productivity it replaces.

This helps explain why specialized machines continue to dominate actual robot sales.

They solve narrow problems economically.

General-purpose robotics is attempting something much harder: building machines capable of solving many problems without becoming prohibitively expensive.

Robotics Is Becoming a Stack

The emerging robotics industry increasingly resembles a technology stack rather than a simple hardware market.

At the bottom are semiconductors, batteries, actuators, motors, sensors and manufacturing capacity.

Above them sit robotic platforms: industrial arms, autonomous mobile robots, surgical systems, drones and humanoid bodies.

Above the hardware sit simulation environments, perception software, mapping systems and motion-control systems.

And increasingly, an AI layer sits on top—foundation models, vision-language-action systems and autonomous agents capable of interpreting what the robot sees and deciding what it should do.

That software layer may ultimately determine how rapidly robotics expands beyond today's predefined applications.

The 2025 sales data suggest that the commercial foundation is already substantial.

Approximately 5 million industrial robots now operate in factories.

More than 600,000 new industrial robots were installed during 2025.

Nearly 250,000 professional service robots were shipped.

More than 117,000 logistics robots entered the market.

Approximately 8,400 surgical robots were sold.

And roughly 7,000 full-size humanoids entered professional or commercial applications.

Meanwhile, consumer robotics expanded even faster in unit terms, with IFR reporting a 37% increase to nearly 34.2 million consumer service robots sold during 2025.

Those numbers describe an industry that has already moved far beyond the research laboratory.

But they also show that the robotics revolution currently unfolding looks different from the one often presented in futuristic demonstrations.

The present belongs primarily to specialized robots.

The next phase may increasingly belong to specialized hardware powered by more general intelligence.

And if today's rapidly developing AI systems eventually learn to reliably understand, reason about and manipulate the physical world, the dividing line between a robot built for one job and a machine capable of learning many jobs could begin to disappear.

That—not the appearance of a human-shaped shell—may be the real transition to watch.

Primary and First-Party Sources