After ChatGPT, Is Physical AI Next? How AI Is Learning to Act in the Real World

For many people, artificial intelligence still means something that lives on a screen.

We ask ChatGPT a question and receive text. Generative AI can create images, summarize documents, write software code, and process audio or video.

A different branch of AI is now receiving much more attention: Physical AI.

The basic idea is straightforward. Instead of limiting AI to digital outputs, Physical AI connects intelligence with cameras, sensors, robots, vehicles, and other machines so that an AI system can perceive its surroundings, reason about what is happening, plan an action, and physically carry it out.

This does not mean that Physical AI is simply “the technology that comes after ChatGPT.” Generative AI will continue developing alongside robotics, autonomous systems, and embodied AI. Physical AI is better understood as a growing convergence between these fields.

South Korea, for example, released a national Physical AI competitiveness strategy in July 2026 and has since been discussing deployment in manufacturing. International standardization work is also beginning to take shape.

What Exactly Is Physical AI?

Physical AI is broader than humanoid robots.

Imagine asking a robot to bring you a bottle from a table.

The robot first needs to identify the bottle and surrounding objects through cameras and sensors. It must estimate where the bottle is, decide how to approach it, plan the movement of its arm and gripper, pick it up, and then use sensor feedback to check whether the action succeeded.

In simple terms:

Perception → Reasoning → Planning → Action → New perception

That feedback loop is at the heart of many Physical AI systems.

NVIDIA describes Physical AI around systems that can understand the real world, reason about it, and take action. Google DeepMind's Gemini Robotics family similarly extends multimodal AI into robotics by translating visual and language information into physical actions.

Generative AI vs. Physical AI

The two categories are not isolated from each other.

In fact, one of the major trends in modern robotics is the use of generative and multimodal models as part of a robot's “brain.”

Why Is Physical AI Getting So Much Attention Now?

Traditional industrial robots are excellent at repeating carefully programmed motions.

The real world, however, is rarely perfectly predictable.

An object may be in a slightly different place. A person may enter the workspace. Lighting can change. A robot may encounter something it has never seen before.

Recent advances in computer vision, multimodal AI, reinforcement learning, sensors, GPU computing, edge AI, simulation, and digital twins are making it increasingly practical to train machines for more flexible behavior.

Simulation is especially important because testing every possible mistake on a real robot is expensive and sometimes dangerous.

NVIDIA, for example, is combining its Cosmos world models, Isaac simulation tools, GR00T robot models, and on-robot computing to build a development stack for training, evaluating, and deploying AI-powered machines.

Physical AI Is Already Here — But Not Everywhere

Some forms of Physical AI are already commercially deployed.

Autonomous driving is a clear example. As of September 2026, Waymo operates fully autonomous ride-hailing services in multiple U.S. cities, although availability and service boundaries vary by location.

Manufacturing is another area where real-world trials are becoming more substantial.

BMW says Figure 02 humanoid robots were tested at its Spartanburg plant for roughly ten months beginning in 2025. According to BMW, the robots supported production involving more than 30,000 BMW X3 vehicles and handled more than 90,000 components.

That does not mean humanoids can now perform every factory job. It demonstrates a much narrower but important point: humanoid robots can already perform selected repetitive tasks under controlled production conditions.

Why Humanoid Robots Get So Much Attention

Humanoid robots are one application of Physical AI, not a definition of it.

Physical AI can also include autonomous cars, drones, warehouse robots, agricultural machines, factory automation, and other intelligent physical systems.

Humanoids receive disproportionate attention because human environments were designed around the human body.

Doors, stairs, tools, shelves, workbenches, and factory lines are generally made for human-sized workers. A robot with roughly human-like proportions could, in theory, work in these spaces without requiring an entirely redesigned environment.

The engineering challenge is enormous, however. A useful humanoid must combine balance, locomotion, perception, reasoning, hand control, real-time computation, and safety.

Boston Dynamics introduced the production version of its all-electric Atlas in January 2026, with deployments planned for Hyundai and Google DeepMind. “Production version,” however, should not be confused with broad, mature adoption across ordinary factories.

Google DeepMind also introduced Gemini Robotics 2 in July 2026, demonstrating whole-body control and dexterous manipulation. The technology remains largely research- and early-access-oriented rather than a mass-market general-purpose robot service.

Why NVIDIA, Google, Automakers and Robotics Companies Are Investing

A capable Physical AI system needs much more than a powerful language model.

It may require:

• cameras and other sensors

• spatial understanding

• multimodal reasoning

• long-horizon planning

• motion and motor control

• low-latency AI processors

• simulation environments

• synthetic and real-world training data

• safety and monitoring systems

That is why semiconductor companies, AI laboratories, automakers, manufacturers, and robotics firms are all participating in the same ecosystem.

The goal is not simply to make robots “smarter.” It is to connect intelligence with reliable physical action.

How Physical AI Could Affect Everyday Life

If the technology continues to improve, Physical AI could expand across many areas:

• robots handling repetitive or physically demanding factory work

• warehouse systems sorting and transporting products

• autonomous vehicles handling more transportation tasks

• agricultural robots monitoring and working with crops

• drones inspecting dangerous infrastructure

• mobile robots operating in hazardous locations

• eventually, service robots helping with selected household or care tasks

The key word is “could.”

These applications are at very different levels of maturity.

Commercial autonomous driving already exists in selected locations. Industrial robots have been common for decades and are becoming more adaptive. General-purpose humanoids operating reliably in unpredictable homes, by contrast, still face much harder technical and economic problems.

The Difficult Problems Have Not Disappeared

A mistake made by an AI chatbot is very different from a mistake made by a moving robot.

A physical machine can collide with a person, damage equipment, misinterpret an object, or behave incorrectly after a sensor failure.

That makes reliability and safety central issues for Physical AI.

Other challenges include:

• hardware cost

• maintenance

• battery life and energy consumption

• sensor errors

• cybersecurity

• privacy

• communication failures

• legal liability after accidents

• employment transitions

• regulation and international standards

In September 2026, ETRI announced Korean participation in a new ITU-T focus group dedicated to embodied AI standardization. The development illustrates how the Physical AI discussion is moving beyond raw performance toward common rules, safety, interoperability, and governance.

Is the Physical AI Era Really About to Arrive?

In one sense, it already has.

Autonomous vehicles and intelligent industrial machines are operating in the real world today, and humanoid robots are moving into increasingly serious pilot programs.

But a robot that can enter almost any environment, understand unfamiliar situations, and perform virtually any human task does not yet exist as a widely deployed commercial product.

The most useful way to understand the transition is this:

Generative AI dramatically expanded what machines can understand and create in the digital world.

Physical AI is about connecting some of that intelligence to perception, movement, and action in the physical world.

That is why one of the next major questions in AI may no longer be only, “What can this model generate?”

It may increasingly become, “What can this AI safely perceive, decide, and do in the real world?”

5. Key Takeaways

• Physical AI is broader than humanoid robotics.

• It connects AI perception and reasoning with actions in the physical world.

• Advances in multimodal AI, robotics, sensors, GPUs, edge computing, and simulation are accelerating development.

• Autonomous vehicles and some industrial applications are already commercially deployed or operating in real-world pilots.

• General-purpose humanoid robots remain much less mature.

• Safety, cost, reliability, cybersecurity, liability, regulation, and standards remain major challenges.

6. FAQ

Is Physical AI the same thing as robotics?

No. Robotics refers to machines and the engineering used to build and control them. Physical AI focuses on adding AI-based perception, reasoning, learning, planning, and adaptive behavior to physical systems.

Is ChatGPT a Physical AI system?

ChatGPT itself primarily operates in the digital domain. However, multimodal and generative AI techniques can be integrated into robotic systems and become part of a Physical AI architecture.

Is Physical AI already being used commercially?

Yes, in some areas. Fully autonomous ride-hailing and intelligent industrial automation provide real examples. General-purpose humanoid robots, however, are still mostly in pilot, research, or limited deployment stages.

Are humanoid robots the main form of Physical AI?

They are one highly visible application, but not the only one. Autonomous vehicles, drones, industrial robots, warehouse systems, and agricultural machines can also incorporate Physical AI.

Will Physical AI replace all human jobs?

Current evidence does not support such a broad conclusion. Automation may reduce the need for some repetitive or hazardous tasks, while also changing or creating work involving deployment, maintenance, supervision, safety, training, and system integration. The outcome will vary substantially across industries and regions.