Arm Wants to Create a Common Language for Robots and Physical AI
The AI race is moving out of computers and into machines that can see, think, move, and act in the physical world. Now Arm wants the robotics industry to agree on a common way to describe just how intelligent those machines really are.
Chip technology company Arm has launched Arm Total Design for Physical AI, bringing together more than 80 companies across artificial intelligence, robotics, automotive technology, semiconductors, cloud computing, sensors, and software.
At the same time, Arm introduced a new Robotics Capability Framework, an effort to create a shared technical language for describing and comparing increasingly intelligent robots and autonomous machines.
The initiative brings together companies including:
AWS,
ECARX,
Hugging Face,
Liquid AI,
NXP,
PlusAI,
PSYONIC,
QNX,
Qwen,
Siemens,
and Unitree Robotics.
Arm says the goal is to reduce one of the biggest obstacles facing physical AI today:
fragmentation.
Different companies currently use different definitions for autonomy, intelligence, perception, reasoning, safety, and robotic capability.
That makes machines harder to compare, systems harder to integrate, and large-scale deployment more complicated.
Arm wants to change that.
What Exactly Is Physical AI?
Most of today's popular AI operates inside screens.
ChatGPT writes.
Gemini answers questions.
Image generators create pictures.
AI coding systems generate software.
Physical AI goes further.
It combines artificial intelligence with machines capable of interacting with the real world.
That includes:
humanoid robots,
industrial robots,
autonomous vehicles,
warehouse machines,
agricultural equipment,
mining systems,
delivery robots,
drones,
and other intelligent machines.
These systems must do more than process information.
They need to:
sense,
reason,
decide,
and then act.
That requires the integration of AI models, software, processors, sensors, actuators, cameras, memory, networking, real-time controls, and safety systems.
Arm says industries including mining, agriculture, manufacturing, transportation, and logistics represent trillions of dollars in global economic activity.
The company estimates that physical AI could create an approximately $200 billion annual compute opportunity during the 2030s. That figure is Arm's own market estimate.
The Robotics Industry Has a Language Problem
Imagine two companies both saying they have an "autonomous robot."
Those robots may actually be very different.
One may simply follow a programmed route.
Another may recognize obstacles and adapt its movement.
Another could understand its environment, plan a new task, manipulate unfamiliar objects, and learn from experience.
Yet all three might still be described as autonomous.
That creates a problem for:
manufacturers,
buyers,
engineers,
regulators,
insurance companies,
integrators,
and investors.
What exactly can the robot do?
How independently can it operate?
What happens when conditions change?
How much human supervision does it require?
How quickly must it respond?
Where does the AI processing happen?
What are the power and memory requirements?
And most importantly:
Can it be trusted?
Arm believes the industry needs a shared vocabulary similar to the SAE Levels of Driving Automation used to classify autonomous vehicles.
Six Levels of Robotic Intelligence
Arm's proposed Robotics Capability Framework currently organizes robotic systems across six levels of sophistication, from RL0 through RL5.
At the lower levels are machines that primarily react to predefined conditions.
As capability increases, systems become more context-aware and capable of handling changing environments.
At the highest levels are increasingly cognitive and potentially self-improving systems.
Arm's framework connects those capabilities with technical requirements including:
latency,
compute placement,
memory,
power consumption,
determinism,
real-time behavior,
and safety.
This is important because intelligence alone does not make a useful robot.
A robot working inside a factory may need to respond within milliseconds.
A surgical robot may have extremely strict safety requirements.
An agricultural robot may need to operate for hours on limited power.
A humanoid may need enormous amounts of perception and AI processing while still balancing power consumption, heat, mobility, and response time.
Different levels of capability therefore create different hardware requirements.
But This Is Not Yet a Final Global Standard
This distinction matters.
Arm is not declaring that the robotics industry has already adopted RL0 to RL5 as a universal standard.
The company describes the Robotics Capability Framework as a starting point for collaboration.
Its initial design was informed by companies and organizations including:
Anaxi Labs,
ANYbotics,
FMC³ Robotics,
Fourier,
GALBOT,
Gravis Robotics,
Lenovo,
McKinsey,
Robotec.ai,
and others.
Arm is now inviting additional companies, researchers, engineers, and industry groups to help refine the framework.
The goal is not necessarily to dictate how every robot must be built.
Arm specifically says the framework is intended to provide common definitions while allowing companies to maintain control over their architectures, technologies, and product differentiation.
That is an important distinction.
Arm wants companies to speak the same language without forcing everyone to build the same robot.
More Than 80 Companies Join Arm's Physical AI Ecosystem
The second major announcement is Arm Total Design for Physical AI.
This expands Arm's collaborative Total Design model beyond cloud AI infrastructure and into robotics, autonomous vehicles, and other intelligent machines.
More than 80 companies are participating across different layers of the technology stack.
The ecosystem connects technologies including:
AI models,
robotics software,
operating systems,
processors,
sensors,
cloud platforms,
virtual development environments,
digital twins,
and hardware manufacturers.
This matters because a robot is not one technology.
A modern intelligent machine might require an AI model from one company, sensors from another, processors from another, real-time software from another, cloud infrastructure from another, and mechanical systems from yet another supplier.
Every integration point creates engineering complexity.
Arm wants Total Design to allow those companies to collaborate earlier.
Building Robots Before the Chips Even Exist
One particularly interesting part of Arm's strategy is the use of virtual platforms and digital twins.
Instead of waiting for physical hardware to be manufactured before software development begins, engineering teams can test software in simulated environments.
Arm has already demonstrated this approach in automotive development.
Arm, AWS, Google, HERE, RemotiveLabs, and Siemens collaborated on an integrated digital cockpit reference system.
Developers could develop, test, and validate automotive software on the Arm Zena CSS platform before the actual silicon became available.
This can dramatically change development timelines.
Traditionally:
hardware gets designed,
chips are manufactured,
prototype systems are assembled,
then software engineers begin testing.
With virtual platforms, several stages can happen simultaneously.
Software developers can begin writing and validating code while the physical hardware is still being designed.
For autonomous machines and robots, where hardware and software are deeply interconnected, that could shorten development cycles considerably.
Why Arm Cares About Robots
Arm is best known for processor architectures used across billions of devices.
Its technology powers smartphones, embedded systems, automotive systems, IoT devices, servers, and increasingly AI systems.
Physical AI could significantly expand demand for computing outside traditional PCs and data centers.
A humanoid robot may contain multiple processors.
An autonomous vehicle can contain hundreds of compute cores.
A factory could eventually operate thousands of intelligent machines.
Warehouses could contain fleets of autonomous mobile robots.
Agricultural operations could deploy AI-enabled tractors, drones, and robotic harvesting systems.
That creates a potentially enormous market for:
processors,
AI accelerators,
memory,
sensors,
software,
networking,
and edge computing.
Arm wants its architecture to become one of the foundations underneath that future.
The AI Race Is Moving Into the Physical World
This is part of a much bigger transformation.
The first major wave of generative AI focused on content.
AI learned to generate:
text,
images,
code,
music,
video,
and voices.
The next phase is increasingly focused on action.
An AI agent can already decide which digital task should happen next.
Physical AI extends that capability into machines.
A warehouse robot decides where to move.
A drone decides how to navigate.
A car decides whether to brake.
A humanoid decides how to pick up an unfamiliar object.
An agricultural robot decides which plant needs attention.
A mining machine adjusts its behavior based on terrain.
Once AI begins controlling physical systems, the consequences become much more serious.
A bad chatbot answer may inconvenience someone.
A bad robotic decision could damage equipment, stop a production line, crash a vehicle, or injure a person.
That is why common definitions, safety requirements, deterministic behavior, real-time response, and trusted computing become increasingly important.
The Bigger Battle May Be About Standards
History shows that technology standards can become extremely powerful.
USB standardized connections.
Wi-Fi standardized wireless networking.
Bluetooth standardized short-range communications.
SAE levels helped establish a common vocabulary for automated driving.
If robotics becomes one of the world's largest industries, whoever helps establish the language and technical foundations around robotics could influence how that industry develops.
Arm is clearly positioning itself to play that role.
But whether its Robotics Capability Framework becomes widely adopted will depend on something Arm cannot decide alone:
whether the rest of the robotics industry agrees.
That is why bringing more than 80 organizations together may ultimately be more important than the framework itself.
What This Means for the Philippines
The Philippines should not look at physical AI as science fiction.
Robotics and autonomous systems could eventually transform industries where the country already has significant economic activity.
Manufacturing
Factories could use increasingly intelligent robots for:
assembly,
inspection,
packaging,
maintenance,
and quality control.
Agriculture
AI-powered machines could assist with:
crop monitoring,
precision spraying,
harvesting,
soil analysis,
and pest detection.
Logistics
Warehouses and ports could use robots for:
sorting,
movement of goods,
inventory management,
container operations,
and last-mile delivery.
Disaster Response
Physical AI could eventually become extremely valuable during:
typhoons,
earthquakes,
landslides,
fires,
and search-and-rescue operations.
Robots can enter environments too dangerous for humans.
Healthcare
Robotic systems could assist with:
rehabilitation,
remote care,
hospital logistics,
prosthetics,
and eventually more advanced surgical applications.
And because the Philippines has a large engineering, electronics, semiconductor, IT, and BPO workforce, there is another opportunity.
We do not have to manufacture every robot ourselves.
Filipinos could participate in the physical AI economy through:
software development,
AI model development,
robotics testing,
voice interfaces,
language localization,
simulation,
digital twins,
data annotation,
electronics,
and system integration.
The opportunity is not only to use robots.
It is to become part of the global supply chain that builds the intelligence behind them.
From AI That Talks to AI That Acts
For the last few years, artificial intelligence has amazed people because machines could suddenly speak, write, draw, and reason.
Physical AI changes the stakes.
The machine no longer simply tells you what it thinks.
It does something about it.
It moves.
It drives.
It lifts.
It flies.
It manufactures.
It delivers.
It interacts with the physical world.
And when AI starts taking action in the real world, industries will need common ways to understand exactly what those machines can and cannot do.
That is the problem Arm is trying to solve.
The next chapter of artificial intelligence may not happen only inside ChatGPT or a data center.
It may happen inside cars, factories, farms, warehouses, drones, and humanoid robots.
And before those machines can truly operate everywhere, the industry may first need to agree on a simple question:
How intelligent is this robot, really?
