Summary: As robotic capabilities and intelligence levels continue to rise, the new challenge facing the industry is how to integrate artificial intelligence (AI) inference, perception, control, safety, and orchestration capabilities into a unified system, enabling them to operate reliably at scale in the real world.
At last week's Arm Everywhere China conference, Arm launched the Robotics Capability Framework and the Arm Total Design for Physical AI program, aiming to establish a unified capability language for the robotics industry and promote industrial collaboration and scaled innovation. Meanwhile, Arm released its latest robotics white paper, analyzing the computing and software foundations required for the continuous evolution of robotic capabilities from a system architecture perspective, providing new references for the industry to understand the relationship between robotic capabilities and underlying system requirements.
The robotics industry is entering a new stage of development. Advances in AI are continuously expanding robots' abilities to perceive, reason, and interact in the physical world, driving robotics technology from proof-of-concept and demonstration stages toward large-scale deployment, and landing in real-world scenarios across manufacturing, logistics, healthcare, agriculture, and many other fields. According to data from the International Federation of Robotics (IFR), the global market size for industrial robot deployments has reached a historic high of $16.7 billion, and this growth trend is expected to accelerate further as robots enter more industries and undertake more complex real-world tasks.
For robotics companies, the current challenge lies in translating successful product demonstrations into commercially viable products that can operate safely, reliably, and cost-effectively in real-world environments. Each new AI capability imposes higher demands on computing power, software, energy efficiency, cost, safety, and system integration. This raises a fundamental system-level challenge: how to equip machines with reliable intelligence, enabling them to perceive, move, adapt to their environment, and execute tasks safely in real time?
Federico Pecora, Global Head of Physical AI Robotics Research at Arm, explores this issue in depth in his newly released technical white paper, "The Missing Scientific Foundations for Robotic Systems." The white paper proposes a system-level approach for the next stage of Physical AI development, clarifying why integrating AI inference, perception, planning, control, safety mechanisms, and orchestration capabilities into a unified platform is essential for robots to operate reliably in the real world.
Prerequisites for robots to operate in the real worldconditions
A robot is essentially a continuously operating autonomous system. As it moves and operates in a changing physical environment, it needs to synergistically integrate capabilities such as camera visual data, LiDAR, mmWave radar, force sensors, motor control, AI inference, motion planning, safety monitoring, and real-time feedback to achieve continuous perception, decision-making, and action execution.
These workloads have varying execution cycles but often need to be performed simultaneously, each with distinctly different computational demands. For instance, motor control requires predictable and deterministic timing; perception processing and AI inference demand high-throughput computing power; motion planning needs access to sensor data and environmental awareness; while safety monitoring requires higher priority and isolation protection.
Meanwhile, various functions need to efficiently share and collaboratively process data to avoid latency, performance bottlenecks, or unpredictable system behavior. Ensuring the seamless collaborative operation of multiple workloads is one of the biggest engineering challenges facing modern robotic systems.
As robots take on increasingly complex tasks, engineering teams need to understand from a system level the impact of each capability on the overall architecture. For example, larger models, more comprehensive world models, or longer-chain continuous inference mechanisms can expand the capability boundaries of Physical AI systems, but this also significantly increases demands on storage systems, memory bandwidth, resource scheduling, and real-time operational control.
Underlying System Foundations for Robots and Physical AI
Whether industrial robots, autonomous vehicles, humanoid robots, or service robots, they all require systems capable of allocating, scheduling, isolating, and coordinating different workloads, while ensuring the predictability of real-time control and safety-critical functions. Although different types of robots have varying computational needs, they all require a system that can effectively coordinate intelligent decision-making with physical execution under real-world constraints.
This makes workload allocation a core decision in robot design. Engineering teams need to determine: which tasks are executed by the CPU, which are handled by accelerators, and which should be deployed in the real-time domain; simultaneously plan data transmission between different functional modules, and define which workloads require priority scheduling and which must be isolated.
AI accelerators are responsible for executing AI models, while the CPU typically handles extensive coordination and scheduling tasks for the entire system: managing sensors, scheduling workloads, transferring data between processors, and maintaining OS operations. Even as AI workloads become increasingly complex, it is essential to ensure that safety-critical software continues to run stably.
Meeting the development needs of next-generation robots requires a computing foundation that balances performance, energy efficiency, safety, and software portability across various workloads. Arm is committed to reducing the complexity of robotic system development and accelerating industry innovation by continuously deepening its research and understanding of robotic systems and architectures, helping engineering teams transform cutting-edge concepts into deployable product capabilities.
Building the Future of the Robotics Industry with Arm
The next stage of development for the robotics industry depends on whether systems can translate AI capabilities into stable and reliable physical execution. Robots not only need to perceive, reason, and adapt to complex environments, but they must also maintain predictable control, efficient energy management, and safe, reliable operation throughout the process.
Arm is helping to make this industry vision a reality by providing a computing foundation for the robotics ecosystem, designed for heterogeneous computing, high energy efficiency, and scalable systems. From high-performance application processing to low-power perception, control, and safety-critical task execution, Arm empowers engineering teams to build robots that achieve higher intelligence, greater energy efficiency, and enhanced reliability in the real world.
Federico Pecora's latest technical paper delves into the emerging research direction of "capability integration" and the critical questions computational architects need to consider and address as Physical AI moves from the lab to actual deployment. The industry outlook is currently very promising: the robotics industry is transitioning from the prototype stage, which showcases impressive technical results, to a phase of stable, reliable, large-scale deployment, helping engineers create intelligent machines capable of large-scale perception, movement, adaptation, and safe operation in real-world environments.
From elucidating the underlying technological logic supporting the evolution of robotic capabilities in the robotics white paper, to promoting a unified capability language across the industry through the Robotics Capability Framework, and further gathering ecosystem innovation power via the Arm Total Design for Physical AI program, Arm is joining hands with industry partners to build the key foundation for the development of Physical AI, accelerating the transition of intelligent robots from technological innovation to large-scale deployment.
– End –