Arm Introduces Physical AI Framework and Robotics Standards for Industrial Automation

The chip designer has brought together more than 80 partner organisations to tackle fragmentation in the physical automation space. Participants in the initiative span software, hardware, and artificial intelligence domains, with notable members including AWS, ECARX, Hugging Face, Liquid AI, NXP, PlusAI, PSYONIC, QNX, Qwen, Siemens, and Unitree Robotics.

Industries focused on physical operations—including mining, agriculture, manufacturing, and global transport—represent trillions of dollars in economic value and present an estimated $200 billion annual compute opportunity by the 2030s. The new framework addresses the need for standardised approaches to system design, integration, and deployment at scale.

Physical systems that combine artificial intelligence models, runtime software, compute silicon, sensors, and actuators require common baselines to reduce integration complexity, streamline compute workloads, and transition from experimental phases to widespread deployment. The initiative seeks to provide these standardised foundations across the ecosystem.

Robotics Capability Framework establishes common language

Robotics lacks a unified approach for describing, comparing, and communicating system capabilities, according to an architectural manifesto published by Arm chief architect Richard Grisenthwaite. This absence of common standards complicates the design, integration, and scaling of robotic systems in industrial settings.

To address this gap, Arm has introduced the Robotics Capability Framework as a foundation for establishing shared technical terminology. The framework draws inspiration from the SAE Levels methodology used in driving automation.

The framework organises robotic systems into progressive tiers reflecting increasing operational sophistication, ranging from reactive configurations to context-aware, cognitive, and self-improving machines. Each tier connects practical applications to machine behaviours, outputs, and hardware requirements.

The capability tiers define specifications for system latency, compute placement, memory allocation, power constraints, determinism, and safety standards. Arm created the initial version drawing on input from the robotics sector, with contributors including Anaxi Labs, ANYbotics, FMC³ Robotics, Fourier, GALBOT, Gravis Robotics, Lenovo, McKinsey, and Robotec.ai.

Virtual platforms support pre-silicon development for autonomous systems

Arm Total Design for Physical AI builds on a collaborative model previously applied to cloud artificial intelligence infrastructure. The programme integrates artificial intelligence models, virtual platforms, digital twins, sensors, compute silicon, and software stacks to enable earlier development and validation cycles.

Autonomous transport and robotics share overlapping technical needs in sensory perception, artificial intelligence processing, real-time control, safety, and power-efficient computing. Arm demonstrated this collaborative approach in the automotive domain working with AWS, Google, HERE, RemotiveLabs, and Siemens.

The automotive partners created an integrated digital cockpit reference solution that allowed software teams to develop, test, and validate complex automotive applications on the Arm Zena CSS platform before physical silicon became available. Arm is now inviting technical contributions from the broader engineering community to refine the Robotics Capability Framework as physical artificial intelligence deployments advance.