Arm is recruiting the robotics industry around shared technical standards, hoping to end the fragmentation that keeps physical AI systems from scaling.
The chip designer launched Arm Total Design for Physical AI on September 8 alongside a Robotics Capability Framework meant to give machines a common way to describe what they can do. More than 80 partners have signed on, including AWS, Hugging Face, NXP, QNX, Qwen, Siemens, Unitree Robotics and PlusAI.
Chief architect Richard Grisenthwaite published a manifesto arguing that robotics lacks a shared vocabulary for comparing system capabilities, making design, integration and deployment needlessly hard. The framework borrows the tiered approach of the SAE levels used in driving automation, grading robots from simple reactive machines up to context-aware and self-improving systems.
Each tier links real use cases to hardware demands such as latency, compute placement, memory, power and safety requirements.
Arm estimates physical industries from mining to agriculture and transport represent a compute opportunity worth around $200B a year by the 2030s, and it wants its chip architectures underneath them. The push extends the Total Design playbook Arm used in cloud infrastructure and automotive, where partners built a digital cockpit reference on the Zena CSS platform.
