Arm Launches Total Design for Physical AI With Over 80 Developers

The automotive and robotics industries are working through many of the same problems: perception, AI inference, real-time control, functional safety, and power-efficient computing. Source: Arm

Arm Holdings PLC has launched Arm Total Design for Physical AI, a program that now counts more than 80 participating companies across the physical AI technology stack. Members include AWS, ECARX, Hugging Face, Liquid AI, NXP, PlusAI, Psyonic, QNX, Qwen, Siemens, and Unitree Robotics, among others.

Dermot O'Driscoll, Arm's vice president of go-to-market and customer solutions for physical AI, told The Robot Report that the program's goal is to bring together companies working in physical AI that are looking for better ways to cooperate. He said Arm drew on its earlier experience building a comparable community in the cloud ecosystem.

O'Driscoll recalled that when he moved into the physical AI space in March, he noticed the same gap: there was no place for companies to share technology and ideas, or to collaborate.

Robotics remains a fragmented industry that draws on many disciplines, and a single finished system may combine hardware and software from numerous vendors. Arm said it hopes to reduce the friction that comes with that fragmentation.

O'Driscoll noted that many teams wanted to build on Arm architecture but struggled to form partnerships or get components to work together. He said it was therefore worth assembling an ecosystem in which Arm could seed its IP and technology while partners contribute their own.

Arm Total Design for Physical AI pulls together software stacks, AI models, sensors, compute hardware, virtual platforms, and digital twins so companies can develop and validate interoperable systems earlier. O'Driscoll described it as a point of coherence spanning silicon vendors, cloud providers, actuation and sensing suppliers, software framework developers, safety and security specialists, robotics firms, and end customers.

Arm also introduced a Robotics Capability Framework, which defines levels of increasing sophistication for robotic systems. O'Driscoll said the framework can help both industry insiders and outsiders understand robots more clearly.

He noted that robots are already widely deployed, but the way the industry talks about and defines them remains fragmented. The initial framework identifies six levels, from zero to five.

At the low end are robots that respond to stimuli and execute commands within fixed rules. At the high end are systems whose behavior evolves as they learn and self-optimize. Arm says it is open to adjusting these levels and wants feedback from the wider industry.

O'Driscoll said the six original levels stood out as the kinds of robots being deployed today and offer a clear picture of where companies are taking the field. He added that the work is better done as an ecosystem and invited others to join.

The framework maps robotic sophistication from reactive systems through context-aware, cognitive, and self-improving ones. Source: Arm

Arm has a long history in automotive, and O'Driscoll said essentially every vehicle on the road today runs at least some of its computing on an Arm chip. That experience showed the company that robotics and automotive manufacturers share requirements around safety criticality, real-time computing, and computer vision.

The framework took inspiration from SAE International's levels of autonomous driving. O'Driscoll said robotics is more complex and therefore needs a multi-dimensional framework, which is what prompted Arm to bring more organization to the space.

On the Arm Total Design side, he added that automotive companies recognize the overlap with robotics and are interested in participating. He said many of them see that technology developed for vehicles could also serve robots, which made it easy for Arm to connect with existing partners.