CSIRO launches AI infrastructure to help robots learn in real time

CSIRO has built new infrastructure that brings powerful AI processing much closer to where data is created — right alongside the robots and sensors that rely on it.

As AI rapidly moves from digital systems into the physical world — powering robots, sensors and real‑time decision‑making — the demand for fast, trusted computing close to where data is generated is growing.

Cloud-based approaches alone are no longer enough for safety-critical and time-sensitive applications, prompting governments, researchers and industry to rethink how AI infrastructure is designed and where it is located.

CSIRO’s compact purpose-built infrastructure named Vetra is based at its Queensland Centre for Advanced Technologies (QCAT) in Pullenvale. It delivers high-performance AI computing in a smaller, modular and sustainable footprint, located where real-world testing and research happen.

Unlike traditional, remote cloud-based data infrastructure, Vetra provides super-fast, onsite processing close, or on the ‘edge’ to where data is generated. This allows robots and sensing systems to respond faster, learn continuously and operate more safely in complex physical environments.

Liming Zhu, Director of CSIRO’s Data61, said Vetra delivers sovereign, trusted AI computing at the edge, close in physical proximity to where data is generated by robots and sensing systems.

“AI is rapidly moving beyond digital systems into the physical world, including robots, infrastructure, sensing and safety-critical environments,” Zhu said. “Vetra enables real‑time physical AI research by bringing high performance computing to the edge, where proximity to data allows systems to respond, learn and operate safely in complex environments in ways that are not possible with cloud only or distant data centre approaches.

“This represents a different form of sovereign AI, where physical location becomes part of the capability itself, establishing a model and associated innovative technologies that can be replicated and exported to other locations where onsite, trusted AI is required.”

Vetra sits alongside Australia’s largest robotics research facility, allowing AI systems to learn directly from real-world testing rather than simulations alone.

Dr Peyman Moghadam, Head of CSIRO’s Embodied AI Cluster, said the infrastructure works alongside CSIRO’s larger supercomputing systems in Canberra as part of an integrated ‘edge‑core‑cloud’ approach, handling immediate, local processing first — before sending data to larger centres for deeper analysis.

“Robots and physical AI systems need to keep learning from the physical world, not just from internet datasets or simulations,” Moghadam said. “Vetra gives us the missing edge layer for this workflow, helping turn real-world robotics data into better, safer and more adaptable AI systems.”

The Vetra infrastructure was delivered with the support of Australian small and medium-sized businesses, including Oper8 Global and XENON, alongside global technology partners.

Vetra includes 48 high-performance graphics processing units (GPUs) which can perform thousands of complex mathematical calculations at the same time on large sets of data. The infrastructure has been designed so it can expand over time as demand grows to meet future research and industry needs.

Image: CSIRO’s new AI infrastructure, Vetra, sits alongside Australia’s largest robotics research facility. Credit: CSIRO.