Key points
- The planned Italian center is the first of three facilities intended to train humanoid robots through repeated real-world and simulated tasks.
- Robot data collected in industrial settings will feed digital twins that run in thousands of scenarios on AMD computing infrastructure.
- Optimized world models will be transferred back to physical robots for validation, creating a recurring reality-simulation-reality training loop.
Context
The project combines physical robot fleets, digital twins, and high-performance computing to train AI models for embodied tasks. AMD describes this initiative as part of its effort to bolster Europe's technological sovereignty in AI and robotics.
What this means for you
Not yet happened
Developers can use the project as a reference for combining robot data, digital twins, simulation and hardware validation, but the training infrastructure will not operate until late 2027.
Who benefits: Developers and architects designing physical AI training systems could benefit from the facility's reference architecture, while others have no immediate need to act.
Share article: