NEURA Robotics and Germany's RWTH Aachen University open a 3,000 square meter robot training gym—the first node of a 10 facility physical AI network, a system of gyms where robots learn from real world motion and touch rather than text scraped from
Physical AI, software that lets robots learn from real-world motion and touch rather than text scraped from the web, is training on a sliver of the data that fed today's language models. NEURA Robotics and RWTH Aachen University are opening a 3,000-square-meter gym for robots on RWTH's Hightech Campus Melaten to start closing that gap.
The Aachen facility is the first of 10 "NEURA Gyms" NEURA says it will stand up across Europe, the United States and China, with roughly half, or about five, targeted to be operational by the end of 2026. A companion center, the TUM RoboGym, is already in motion at Munich Airport with TU Munich's robotics institute, covering more than 2,300 square meters.
Each gym pairs physical practice with high-fidelity simulation and feeds an open, cloud-based development platform NEURA calls the "Neuravese," where partners can train and validate robots before industrial deployment. The company is backing the network from a Series C of up to $1.4 billion announced earlier this year.
NEURA says real-world friction and variability still cannot be simulated away, and that the gap with text-trained models is the bottleneck for humanoids and industrial robots. The company has not disclosed customers or independent benchmarks for the platform, and no third-party reporting yet confirms the global buildout. Eight more sites, and the data to fill them, are still to come.