NEURA is building gyms to train robots for the real world

NEURA Robotics is building a chain of robot training gyms to give humanoids real-world experience, starting with a 3,000-square-metre centre at RWTH Aachen University opening by the end of 2026. The German firm argues that the biggest bottleneck in Physical AI is not intelligence but experience, and plans to have five of ten planned gyms operational across Europe, the United States and China by the end of 2026. Founder and CEO David Reger said the gyms pair physical practice with simulation to generate training data for the Neuraverse cloud platform, backed by a $1.4bn Series C from NVIDIA, Qualcomm, Amazon and Bosch.

NEURA Robotics is opening ‘gyms’ where robots train under real-world conditions, starting with two German universities. Its bet is that the thing holding humanoids back is not intelligence, but a shortage of real experience. The German robotics firm NEURA has a theory about why robots still fumble in the real world. The problem is not that they are too stupid. It is that they have not practised enough. Its answer is a chain of gyms, where machines work out under real conditions. NEURA has announced https://neura-robotics.com/neura-robotics-rwth-aachen-neura-gym-physical-ai/ a new training centre with RWTH Aachen University, one of Germany’s top engineering schools. The NEURA Gym RWTH Aachen will cover about 3,000 square metres at the Hightech Campus Melaten and open by the end of 2026. Around 20 of the university’s institutes will plug into it. It is one node in a wider build-out. Ten gyms are in development, and five should run by the end of 2026 across Europe, the United States and China. A second German flagship, the TUM RoboGym, opens at Munich Airport later this year with the Technical University of Munich, backed by a €17m commitment. Why robots need a gym The logic starts with a data gap. Large language models learn from trillions of words scraped off the web. Robots get a tiny fraction of that, and it is far harder to gather. Simulations help, but they cannot fully copy the friction and mess of a real factory floor. So NEURA wants robots to log real repetitions. The gyms pair physical practice with high-fidelity simulation, then pipe the resulting training data https://thenextweb.com/news/worldmodeldata-7m-seed-gaming-world-models into the Neuraverse, its cloud platform linking robots, developers and clients. “The biggest bottleneck in Physical AI is no longer intelligence, it’s experience,” said founder and chief executive David Reger. A European bet The siting is deliberate. Two German university flagships, NEURA argues, make the country the backbone of Europe’s physical-AI infrastructure https://thenextweb.com/news/japan-noetra-nvidia-rubin-ai-factory-frontia . RWTH’s rector, Ulrich Rüdiger, framed it as “securing Germany’s leading position in the global AI competition.” That pitch lands in a fraught moment. The company casts itself as building Physical AI “from Europe,” a sovereignty message as the United States and China pour money into humanoid https://thenextweb.com/news/humanoid-152m-series-a-robotics-unicorn-bosch robots. It has the money to back the talk. Its $1.4bn Series C https://thenextweb.com/news/neura-robotics-1-4b-series-c-physical-ai , closed in June, drew in NVIDIA, Qualcomm, Amazon and Bosch. The catch The vision is still mostly blueprints. Most of the gyms do not exist yet, and the flagships open late this year at the earliest. The model also asks industrial partners to feed in hard-won know-how, though NEURA says each one keeps control of its own data. The harder question sits one level up. Humanoids remain largely a demo-stage bet on physical AI https://thenextweb.com/news/onsemi-synaptics-deal-physical-ai , and reps in a gym are not the same as reliable work on a production line. Still, if experience is the real moat, then whoever builds the gym owns the ground the robots learn on. Get the TNW newsletter Get the most important tech news in your inbox each week.