Giant 25,000-Square-Foot Robot Gym in Germany to Train Hundreds of Humanoid Robots

The partners are investing USD 19.8 million (EUR 17 million) in the training hub, with NEURA Robotics contributing USD 12.8 million (EUR 11 million) for robots and infrastructure.

“In return, NEURA Robotics participates in our research,” Achim Lilienthal, PhD, a computer science professor at TUM, claimed. “The interaction between high-end robotics technology and cutting-edge academic research in artificial intelligence will give development a huge boost.”

The future robot gym

Lilienthal will reportedly co-lead the initiative along with colleague Lorenzo Masia, PhD, a professor of intelligent biorobotic systems, and executive director of the Munich Institute of Robotics and Machine Intelligence (MIRMI).

The duo aims to speed up progress in humanoid robotics and embodied artificial intelligence (AI). As per the researchers, the RoboGym will function as a training ground where robots can repeatedly practice tasks while learning from humans.

“Humanoid robots have long since left the realm of science fiction,” Thomas F. Hoffman, PhD, TUM president explained. “In the near future, they will become an integral part of everyday life and support humans in many tasks.”

Humanoid robots are the main focus of the TUM RoboGym (powered by NEURA). Credit: NEURA Robotics

At the 25,000-square-foot training site, the team plans to develop AI-supported methods that allow robots to acquire general skills. These abilities can later be applied to specific real-world situations.

The skills could include tasks such as folding boxes, assembling components, or even manipulating objects, activities that still remain challenging for many robotic systems.

“The decisive competitive factor in intelligent robotics is no longer mechanics, but data,” David Reger, NEURA Robotics founder and CEO, reported. “Those who have high-quality, realistic training data set the pace.”

The smart robotics race

As per Lilienthal, the center will address the lack of high-quality training data, one of the biggest challenges in embodied AI. He said that unlike language models such as ChatGPT, embodied AI can’t use vast amounts of online data for training.

Meanwhile, web videos featuring robot arms carrying out specific tasks are hard to find. Also, they rarely capture the complexity of robotic manipulation. What’s more, simulations also remain too imprecise.

Lilienthal emphasized that robots require physical, real-world data to understand movement and interaction with objects. This large dataset of movements is now being developed in RoboGym by human trainers.

David Reger, NEURA Robotics founder and CEO. Credit: NEURA Robotics

“This is why we need a training center where people can teach robots individual skills, such as folding a box or assembling components,” Lilienthal said in a press release. “In RoboGym, robots will learn general capabilities that they can then independently apply to specific tasks.”

The facility is expected to host hundreds of robots. Apart from research, the site will also serve as a training ground for students and engineers. Masia believes the project will help strengthen Europe’s position in robotics by providing one of the most advanced robotics infrastructures around the globe.

“European sovereignty is extremely important in times of geopolitical competition between East and West,” he concluded. The site will give scientists and students a unique hub to develop and test new robotics and AI approaches.