Largest German gym for humanoid robots being built in Munich

The Munich Institute of Robotics and Machine Intelligence (MIRMI) at TU Munich (TUM), together with the manufacturer NEURA Robotics, plans to establish the world's largest research and training center for robotics in the scientific field. The partners intend to further develop and train hundreds of robot systems for everyday applications there with AI support, thereby creating a counterweight to competitors from China and the USA. A focus of this work will be on humanoid robots.

The new robotics center is being built in the TUM Convergence Center at Munich Airport, and will be headed by MIRMI professors Lorenzo Masia and Achim Lilienthal. NEURA Robotics is contributing the main share with 11 out of 17 million euros in total investment, with the largest financial portion needed to procure the robots and maintain the hardware. "In return, NEURA Robotics participates in our research," writes Lilienthal in a statement from TUM. "In the very near future, humanoids will be an integral part of our everyday lives and will support people in many tasks," predicts TUM President Thomas Hofmann. Together, they want to accelerate this development and, in addition to individual functions, ensure the safety of robots in direct interaction with humans.

Humans Needed as Trainers

Unlike large language models, for whose training developers find almost inexhaustible data on the internet, hardly any videos can be found there that observe, for example, robot arms performing specific tasks. Furthermore, it is unclear how well movements can be read from videos. Simulations, in turn, are often too imprecise. Moreover, they cannot reliably reproduce physical subtleties such as friction forces in the real world. "These are the reasons why a training center is needed where humans teach robots individual skills," writes Lilienthal. This could involve, for example, folding a box or assembling parts.

With an area of 2300 square meters, the new scientific robotics center is considered one of the largest worldwide. The partners intend to make the majority of the data generated during training available to the robotics community in an open ecosystem. In the RoboGym, the robots are to learn general skills that they can then independently transfer to suitable tasks. The prerequisite for this is a large number of datasets that human trainers must first create and practice with the robots.

(agr)