Suche
Newsletter abonnieren
News
Branchenverzeichnis
Sonderhefte/Specials
Shuttle-Profis
Neura Robotics: New "Robo Gym" to Train Robots for Industrial Tasks
Neura Robotics and TU Munich Open Training Center for Robots
Neura Robotics and the Technical University of Munich inaugurated the joint research and training center "TUM Robo Gym powered by Neura" at the Lab Campus of Munich Airport on October 22, 2026. On more than 2,300 square meters, robots are to be trained for industrial and other practical tasks. The partners are investing around 17 million euros in the collaboration, initially planned for three years. At the opening, the robotics manufacturer also presented a new generation of its humanoid robot, the "4NE1 Gen 4."
The center was established in collaboration with the Munich Institute of Robotics and Machine Intelligence at the Technical University of Munich. It aims to combine scientific research with specific business requirements. The focus is on developing robotic capabilities using artificial intelligence, which can capture and respond to physical processes. This approach is referred to as Physical AI.
Companies provide real-world challenges and their knowledge of industrial processes. Researchers and development teams investigate how robots can learn the respective tasks and execute them reliably under different conditions.
Robotics Training for More Than Ten Industries
Since September, practical workflows have been trained in the "TUM Robo Gym". According to the project partners, companies from more than ten industries are participating, including the automotive industry, pharmaceutical industry, medical technology, packaging industry, logistics, agriculture, and care.
The tasks so far include handling boxes, filling bottles, and loading and unloading wood components. The participating companies can investigate the applications of various robotic systems, determine the required training effort, and test applications under practical conditions.
Different forms of robot construction are available for this. In addition to humanoid robots, mobile assistance robots like "MiPA" and cognitive robotic arms are used. The choice of the specific system depends on the requirements of the task to be processed.
The shared infrastructure is intended to enable companies to investigate robotic applications before expanding their use in their own operations. Both established industrial companies and young technology companies can collaborate with scientific teams.
Three Approaches for Training Artificial Intelligence
An essential component of the "Robo Gym" is the collection and processing of data on physical workflows. The center combines three approaches for training the robotic systems: real robot data, human demonstrations, and simulation-based training methods. In collecting real robot data, various systems perform specific tasks. Their sensors record movements, contacts, and the results of individual actions. The information obtained is used to analyze the behavior of the robots during task execution.
Additionally, humans demonstrate certain activities. Sensory data suits, motion capture systems, and methods for remote control of robots are used. In this way, human movements and interactions are intended to be usable as training data.
The third approach uses virtual training environments. With simulations, real processes can be expanded with additional situations and altered conditions. Neura refers to these methods for expanding the data foundation as a "Data Amplifier." The skills trained with these are then to be tested on real robotic systems.
The training infrastructure is connected to the software platform "Neuraverse." It provides tools for data processing and the development of artificial intelligence. For collaborative research, training data is structured and used for developing and investigating corresponding models.
4NE1 Gen 4 Developed for Mass Production
Parallel to the opening, Neura Robotics presented the fourth generation of its humanoid robot "4NE1." The new "4NE1 Gen 4" was developed for future mass production and is, according to the company, in the corresponding preparation phase. Compared to the previous generation, the weight was reduced by 20 percent. At the same time, the robot features a slimmer design and a black textile outer skin into which sensors are integrated. The new design is geared towards applications in close proximity to humans.
The handling of objects has also been revised. Advanced robot hands are intended to allow more precise gripping movements. Additionally, the "Smart Limbs" technology enables individual limbs to have their own reaction functions. As a result, they are supposed to react immediately to external influences without requiring each adjustment to be centrally controlled. The changed motion control is intended to enable the robot to perform its movements more fluidly and respond to changes in its environment.
Neuron AI processes different sensor data
For the control of the new generation of robots, Neura Robotics relies on the self-developed model "Neuron AI". This is a so-called Perception-Language-Action model, which processes information about the environment and derives motion actions from it.
According to the manufacturer, in addition to visual information, the system also considers touches, forces, acoustic signals, and internal state data of the robot. This combination is said to be particularly important for industrial handling tasks where visual information alone is not sufficient.
An application example is the assembling of components. A robot must not only recognize the position of individual components but also determine, if necessary, whether a connection has been correctly established based on force and touch information.
Neuron AI is trained with real robot data, human demonstrations, and simulations. Capabilities that a robot system has learned are to be made available to other connected robots via Neuraverse.
The 4NE1 Gen 4 is already in use at the manufacturer's training facilities. Companies can test tasks with the new generation there and develop appropriate training routines. The announcement does not contain any information regarding the general market availability or sales prices.
17 million euros for joint research infrastructure
The collaboration between Neura Robotics and the Technical University of Munich is initially planned for three years. The partners are investing a total of around 17 million euros. Of this, approximately eleven million euros are allocated to Neura Robotics. The manufacturer contributes robotics systems, technical infrastructure, computing power, and operating personnel. The university participates with space, scientific expertise, and other resources for research and operation.
The center is available to researchers, students, and companies as a joint development and training environment. Scientific approaches can be examined there based on practical tasks. At the same time, requirements from industrial applications are intended to provide new research questions for robotics and artificial intelligence. The technical infrastructure is thus geared both towards the development of new robotic capabilities and their verification under application-oriented conditions.
International network of training centers planned
The "TUM Robo Gym" in Munich forms the starting point for a planned international network of training facilities by the manufacturer. Further Neura Gyms are to be established in Europe, the United States, and China. In Germany, an additional facility is planned in cooperation with the Rheinisch-Westfälische Technische Hochschule Aachen.
Across locations, training data and developed capabilities are to be made usable for other connected robot systems. In this way, Neura aims to reduce the effort required for learning new tasks and to bring together results from different training environments.
Translated automatically from German.
Branchenverzeichnis
