NEURA unveils its 4NE1 Gen 4 humanoid as it opens the TUM RoboGym in Munich

What opened in Munich

NEURA Robotics and the Technical University of Munich (TUM) opened the TUM RoboGym powered by NEURA on 22 October 2026. According to NEURA's announcement, the center covers more than 2,300 square meters at LabCampus, the innovation campus next to Munich Airport, and was built with TUM's Munich Institute of Robotics and Machine Intelligence, known as MIRMI. NEURA, which is based in Metzingen near Stuttgart, used the opening to launch the fourth generation of its humanoid robot, 4NE1 Gen 4. The name is read as "For Anyone".

TUM's own press release describes the RoboGym as one of Europe's most advanced training and research centers for embodied AI. At the start, around 100 humanoid and service robots will learn specific skills and processes in 48 training cells. The site also has substantial computing power for AI model training and simulation, and TUM says the current focus is on industrial applications and intralogistics, the movement of goods inside factories and warehouses.

The opening drew Bavaria's two senior ministers for the field, Markus Blume for science and Hubert Aiwanger for economic affairs, along with TUM president Thomas Hofmann and NEURA founder and chief executive David Reger. Hofmann called the center "unique in Germany and Europe" and said working with a German robot maker sends a signal for Germany as an industrial location "in geopolitically uncertain times".

4NE1 Gen 4: what NEURA says is new

NEURA describes 4NE1 Gen 4 as designed for series production rather than as a lab prototype. The company lists four main changes. The body is slimmer and 20 percent lighter than the previous generation. It is covered in a black textile outer skin, which NEURA says makes the robot look more approachable next to people and can be customized more easily. The hands have been redesigned for more accurate grips.

The fourth change is what NEURA calls Smart Limbs: reflexes are built into the limbs themselves, so each limb can react on its own without waiting for a central controller. The company says this lets the robot react instantly to its surroundings and move more fluidly. NEURA has not published a technical paper on how the reflexes work or how they interact with the main control model.

The robot's software brain is Neuron AI, the model AI Lookout covered when NEURA introduced it on 14 October. NEURA calls it a Perception-Language-Action model: it turns what the robot perceives directly into movement, and it takes in touch, force, sound and the robot's own state alongside vision. NEURA's example is an industrial one: a robot that needs to tell whether a fastener has clicked into place can listen and feel for it rather than only look.

NEURA did not publish a price, a shipping date or a full specification sheet for Gen 4 with the launch. It says the robot is being prepared for series production, is already working in the NEURA Gyms, and was also shown at NVIDIA's GTC conference in Berlin this week. Companies training tasks at the Gyms will be among the first to use it.

How the RoboGym trains robots

The central idea is that robots learn physical skills the way people do: by practicing. NEURA's release lists three sources of training data. The first is real-world robot data, where different robots perform tasks and their sensors capture movements, contacts and outcomes. The second is human demonstration, using sensor-equipped data suits, motion-capture systems and teleoperation to turn a worker's movements into training data. The third is simulation, which NEURA calls its Data Amplifier: virtual environments add variations, and the skills are then checked on real robots.

TUM adds that all of this flows into a curated cloud data lake, which it describes as a central scientific resource for research on generalizable robot skills, safe human-robot interaction and AI models for robotics. The computing side includes GPU nodes, networks and storage tied into NEURA's software platform, the Neuraverse, and is used for simulation, reinforcement learning and training large models.

The Neuraverse is also how skills spread. NEURA says a skill learned by one robot becomes available to other connected robots, so every new Gym adds to what the whole fleet can do. Example tasks named by the two partners include folding boxes and placing parts in them, filling bottles, handling cardboard cartons, loading and unloading wooden components, assembling parts and spotting defective parts. TUM also showed researchers steering a NEURA robot with gestures.

The question NEURA says sits at the center of all this is practical: how quickly a robot can learn a new task, and how reliably it can then perform it. Neither partner has yet published numbers on either.

Who is paying, and who trains there

The money is spelled out more clearly than usual for this kind of announcement. NEURA says the joint initiative is set up for an initial three years and that the partners are investing around 17 million euros in total, with NEURA contributing about 11 million. NEURA supplies the robots, technical infrastructure, computing power and operating staff. TUM provides the space, scientific expertise and research resources.

Partners have been training workflows at the site since September. NEURA says companies from more than ten industries are involved, including automotive, pharmaceuticals, medical technology, packaging, logistics, agriculture and care, and that capacity in Munich is already in strong demand. It did not name the companies. The pitch to them is that they can test an application, work out what training it needs and compare robot types before buying, choosing between humanoids, mobile assistants such as NEURA's MiPA and cognitive robot arms.

Reger framed the RoboGym as a way for Germany's Mittelstand, its mid-sized manufacturers, to turn their craft knowledge into robot skills. He also said companies should be able to protect that know-how and decide for themselves where and under what conditions they make it available to others. How that works when the Neuraverse is designed to share skills across robots is one of the details buyers will want in writing.

Why physical AI needs a gym

TUM's release makes the core argument plainly. Chatbots learn from vast amounts of internet text, but robots that act in the world need data from real physical interactions, and that data is scarce. A system that has to grip a delicate part or notice that two pieces did not join properly has to learn how objects, materials and forces behave, and there is no web-scale archive of that.

That is why several robot makers are building training facilities rather than only models. AI Lookout has reported on rivals taking different routes this month, from Agility Robotics standardizing Digit 5 on NVIDIA IGX Thor to Agile Robots' Agile ONE SN reference humanoid. NEURA's bet is that a network of shared gyms, fed by many customers' tasks, will produce more useful data than any single company could collect.

TUM's scientific coordinator for the site, Achim Lilienthal, said close and flexible collaboration with industry is "the only way to systematically and rapidly develop research prototypes into market-ready products" and called the RoboGym a pioneering pilot project for the university. Lorenzo Masia, who directs MIRMI and the RoboGym, said the long-term goal is a dynamic robotics industry that helps drive growth in the German economy.

How it fits Germany's robotics push

Munich is meant to be the first of many sites. NEURA says further Gyms are being set up in Europe, the United States and China, and that another German Gym is being built with RWTH Aachen. The timing also lines up with national policy: days earlier, AI Lookout reported that Germany signed a physical AI pact with NVIDIA, with federal ministries committing to develop open models for physical and industrial AI.

Bavaria's ministers leaned into that framing. Blume said robots "need training before they can help us in our everyday lives" and that LabCampus is "creating their gym", adding that NEURA deliberately chose Bavaria and TUM. Aiwanger pointed to the shortage of skilled workers and said robots that move materials, assemble components and take over physically demanding routine work could ease the load on employees in factories, trades and care.

What NEURA has not said yet

Several practical questions are open. NEURA has not given a list price or delivery schedule for 4NE1 Gen 4, has not said how many units it plans to build, and has not published payload, battery life or speed figures for the new generation alongside the launch. The 20 percent weight reduction is relative to the previous model, whose figures should not be assumed to carry over.

There are also no published results from the RoboGym yet: no measure of how many tasks have been trained, how long training takes or how reliable the learned skills are. The governance of the shared data lake, including what TUM researchers can publish from partner data and how company know-how is kept separate, has only been described in general terms.

What to watch next

Watch for the first named industrial customers, any published benchmarks from the TUM and NEURA research teams, Gen 4 pricing and delivery dates, and the opening dates for the Aachen Gym and the planned sites in the United States and China. If NEURA can show that a skill trained in Munich runs reliably on a customer's line elsewhere, the gym model will have proved its main claim.