NEURA 4NE1 Gen 4: What Changed and How TUM RoboGym Trains It

NEURA Robotics introduced 4NE1 Gen 4 on October 22 alongside the opening of TUM RoboGym near Munich Airport. The humanoid is already working inside training facilities, but is still being prepared for series production. For companies considering robotics, the immediate opportunity is testing tasks rather than ordering a finished general-purpose worker.

What changed in 4NE1 Gen 4

According to NEURA's announcement, the new generation is 20% lighter than its predecessor, with a slimmer body and textile exterior. Updated hands target more precise handling, while its Smart Limbs put reflex responses inside individual limbs.

These are manufacturer claims, not independent measurements of performance. Lower weight and more capable hands are useful changes, but neither tells a buyer how many successful cycles the robot can complete before someone intervenes.

The release does not provide a purchase price or firm customer delivery date. Operating in a training facility is different from being available for routine production deployment. That distinction matters more than the launch label.

What TUM RoboGym actually provides

TUM's facility Q&A describes 48 training cells containing approximately 100 robots. The mix includes humanoids, service robots and robotic arms. Researchers can use the center, while external universities, institutes and companies can request access.

The center collects physical interaction data and human demonstrations for training robotics models. Its stated goal goes beyond teaching isolated tasks: the partners want learned manipulation skills to transfer to new applications and different robot platforms. Companies can work with NEURA on application-specific training; research collaborations go through TUM.

That gives a potential customer a more useful starting point than a demonstration video. A packaging company could bring a representative box-folding task, define what counts as a completed cycle and examine where the system needs help. This is an example of how to evaluate the offering, not a claim that a particular customer has done so.

How Neuron AI fits into the robot

4NE1 Gen 4 uses Neuron AI, NEURA's Perception-Language-Action model. The model was announced separately on October 14, so the software itself is not a new October 22 release.

NEURA says it combines vision, touch, force, sound, language and robot-state information with memory. Training draws on robot interactions, simulated variations and human demonstrations. The company describes a shared foundation across five robot embodiments, with customer fine-tuning and skill deployment through Neuraverse. Customers can keep their trained skills private.

For a box-folding task, this suggests a useful evaluation question: can the system detect an incomplete fold or a bad grip, rather than simply repeat the expected movement? The value of additional sensing depends on whether it changes the robot's response when the task goes wrong.

What buyers still need to measure

The announcements do not establish independent task success rates or production economics for Gen 4. Those gaps should shape a pilot. Measure interventions per shift, successful cycles, recovery time and the effort needed when materials or layouts change.

A successful run on one carefully arranged workstation is a starting point. A stronger result would show the same task holding up across changed object positions, worn materials and interruptions, without collecting an entirely new dataset each time.

The useful test for this combined hardware and training launch is therefore fairly concrete: does the next customer's task become quicker to teach, validate and maintain because previous robots have already learned something relevant? That is where shared robotics training would start to earn its place in a deployment budget.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn complex AI launches and product updates into clear stories developers and buyers can understand and share.

Book a 30-minute call.