Germany’s Neura Robotics Launches Its Own A.I. Model for Robots
Trending Topics Jakob Steinschaden
Neura Robotics has built its own AI model for robots, trained in Germany. It uses touch, sound and memory too, so the machines aren’t just staring at the world.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Neura Robotics has decided its robots should not rely on borrowed brains. The Metzingen, Germany company has unveiled Neuron AI, a model it says was developed and trained entirely in Germany and built to give robots more than vision alone. Cameras stay in the mix, but so do touch, hearing and memory.
Neura calls the system a perception-language-action model, or PLA. It pulls together 3D images, touch and force data, sounds, the robot’s own state and language, then turns that into movement. The company’s point is simple enough: a camera can tell you shape and colour, but it cannot tell you whether a lock has actually clicked open or whether a welding seam sounds right.
Memory is the other big piece. Neuron AI is meant to keep track of objects that move out of sight and remember the effects of its own actions over longer stretches of time. David Reger, Neura’s founder and chief executive, says an intelligent robot needs a brain that can combine its senses and remember what happened. Neura is also betting that one model can serve all of its robots, across five form factors and different grippers and hands.
The training recipe mixes three kinds of data: robots doing tasks in Neura’s training halls, called Neura Gyms, and in real-world settings; simulations that create more task and environment variations; and human movement captured by sensor-packed suits. Customers can then train Neuron AI on their own tasks through the Neuraverse platform and push new skills across entire robot fleets. Neura says its main metric is “time-to-skill” — how quickly a robot learns a task reliably.
Next week, at Nvidia’s GTC developer conference in Berlin, Neura plans to show its humanoid robot 4NE1 spending 11 minutes fully autonomously sorting laundry, loading a washing machine and then moving the laundry to a dryer. That demo is doing a lot of work for the company. So is the larger strategy: make its own model, keep more control over development and pricing, and avoid handing sensitive industrial data to someone else.
There is plenty of money behind the bet. Nvidia and Amazon joined Neura’s Series C round of up to €1.2 billion, led by the stablecoin company Tether in the middle of the year, and media reports have placed the valuation at about €6 billion. But Neura has not published benchmark results for Neuron AI, so for now the company is asking the market to trust the pitch before it sees the scoreboard.
My take — AI-written commentary, not fact-checked reporting
The cleanest move in robotics right now is to own the brain, not rent it from a Silicon Valley neighbour with a shiny demo. Neura is doing the sensible, slightly stubborn German thing: treat industrial know-how like industrial know-how, not training fodder. That seems less sexy than a giant U.S. platform, which is probably why it has a chance.
Read more about this at: Trending Topics
Related stories
NEURA Robotics acquires Bosch Rexroth’s ACTIVE Shuttle to expand Physical AI ecosystem
Tech.eu · 2 months ago ·
43
NEURA Robotics acquires ADLATUS to bring Physical AI to autonomous cleaning
Tech.eu · 1 month ago ·
20
Neura Robotics Puts Robots to Work Pushing Hospital Beds
Trending Topics · 1 week ago ·
33