Neura Robotics launches Neuron AI physical AI model

Neura Robotics is positioning Neuron AI as a foundation model trained across five of its own robot embodiments and three data sources: real-world robot data from its Neura Gyms, simulation-amplified data through its Data Amplifier, and human demonstration data collected with sensor-equipped suits. That mix is meant to carry both physical interaction and human know-how into a shared model, which Neura can then push to fleets through its Neuraverse platform and allow customers to fine-tune for their own tasks.

The company also frames memory as a specific architectural adaptation rather than a generic feature, tied to object permanence — remembering an object after it leaves view — and to tracking the consequences of a robot's own actions over time. Neura's Time-to-Skill metric, which measures how quickly a robot learns a new task and performs it reliably, is how it intends to track progress for industrial customers.

Neura is anchoring the model in its home market. It says Neuron AI was developed and trained in Metzingen, Germany, and cites data partnerships with industrial companies including Schaeffler and Bosch alongside its Gym training. The upcoming demonstration at Nvidia GTC Berlin 2026, where the humanoid 4NE1 is set to run an eleven-minute laundry-handling sequence without human intervention, appears intended to put planning, multisensory perception, memory and physical action into one visible example for visitors.