Neura Robotics unveils Neuron AI, one robot model that adds touch, sound and memory to vision and language across five of its robot designs

Neura Robotics of Metzingen, Germany, introduced Neuron AI on 14 October, calling it “its multimodal foundation model for Physical AI”. The model takes in 3D vision, touch and force readings, audio, the robot's own state and spoken or written instructions, and a memory function lets it keep track of objects after they move out of view.

Neura says it developed and trained the model in-house. The training data comes from robots working in its “NEURA Gyms”, from data multiplied in simulation, and from people wearing sensor suits; Schaeffler and Bosch are named as data partners. One shared model runs across five Neura robot designs with different grippers and hands. Customers can fine-tune it and roll new skills out across their fleets through Neura's Neuraverse platform.

“An intelligent robot needs a brain that brings its senses together and remembers what happened,” said founder and chief executive David Reger.

The release gives no benchmark figures, no price and no date for general availability. Its first public test is set for Nvidia's GTC Berlin conference on 20–22 October, where Neura says its 4NE1 humanoid will run an eleven-minute fully autonomous laundry routine: sorting clothes, loading and unloading a washer, and moving laundry to a dryer.

Neura is building its own robot brain rather than licensing one, the opposite route to the one Physical Intelligence offered other companies on 14 October, when it opened an API beta for its robot models.

• Claimed Neuron AI combines 3D vision, touch and force, audio, robot state, language and memory in one model shared across five robot designs. Neura Robotics
• Claimed The 4NE1 humanoid will run an eleven-minute autonomous laundry routine at Nvidia GTC Berlin, 20–22 October. Neura Robotics
• Confirmed The release gives no benchmark, price or availability date. Neura Robotics