NEURA introduces robot AI built to combine sight, touch, sound and memory

NEURA Robotics introduced Neuron AI on October 14, 2026, aiming to give its robots more than a camera-based view of their surroundings. The company says the model combines vision, touch, force, sound and language with memory, helping machines act on physical feedback and retain context when objects move out of sight.

Connecting the senses to the next movement

NEURA describes Neuron AI as a foundation model for physical AI: a shared model for machines that perceive and act in the physical world. Its Perception-Language-Action architecture connects sensory information to movement. Rather than handling each sensor separately, it aligns 3D vision, tactile and force data, audio, the robot’s own state and language in one shared representation over time.

The company illustrates the purpose with tasks where appearances are not enough. A camera can locate a lock, but sound and touch help confirm whether it has opened. In welding, the sound of the process can provide information about seam quality. NEURA presents these as reasons to combine senses, rather than rely on images alone.

Memory addresses a different problem: keeping track of what happened between movements. NEURA says it adapted the architecture and uses dedicated training datasets to help the model retain relevant context. That includes object permanence—remembering an object when it is no longer visible—and tracking the consequences of the robot’s own actions.

Three training sources, five robot designs

NEURA says it developed and trained Neuron AI in-house in Metzingen, Germany. The training combines three sources of physical experience, intended to support a shared foundation across five NEURA robot designs with different grippers and hands:

• Real robot experience: Machines working in NEURA’s training facilities, called NEURA Gyms, and other real-world environments generate synchronized vision, touch, force, audio and movement data.
• Simulated variation: The company’s Data Amplifier and simulation tools expand training material with variations of tasks and environments.
• Human demonstrations: People wearing sensor-equipped suits record movements and task demonstrations, adding human expertise to the training pipeline.

Data partnerships with Schaeffler and Bosch complement that training. NEURA says the Gyms connect data collection, skill training and benchmarking before deployment, with the resulting data and skills feeding subsequent model releases. Its Neuraverse platform handles deployment and updates across robot fleets and gives customers tools to fine-tune the model for their own tasks.

Customers can keep their robot skills private. If they choose to share skills or deployment data, NEURA says those contributions can support future model development. The company identifies time-to-skill—how quickly a robot learns a new task and performs it reliably—as a central measure of progress.

Laundry as a test of task coordination

NEURA plans to demonstrate the current model through an 11-minute laundry-handling sequence by its 4NE1 humanoid at NVIDIA GTC Berlin. The company says the robot will sort laundry, load a washing machine, close and reopen it, unload it and transfer the laundry to a dryer without human intervention. That is a planned demonstration, not a completed performance result.

The sequence focuses on handling and coordination, rather than any single movement. NEURA says the robot must carry context through the process, finish each step and transition to the next, bringing planning, multisensory perception, memory and action together. Visitors can see the demonstration from October 20–22, 2026, at Station-Berlin, Hall 3, Booth 3031.