NEURA's Neuron AI adds touch, sound and memory to the robot foundation model

In short: Most robot foundation models today are vision-language-action (VLA) models: they mostly look at the current camera frame. NEURA Robotics' new Neuron AI adds touch, force, audio and memory, so a robot can act on what the camera can't see and remember objects that leave its view.

What NEURA announced

On October 14, 2026, Metzingen-based NEURA introduced Neuron AI. Instead of treating each sensor separately, the model encodes 3D vision, tactile and force data, audio, robot state and language, aligns them in one shared space over time, and outputs a single stream of actions. NEURA calls this a Perception-Language-Action (PLA) architecture.

Why extra senses matter

NEURA's own examples make the point clearly:

• Opening a lock: only sound and touch confirm it is really open.
• Grasping unseen objects: touch lets a robot grip something its camera can't see.
• Welding: seam quality can be judged from the sound of the process.

Why memory matters

A robot works in a continuous physical world. It needs object permanence, remembering things that are out of sight, and a sense of cause and effect over long tasks. NEURA says it changed the model architecture to add memory and trains it on datasets designed to use it.

Three kinds of training data

1. Real robot data from NEURA Gyms, its training facilities, with synchronized vision, touch, force, audio and motion.
2. Simulation-amplified data from its Data Amplifier and simulation stack.
3. Human demonstrations captured by people wearing sensor suits.

The data trains one shared model across five NEURA robot embodiments. Industrial partners including Schaeffler and Bosch contribute data.

The demo to watch

At NVIDIA GTC Berlin 2026 (October 20–22), NEURA plans to show its humanoid 4NE1 doing an eleven-minute laundry sequence: sorting, loading a washer, closing and reopening it, unloading and moving clothes to a dryer, with no human intervention. The hard part is holding context across many steps, not any single motion.

Why it matters

The race in robot AI is moving from "can it see and follow an instruction" to "can it finish a long, contact-rich job." Multisensory input and memory are two of the missing pieces. NEURA, which announced a Series C of up to €1.2 billion in June 2026, is betting on owning the whole stack: robots, hands, sensors, data and model. Independent results beyond company demos have not been published yet.