NEURA Robotics launched Neuron AI on Wednesday, its own robot foundation model that adds touch, hearing and memory to camera vision. The model combines sensory inputs with language and the robot’s internal state to guide physical actions. NEURA has published no benchmark for the model, and its first public test is an 11-minute laundry run at Nvidia GTC Berlin next week.

“An intelligent robot needs a brain that brings its senses together and remembers what happened,” said David Reger, NEURA’s founder and CEO.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

What the model senses

NEURA calls its architecture Perception-Language-Action, or PLA, distinguishing it from vision-language-action models focused on camera inputs. It processes depth images, tactile and force-torque data alongside sound. Memory lets the robot remember objects once they are out of sight and keep track of the consequences of its own actions over time, often without seeing those consequences.

In its announcement, NEURA describes how the model would use its senses in examples that include confirming that a lock has opened through its click and resistance, gripping an unseen object by touch and judging a welding seam by sound.

Training and fleet deployment

Developed and trained in-house in Metzingen, Germany, Neuron AI draws on robot data from NEURA Gyms and real environments. Its Data Amplifier adds simulated variations of tasks. People wearing sensor-equipped suits contribute demonstrations, while Schaeffler and Bosch provide data partnerships.

The model covers five NEURA robot types: 4NE1 Gen 3, 4NE1 Gen 4, a wheeled 4NE1 and two further bimanual platforms. It runs locally without requiring a cloud connection.

Through Neuraverse, customers can fine-tune the model and distribute skills across fleets. They can keep process know-how private or share skills and deployment data for future models. Schaeffler agreed in November 2025 to deploy a mid-four-digit number of 4NE1 units by 2035.

An in-house model

NEURA was an early adopter of Nvidia’s Isaac GR00T models. Nvidia invested in NEURA’s Series C financing round of up to $1.4 billion, announced June 10, 2026, led by Tether, with Amazon also among the investors. The Neuron AI demonstration runs at Nvidia’s own conference. Munich rival Agile Robots relies on a Google DeepMind partnership for its industrial robots.

No benchmark yet

As of Wednesday’s launch, the capabilities are NEURA’s own description, NEURA has published no Neuron AI benchmark results, and no independent model evaluation has yet been published. It names Time-to-Skill, the time required to learn a task and perform it reliably, as its central metric, but has released no figures for learning times, success rates or adaptation costs.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Google’s published figure for Gemini 4 Argon, positioned for robot control, is 87 percent on the ERQA spatial-reasoning benchmark. That is Google’s own result. Touch, sound and memory, the capabilities NEURA is selling alongside camera vision, are outside the scope of what that benchmark measures.

The laundry demonstration

In June 2026, four months before Neuron AI was announced, a 4NE1 Gen 3.5 running Qualcomm’s Dragonwing reference design collapsed during a live Computex keynote in Taipei, falling onto its knees.

At Nvidia GTC Berlin, October 20-22, 2026, a 4NE1 running the current Neuron AI version is scheduled to perform an 11-minute laundry sequence without human intervention. It will sort clothes, load a washer, close and reopen it, unload it and transfer laundry to a dryer.

NEURA is taking refundable €100 reservations for the 4NE1 Gen 3.5, with availability promised for the end of 2026.

Frequently Asked Questions

What is Neuron AI?

Neuron AI is NEURA Robotics' own foundation model for robots, launched October 14, 2026. It combines 3D vision, tactile and force data, sound, the robot's internal state and language with memory. NEURA calls the architecture Perception-Language-Action, or PLA.

How is Neuron AI different from vision-language-action models?

NEURA says vision-language-action models rely on camera inputs, while Neuron AI adds touch, sound and memory. Memory lets a robot remember objects once they are out of sight and keep track of the consequences of its own actions.

Has NEURA published benchmark results for Neuron AI?

No. As of the launch, NEURA had published no Neuron AI benchmark results, and no independent evaluation had been published. It names Time-to-Skill as its central metric but has released no figures for learning times, success rates or adaptation costs.

Which robots run Neuron AI?

The model covers five NEURA robot types: 4NE1 Gen 3, 4NE1 Gen 4, a wheeled 4NE1 and two further bimanual platforms. It runs locally on the robot without requiring a cloud connection.

When will Neuron AI be shown publicly?

At Nvidia GTC Berlin, October 20-22, 2026, a 4NE1 humanoid running the current Neuron AI version is scheduled to perform an 11-minute laundry sequence without human intervention, from sorting clothes to moving them into a dryer.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Unitree Founder Says Humanoid Robots' ChatGPT Moment Is Up to 10 Years Away
Wang Xingxing, the founder of Unitree, said the humanoid robot industry's "ChatGPT moment" could take as long as 10 years. The breakthrough depends on robots learning to handle unfamiliar tasks withou
The Implicator
Germany's Soofi S AI Model Tops All Open-Source Rivals on German Benchmarks
A German research consortium coordinated by the KI Bundesverband released Soofi S, an open-source German-English foundation model, this week, according to its pretraining report. In the team's tests,
The Implicator
Hugging Face Buys Robot Maker to Push Open Source AI Hardware
The acquisition centers on Reachy 2, a $70,000 humanoid robot that runs on open source software. Researchers at Cornell and Carnegie Mellon already use it to test AI applications. Research Roots Pol
The Implicator

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai