See, touch, remember: NEURA Robotics launches its own AI brain for robots
The Metzingen-based company unveils Neuron AI – a foundation model that combines vision, touch, sound and language and remembers what the robot has already done.
Photo: Web Summit / Wikimedia Commons, CC BY 2.0 (cropped and resized)
Video: YouTube · NEURA Robotics · youtube.com
German robot maker NEURA Robotics of Metzingen on Wednesday unveiled Neuron AI, a so-called foundation model for "physical AI", meaning AI that acts in the real world. It links 3D vision, touch, audio, the state of the robot's body and language with a memory. This is meant to let a robot know, for example, where an object is even when it cannot currently see it, and which work steps have already been completed. NEURA calls the approach "Perception-Language-Action".
According to the company, the model was developed and trained in Metzingen. The training data come from real robot deployments in the so-called Neura Gyms, from simulations and from demonstrations by people wearing sensor suits; Bosch and Schaeffler contribute as data partners. Neuron AI is to run on five robot types with different hands and grippers. Customers can adapt the model on the Neuraverse platform, and can keep the skills they teach private.
"An intelligent robot needs a brain that brings its senses together and remembers what happened," said founder and CEO David Reger. At Nvidia's GTC conference in Berlin from 20 to 22 October, NEURA plans to have its humanoid robot 4NE1 autonomously sort laundry, wash it and move it to the dryer for eleven minutes.