NEURA Robotics Neuron AI development

NEURA Robotics, based in Metzingen, Germany, has introduced Neuron AI, a multimodal foundation model designed for Physical AI. The technology utilizes a Perception-Language-Action (PLA) architecture to integrate 3D vision, tactile and force sensing, audio, language, and robot status into a single shared space.

Unlike traditional camera-centric systems, Neuron AI incorporates multisensory data and memory. This allows robots to process tactile and acoustic feedback to understand physical nuances, such as the sound of a lock opening or the resistance of a screw. The inclusion of a memory function enables robots to understand cause and effect and track objects outside their immediate field of view.

The model was trained using real-world robot data, simulated data, and human demonstrations. NEURA Robotics intends to deploy this technology across its robot embodiments, including its humanoid robot, 4NE1, which has been used to demonstrate autonomous task completion at NVIDIA GTC Berlin. Customers may also customize the model for specific tasks through the Neuraverse platform.

To enhance these capabilities, NEURA Robotics has entered a partnership with Austrian AI start-up NXAI. The collaboration aims to integrate NXAI’s xLSTM architecture into Neuron AI to provide robots with advanced working and episodic memory. This development is intended to allow robots to track tasks over long periods, such as remembering if a specific component has been installed. Furthermore, NXAI plans to offer its xLSTM-based models on the Neuraverse platform for user training and testing. NXAI is led by Sepp Hochreiter, a professor at Johannes Kepler University Linz, who co-developed the Long Short-Term Memory (LSTM) architecture in the 1990s.