Neura Robotics Unveils ‘Neuron AI’ Foundation Model for Physical AI
aiinsider ·
Neura Robotics has introduced Neuron AI, a multimodal foundation model designed to enhance physical AI by integrating vision, touch, audio, robot-state data, language, and memory. Based in Metzingen, Germany, the company utilizes a Perception-Language-Action architecture that allows robots to maintain context and track objects even when they leave the camera's field of view. The model is trained using a combination of real-world robot data from Neura Gyms, simulation tools, and human demonstrations captured via sensor-equipped suits. This approach enables robots to connect cause and effect over longer sequences rather than reacting solely to immediate visual input. Neura plans to demonstrate these capabilities at Nvidia GTC Berlin, where its 4NE1 humanoid will perform an autonomous 11-minute laundry sequence. The demonstration includes sorting clothes, operating a washing machine, and transferring items to a dryer without human intervention. Additionally, the company connects model development with its Neuraverse platform to distribute updates across robot fleets while allowing customers to fine-tune models for private use.
aiinsider