Neura Robotics unveils Neuron AI multimodal model

German robotics company Neura Robotics 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, touch, hearing, language, and the robot's own state with a specialized memory function.

Unlike traditional camera-centric systems that focus on geometry and color, Neuron AI allows machines to process tactile and acoustic feedback. This enables robots to perform complex tasks where visual data alone is insufficient, such as sensing the resistance of a screw, hearing a connector click into place, or identifying a successful weld by its sound. The inclusion of a memory function allows robots to understand cause and effect and remember objects even when they are outside the immediate field of view.

Developed and trained in Metzingen, Germany, the model is intended to allow robots to execute autonomous tasks and transfer learned skills across different robot types. The company demonstrated the technology using its humanoid robot, 4NE1, at NVIDIA GTC Berlin.