NEURA Robotics launches Neuron AI, a Perception-Language-Action foundation model that fuses 3D vision, touch, audio and memory across five robot embodiments

NEURA Robotics introduced Neuron AI on October 14, 2026, a foundation model for Physical AI developed and trained in Metzingen, Germany. Instead of the vision-led VLA approach, its Perception-Language-Action (PLA) architecture encodes 3D vision (point clouds and depth), tactile and force-torque data, audio, robot state and language into one shared space, aligns them over time, and produces a single action stream. Memory is built into the architecture so the robot keeps track of objects it can no longer see. Training combines real robot data from NEURA Gyms, simulation-amplified data, and human demonstrations captured with sensor suits, with industrial data partners including Schaeffler and Bosch. The model runs locally on the robot, without the cloud, across five embodiments (4NE-1 Gen 3, Gen 4, a wheeled 4NE-1 and two bimanual platforms). NEURA has not published quantitative benchmarks; it frames progress as Time-to-Skill without figures. Access is through NEURA sales; customers fine-tune skills and deploy them to fleets via Neuraverse. A fully autonomous 11-minute laundry demo runs at NVIDIA GTC Berlin, October 20–22.