NEURA Robotics launches Neuron AI multimodal model for physical tasks

Neuron AI is trained on three data pillars: real-world robot interactions, simulated variations, and human demonstrations from sensor-equipped suits. It supports five robot embodiments, including the humanoid 4NE-1 Gen 3 and 4NE-1 Gen 4, and runs locally on robots for real-time decision-making without cloud dependency. NEURA’s Neuraverse platform allows customers to fine-tune the model for specific tasks, with progress measured by Time-to-Skill—how quickly a robot learns and executes a new task reliably. The company will demo Neuron AI at NVIDIA GTC Berlin 2026, where a 4NE-1 robot will autonomously handle an 11-minute laundry sequence, showcasing its ability to maintain context across sequential steps.