Neuron AI
A multimodal Physical AI foundation model released by Neura Robotics that fuses five input modalities: 3D vision, touch, audio, robot state, and language. It adopts a Perception-Language-Action architecture and incorporates a memory mechanism.
Timeline (last 90 days)
Neuron AI is not an openly downloadable model or API; it is deployed on NEURA's own robot series and tuned through the Neuraverse platform.
Neuron AI encodes vision, touch, sound, and proprioception early into a shared time-aligned representation, then conditions it into a single action stream, achieving native multimodal fusion.
Neuron AI is trained on five robot forms, covering the MAiRA, MiPA, MAV+, and 4NE1 robot series.
Neuron AI is trained on three types of data sources—real physical experience, simulated variation, and human experience—where real data comes from NEURA Gyms and real environments, simulated data comes from Data Amplifier and the simulation stack, and human data comes from human demonstrations wearing sensor data suits.
Neuron AI has built-in memory, which NEURA says is based on causality and object permanence, and for this it reshaped the architecture and built dedicated datasets for training.
Neuron AI runs locally on the robot body, making real-time decisions with no server round trip, still operating when disconnected, and production data is controlled by the customer.
At NVIDIA GTC Berlin 2026, NEURA will demonstrate the 4NE1 humanoid robot running Neuron AI, completing an eleven-minute-long laundry handling sequence with no human intervention throughout.
Neuron AI currently has no independent benchmark tests, no third party has evaluated its manipulation success rate, NEURA has not published comparisons with VLA baselines, and it has not disclosed model specifications, pricing, or availability information.
Neura Robotics released Neuron AI, a multimodal Physical AI foundation model that fuses five input modalities—3D vision, touch, audio, robot state, and language—and introduces a memory mechanism.
All Facts (9)
Neuron AI is not an openly downloadable model or API; it is deployed on NEURA's own robot series and tuned through the Neuraverse platform.
50%UnverifiedNeuron AI encodes vision, touch, sound, and proprioception early into a shared time-aligned representation, then conditions it into a single action stream, achieving native multimodal fusion.
50%UnverifiedNeuron AI is trained on five robot forms, covering the MAiRA, MiPA, MAV+, and 4NE1 robot series.
50%UnverifiedNeuron AI is trained on three types of data sources—real physical experience, simulated variation, and human experience—where real data comes from NEURA Gyms and real environments, simulated data comes from Data Amplifier and the simulation stack, and human data comes from human demonstrations wearing sensor data suits.
50%UnverifiedNeuron AI has built-in memory, which NEURA says is based on causality and object permanence, and for this it reshaped the architecture and built dedicated datasets for training.
50%UnverifiedNeuron AI runs locally on the robot body, making real-time decisions with no server round trip, still operating when disconnected, and production data is controlled by the customer.
50%UnverifiedNeuron AI currently has no independent benchmark tests, no third party has evaluated its manipulation success rate, NEURA has not published comparisons with VLA baselines, and it has not disclosed model specifications, pricing, or availability information.
50%UnverifiedAt NVIDIA GTC Berlin 2026, NEURA will demonstrate the 4NE1 humanoid robot running Neuron AI, completing an eleven-minute-long laundry handling sequence with no human intervention throughout.
50%UnverifiedNeura Robotics released Neuron AI, a multimodal Physical AI foundation model that fuses five input modalities—3D vision, touch, audio, robot state, and language—and introduces a memory mechanism.
50%