Neura Robotics has launched Neuron AI, a multimodal foundation model for physical AI that combines 3D vision, touch, audio, robot state and language with memory. Its Perception-Language-Action (PLA) architecture links what a robot sees, hears and feels with what it has already done, converting multisensory information into physical action. Memory lets robots retain context when objects leave the camera's view, supporting object permanence and cause-and-effect reasoning.
Key takeaways
- Multisensory perception addresses tasks cameras cannot solve alone: sound and touch confirm a lock is open, touch enables grasping unseen objects, and weld-seam quality can be judged from process audio.
- Training draws on three sources: real-world robot data from Neura Gyms and real environments, simulation-amplified data via the Data Amplifier, and human demonstration data from sensor-equipped suits, spanning five robot embodiments with different grippers and hands.
- The Neuraverse platform deploys and updates skills across fleets and lets customers fine-tune the model, keeping skills private or sharing them to support the wider ecosystem.
- Time-to-Skill — how quickly a robot learns a task and performs it reliably — is positioned as the central progress metric for industrial adoption.
- Neuron AI was developed and trained in-house in Metzingen, Germany, with data partnerships including Schaeffler and Bosch.
At Nvidia GTC Berlin 2026, Neura will demonstrate its humanoid robot 4NE1 performing an eleven-minute laundry-handling sequence with the current Neuron AI version — sorting, loading a washing machine, closing and reopening it, unloading and transferring to a dryer — without human intervention, testing planning, multisensory perception, memory and action in one continuous task.
Source: Robotics & Automation News · Published 2026-10-14T08:25:59 · “Neura Robotics launches multimodal AI model that gives robots perception and memory”