NEURA Robotics introduces Neuron AI foundation model

NEURA describes Neuron AI as combining 3D vision, touch, audio, robot state and language with memory.[2][1] Its Perception-Language-Action architecture, the company says, connects sensory inputs with a robot’s past actions and turns that information into physical action.[2] NEURA says the model aligns different inputs in a shared space over time and converts them into one stream of action.[2] The company says decisions happen in real time on the robot, without relying on the cloud.[3]

NEURA says cameras capture geometry, color and position, but do not provide information about contact, force or sound.[2] The company gives examples of those additional signals: sound and touch can confirm whether a robot has opened a lock, while touch can help it grasp objects it cannot see.[2] NEURA also says a welding seam’s quality can be judged from the sound of the process.[2] It says robots need to track objects and the consequences of their actions over time, often without seeing them.[2]

NEURA says it adapted the architecture to add memory and trains it on dedicated datasets that use this capability.[2] The company says its Data Amplifier and simulation stack expand training data with task and environment variations, and that Neuron AI is trained across five of its robot embodiments.[2] Customers can fine-tune the model for their tasks and deploy or update capabilities across fleets through Neuraverse; NEURA says customers choose whether to share skills or deployment data.[2]

Visitors can see a demonstration at Station-Berlin from October 20 to 22.[2] NEURA said its 4NE1 demonstration would involve an eleven-minute laundry-handling sequence, with unloading laundry and transferring it to a dryer without human intervention.[2] The company says its Gyms connect data collection, skill training and benchmarking before deployment, and that industrial data partnerships including Schaeffler and Bosch complement training there.[2] However, NEURA had not published benchmark results for Neuron AI, leaving independent comparisons with other robotics models pending.[4]