Nature Will launches Natura-1 model for dexterous robot hand control
The AMW Read
The launch adds a hand-focused model to embodied AI, but its capabilities and wider impact remain unverified.
Nature Will launches Natura-1 model for dexterous robot hand control
Nature Will has launched Natura-1, a 50-billion-parameter model designed for dexterous robot hand control. The company says the model was pretrained on more than 1 million hours of egocentric human data. It combines language, images, video, and action in a mixture-of-experts architecture and operates on a 100-millisecond control loop. Nature Will says it can follow natural-language instructions and perform manipulation tasks on real hardware without task-specific examples.
The release adds a hand-focused model to the emerging market for embodied AI systems. Fine manipulation requires a model to connect instructions and visual information with physical actions across changing conditions. Nature Will says Natura-1 can sustain workflows and account for spatial constraints, relationships among objects, and instructions specifying what not to do. The source provides no independent benchmark results or task success rates, leaving the breadth and reliability of those capabilities unverified.
For robotics builders and investors, a practical test is whether Natura-1 can repeat useful manipulation tasks reliably on real hardware. Evaluation should examine performance across changing objects and task conditions, as well as the duration of continuous workflows. The reported training volume and model size describe the approach, but they do not establish deployment readiness or commercial demand.