#United States #San Francisco #Physical AI #Nature Will #Natura-1

Nature Will Unveils Natura-1: A New Era in Physical AI

On October 13, 2026, Nature Will, a leading physical AI laboratory, unveiled its highly anticipated model, Natura-1. This groundbreaking development represents the first general foundation model tailored for physical applications, boasting a remarkable 50 billion parameters. This model is engineered to excel in dexterous hand control and can achieve zero-shot instruction following—an innovative approach aimed at revolutionizing advanced robotic interactions in the physical realm.

Pre-training and Capabilities
Natura-1 has been pre-trained using over one million hours of egocentric human data, equipping it with the ability to follow complex instructions and adapt its manipulative actions to shifting task conditions. The model is designed to bridge human intent with physical execution, creating a seamless interface for robotic dexterity. In a world that increasingly relies on automation and robotic assistance, the ability to manipulate tools and navigate the physical environment effectively becomes an invaluable asset.

The significance of Natura-1 lies in its unique capability to learn generalization and fine motor control simultaneously, thereby enabling applications across various fields that demand precise and delicate manipulations. These include medical robotics, manufacturing, and even everyday assistance devices. By employing a mixture-of-experts model that integrates language, images, video, and action, Natura-1 operates at an astounding 100-millisecond control loop. This allows it to unify understanding from language inputs and convert them into high-degree-of-freedom actions, transforming verbal directives into tangible manipulations.

A Vision for the Future
"We believe general models that can handle every task must exist, and those models must utilize dexterous hands," stated Ning Ding, the CEO of Nature Will. He emphasized that Natura-1 is merely the first step toward models that can learn directly from human experience, effectively scaling up from this foundational knowledge to cater to any practical application. The model’s long-horizon execution ability, alongside zero-shot instruction following, cultivates its versatility and applicability across countless scenarios.

Natura-1 marks a significant milestone as Nature Will’s inaugural physical foundation model capable of executing general-purpose dexterous manipulation with real hardware. This development serves as a beacon of progress in integrating intelligence and physical interaction, proving that advanced general foundation models can interact meaningfully with the real world. The model has been meticulously designed for scalability, allowing for autonomous execution of sustained workflows and responsiveness to complex natural language commands. It excels in reasoning through spatial constraints while accounting for multi-object dynamics and potential challenges—all essential for real-world applications.

Conclusion
As the landscape of artificial intelligence continues to evolve, Nature Will's commitment to building physical AGI is an embodiment of future possibilities. By forging ahead with innovations like Natura-1, the organization not only paves the path for enhanced robotic capabilities but also sets the stage for genuine interaction between humans and machines. To learn more about the workings and implications of Natura-1, please visit Nature Will's official blog.

About Nature Will

Nature Will is a forward-thinking research organization dedicated to developing physical artificial general intelligence. With a focus on training general-purpose foundation models, the team aims to create systems capable of harmonizing perception, prediction, reasoning, and action within real-world settings. Their research encompasses data acquisition, model training, infrastructure development, simulation, and reinforcement learning. Under the leadership of Founder and CEO Ning Ding, the company is driven by a passion for merging AI with tangible, human-like experiences.