Galbot Unveils Dual Breakthroughs in Embodied AI: DexNDM and NavFoM Revolutionize Dexterous Manipulation and Autonomous Navigation

BEIJING, Nov. 16, 2025 /PRNewswire/ -- Galbot is proud to announce two major technological advancements in embodied intelligence: DexNDM, a neural dynamics model designed to revolutionize robotic dexterous manipulation, and NavFoM, the world's first cross-embodiment, cross-task navigation foundation model. Developed in collaboration with Tsinghua University, Peking University, University of Adelaide, and Zhejiang University, these innovations mark a significant leap forward in robotics, enabling robots to autonomously perform complex tasks across diverse environments.

DexNDM: A Game-Changer in Dexterous Manipulation

DexNDM (Dexterous Hand Neural Dynamics Model) is a groundbreaking neural dynamics model that empowers robots to perform high-precision in-hand rotations of objects with varying sizes and geometries. From tiny, delicate parts to large, irregular objects, DexNDM adapts to dynamic forces and varying wrist orientations, solving a longstanding challenge in dexterous manipulation. This innovation pushes the boundaries of what robots can achieve in fine motor skills, offering unparalleled flexibility and adaptability.

Key highlights of DexNDM:

• Unprecedented Generalization Across Object Types: Unlike traditional models limited to specific object sizes or shapes, DexNDM can handle a wide range of objects, from microelectronics to larger, more complex geometries (such as books or elongated tools), breaking past limitations in robotic manipulation.
• Multi-Axis, Multi-Posture Rotation: DexNDM excels at rotating objects along multiple axes, regardless of the hand's orientation. Whether the hand is facing up, down, or sideways, the model ensures stable and precise rotations. This capability allows robots to perform complex tasks, such as adjusting the grip and rotating objects for precise insertion or assembly, even in hard-to-reach positions.
• Highly Dexterous Teleoperation System: Galbot has integrated DexNDM into a teleoperation system, enabling robots to perform complex tasks such as screw tightening, hammering nails, and long-range assembly. This system allows human operators to issue high-level commands while the robot autonomously handles precise finger movements, ensuring unmatched flexibility and robustness.
• Expert-to-Generalist Learning Paradigm: DexNDM uses a unique expert-to-generalist learning approach. The model first trains specialized policies for handling specific tasks and object types and then distills these policies into a unified, adaptable strategy. This approach allows DexNDM to seamlessly perform tasks across various robot types and environments with minimal fine-tuning.

DexNDM represents a significant milestone in the journey from robotic dexterity to real-world productivity, enabling robots to perform tasks like tool use, assembly, and fine-grained manipulation in industries like manufacturing, healthcare, and logistics.

NavFoM: A Leap in Autonomous Navigation

NavFoM (Navigation Foundation Model) is the world's first cross-embodiment, cross-task navigation foundation model that enables robots to autonomously navigate across a wide range of environments and tasks. Developed by Galbot and its research collaborators, NavFoM consolidates fragmented navigation technologies into a unified system capable of seamlessly integrating perception, understanding, decision-making, and action.

Key highlights of NavFoM:

• All Environments: NavFoM supports both indoor and outdoor navigation, enabling zero-shot operation in unseen environments without requiring pre-mapping or additional training data. Robots powered by NavFoM can navigate new environments without prior knowledge, significantly reducing deployment time and effort. Whether navigating a cluttered warehouse, crowded public space, or dynamic outdoor terrain, NavFoM can seamlessly adapt to each scenario.
• Cross-Task: NavFoM can perform a wide range of navigation tasks, from target-following (e.g., following a person) to autonomous navigation (e.g., navigating through a busy urban area). It also works with natural language commands, allowing human operators to issue simple instructions, such as "follow that person" or "navigate to the red car."
• Cross-Embodiment: One of NavFoM's standout features is its ability to adapt to various robot types, including quadrupeds, wheeled humanoids, drones, and even vehicles. This cross-embodiment adaptability ensures that NavFoM can be applied across different robot forms, enhancing scalability and development efficiency. Whether you're navigating an indoor robot, a drone, or an autonomous vehicle, NavFoM delivers consistent performance.
• Unified Navigation Paradigm: NavFoM introduces a novel method for processing video streams and text commands to generate precise action trajectories, unifying traditionally fragmented navigation models. This allows robots to see, understand, and act in real time across tasks and environments. For instance, a quadruped robot's ability to avoid crowds in a shopping mall can inform a drone's ability to navigate obstacles in the air, increasing efficiency and accuracy in unfamiliar environments.

With NavFoM, Galbot has redefined the logic of navigation. Robots powered by NavFoM can perceive their surroundings, plan paths, and make real-time decisions in completely unknown environments with minimal human intervention.

Real-World Applications: Unlocking New Possibilities with DexNDM and NavFoM

Both DexNDM and NavFoM represent dual breakthroughs in embodied AI, pushing the boundaries of robotic capabilities in manipulation and navigation. DexNDM enhances robots' dexterous manipulation abilities, enabling high-precision, multi-step tasks involving tools and fine manipulation, while NavFoM redefines how robots understand and navigate their environments with autonomy and adaptability.

Together, these models provide a comprehensive solution for real-world applications across industries, from assembly and tool use in manufacturing to navigation and autonomous decision-making in dynamic environments.

A Future Powered by Embodied Intelligence

Galbot's NavFoM and DexNDM bring us closer to a world where robots not only understand their environment but can perform complex tasks with precision and autonomy. These innovations are poised to redefine industries, including manufacturing, logistics, healthcare, retail, and beyond, enabling robots to operate in real-world environments with unprecedented adaptability and efficiency.

For more information on DexNDM, visit the project website: www.meowuu7.github.io/DexNDM

For more information on NavFoM, visit the project website: https://pku-epic.github.io/NavFoM-Web/

About Galbot

Founded in 2023, Beijing Galbot Co., Ltd. is a global leader in embodied AI and general-purpose robotics. Galbot's flagship product, the Galbot G1, has been widely deployed, achieving over one year of proven, stable real-world operations across industries such as manufacturing, logistics, healthcare, and retail. The Galbot G1 is currently operating more than 20 fully autonomous Galbot Stores across over 20 cities, with plans to expand to 100 stores by the end of the year.

With over $400M in funding, Galbot is transforming industries by setting new benchmarks in automation, delivering solutions that drive operational efficiency, scalability, and productivity across global sectors.