NVIDIA’s Open-Source Innovations Propel Robot Learning with OpenUSD

Ted Hisokawa
Sep 30, 2025 14:31

NVIDIA introduces advancements in open-source physics engines and development frameworks, leveraging OpenUSD to enhance robot learning and simulation capabilities.

NVIDIA has unveiled groundbreaking advancements in open-source physics simulation and development frameworks that are set to accelerate robot learning. These developments, presented at the Conference on Robot Learning, leverage the Universal Scene Description (OpenUSD) to create scalable, interoperable data standards for advanced robot development, according to NVIDIA’s blog.

The Role of OpenUSD in Robotics

OpenUSD serves as a foundational framework that allows developers to build physically accurate virtual worlds. This environment enables robots to practice and refine their skills before applying them to real-world tasks. By using a “sim-first” approach, developers can conduct parallel training of numerous robot instances using both real and synthetic data within simulation environments.

Newton Physics Engine and Isaac GR00T Model

A significant highlight of NVIDIA’s announcement is the Newton Physics Engine. This open-source, GPU-accelerated engine, codeveloped by Google DeepMind, Disney Research, and NVIDIA, is managed by the Linux Foundation. It allows robots to learn complex tasks more precisely and integrates seamlessly with robot learning frameworks such as MuJoCo Playground and NVIDIA Isaac Lab.

Additionally, the Isaac GR00T N1.6 model is designed to help humanoid robots perform humanlike tasks in the physical world. This model integrates NVIDIA Cosmos Reason, a vision language model that acts as the robot’s cognitive brain, transforming ambiguous instructions into actionable plans.

Industry Adoption and Implementation

Leading robotics companies, including Agility Robotics, Lightwheel, and Universal Robots, are adopting NVIDIA’s simulation technologies to advance their development and deployment of physical AI. For instance, Agility Robotics uses NVIDIA Isaac Lab to train control models for its Digit robot, while Universal Robots employs the NVIDIA Isaac platform to create interoperable digital twins for validating safety protocols.

Moreover, the Lightwheel Simulation Platform, built on NVIDIA Omniverse, aids in developing simulation-ready assets that streamline asset discovery and create accurate digital twins to enhance robotics training and simulation workflows.

Community Engagement and Future Prospects

NVIDIA’s open frameworks and libraries are gaining traction within the robotics community. Community members are utilizing platforms like Isaac Sim and Isaac Lab for innovations in robotics navigation and control. NVIDIA continues to support developers through resources and community engagement, fostering a collaborative environment for advancing robot learning with OpenUSD.

Image source: Shutterstock