NVIDIA, a leading AI computing technology company, is accelerating global robotics technology innovation by releasing new open-source models and simulation libraries for robotics research and development.

NVIDIA today announced that its open-source Newton Physics Engine is now available in NVIDIA Isaac™ Lab, along with the NVIDIA Isaac GR00T N1.6 inference vision language action (VLA) model and new AI infrastructure for robotics.

This update includes key technologies to enhance robots' reasoning capabilities and adapt to physical environments.

Newton, a GPU-accelerated open-source physics engine jointly developed by NVIDIA, Google DeepMind, and Disney Research, can precisely simulate the movements of humanoid robots with complex joints and balance, and is being adopted by leading research institutions around the world, including ETH Zurich, Technical University of Munich, and Peking University.

The open source foundation model 'Isaac GR00T N1.6' is based on Vision-Language-Action (VLA) with human-like reasoning ability, enabling robots to understand ambiguous instructions.It is designed to be able to adapt to new situations.

The integrated 'Cosmos Reason' has been downloaded over 1 million times and has ranked #1 on the Physical Reasoning leaderboard.

Data generation for robot training has also been enhanced.

NVIDIA provides large-scale training data based on text, image, and video prompts through the Cosmos World Foundation Model (WFM), and supports the rapid generation of high-quality synthetic data through Cosmos Predict 2.5 and Transfer 2.5.

Dexterous grasping technology is also attracting attention.

A new workflow in Isaac Labs 2.3 enables robots with multi-joint hands and arms to learn to grasp objects in a variety of environments using an automated curriculum.

Boston Dynamics' Atlas robot has greatly improved its manipulation capabilities through this.

NVIDIA also unveiled AI infrastructure to support robotics workloads.

High-performance hardware such as the GB200 NVL72, RTX Pro Server, and Jetson Thor provide the foundation for complex inference and real-time interaction, and are being adopted by major companies such as Google DeepMind, Meta, and LG Electronics.

At this year's CoRL, Stanford University's BEHAVIOR project and Peking University's tactile robotics simulation platform 'Taccel' also received attention.

NVIDIA is standardizing simulation-based robotics technology assessment through its open-source policy evaluation framework, Isaac Lab-Arena, setting a new standard for robotics research.