NVIDIA enhances robotics research with new models and simulation tools

NVIDIA announced new developments in robotics at the CoRL conference, including the open-source Newton Physics Engine and the Isaac GR00T N1.6 reasoning model.

The Newton Physics Engine, co-developed with Google DeepMind and Disney Research, is launched in NVIDIA Isaac Lab. This tool aims to aid researchers by providing capabilities for developing adaptable robots.

In addition, the Isaac GR00T N1.6 model introduces humanlike reasoning, allowing robots to decipher complex instructions and execute tasks by leveraging prior knowledge. The latest offering encourages the integration of Artificial Intelligence (AI) into robotic systems.

Leading educational institutions such as Stanford University and ETH Zurich have started utilizing NVIDIA’s accelerated computing technologies to further robotics research. Multiple robot developers, including Agility Robotics and Boston Dynamics, are adopting NVIDIA's platforms in their projects.

NVIDIA also introduced the Cosmos foundation models, designed to enhance the training processes of physical AI by enabling developers to generate diverse datasets. Upcoming models, Cosmos Predict 2.5 and Cosmos Transfer 2.5, are expected to simplify complexity in data generation and improve efficiency.

Furthermore, NVIDIA presented a new dexterous grasping workflow for the Isaac Lab platform, which trains multi-fingered robotic hands using automated curricula. This technique aims to improve manipulation capabilities in complex environments.

To support demanding workloads, NVIDIA announced AI infrastructure like the GB200 NVL72 and RTX PRO servers, designed specifically for robotic applications. These advancements reflect NVIDIA’s commitment to fostering progress in robotic Research and Development (R&D).