Recently, NVIDIA unveiled a series of innovations in robotics and physics AI at the CoRL conference: The open-source physics engine Newton beta is now available through Isaac Labs. Co-developed by Google DeepMind and Disney Research, this engine leverages the Warp and OpenUSD frameworks to simulate complex robot actions such as walking on snow and manipulating objects, enabling transfer from simulation to reality.
Simultaneously, the open-source robotics foundation model Isaac GR00T N1.6 is set to launch on Hugging Face. Its integrated physics AI inference model, Cosmos Reason-which has surpassed 1 million downloads and ranks first on Hugging Face's physics reasoning leaderboard-converts ambiguous commands into actionable plans and can be deployed via NVIDIA NIM. Furthermore, the Cosmos World Foundation Model (WFM), with over 3 million downloads, has been updated to support training data generation via text and images.
Isaac Lab 2.3 Developer Preview introduces dexterous grasping workflows, enhancing manipulation capabilities for Boston Dynamics' Atlas robot. Companies like Agility Robotics and Figure AI have adopted these technologies. NVIDIA also launches AI infrastructure including GB200 NVL72 and Jetson Thor, reshaping robotics R&D through open-source and accelerated computing to advance physical AI into a new era of human-robot collaboration.