“The age of generalist robotics is here,” said Jensen Huang, NVIDIA founder and CEO, during his keynote at GTC 2025. This came after the launch of NVIDIA Isaac GR00T N1 as the world’s first open, end to end customizable foundation model for humanoid reasoning and capabilities. Although intended to speed up the creation of humanoid robots, GR00T N1 is a milestone to bring AI closer to the actual world.
Humanoid robots have long been imagined as capable bodied companions in human worlds, but breakthrough has been plagued by disillusionment. Building general purpose robots capable of learning to handle sets of tasks takes great training data, computationally intensive loads, and generalizable models to treat unstructured environments. NVIDIA’s GR00T N1 attempts to surpass the weaknesses of its predecessors with its two system approach patterned after human thought processes.
System 1 is a quick response system, similar to human reflexes, whereas System 2 takes an analytical, conscious route to decision making. System 2, fueled by a vision language model, reads environmental commands and signals and transmits action plans to System 1, which converts them into accurate movements. This coordination allows GR00T N1 to generalize on reaching, object manipulation, and hand-to-hand transfer of objects with high accuracy.
Huang showed how GR00T N1 was put to action in the keynote with 1X’s humanoid robot, NEO Gamma, that was cleaning the house on its own using a post trained policy. “The future of humanoids is about adaptability and learning,” stated Bernt Børnich, CEO at 1X Technologies. “NVIDIA’s GR00T N1 provides a significant boost to robot reasoning and skills. With minimal post-training data, we fully deployed on NEO Gamma advancing our mission of creating robots that are not just tools, but companions capable of assisting humans in meaningful, immeasurable ways.”
The training data strategy behind GR00T N1 is revolutionary as well. NVIDIA is merging internet-scale human video data, synthetic data captured from the Omniverse platform, and real-world robot data teleoperated. The multi-layer solution is fixing the shortage of demonstration data generated by humans in the real world, which takes time and efforts to acquire. Based on the Isaac GR00T Blueprint, NVIDIA generated 780,000 simulated trajectories nine months’ worth of human demonstration data in just 11 hours and improved the performance of GR00T N1 by 40% over using real data by itself.
GR00T N1 base model is being customized by top robotics engineers such as Boston Dynamics, Agility Robotics, Mentee Robotics, and NEURA Robotics. GR00T N1 is also being optimized for particular applications and environments even more by engineers with the help of post training utilizing PyTorch scripts and NVIDIA DGX Spark systems.
In addition to humanoid robots, NVIDIA launched Newton, an open-source physics engine that was jointly developed with Google DeepMind and Disney Research. Newton, built for learning robots, is developed on the basis of the NVIDIA Warp framework and includes simulation platforms like MuJoCo and Isaac Lab. Disney Research will leverage Newton to augment its robotic character technology that makes expressive robots like Star Wars-themed BDX droids come to life. “This collaboration will allow us to create a new generation of robotic characters that are more expressive and engaging than ever before and connect with our guests in ways that only Disney can,” stated Kyle Laughlin, SVP of Walt Disney Imagineering R&D.
Newton’s differable physics functionality bridges the “sim to real” gap the disparity between simulated and real robot action. The capabilities enable developers to simulate contact with objects that can be deformed, like food, cloth, and sand, so robots can sense hard environments. DeepMind’s MuJoCo-Warp, an extension of MuJoCo, also speeds up robotics workloads over 70x for humanoid simulations.
GR00T N1 and Newton are NVIDIA Cosmos World Foundation Models (WFMs) that offer developers an unprecedented level of control capability in world generation via intellect and AI. Cosmos Transfer WFMs support structure to video processing in attempts to supply input to real sounding synthetic data as a facilitator of smooth AI training opportunities. Cosmos Predict WFMs go a step further and perform multimodal input for action in between prediction and motion paths to use for world generation. These technologies are already being used in companies such as Nexar and Agility Robotics since they bring autonomy to their autonomous devices.
In medicine, NVIDIA launched its Isaac for Healthcare platform, which uses robots and AI to tackle imaging deficiencies around the world. Based on NVIDIA’s DGX, Omniverse, and Holoscan platforms, Isaac for Healthcare allows autonomous medical imaging equipment to virtually simulate patient positioning, scanning, and quality control. “The healthcare industry is one of the most important applications of AI, as the demand for healthcare services far exceeds the supply,” said NVIDIA VP of Healthcare Kimberly Powell. The innovators like Moon Surgical and Neptune Medical are investigating ways to utilize it for better patient outcomes and greater access to diagnostic imaging.
With the introduction of GR00T N1, Newton, and Cosmos WFMs, NVIDIA is revolutionizing the future of physical AI. Not only do these technologies push the boundaries in humanoid robotics but also promise developments towards applications in entertainment, healthcare, and autonomous systems. As Huang himself aptly quoted, “With Nvidia Isaac GR00T N1 and new data-generation and robot-learning frameworks, robotics developers everywhere will open the next frontier in the age of AI.”
To know more about these innovations, check out the interactive demos on GitHub and Hugging Face or learn more about GR00T N1 and Newton on NVIDIA’s API catalog and Vertex AI Model Garden. Isaac GR00T N1, Newton physics engine, and Cosmos WFMs are available for developers worldwide.
