May 20, 2012 - At Computex Taipei 2025, theNvidiaChief Executive OfficerJen-Hsun HuangAnnouncing a significant expansion of its robotics development infrastructure.roll outHumanoid RobotA new version of the base model, Isaac GR00T N1.5, is described as "a core building block for the next generation of the industrial revolution".
The highlight of the model is the dramatically improved training efficiency. While traditional manual data collection usually takes nearly three months, the new model generates synthetic data through the GR00T-Dreams tool, and the development of a new version can be shortened to 36 hours, which significantly alleviates the data bottleneck that constrains the development of humanoid robots.
1AI cites a blog post that describes how GR00T-Dreams works in an ingenious way: by first fine-tuning the Cosmos Predict world model for a specific robot, it can generate multi-environment task demo videos from just a single image, and extract "action tokens" for training.
This model of "generating AI training data with AI" allows the robot to learn new behaviors without relying on time-consuming and laborious physical tests. Real-world testing shows that the N1.5 model has significantly improved in new environment adaptation, workspace configuration, and command recognition.
Boston Dynamics, Agility Robotics, Foxconn and other headline companies have now accessed the NVIDIA Isaac platform. In order to support the development needs, NVIDIA synchronously launched Cosmos Reason world model, Isaac Sim 5.0 open source simulation tools, and jointly launched RTX PRO 6000 workstations with Cisco, Dell and other manufacturers.