NVIDIA accelerates the era of robots, GR00T N1.5 makes its debut

At the Taipei International Computer Show (COMPUTEX) 2025 held today, NVIDIA CEO Jensen Huang officially released a major upgrade of humanoid robot technology - Isaac GR00T N1.5 and its key data generation engine GR00T-Dreams blueprint, marking a key step for NVIDIA in the general robot intelligence model.

GR00T N1.5 is the first version update since NVIDIA released Isaac GR00T last year. As an open, general, and fully customizable humanoid robot skills and reasoning basic model, this version significantly improves the robot's task success rate in typical scenarios such as logistics and manufacturing, and will be officially launched on the Jetson Thor platform later this year.

At the release site, Huang Renxun emphasized: "Human demonstrations are not scalable - we only have 24 hours a day." In order to break through the bottleneck of traditional data collection, NVIDIA simultaneously launched the GR00T-Dreams blueprint, which can automatically generate massive amounts of high-quality synthetic motion data from a single image, and greatly optimize the behavior learning speed by compressing action labels.

According to reports, the data generated by GR00T-Dreams in just 36 hours is enough to support the training and development of GR00T N1.5, while in the past, this process usually took nearly three months of human demonstration and data preparation.

The key highlights of this release include:

• GR00T N1.5 basic model: a more capable general skill execution system that supports highly customized and modular training.
• GR00T-Dreams blueprint: combining image generation and motion compression technology to achieve large-scale, efficient synthetic training.
• Jetson Thor compatible deployment: laying the technical platform for the future large-scale humanoid robot landing.

NVIDIA said that GR00T N1.5 and GR00T-Dreams will become important pillars of the Isaac humanoid robot platform, promoting automation from factory floors to more complex real-world environments.