Nvidia has introduced new tools under its Isaac platform to help companies build and train humanoid robots more efficiently using synthetic data and cloud computing.
The announcement includes Isaac GR00T N1.5, an updated version of its foundation model for teaching robots how to understand and carry out physical tasks. Nvidia also launched Isaac GR00T-Dreams, a guide for generating large amounts of synthetic motion data, and highlighted the use of its latest Blackwell systems to speed up robot training and simulation.
These technologies were unveiled during Nvidia’s presentation at Computex 2025. The Isaac GR00T-Dreams blueprint lets developers use a single image to generate videos of robots performing tasks in different environments. These videos are then broken down into data pieces called action tokens, which help train robots without needing massive real-world data collection.
Robotics companies tap Isaac platform for development
Companies such as Agility Robotics, Boston Dynamics, Fourier, Foxlink, Galbot, Mentee Robotics, NEURA Robotics, Skild AI, XPENG Robotics and others are now using Nvidia’s Isaac tools to build and test their humanoid robots.
Nvidia’s researchers used GR00T-Dreams to produce synthetic data and update GR00T to version N1.5 in just 36 hours, a process that could have taken months using manual data collection. The updated model can adjust to new work environments and better follow user instructions for tasks like sorting and placing objects.
Early users of the GR00T N models include AeiRobot, Lightwheel and NEURA Robotics. AeiRobot uses the tools to train its ALICE4 robot to follow spoken instructions for handling items in factories. Foxlink is improving its robots’ performance, and Lightwheel is using synthetic data to test faster deployment in industrial settings.
Simulations and tools reduce risks and costs
To help developers avoid the high costs of collecting and testing robot data in real life, Nvidia also released new simulation and data-generation tools. These include Cosmos Reason and Cosmos Predict 2 for generating training data, as well as Isaac GR00T-Mimic for creating motion sequences using a few human demonstrations.
Other resources now available include a dataset of 24,000 robot motion sequences, Isaac Sim 5.0 for robot simulation, and Isaac Lab 2.2 for testing GR00T models.
Manufacturers like Foxconn and Foxlink are already using these blueprints in their training systems. Agility Robotics, NEURA Robotics, and others are training their robots using Nvidia’s simulation tools.
Hardware makers such as Dell Technologies, Lenovo, and Supermicro are offering new servers and workstations with Nvidia RTX PRO 6000, supporting a wide range of robot development tasks from training to deployment.
Get the latest before it trends. Follow Back End News on LinkedIn, Facebook, X, YouTube, and TikTok for updates and in-depth coverage across the tech and security landscape.