Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

The Isaac GR00T workflow for synthetic data and the NVIDIA Cosmos world base model power the development of general-purpose humanoid robots.

On 6 January, NVIDIA Founder and CEO Jen-Hsun Huang announced at CES that NVIDIA has officially launched the NVIDIA Isaac GR00T Blueprint for Synthetic Motion Generation, which helps developers generate massive amounts of synthetic motion data to train humanoid robots through imitation learning.

Officially, the humanoid robotics market is expected to reach $38 billion over the next two decades. To meet this huge demand, especially in the industrial and manufacturing sectors, NVIDIA has released a series of robotics base models, data pipelines and simulation frameworks to accelerate the development process of next-generation humanoid robots.

Imitation learning is a subset of robotics learning that enables humanoid robots to acquire new skills by observing and imitating demonstrations by human experts. Collecting these extensive, high-quality datasets in the real world is tedious, time-consuming, and often prohibitively expensive. With the Isaac GR00T Blueprint for Synthetic Motion Generation, developers can easily generate massive synthetic datasets with only a handful of human demonstrations.

Starting with the GR00T-Teleop workflow, users can capture human movements in a digital twin environment with Apple Vision Pro. These human actions are recorded as the gold standard and learnt by the robot imitating them in the simulation environment.

The GR00T-Mimic workflow then expands the captured human demonstrations into a larger synthetic motion dataset. Finally, the GR00T-Gen workflow, built on the NVIDIA Omniverse and NVIDIA Cosmos platforms, exponentially augments this dataset with domain randomisation and 3D lifting techniques.

This dataset can then be used as input for robot strategies in the NVIDIA Isaac Lab, an open-source modular framework for robot learning, to teach robots how to move and interact efficiently and safely in their environments.

NVIDIA also unveiled the Cosmos platform at CES, which features a series of open, pre-trained world base models designed for generating the world states needed for physically-aware video and physics AI development. It includes autoregressive and diffusion models that come in a variety of model sizes and work with multiple input data formats. The models are trained on 1800 trillion data units, including 2 million hours of video captured by autopilots, robots, drones, and synthetic data.

In addition to helping generate large datasets, Cosmos can bridge the gap between simulation and reality by extending images from 3D to real scenes. Integrating Omniverse, an application programming interface and microservices development platform for building 3D applications and services, with Cosmos is critical, providing key safeguards through its highly controllable, physically accurate simulations that help minimise the illusionary problems common to world models.

The NVIDIA Isaac GR00T, Omniverse, and Cosmos are helping to make huge leaps in physics AI and humanoid robotics innovation. Major robotics companies, including Boston Dynamics and Figure, have already begun to adopt the Isaac GR00T and are demonstrating results.

Software and hardware makers of humanoid robots, as well as robotics manufacturers, can also apply for early access to NVIDIA's Humanoid Robotics Developer Programme.