Verified, then logged.
Sun 22 Nov 2026 · 00:20 KST / 15:20 UTC한국어 | ENSearch /

01 / AI · Analysis

Nvidia sets December release for GR00T 2 robot AI weights

The model predicts actions and environmental changes together, while datasets and tools have a separate release schedule from weights and robot-specific training methods.

An unbranded robotic arm pours dry beans from a metal cup into a bowl while an overhead camera observes the changing scene on a laboratory workbench. AI-generated image

Nvidia unveiled its GR00T 2 robot AI model and scheduled its weights for release in December for use with individual robots, according to DigitalToday, a Korean technology news publication.

DigitalToday reported that the human demonstration dataset and related tools are already available. Their availability marks a separate stage from the release of the components needed to adapt the model to individual robots.

The publication described GR00T 2 as a model that learns from videos of human actions to help robots perform precise tasks. Its central feature is the joint prediction of actions and the environmental changes those actions cause.

Joint prediction of actions and environmental changes

DigitalToday described GR00T 2 as a world action model, or WAM. The term refers to a model that predicts both a robot's actions and the resulting changes in its surroundings.

According to the report, its inputs are recently captured video, task instructions and the robot's state. Videos of human actions provide training material, while the robot's state also informs its predictions.

Separate release stages for robot adaptation

DigitalToday put the human demonstration dataset at 100,000 hours. It reported that methods for training and tuning the model for individual robots are scheduled for release alongside the weights in December.

According to the report, the model weights—the learned parameters produced by training—and the robot-specific training and tuning methods remain unreleased. The December release is therefore the next stage in making the announced model available for adaptation to individual robots.