NVIDIA Generalist Embodied Agent Research Lab

NVIDIA Research's group for general-purpose embodied agents, the team behind the GR00T humanoid model.

GEAR is the NVIDIA Research team working on general-purpose embodied agents, co-led by Jim Fan (a Stanford PhD) and Yuke Zhu (also a professor at UT Austin); it reportedly became a formal team in early 2024, with the goal of building foundation models for agents in both virtual and physical worlds. Before GEAR, its members had worked on the Minecraft agent Voyager, MineDojo, and Eureka, which uses a large language model to write reward functions. Since forming, the team's flagship project has been the GR00T humanoid foundation model, from N1 (open-sourced in March 2025) through N1.7 (April 2026), alongside related work such as the DreamGen synthetic-data pipeline, the SONIC whole-body controller, EgoScale human-video pretraining, and the DreamZero world-action model.

ExampleDevelopers can download the GR00T N1.x weights and fine-tune them on a few dozen demonstrations from their own robot to get a policy that follows instructions.

Also called

NVIDIA GEAR Lab, GEAR Lab, GEAR

Related

NVIDIA Isaac GR00T N1 · NVIDIA · Eureka · Voyager · DreamGen · SONIC

Sources

NVIDIA GEAR Lab
GR00T N1 paper

As of

2026-04