GR00T N1.6

A reference generalist VLA model for humanoid robots (NVIDIA): a VLM backbone (a Cosmos-2B variant) plus a 32-layer diffusion transformer, trained on thousands of hours of teleoperation data across multiple robot bodies. It embodies the 2025-2026 consensus: a pretrained VLM plus an action-generation module. Showcased by NVIDIA at CES 2026.