GR00T N1.7
NVIDIAA 32-layer DiT takes the VLM's output and live robot state, then denoises them into precise motor commands in real time.
2 more from this model
diffusion transformer head that denoises continuous actions
implemented as a diffusion transformer (DiT)