GR00T N1.6

Research organization / teamNVIDIAAI model developer

NVIDIA's cross-embodiment vision-language-action model takes language and images as input to perform robot manipulation tasks.

A newer version is available: GR00T N1.7 ↗

View model weights Official code Official code Model family resource

Access and availability

Code: Public access
Model files: Public access
Training data: Partly available

Training / fine-tuning code

Access: Public access
Verified on · 2026-10-27
Code license: NVIDIA License

Training data

Some training datasets are public; this does not establish access to the complete training mixture.

GR00T X-Embodiment Sim

Model files

Model files
File	Access	Weights license

At a glance

Model type: Vision-language-action
Input: Language instructions / RGB images / Robot proprioceptive state
Output: Continuous robot actions
Released: December 15, 2025

isaac-gr00t-n16 official NVIDIA model card View source

Related robots

Technical details

Architecture

Flow matching

Functional purpose

Action & control

Training data

GR00T X-Embodiment Sim
Fine-tuningAvailable
CC BY 4.0

Before you use this version: Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.

Submit a source or correction

GR00T-N1.6-3B	Access Public access	Weights license NVIDIA License (non-commercial research/evaluation)	Commercial use ProhibitedTerms Verified on · 2026-10-23 Conditions The public NVIDIA License §3.3 limits use of this work and derivatives to non-commercial research or evaluation. NVIDIA Corporation and its affiliates have an explicit commercial-use exception. Separate commercial permission for other users was not verified. Applies only to this checkpoint, not other models or APIs.

Robot

G1

Related versions 3

GR00T N1.7 ↗Support statement · Simulation / software configuration

NVIDIA documents a pretrained G1 configuration tag for the N1.7 base model. · Official G1 simulation workflow fine-tunes GR00T N1.7 on apple-to-plate demonstrations.

REAL_G1 uses relative end-effector and joint-state modalities. Input data must match the pretrained embodiment; inference hardware is not established by this configuration table.

Simulation/software configuration only; this source does not establish real-robot validation or factory-installed software.

GR00T N1 ↗Real-robot experiment

NVIDIA reports G1 post-training and research evaluation of GR00T N1 against N1.5.

G1 teleoperation data is used for post-training. The source reports N1.5 outperforming N1; it does not establish equal performance or commercial deployment.

GR00T N1.5 ↗Real-robot experiment

Post-trained GR00T N1.5 evaluated on G1 in target-object pick-and-place experiments.

Research setup; no claim of standard configuration.

GR00T N1.6

Access and availability

Training / fine-tuning code

Training data

Model files

At a glance

Related robots

Relationship evidence

Technical details

Architecture

Functional purpose

Training data

G1

Relationship evidence

Relationship evidence

Relationship evidence

GR-1

Relationship evidence

AI Worker

Relationship evidence

Relationship evidence