NVIDIAUnited States2026

Isaac GR00T N1.7

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

RobotsG1by Unitree Robotics C

Key figures as stated by the developer

6 of 6 stated

Model type ⁠A vision-language-action
Controls ⁠A arms and hands
Inputs ⁠A camera language robot state
Access ⁠A open download
Parameters ⁠B 3,000,000,000
Base model ⁠A Cosmos-Reason2-2B (Qwen3-VL architecture)

How to read the figures

14kgA30.9lbA: The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb: Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (6)

Model type

Developer wording: "NVIDIA Isaac GR00T N1.7 is an open vision-language-action (VLA) model for generalized humanoid robot skills.".

Controls

NVIDIA says the model performs manipulation tasks ("This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments."). The outputs are actions: "Continuous-value vectors correspond to different motor controls on a robot, which depends on Degrees of Freedom of the robot embodiment."

Inputs

"This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments." The model card lists the state input as "State: Robot Proprioception".

Access

"You are welcome to download the model, explore the codebase, and build on the stack, with full support and stability guarantees."

Parameters

The model card gives one total under Number of Model Parameters: "3,000,000,000".

Base model

The vision-language backbone is "Cosmos-Reason2-2B (Qwen3-VL architecture), replacing the Eagle backbone used in N1.6".

Full specifications

9 of 12 fields stated

The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field	Value	Source and caveats
Input and output
Model type	`vision-language-action`	A Developer wording: "NVIDIA Isaac GR00T N1.7 is an open vision-language-action (VLA) model for generalized humanoid robot skills.".
Inputs	`camera` `language` `robot state`	A "This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments." The model card lists the state input as "State: Robot Proprioception".
Controls	`arms and hands`	A NVIDIA says the model performs manipulation tasks ("This cross-embodiment model takes multimodal input, including language and images, to perform manipulation tasks in diverse environments."). The outputs are actions: "Continuous-value vectors correspond to different motor controls on a robot, which depends on Degrees of Freedom of the robot embodiment."
Degrees of freedom controlled	not stated	Note No number is given; it depends on the robot: "Continuous-value vectors correspond to different motor controls on a robot, which depends on Degrees of Freedom of the robot embodiment." ⁠B
Size and speed
Parameters	3,000,000,000	B The model card gives one total under Number of Model Parameters: "3,000,000,000".
Control rate	not stated	Note The model card reports inference speed per GPU, not a control rate. ⁠B
Where it runs and what it is built on
Runs where	not stated	Note NVIDIA names GPUs and Jetson boards for inference but does not say whether the model runs on the robot: "1 GPU with 16 GB+ VRAM (e.g., RTX 4090, L40, H100, Jetson AGX Thor/Orin, DGX Spark)." ⁠A
Base model	Cosmos-Reason2-2B (Qwen3-VL architecture)	A The vision-language backbone is "Cosmos-Reason2-2B (Qwen3-VL architecture), replacing the Eagle backbone used in N1.6".
Openness
Access	open download	A "You are welcome to download the model, explore the codebase, and build on the stack, with full support and stability guarantees."
License	NVIDIA Open Model License Agreement	B The model weights are released under the "NVIDIA Open Model License Agreement"; the code on GitHub is Apache 2.0. The GitHub introduction also says "GR00T N1.7 is fully commercially licensable under Apache 2.0."
The developer’s own claims
Learns new tasks	needs retraining	A Developer claim: the base model is for "zero-shot inference on pretrain embodiments, or finetune for new tasks". "It is adaptable through post-training for specific embodiments, tasks and environments."
Works across robots	yes	C Developer claim: "This cross-embodiment model takes multimodal input". The base model's pretrain tags name several robots, among them "Real-world Unitree G1 (relative EEF + joint)" and "Real-world R1 Pro Sharpa (relative EEF)". Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.

Notes

NVIDIA describes Isaac GR00T N1.7 as an open vision-language-action model for generalized humanoid robot skills.
NVIDIA pretrained N1.7 on 20K hours of human video alongside robot demonstrations.
NVIDIA published N1.7 on GitHub as an early-access release on 18 April 2026; the repository now calls it a General Availability release.

Sources

A Manufacturer product page. github.com/NVIDIA/Isaac-GR00T · Retrieved 2026-10-13
B huggingface.co/nvidia/GR00T-N1.7-3B · Retrieved 2026-10-13
C github.com/NVIDIA/Isaac-GR00T/blob/main/getting_started/policy.md · Retrieved 2026-10-13

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.