HomeFull speed ahead. Keep track of advancements in every frontier

TheTechno Optimist.

Vera RubinNVIDIA—Isaac GR00TNVIDIA—Vera RubinNVIDIA—Vera RubinNVIDIA—Isaac GR00TNVIDIA—NVIDIA—Vera RubinNVIDIA—NVIDIA—
Isaac GR00T
Courtesy of NVIDIA · Press kit, editorial use (opens nvidianews.nvidia.com)

Isaac GR00T

by NVIDIAUS

PilotStage 3 of 5

Open GR00T models and Jetson Thor computers are used by many humanoid makers.

Updated 16 Mar 2026Checked 25 Sep0 updates this week

Milestones

Next · GR00T N2 release, which NVIDIA says will ship by the end of 2026
  1. Project GR00T announced18 Mar 2024Complete.
  2. GR00T N1 open humanoid model18 Mar 2025Complete.
  3. GR00T N1.5May 2025Complete.
  4. Jetson AGX Thor developer kit available25 Aug 2025Complete.

Most important updates

  • 16 Mar 2026
  • 5 Jan 2026
  • 29 Sep 2025

Upcoming

  1. Q4 2026GR00T N2 release, which NVIDIA says will ship by the end of 2026 (next)

Current obstacles

  • Sim-to-real gapSkills learned mostly in simulation still work worse on real robots, sensors and objects.

Physics limits

  • There is no internet of robot actionsLanguage models learned from trillions of words online. No comparable record of physical actions exists, so robot data must be collected by teleoperation or simulated, which is slow and costly.
  • Simulation never matches reality exactlyFriction, contact, soft materials and sensor noise are hard to model precisely. Policies trained in simulation learn to exploit its errors and then fail on real hardware: the sim-to-real gap.
  • Fast control on a battery budgetBalance and grasping need decisions every few milliseconds, but large models take tens of milliseconds per step, and a humanoid can spare only about 100 W for its computer.

How it works

3 parts
A Jetson Orin NX module on a carrier board: the smaller forerunner of Jetson Thor, the robot-mounted computer GR00T models run on
A Jetson Orin NX module on a carrier board: the smaller forerunner of Jetson Thor, the robot-mounted computer GR00T models run onPhoto: 4300streetcar · CC BY 4.0 (opens commons.wikimedia.org)
Model

See, understand, act

A vision-language model reads camera images and an instruction; a faster action module turns its output into smooth joint motions many times a second.

Data

Real, human and simulated data

Training mixes robot recordings, videos of people and demonstrations multiplied in Isaac Sim and Cosmos world models, because real robot data is scarce.

Onboard

Jetson Thor

A Blackwell-based computer inside the robot (up to about 2,070 teraflops at 4-bit, 40–130 W) runs the model locally, without a network link.

Update log

3 updates

Mon 16 Mar

  • Major: PressSoftware

Mon 5 Jan

  • Major: PressSoftware

Mon 29 Sep 2025

  • Minor: PressSoftware

About NVIDIA

The team behind Isaac GR00T

NVIDIA

AI GPUs, rack systems, CUDA, networking

NVIDIA designs the GPUs, networking and software behind most AI training and inference. Its CUDA software made GPUs the default for AI. In October 2025 it became the first company worth $5 trillion.

  • Vera Rubin in production in 2026; cloud partners get it from the second half of the year.
  • Licensed Groq's inference technology and hired its leaders for about $20B (Dec 2025).
  • Plans to invest up to $100B in OpenAI, tied to at least 10 GW of NVIDIA systems (Sep 2025).

nvidia.com (opens nvidia.com)Company profile →

Founded
199333 yrs
Headquarters
United States
Status
Public
Listed
NVDA (opens google.com)NASDAQ
Valuation
$5.4Tmarket capSep 2026
Staff
~36,000Jan 2025
Works in
ComputingAI chipsAIRobotics
Coverage
3 programs · 21 updateslatest 16d agochecked 25 Sep
Customers
OpenAICoreWeave
Partners
IntelGroq
Suppliers
TSMCSK hynix
People
Jensen Huang (opens en.wikipedia.org)Co-founder, President and CEOColette Kress (opens en.wikipedia.org)CFOBill Dally (opens en.wikipedia.org)Chief Scientist
Primer