New Report

The Humanoid Robot Supply Chain

Supplier Strategy and Market Positioning 2026–2027

New! 2026 Humanoid
Robot Market Report

198 pages of exclusive insight from global robotics experts — uncover funding trends, technology challenges, leading manufacturers, supply chain shifts, and surveys and forecasts on future humanoid applications.

Featuring insights from Aaron Saunders, Former CTO of Boston Dynamics,
now Google DeepMind

Get the Report

Isaac GR00T

Name: Isaac GR00T
SKU: 11911
Availability: BackOrder

$ 95 000

NVIDIA

The NVIDIA Isaac GR00T Reference Humanoid Robot is the first open humanoid platform built for academic research. It combines a Unitree H2 Plus body, dual Sharpa Wave five-finger hands, NVIDIA’s Jetson AGX Thor on-board AI compute, and the full Isaac GR00T software stack into one integrated system. Available from Unitree in late 2026 for research and university labs.

This product is currently not available for purchase.

Compare

Navigation

Manipulation

Humanoid.Guide
Skill Score

(Only if you are this robot manufacturer)

Specifications and details:

Availability	Limited production / pilot deployments
Nationality	US
Website	https://www.nvidia.com
Degrees of freedom, overall	75
Degrees of freedom, hands	44
Height [cm]	180
Max speed (km/h)	7
Strength [kg]	15
Weight [kg]	68
Runtime pr charge (hours)	3
Safe with humans	Yes
CPU/GPU	NVIDIA Jetson AGX Thor T5000 — Blackwell-class GPU with 2,070 FP4 TFLOPS, 14-core Arm CPU, 128 GB unified memory, configurable 40–130 W
Ingress protection	Estimate IP30–IP40 (dust-/touch-protected, not water-rated)
Camera resolution	Resolution not officially disclosed — head-mounted stereo camera (140° H × 102° V FoV), wrist cameras for close-range manipulation, plus IMU
Connectivity	Bluetooth 5.2, Ethernet, microphone array and speakers for voice interaction, USB, Wi-Fi 6
Operating system	Not explicitly stated — runs NVIDIA Isaac ROS middleware on a Linux-based stack (standard for Jetson Thor); GR00T models served on top
LLM integration	Yes — Isaac GR00T includes open Vision-Language-Action (VLA) foundation models (GR00T N1 family) for humanoid reasoning, language understanding, and multitask behavior
Latency glass to action	Not officially disclosed — Jetson Thor's 2,070 FP4 TFLOPS is designed for real-time on-robot inference, but no end-to-end latency figure published
Motor tech	Brushless DC servomotors (Unitree's quasi-direct-drive design) — arm torque up to 120 Nm, leg torque up to 360 Nm
Gear tech	Quasi-direct-drive with low-ratio reducers (Unitree's signature approach); exact reducer type not disclosed
Main structural material	Not officially disclosed — qualified estimate aluminum alloy with composite covers (standard for Unitree H-series humanoids)
Number of fingers	10
Main market	Academic research, frontier humanoid robotics R&D
Walking Speed [km/h]	3
Manufacturer	NVIDIA

Description

NVIDIA unveiled the Isaac GR00T Reference Humanoid Robot at GTC Taipei on May 31, 2026. The reference design serves as the first open platform for advanced humanoid research. Rather than a consumer product, this machine targets academic researchers and frontier robotics labs. Indeed, institutions like ETH Zurich, Stanford, UC San Diego, and Ai2 have already signed up to use it. The package pairs a Unitree H2 Plus humanoid body with Sharpa Wave five-finger hands. Then NVIDIA’s Jetson Thor computer and the Isaac GR00T software stack complete the brain.

Robotics teams currently face a fragmented process. They juggle hardware integration, data collection, simulation, training, evaluation, and deployment across separate stacks. As a result, frontier humanoid work is hard for any single lab to assemble from scratch. NVIDIA built the reference robot specifically to fix this. It gives researchers a single, open foundation. The bundle ties together robot body, dexterous hands, sensors, and on-board AI compute. Consequently, teams can move faster from initial bring-up to skill development and real-world testing.

The body stands about 1.8 meters tall and weighs roughly 68 kilograms. Its 31 body joints, combined with 22 finger joints per hand, deliver a total of 75 degrees of freedom. A head-mounted stereo camera with a wide field of view handles long-range vision, while wrist cameras manage close-range manipulation. Inside, NVIDIA’s Jetson AGX Thor T5000 chip runs the heavy AI workloads. It packs a Blackwell-class GPU, 128 GB of unified memory, and over 2,000 FP4 teraflops of AI performance. Furthermore, the robot can lift up to 7 kilograms continuously and peak at 15 kilograms.

I’m Olivia, Humanoid Analyst at Humanoid.Guide

My mission, together with the team, is to help your organization understand the landscape, compare solutions, and move toward successful robot deployments.

Buy-a-Humanoid™

NVIDIA wraps the hardware in its full Isaac GR00T software stack. The platform covers data capture, simulation, training, evaluation, and on-robot deployment. Isaac Sim and Isaac Lab let teams test policies in virtual environments first. Then NVIDIA Isaac ROS middleware moves trained policies onto the physical robot for live runs. Importantly, researchers keep ownership of their robot data, training data, and logs throughout the process. Unitree will sell the H2 Plus robot to research customers in late 2026. NVIDIA has not announced a unit price yet.

Download the Humanoid Robot Market Report here

Website: https://nvidianews.nvidia.com/news/nvidia-open-humanoid-robot-reference-design