Real Robot Workflow#

This section covers the complete real-hardware pipeline:

  • Preparing the Unitree G1

  • Teleoperating the G1 to collect demonstration data

  • Exporting data for training

  • Fine-tuning GR00T 1.7

  • Deploying the trained policy on the physical robot

Teleoperation environment for this task.

Teleoperation environment for this task.#

Warning

Working with a physical humanoid robot introduces real safety considerations. Read the G1 Introduction and Safety lesson in full before powering on the robot.

Having at least two people work on this workflow is recommended:

  • One person for teleoperation

  • One person to monitor the robot for safety purposes

Note

This workflow is fully independent of the Simulation Workflow. You can complete it on its own if you have G1 hardware available.

Tip

If you want a starting point before collecting or training your own assets, see the real robot model and dataset links.

Unitree G1 Hardware Overview#

This workflow uses a Unitree G1 humanoid with Dex3-1 hands, a head-mounted Intel RealSense camera, and a Jetson AGX Thor connected either externally or through the Unitree Thor backpack.

The Thor runs Isaac ROS for all controllers and policies. The AGILE WBC controls lower body stability, while an operator or the GR00T policy commands the upper body for manipulation.

Workflow Overview#

#

Tutorial

Supported compute platform

1

Set Up Isaac ROS on Thor

Jetson AGX Thor

2

Set Up the G1 Hardware

Jetson AGX Thor

3

Teleoperate the Unitree G1

Jetson AGX Thor

4

Record Demonstrations

Jetson AGX Thor

5

Convert MCAP Bags to a LeRobot Dataset

x86_64 or Jetson AGX Thor

6

Fine-Tune GR00T 1.7

x86_64

7

Export the Policy with LEAPP

x86_64

8

Deploy the Fine-Tuned Policy

Jetson AGX Thor

Learning Objectives#

By the end of this section, you’ll be able to:

  • Apply workflow safety checkpoints before teleoperation, recording, and deployment.

  • Teleoperate the physical G1 through AGILE and record real-world pick-and-place demonstrations.

  • Export real demonstration data from MCAP to LeRobot format.

  • Fine-tune GR00T 1.7 on real-world data and prepare LEAPP deployment assets.

  • Deploy a trained manipulation policy to the G1 via NVIDIA Thor and evaluate on hardware.