Skip to main content
GR00T checkpoints carry their robot embodiment and input contract in upstream metadata. Use the inspection output for the checkpoint you deploy; when it exposes multiple complete embodiments, pass the upstream embodiment_tag shown by inspection. A checkpoint with one complete embodiment needs no selection.
1

Install and sign in

Create an environment and install servo-client:
your computer
Sign in:
your computer
Open the login link in a browser. servo whoami confirms the active account.
2

Connect Hugging Face

Connect a Hugging Face account in Servo account settings before inspecting or importing a checkpoint, including public repositories.
3

Enroll the robot

On the robot computer, enroll it with the name robot-1:
robot computer
Enroll it; the command prints a code and waits for approval:
robot computer
An organization admin approves the printed code on your computer:
your computer
The robot computer reports Enrolled robot robot-1. when it is ready.
4

Inspect the checkpoint

Inspect the Hugging Face repository you want to serve:
your computer
This example uses the repository’s main revision; for a reproducible deployment, use the same immutable commit ID in the inspect and deploy commands.Use the checkpoint’s reported image keys, dimensions, state and action shapes, and history offsets; if inspection lists embodiment_tag choices, select one exact upstream value with --checkpoint-option:
your computer
The tag is part of the checkpoint’s own GR00T configuration. Servo does not define a separate embodiment enum. Omit the option when inspection reports a single complete configuration.
5

Deploy the checkpoint

Deploy the same repository and revision you inspected, passing the same option if the checkpoint requires a choice:
your computer
For a checkpoint with multiple complete embodiments, pass the selected upstream tag:
your computer
The first deployment imports and prepares the checkpoint. Later deployments of the imported model reuse that preparation.
6

Predict an action chunk

On the robot computer, download and run the shared sample, which builds arrays from the endpoint’s published input contract:
robot computer
Prediction returns actions; your application remains responsible for safety checks and sending commands to the robot.
7

Stop the deployment

Stop the endpoint when you are done:
your computer