> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orchestrallabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# GR00T

> Inspect and deploy GR00T N1.7, with its camera and state inputs and action output.

## Checkpoint

```bash title="Your computer" sample=verified theme={null}
orchestral models inspect gr00t hf://nvidia/GR00T-N1.7-LIBERO@2ea293aa20ba7cf5bbf3ba17a5fbcb1a01cbfe21 --options embodiment_tag=libero_sim
```

```text title="Sample output" theme={null}
Inputs:
  image: RGB uint8 (256, 256, 3)
  wrist_image: RGB uint8 (256, 256, 3)
  state: float32 (8,)

Outputs:
  actions: float32 (16,7)
```

## Inputs and output

| Direction | Name | Shape | Type |
| - | - | - | - |
| Input | `image` | `(256, 256, 3)` | RGB `uint8` |
| Input | `wrist_image` | `(256, 256, 3)` | RGB `uint8` |
| Input | `state` | `(8,)` | `float32` |
| Output | `actions` | `(16, 7)` | `float32` |

## Deploy and predict

Deploy the checkpoint URI directly. Orchestral resolves, imports and prepares it before bringing up the serving deployment.

```bash your computer theme={null}
orchestral deployments create hf://nvidia/GR00T-N1.7-LIBERO@2ea293aa20ba7cf5bbf3ba17a5fbcb1a01cbfe21 --options embodiment_tag=libero_sim --robot robot-1 --wait
```

```text title="Sample output" theme={null}
jolly-tangerine-snail-rdik  ACTIVE · ready
GR00T · robot-1
Timeline: requested 12s ago → starting worker 8s ago → ready now
```

```python robot computer theme={null}
import numpy as np
import orchestral as ol

client = ol.Orchestral()
deployment_id = input("Deployment ID from the CLI output: ").strip()
deployment = client.deployments.get(deployment_id)

inputs = {
    "image": np.zeros((256, 256, 3), dtype=np.uint8),
    "wrist_image": np.zeros((256, 256, 3), dtype=np.uint8),
    "state": np.zeros((8,), dtype=np.float32),
}

with deployment.session() as session:
    result = session.predict(inputs=inputs, instruction="pick up the object")

print(result.actions.shape)  # (16, 7)
```

The selected GR00T processor defines the camera keys. Inspection prints the state and action key
orders with their widths. Supply the flat state vector in that order; returned action columns
follow the action key order. Widths come from the saved processor's `statistics.json`.
Zero-width entries are inactive and are omitted from the advertised inputs and outputs; relative
normalization bounds do not add action columns. For example, the base checkpoint's XDOF
embodiment has 14 active state values and 14 active action values per step.

The LIBERO checkpoint contains multiple embodiment tags, so select `libero_sim` when inspecting,
importing or deploying it. For other multi-embodiment checkpoints, pass the exact metadata tag
with `--options embodiment_tag=...`. Select a policy within a checkpoint with
`--options checkpoint_subdir=...`. Inspection reports each camera's required frame offsets.
Pass a single `HWC` image for a camera with offset `[0]`; pass a `THWC` array for a camera
with multiple offsets, ordered as listed by inspection. Camera names and counts come from the
selected checkpoint, so they can differ between embodiments.

The runtime setting `execution_capture=none` selects reference PyTorch execution with the same
checkpoint, BF16 precision, and saved sampling steps.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.