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

# PI05

> Vision-language-action policy inference for robot manipulation

Run PI05 policy inference on robot observations. Predicts action trajectories from camera images and robot state.

## Parameters

<ParamField body="base_rgb" type="str | PIL.Image | np.ndarray" required>
  Base camera RGB image.
</ParamField>

<ParamField body="wrist_rgb" type="str | PIL.Image | np.ndarray" required>
  Wrist camera RGB image.
</ParamField>

<ParamField body="joints" type="np.ndarray | list" required>
  Joint positions.
</ParamField>

<ParamField body="gripper" type="np.ndarray | list" required>
  Gripper position.
</ParamField>

<ParamField body="state" type="np.ndarray | list" required>
  Full robot state (concatenation of joints and gripper).
</ParamField>

<ParamField body="prompt" type="str">
  Optional task prompt (e.g. `"pick up the cup"`).
</ParamField>

<ParamField body="timeout" type="float | None">
  Optional HTTP timeout.
</ParamField>

## Returns

`dict` containing:

* `actions` — Numpy array of shape `(horizon, action_dim)` with predicted actions
* Additional fields from policy inference (e.g. `state`, `policy_timing`)

## Example

```python theme={null}
from grid_cortex_client import CortexClient
import numpy as np

client = CortexClient()
base_img = np.random.randint(0, 255, (224, 224, 3), dtype=np.uint8)
wrist_img = np.random.randint(0, 255, (224, 224, 3), dtype=np.uint8)
joints = np.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6])
gripper = np.array([0.5])
state = np.concatenate([joints, gripper])

action = client.run(
    model_id="pi05",
    base_rgb=base_img,
    wrist_rgb=wrist_img,
    joints=joints,
    gripper=gripper,
    state=state,
    prompt="pick up the cup",
)
print(action["actions"].shape)  # (horizon, action_dim)
```