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

# Online Controller Lifecycle node

> Start or stop a twin's assigned online controller, optionally steering an RL policy toward a scene object's current pose.

The **Online Controller Lifecycle** (`online_controller_lifecycle`) node starts or stops the online controller assigned to a twin. On a manual (or any) trigger it deploys the twin's primary controller policy, optionally auto-stopping after a set duration.

<Note>
  **Palette category:** Actuation. Uses the twin's **assigned** controller policy — assign one on the twin first.
</Note>

## Steer an RL policy toward an object pose

For RL-task policies you can have the node **collect a target object's pose at trigger time** and pass it to the policy as its startup goal (inference command). This is the "pose → policy" pattern: run the workflow and the policy is steered toward the target.

Point the node at the object whose pose is the goal:

* `target_object_external_id` — a scene object (e.g. a procedural primitive) in the workflow's environment, or
* `target_twin_uuid` — a twin whose world pose is the goal.

The node reads that entity's **authored** pose each run (the scene object's authored primitive position, or the twin's stored world pose) and starts the policy with an `object_position` inference command. This requires the policy's RL task to declare an `object_position` inference command; set `command_name` when the task declares more than one.

<Note>
  When resolved inside this node, the `target_*` params use the **authored** pose, not the live simulator pose. To seed the policy with a target's **live** sim pose, use a [Get Pose](/feature-reference/workflows/get-pose) node upstream (it prefers the live streamed pose) and wire its `position` output into `object_position`.
</Note>

### Or wire the pose from a Get Pose node

Instead of resolving the pose inside this node, you can keep the pose-reading step as a separate, visible node: put a [Get Pose](/feature-reference/workflows/get-pose) node between the trigger and this node, and wire its `position` output into this node's **`object_position`** input. The mapped pose takes precedence over the `target_*` params.

## Live vs. simulation starts

The node does **not** take a mode input — it follows the **workflow's execution target** (the Live/Simulation toggle on the workflow), the same convention the recorder and navigation nodes use. When the workflow runs **live**, the controller drives the physical robot on the twin's bound edge; a live start against a twin whose edge is online requires `confirm_live` to be `true`, otherwise the safety gate rejects it. For live RL-task dispatches you can also pick the controller↔plant `transport` (`mqtt` default, or `zenoh` for the low-latency edge DataBus when the driver declares it); simulation runs ignore transport and `confirm_live`.

## Inputs

| Field                       | Type    | Required | Description                                                                                                                                                              |
| --------------------------- | ------- | -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| `twin_uuid`                 | string  | Yes      | Twin whose assigned controller is started/stopped.                                                                                                                       |
| `action`                    | string  | Yes      | `start` or `stop`.                                                                                                                                                       |
| `duration_seconds`          | number  | No       | Auto-stop after N seconds (0 = run until stopped).                                                                                                                       |
| `runtime_params`            | object  | No       | Parameters passed to the controller at startup.                                                                                                                          |
| `device`                    | string  | No       | Inference device override (`cpu`/`gpu`).                                                                                                                                 |
| `transport`                 | string  | No       | Live dispatch wire transport: `mqtt` (default) or `zenoh` (edge low-latency DataBus). Ignored in simulation. Mode itself is inferred from the workflow execution target. |
| `confirm_live`              | boolean | No       | Acknowledge this may drive a real robot. Required to start when the twin's edge is online; otherwise the start is rejected.                                              |
| `object_position`           | array   | No       | RL policies: explicit `[x, y, z]` goal (e.g. wired from a Get Pose node's `position`). Takes precedence over `target_*`.                                                 |
| `inference_command`         | object  | No       | RL policies: a fully-formed startup inference command payload, used verbatim. Highest precedence.                                                                        |
| `target_object_external_id` | string  | No       | RL policies: scene object whose authored pose seeds the inference command. Mutually exclusive with `target_twin_uuid`.                                                   |
| `target_twin_uuid`          | string  | No       | RL policies: twin (in the workflow's environment) whose stored world pose seeds the inference command.                                                                   |
| `command_name`              | string  | No       | RL policies: inference command term to seed (defaults to the task's single command).                                                                                     |