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

> ## Agent Instructions
> Treat the Dropbear Python SDK and CLI as the supported public integration surfaces.
> Prefer context-managed sessions and model-specific observation helpers.
> Do not infer or recommend internal control-plane APIs.

# Browser simulation

> Run a curated LIBERO task from the dashboard without installing a local simulator or connecting hardware.

The browser simulation is the quickest way to see Dropbear allocate a policy,
start a simulator worker, stream two camera views, and report control-loop
diagnostics. It does not require local hardware.

## Start a run

<Steps>
  <Step title="Open the simulator">
    Sign in and open the
    [dashboard simulator](https://dropbear.dreamscalelabs.com/dashboard/sim).
  </Step>

  <Step title="Choose a model and task">
    Select a live LIBERO model and one curated task preset. The Run card shows
    the maximum step budget.
  </Step>

  <Step title="Run simulation">
    Select **Run simulation** once. Keep the page open while startup progresses.
  </Step>
</Steps>

## Understand startup

A cold run can take several minutes because the dashboard may start both an
inference worker and a simulator worker. The progress panel moves through:

1. **Spinning up inference worker**
2. **Spinning up sim worker**
3. **Starting stream**
4. **Video live**

Do not repeatedly press Run or refresh while those phases are advancing. The
page shows the selected region and startup status when available.

If startup reports an error, copy the visible run/session identifiers before
retrying and use [Troubleshooting](/troubleshooting).

## Read the run

During a healthy run you should see:

* live **Scene cam** and **Wrist cam** frames;
* a status of `running`;
* step and policy-call progress;
* mean policy round-trip time;
* action-buffer health over the run;
* selected region and target diagnostics.

The terminal result is `complete`, `step limit`, or a visible failure message.
Use **Run again** only after the prior run reaches a terminal state.

## Cancel or leave

Changing the task, starting a new completed run, or leaving the simulator page
requests cancellation of the active dashboard run and closes its stream. A
browser or network crash may prevent the final request from reaching Dropbear,
so verify cleanup in
[Usage](https://dropbear.dreamscalelabs.com/dashboard/usage) or with:

```bash theme={null}
dropbear sessions list
dropbear sessions stop <session-id>
```

Open sessions consume prepaid credits. A finished or cancelled run should no
longer appear as an open session.

## Continue with Python

The browser path owns the simulator and control loop. To integrate your own
robot or environment, install the SDK and follow the
[Python quickstart](/quickstart), where your code owns sensing, validation,
actuation, and safety.