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

# Record a dataset

> Collect high-quality demonstration episodes for training a manipulation policy.

Training a manipulation policy starts with data. You **teleoperate** the robot — control its arm remotely with the leader arm — through the task you want it to learn, and the robot records synchronized camera images, joint positions, and velocities at 30 Hz. The more consistent and varied your demonstrations, the better the resulting policy.

You can record from the phone app or the [web app](/robots/web-app) — pick your surface in the tabs below.

## Create a physical skill

Before you can record, you need a skill entry to hold your dataset.

<Tabs>
  <Tab title="Phone app">
    <Steps>
      <Step title="Open the Skills screen">
        Tap **Skills** and switch to the **Physical** tab.
      </Step>

      <Step title="Create the skill">
        Tap the **+** button, enter a name in **Enter Action name here**, and tap **Create Physical Skill**. The app opens the episode recorder directly.
      </Step>
    </Steps>

    If the skill already exists, open **Skills**, go to the **Datasets** tab, and select it.
  </Tab>

  <Tab title="Web app">
    <Steps>
      <Step title="Open the Collect page">
        Open the web app and switch to the **Collect** page from the icon rail.
      </Step>

      <Step title="Create the skill">
        In the skill picker, choose **+ New skill…**, pick **Learned skill**, name it, and create it. The recorder points at the new skill automatically.
      </Step>
    </Steps>

    If the skill already exists, select it in the **Collect** page's skill picker.
  </Tab>
</Tabs>

## Record episodes

<Tabs>
  <Tab title="Phone app">
    Inside the skill page's **Record** tab you'll see an episode counter, the current episode list, and a **Record Episode** button.

    <Steps>
      <Step title="Open the recorder">
        Tap **Record Episode** at the bottom of the Record tab.
      </Step>

      <Step title="Enable arm streaming">
        On the recorder screen, toggle **Arm** on. The **Record** button is enabled only when arm updates are flowing (you'll see an MPS value appear).
      </Step>

      <Step title="Capture the demonstration">
        Press **Record**, teleoperate the robot through the task, then press **Stop** when done.
      </Step>

      <Step title="Save or discard">
        After stopping, choose **Save** to keep the episode or **Discard** to drop it.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Web app">
    The **Collect** page is the teleop cockpit with a recording HUD on top.

    <Steps>
      <Step title="Engage the leader arm">
        Plug the leader arm into the computer, click **Connect arm**, then **Engage follow**. **Record** stays disabled until the arm is engaged and streaming.
      </Step>

      <Step title="Capture the demonstration">
        Press **Record**, teleoperate the robot through the task, then press **Stop**.
      </Step>

      <Step title="Save or discard">
        Choose **Save** to keep the episode or **Discard** to drop it.
      </Step>
    </Steps>

    The head is locked during recording so the camera viewpoint stays identical across episodes (the policy needs a consistent view). Leader-arm teleop needs Chrome or Edge — see the [web app overview](/robots/web-app).
  </Tab>
</Tabs>

Repeat until you have enough episodes. See the tips below for guidance on how many you need.

## Tips for high-quality data

<AccordionGroup>
  <Accordion title="How many episodes do I need?">
    We recommend **at least 30 episodes of good quality** as the minimum for a simple task (pick up one object from a fixed area); **50–80** is a comfortable starting point. For tasks that require generalization across object positions, lighting, or object types, aim for **150+ episodes**. More diverse data almost always improves robustness — and quality beats quantity: 30 clean episodes train better than 80 sloppy ones.
  </Accordion>

  <Accordion title="Be consistent with start poses">
    Begin each episode with the arm in roughly the same position. Large variation in start poses confuses the policy — it spends capacity learning where to begin instead of how to complete the task.
  </Accordion>

  <Accordion title="Demonstrate smoothly and deliberately">
    Avoid jerky movements, hesitations, or mid-episode corrections. The policy learns to imitate exactly what you do, including mistakes. If you fumble, discard the episode and re-record.
  </Accordion>

  <Accordion title="Vary object placement slightly">
    Move the target object by 2–5 cm between episodes. This teaches the policy to use vision instead of memorizing a single position.
  </Accordion>

  <Accordion title="Control lighting and background">
    Keep lighting and background consistent across your recording session. Dramatic changes between episodes (e.g., a bright window that appears only in some episodes) add noise the policy can't resolve.
  </Accordion>

  <Accordion title="Recording with base movement">
    If your task involves driving the robot while manipulating, enable base movement in the recorder. The episode will additionally capture `cmd_vel` and odometry data. See [Dataset format](/training/dataset-format#additional-fields-for-mobile-tasks) for details on the extra fields.
  </Accordion>
</AccordionGroup>

## Review your dataset

Play your episodes back to verify quality, and delete any that look wrong — inconsistent demonstrations hurt more than a smaller dataset helps.

<Tabs>
  <Tab title="Phone app">
    Return to the skill's **Record** tab to inspect your episodes — each card shows an index and timestamp. Tap any episode to open **Episode Replay** and play back the recorded actions and camera feeds.
  </Tab>

  <Tab title="Web app">
    Open the **Datasets** page and select the skill. Each episode shows a thumbnail, time, and a ready/preparing dot — click a **ready** episode to replay it (per-camera video with a synced joint graph). Label outcomes and delete bad takes here.

    <Tip>
      Freshly recorded episodes show **preparing** while the robot encodes them to video in the background; they flip to **ready** on their own.
    </Tip>
  </Tab>
</Tabs>

## Upload to the cloud

<Tabs>
  <Tab title="Phone app">
    <Steps>
      <Step title="Open the Record tab">
        Navigate to the skill's **Record** tab. You'll see an **Upload** button and a sync status indicator.
      </Step>

      <Step title="Start the upload">
        Tap **Upload**. The app compresses your episodes and uploads them to Innate's training servers. A progress indicator shows the current stage.
      </Step>

      <Step title="Confirm sync status">
        When the upload finishes, the sync badge turns green. You can verify this on the **Train** tab — the dataset card should show the correct episode count and a green sync status.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Web app">
    There's no separate upload step — the web app uploads the dataset automatically when you start a training run from the **Training** page. Only changed files are transferred, so it's quick to re-run after adding episodes.
  </Tab>
</Tabs>

<Tip>
  Only new or changed episodes are transferred — you don't need to re-upload the entire dataset when you add more.
</Tip>

## Next steps

Your dataset is ready to train:

<CardGroup cols={2}>
  <Card title="Train a policy" icon="brain" href="/training/train-act-policy">
    Configure hyperparameters and launch training on Innate's cloud.
  </Card>

  <Card title="Dataset format" icon="database" href="/training/dataset-format">
    Understand what's inside each episode file.
  </Card>
</CardGroup>