> ## 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.

# Robotic arms

> Build manipulation automations with Cyberwave, using any type of robotic arm: from the hobbyist SO-101 to the UR7e

export const EdgeSetup = ({exclude = []}) => {
  const showMac = !exclude.includes("mac");
  const showLinux = !exclude.includes("linux");
  const macCode = `curl -fsSL https://cyberwave.com/install.sh | bash\ncyberwave pair`;
  const linuxCode = `curl -fsSL https://cyberwave.com/install.sh | bash\nsudo cyberwave pair`;
  const linuxTip = <Tip>
      First time on a Raspberry Pi? See{" "}
      <a href="/feature-reference/edge/raspberry-pi">Raspberry Pi Setup</a>.
      <br />
      First time on a Jetson Orin Nano? See{" "}
      <a href="/feature-reference/edge/jetson-orin-nano">
        Jetson Orin Nano Setup
      </a>
      .
    </Tip>;
  if (showMac && showLinux) {
    return <Tabs>
        <Tab title="Mac">
          <CodeBlock language="bash">{macCode}</CodeBlock>
        </Tab>
        <Tab title="Linux">
          {linuxTip}
          <CodeBlock language="bash">{linuxCode}</CodeBlock>
        </Tab>
      </Tabs>;
  }
  if (showMac) {
    return <CodeBlock language="bash">{macCode}</CodeBlock>;
  }
  if (showLinux) {
    return <>
        {linuxTip}
        <CodeBlock language="bash">{linuxCode}</CodeBlock>
      </>;
  }
  return null;
};

Cyberwave treats every arm the same: drop a digital twin into an [environment](/get-started/key-concepts#environments), pair the hardware, and start sending commands from the dashboard, the [Python SDK](/tools/python-sdk), or an AI controller. The same workflow scales from hobbyist desk arms to industrial cells.

***

## Pick your arm

Cyberwave supports **hundreds of robotic arms** out of the box, from the hobbyist [SO-101](https://cyberwave.com/the-robot-studio/so101) ([setup guide](/hardware/so101/get-started)) and the industrial [Universal Robot UR7e](https://cyberwave.com/universal_robots/UR7) ([setup guide](/hardware/universal-robot-7/get-started)) to everything in between — browse the full lineup at [cyberwave.com/catalog/tag/robotic-arm](https://cyberwave.com/catalog/tag/robotic-arm).

Every catalog page bundles the **bill of materials**, supported drivers, and **troubleshooting** specific to that arm — start there whenever you're unboxing new hardware.

***

## Set up an arm in 4 steps

<Steps>
  <Step title="Create an environment and add the arm">
    From the [dashboard](https://cyberwave.com/dashboard), click **New Environment**, then **Add from Catalog** and search for your arm (e.g. `SO-101`, `UR7e`). Position the twin to match your real workspace.

    <img src="https://mintcdn.com/cyberwave/evaJEwUgMR7XmOzz/images/digital-twins-catalog.png?fit=max&auto=format&n=evaJEwUgMR7XmOzz&q=85&s=ed5aca99014fd5f2f6361d73ecfb0361" alt="" width="1788" height="1366" data-path="images/digital-twins-catalog.png" />
  </Step>

  <Step title="Add cameras and other sensors">
    Click **Add from Catalog** again and add a `Standard Camera` (or any other sensor twin). Dock it to the arm's wrist or place it overhead — the twin hierarchy mirrors your physical mount. With the camera attached you're already set up for [data collection](/use-cyberwave/environment-editor/replay).
  </Step>

  <Step title="Pair the hardware">
    On any device connected to the arm (laptop, Raspberry Pi, Jetson), install the CLI and pair:

    <EdgeSetup />

    The CLI auto-detects the arm, installs the right driver, and links it to the digital twin. See [Cyberwave Edge](/edge/overview) for details.
  </Step>

  <Step title="Calibrate and assign a controller">
    Open the twin in the dashboard, run **Calibrate** (required once per arm), then **Assign Controller** — pick **Keyboard**, **Local Teleop**, an AI policy, or a custom controller.

    <img src="https://mintcdn.com/cyberwave/A30GjvRh58-wSMD4/images/so101-assign-controller.png?fit=max&auto=format&n=A30GjvRh58-wSMD4&q=85&s=69cd4feda6f1bc8cf269bc1bc5bfbb58" alt="" width="5088" height="4646" data-path="images/so101-assign-controller.png" />
  </Step>
</Steps>

***

## Control it from Python

Once the twin is paired, the [Python SDK](/tools/python-sdk) gives you the same API for any arm in the catalog:

```python theme={null}
from cyberwave import Cyberwave

cw = Cyberwave()
arm = cw.twin("the-robot-studio/so101")   # or "universal_robots/UR7"

with cw.affect("live"):
    arm.joints.set("shoulder_pan", 30, degrees=True)
    arm.joints.set("elbow_flex", 45, degrees=True)
    arm.joints.set("gripper", 0, degrees=True)

frame = arm.capture_frame("numpy")        # grab the wrist camera frame
```

Swap `cw.affect("live")` for `cw.affect("simulation")` to run the exact same code against the digital twin — see [Hello Robot](/get-started/hello-robot) and the [simulation capability page](/capabilities/simulation).

***

## Collect data, train, deploy

Once an arm + camera pair is recording, you're already producing datasets.

<Steps>
  <Step title="Record episodes">
    Switch to **Live Mode** in the dashboard and hit **Start Recording** while you teleoperate. Trim the timeline into episodes and export as a LeRobot-format [dataset](/use-cyberwave/environment-editor/replay).
  </Step>

  <Step title="Train an RL or VLA policy">
    Train directly from the dashboard or with your own pipeline. See the [SO-101
    voice pick-and-place](/tutorials/so101-voice-pick-and-place), [SmolVLA
    training](/use-cyberwave/ml-models/smolvla-training), and [OpenVLA-OFT
    training](/use-cyberwave/ml-models/openvla-oft-training) tutorials.
  </Step>

  <Step title="Deploy on edge or cloud">
    Assign the trained policy as a [controller](/use-cyberwave/ml-models). Cyberwave decides whether the model runs on the edge device next to the arm or in a [cloud node](/tools/vla-cloud-node) — your code doesn't change.
  </Step>
</Steps>

***

## Safety, simulation, and orchestration

* **Human takeover anytime.** When an [alert](/edge/drivers/alerts) fires (collision risk, drift, anomaly), swap the AI policy back to **Local Teleop** or **Keyboard** with one click — the twin keeps streaming so the handover is seamless.
* **Validate in simulation first.** Every environment exports to MuJoCo and a browser playground — run the same policy against the twin before letting it touch hardware. See [Simulation](/capabilities/simulation).
* **Compose multi-step tasks.** Wire perception, planning, and motion into repeatable sequences with [Workflows](/use-cyberwave/workflows) — e.g. `camera frame → object detection → VLA policy → gripper primitive` — and fall back to Python whenever a step needs custom logic.

***

## Where to go next

<CardGroup cols={3}>
  <Card title="Manipulation Capability" icon="hand" href="/capabilities/manipulation">
    The full picture of what arms can do on Cyberwave.
  </Card>

  <Card title="SO-101 Tutorial" icon="microchip" href="/hardware/so101/get-started">
    Step-by-step for the hobbyist arm.
  </Card>

  <Card title="UR7e Tutorial" icon="robot" href="/hardware/universal-robot-7/get-started">
    Industrial setup with ROS 2 and the MQTT bridge.
  </Card>
</CardGroup>