Manage devices and robot access

Platform architecture

The Embodied Intelligence Platform uses a three-layer architecture to connect the cloud to robot devices:

Devices communicate with the platform by using the MQTT protocol. The platform assigns independent MQTT connection information (host, client ID, username, and password) to each device.

Add a device

1. Log on to the Embodied Intelligence Platform.

2. In the left-side navigation pane, click Devices.

3. Click Add Device.

4. In the dialog box, configure the following parameters:

   Parameter	Required	Description
   Device name	Required	A custom device name. We recommend that you use a meaningful naming convention for easy management.
   Device type	Required	Select a type from the drop-down list: Mobile robot, Industrial robotic arm, Drone, or Other.
   Description	Optional	Additional information about the device.

5. Click OK.

   After the device is added, click Details in the device list to view the MQTT connection information, including the MQTT host, client ID, username, and password. This information is required when you configure adb-robot-client on the device.

View device status

The device management page displays the real-time status of all registered devices in a list:

Click Refresh to update the device status.

LeRobot ecosystem robot access

For robots that are already compatible with the LeRobot framework (such as Unitree G1), you can connect them to the platform without any modifications by using the adb-robot-client SDK.

pip install adb-robot-client

adb-robot-client \
  --mqtt \
  --mqtt.broker_host <MQTT host> \
  --mqtt.broker_port <MQTT port> \
  --mqtt.robot_id <Robot ID> \
  --mqtt.client_id <MQTT client ID> \
  --mqtt.username <MQTT username> \
  --mqtt.password <MQTT password>

New robot access

For proprietary robots that are not part of the LeRobot ecosystem, you must complete the following development tasks before they can be connected to the platform:

1. Implement the LeRobot Robot interface: Implement the robot control interface according to the LeRobot framework specifications, including joint state reading and action execution.

2. Develop the access agent: Develop a device-side agent program based on the adb-robot-client SDK to handle MQTT commands dispatched by the platform.

Remote task dispatch

After a device is online, administrators can remotely dispatch the following tasks through the platform:

• Data collection tasks: Create tasks on the data collection page and assign them to devices. The platform sends collection commands to devices through MQTT (establish connection, confirm, start recording, stop recording, and disconnect). For more information, see Collect and manage data.

• Inference tasks: After a trained model is deployed as an inference service on the model deployment page, the platform can dispatch inference tasks to devices. After receiving a command, the device calls the inference service for real-time inference.

Manage devices

In the device list, you can perform the following operations on existing devices:

• Details: View the MQTT connection information of the device.

• Edit: Modify the device name, type, and description.

• Delete: Remove the device.

Related documents

• Create and log on to the Embodied Intelligence Platform

• Collect and manage data

• Train and deploy built-in models