How can I use the ZED with Docker on a robot with no internet connection?

An internet connection is normally required only the first time you open a ZED camera with an application based on the ZED SDK, to download its factory calibration file, and the first time you use an AI model, to download it.

Note: cameras produced after May 2026 (except the ZED Mini) also store the factory calibration on their internal EEPROM, which is read by ZED SDK v5.3 or newer without an internet connection. See How do I recalibrate my ZED stereo camera?

Every time a container is launched, it behaves as if the camera were connected for the first time on a new device: the ZED SDK does not find the calibration file and the AI models, and tries to download them again.

Calibration files

Mount the host folder /usr/local/zed/settings/ as a volume, so that the container uses the calibration files stored on the host machine:

docker run --runtime nvidia -it --privileged \
  -v /usr/local/zed/settings/:/usr/local/zed/settings/ \
  <docker_image> sh

Replace <docker_image> with the name of your image. If the host has no internet connection either, download the calibration file from another computer, as described in this article, and copy it into the host settings folder.

AI models

Robots that use AI modules of the ZED SDK (NEURAL depth modes, Object Detection, Body Tracking) also need the optimized AI models. Mount the host folder /usr/local/zed/resources/ as a volume:

docker run --runtime nvidia -it --privileged \
  -v /usr/local/zed/settings/:/usr/local/zed/settings/ \
  -v /usr/local/zed/resources/:/usr/local/zed/resources/ \
  <docker_image> sh

The models must be optimized on the same GPU model and with the same ZED SDK version used in the container; see How can I optimize the ZED SDK AI models manually?

For GMSL2 cameras (ZED X family) additional volumes and devices are required by the ZED X Driver; see the Docker documentation.