Hey MarilleTielens - I don’t know of a drop-in example for this exact ground-frame setup. The v3 AprilTag example provides IDs and tag corners; pose must then be calculated on the host.
I’d calibrate once using all four floor tags’ known corner coordinates to obtain the camera-to-ground transform, then transform each object-tag pose into that frame. For a cube/rig, solve all visible tag corners together rather than averaging individual tag poses. Make sure the mono stream is undistorted and the intrinsics match the actual output resolution/crop—this is a common source of unstable results.
Note that on OAK-D Lite (RVC2), AprilTag detection runs on the host. You can also use our Luxonis skills with Codex or Claude to help build and debug the application.
Thanks,
Oskar