Closed bmegli closed 1 year ago
After installing ZED SDK 4.0.3 I can no longer reproduce this problem from ROS side.
I am not sure it is SDK or GMSL grabber driver or something else that fixed the problem.
If I don't see it again soon I will close the issue.
When one of the cameras is not reachable it is possible that the argus service is frozen for some reason.
You can recover the cameras by restarting the service:
sudo service nvargus-daemon restart
Thanks.
I can no longer reproduce the problem also with ZED_Depth_Viewer
So somehow installing ZED SDK 4.0.3 or GMSL driver fixed it
Preliminary Checks
Description
Related to GMSL cameras (ZED-X, ZED-XM)
https://github.com/stereolabs/zed-ros-wrapper/blob/50dda077a5ea72b7f8c00faad5746938331e46c2/zed_wrapper/params/zedx.yaml#L8
Decreasing
grab_frame_rate
to 30 fps makes starting multiple GMSL cameras at the same time unstable.Steps to Reproduce
See
Anything Else
section for nowExpected Result
Changing
grab_frame_rate
not affecting cameras startup stability.Actual Result
Decreasing
grab_frame_rate
to 30 fps makes starting multiple GMSL cameras at the same time unstable.Starting only 1 camera works as expected.
Waiting for first camera to finish init before second camera helps a bit but is still hit or miss
When both cameras start then they work reliably, it is only the start that is affected.
Warnings
Camera 1 (eventually succeeds)
Camera 2 (eventually fails)
ZED Camera model
ZED-X, ZED-XM
Note - impossible to select ZED-X/ZED-XM, in the form, edited afterwards
Environment
Both are running with Neural depth
Jetson AGX Orin
Resulting from
Anything else?
Workaround
Keep
grab_frame_rate
at 60 fpsOther notes
ZED_Depth_Viewer
It is possible to trigger similar condition with 2x ZED_Depth_Viewer + point at different GMSL cameras + neural depth + playing with framerate (which restarts the cameras)
So the real problem is below ROS layer.
First
Second