Enable camera in headless model experience memory leak

Hi @kellyg ,

I set enable_camera to True in Repo OmniIsaacGymEnvs, task FrankaCabinet without changing anything else and found out that the memory keeps increasing as training goes on. The training stopped when Python got killed due to a memory error. Here are some screenshots:




Is there a bug in this module? And how can I fix it?

Best,
Chay

1 Like

Apologies, I’d like to provide some additional clarification. Currently, the environment is being executed on a server within a docker container, and as a result, there is no access to a monitor. To ensure the successful execution of the script, the configuration has been set with headless: True

Hi Chay, thanks for reporting the issue. I tested the FrankaCabinet environment with enable_camera set to True using our latest internal build and did not observe an increase in memory. It is likely that the issue has already been fixed and will be made available in the next Isaac Sim release.

Hi kellyg,

Upon reviewing my repository, I found that another env was imported via the task_util.py which resulted in this memory issue.

These two lines of code I believe are responsible for increasing memory.

from omni.isaac.core.utils.extensions import enable_extension
enable_extension("omni.kit.window.viewport")

Though I found another way to create multiple cameras, I’m still curious about how this would lead to the memory leak problem.,

Best,
Chay