FAQ / Troubleshooting

Getting UnityTimeOutException upon submission (discussion here)

• Try increasing timeout_wait when creating ObstacleTowerEnv from original 30 to 300 or even higher.
• Start your agent code by first creating the ObstacleTowerEnv object before anything else (including larger imports). If you create the environment "too late", it may have issues connecting to the OT binary running for evaluation.

Getting "No module named ..." for common packages (e.g. Numpy) (discussion here)

• If you are using older version of the run.sh example, remove/comment out part with source activate base.

Issues with aicrowd-repo2docker

• Try updating aicrowd-repo2docker (pip install -U aicrowd-repo2docker)
• aicrowd-repo2docker ships with Python 3.7 by default. mlagents-env does not (as of writing) work with Py37. You can fix this by adding python-3.6 to runtime.txt in root of your submission code directory. More details here.

Zero score in leaderboard

• Make sure you have debug mode disabled in aicrowd.json. Debug mode forces leaderboard score to zero.