Hi, For an environment we are building, we think the best approach is to use a Conv3D as the agent's observation. Is this possible using the ml-agents pipeline? Can one define custom observation spaces? I read through the documentation and couldn't find anything about 3D CNNs. Thanks in advance!
To some extent. Regarding CNNs please see https://forum.unity.com/threads/cnn-with-4d-observations.942250/#post-6155466