hi I’m beginner of unity ml.
I tried hallway examples and pushblock, but I fail to train a visual hallway or visual pushblock example.
in the official docs, The visual observation versions of this environment(visual hallway or visual pushblock) does not train with the provided default training parameters.
So I understood that the Yaml file needs to be changed. I tried many different ways, but all failed. If anyone succeeds in learning through visual observation,Please could you share the whole project and code with me?
I want to know where my method is wrong.
Hi @hestia_p ,
I’ve reached out to the research team and they have said that the visual hallway example doesn’t train with our trainers, but visual pushblock “should” train.
Depending on the version of the c# package you have, there would be a note in the directory that notifies you that you may have difficulty training them.
@[christophergoy]( I can't train a visual hallway or visual pushblock example members/christophergoy.936881/)
hi I understood that I can learn the visual push block by using the given yaml file(config/ppo/visualPushblock.yaml) and “ml-learn” because your trainer can learn the visual push block. Is it right?
If I understand correctly, I got this error when I try to learn again.
mlagents_envs.exception.UnityEnvironmentException: Environment shut down with return code 0.
also I want to know more about the C# version when you developed this example and the other environmental version that I need to match.
Could you explain the visual examples in more detail? I’m having a hard time studying because there aren’t any guidebooks known step by step on unity forums and the Internet.
We no longer support the visual pushblock and visual hallway examples. However, you shouldn’t be getting that error - does it only happen with the visual examples or the normal versions also getting the environment exception?
it happened only visual examples in macos.
the normal versions also getting the environment exception in windows and ubuntu.
the error msg is 2 types. the one is
mlagents_envs.exception.UnityEnvironmentException: Environment shut down with return code 0.
the other is mlagents_envs.exception.UnityEnvironmentException: Environment shut down with return code -6(sigabt)
I read already your comment about [MLA-1712] Make UnityEnvironment fail fast if the env crashes by chriselion · Pull Request #4880 · Unity-Technologies/ml-agents · GitHub.
I try to it 14version, unity 2019.4.18fi, and I have a gpu.
Can you try running the GridWorld example? If it works, there is something wrong with the scene. It’s possible that it’s missing a component or two from the agent, as we haven’t supported those scenes in a while.