GridSensor shape mismatch

[mbaske]

Describe the bug
I'm trying to retrain the GridFoodCollector demo and get a python error indicating a wrong observation shape.
I think the sensor observes 6 channels, while the PNG-compressed data only has 3.

Console logs / stack traces
mlagents_envs.exception.UnityObservationException: Decompressed observation did not have the expected shape - decompressed had (40, 40, 3) but expected [40, 40, 6]

Environment (please complete the following information):

• Unity Version: 2020.1.4f1
• Windows 10
• ML-Agents C# from master branch
• ML-Agent python 0.19.0

[surfnerd]

hi @mbaske,
I am unable to reproduce your issue. Can you give me more information?

[surfnerd]

i used mlagents-learn ./config/ppo/FoodCollector.yaml to run my training session.

[surfnerd]

Hi @mbaske,
Did you happen to update the python package when you tried out these changes?

[surfnerd]

We will need to bump the communication minor version as this feature was recently added.

[mbaske]

Hi @surfnerd, here is my version info:
ml-agents: 0.19.0,
ml-agents-envs: 0.19.0,
Communicator API: 1.0.0,
TensorFlow: 2.2.0
Do I need to upgrade any of this?

[surfnerd]

Yes, you will need to install the python packages from master.

[surfnerd]

from the root directory of the repo:

$ python3 -m pip install -e ./ml-agents-envs/
$ python3 -m pip install -e ./ml-agents/

[mbaske]

Thank you!
I updated the python files and I'm not getting the UnityObservationException any longer.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.