[Question] ValueError: the actions value are set to NaN after 10 steps

[DavidLudl]

Required prerequisites

• I have read the documentation https://omnisafe.readthedocs.io.
• I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
• Consider asking first in a Discussion.

Questions

Hello. Thank you for your effort of this interesting project.

Currently, I am working with my costomized environment. I followed the tutorial of [Environment Customization From Scratch](url) . My costomized env has a fixed steps for each epochs, 96 steps per epoch.

I have run the single step check of my env, and the results wents well. However, when I run the training loop, I found that after 10 steps, the actions values became NaN and raise the error:

After reading the exising discussions, bugs and questions, I found my question is very similar to the [https://github.com//issues/165 bug](url), that the problems may lies in the configuration. However, after I checked my configuration, there are no value that control the early termination after 10 steps. Here is my configuration:

config.json
Therefore, I would like to ask if there is any internal parameters of the algorithms that, for example, trunicate the env after 10 steps?

Thank you for your help.

[Gaiejj]

Sorry for late reply! Maybe you can set an internal steps counter in your environment. When it exceeds 10 steps, your step() function would return truncated as True.

[DavidLudl]

Sorry for my late reply and thank you for your advice. I have finally located the bug in my environment. It was because I implemented some fractional exponenets expression and it raised the error during the training. Now I have it fixed. Thank you.