[Question] The difference in results between "./simulate" and when run in a python file

Hello,

I am having trouble with different results when I run the .`/ simulate` and python files.

I am simulating the release of balls when a cup is rotated.

When I run `./ simulate` and manually rotate the cup, the results are close to the real environment as bellow.

https://user-images.githubusercontent.com/63595993/203564412-6b78f582-95c1-4f1b-99df-930f92be69a2.mp4

However, when I load the exact same xml, obj files and rotate the container by a python file, I got wrong results as bellow.

https://user-images.githubusercontent.com/63595993/203564492-408edfde-a25b-447c-ae1f-a65bbdf58edb.mp4

Specifically, when the container was rotated, the sphere inside the container rotated unnecessarily, and the simulation was performed as if the lower sphere was pushed out.
This did not occur in real or simulate environments.

https://user-images.githubusercontent.com/63595993/203564709-07eda952-8e99-4fb9-b0c6-92f40378704c.mp4

https://user-images.githubusercontent.com/63595993/203564804-ae760111-82ec-413c-87a1-3d11cd506d46.mp4

In particular, we are using the same xml and obj files as in simulte, and we do not think there is anything wrong with the files themselves, but we do not know the cause.

I am attaching all files.
[question.zip](https://github.com/deepmind/dm_control/files/10076156/question.zip)

We are very sorry for the trouble and would appreciate your feedback.
Thank you in advance.

Start by unpacking question.zip and comparing the ./simulate run with the Python-file invocation, including how the XML and OBJ files are loaded and how rotation is applied. Reproduce the differing sphere behavior, then document the cause and the conditions needed for both entry points to produce consistent results.

Written by the indexing model from the issue text.