Errors in Windows environment

[aoiohta]

Hello, AI45Lab.

Thank you very much for your excellent work on this benchmark. I'm currently using your platform for my research and have encountered a persistent issue.

I have tried setting up the environment on my Windows laptop with NVIDIA GPU, but the following error persists. I am confusing that if it is impossible to set up your environment on windows.

Any guidance on this would be greatly appreciated. Thank you very much for your time and support.

Steps to Reproduce

1. Follow the procedure of Installation 2. 3. and 4.
2. changed the directory structure as follows, since bddl could not be loaded in original setup

3. modified "bddl/setup.py" as follows:

from setuptools import setup, find_packages

# TODO package data

with open('README.md', 'r') as fh:
   long_description = fh.read()

setup(
   name='bddl',
   version='3.5.0',
   author='Stanford University',
   long_description_content_type='text/markdown',
   long_description=long_description,
   url='https://github.com/StanfordVL/bddl',
   zip_safe=False,
   package_dir={"": "src"},
   packages=find_packages(where="src"),
   install_requires=[
       'pytest',
       'numpy',
       'networkx',
       'jupytext',
       'future',
       'nltk~=3.7'
   ],
   package_data={"bddl": ["generated_data/*.json"],},
   include_package_data=True
)

4 bash entrypoints/validate_gt.sh
5 see the error

error

1. it seems like some tasks are not included in the dataset (for example, as attached:

benchmark_store_cleaner_in_cabinet__with__jar__of__jam___Beechwood_0_int_example_20260508-13-22-57.log

)
2. for the task successfully loaded tasks: found many errors. (as attached: benchmark_cleaning_rainboots___Wainscott_1_int_example_20260508-12-06-57.log)

benchmark_cleaning_rainboots___Wainscott_1_int_example_20260508-12-06-57.log

My environment

• OS: Windows 11
• NVIDIA GeForce RTX 3080 Ti laptop
• For other environments I can prepare, I can access to H200 GPU cluster (but I know isaac sim does not support H200 or A100 GPUs).

[ursulalujun]

Hi, aoiohta. Thanks for reporting this issue.

1. The file should be included in the scenes.tar.gz we uploaded on GitHub / Hugging Face: https://huggingface.co/datasets/Ursulalala/IS_Bench_dataset/resolve/main/scenes.tar.gz

Specifically, after extraction, you should be able to find: scenes/Beechwood_0_int/json/Beechwood_0_int_task_store_cleaner_in_cabinet__with__jar__of__jam_0_0_template.json Please check whether the archive was downloaded and extracted completely.

2. OmniGibson itself may print many non-fatal errors or warnings. If the scene can still be rendered normally and training/evaluation can proceed, these messages can usually be ignored.

From your log, the fatal error seems to be: AssertionError: Got mismatch in particle setting size: 2233, vs. number of expected particles 2235! This issue may behave differently on different GPUs. You can refer to this related discussion: StanfordVL/BEHAVIOR-1K#253

A possible workaround is to try other supported scene JSON files for the same task. If the issue still persists for a specific task-scene pair, you may need to filter it out.