Request for the official environment configuration (PyTorch/CUDA versions)

Hi authors,

Thank you for releasing this great work. I am trying to reproduce the results and deploy the SpatialAgent system on my server, but I encountered some compatibility issues during environment setup.

I followed the provided requirements-agent.txt to install the dependencies. However, I noticed that the PyTorch version was not explicitly specified. Since uniception==0.1.7 depends on torch, pip automatically installed the latest available version:

• PyTorch: 2.13.0
• CUDA runtime: 13.0

When launching the GPU server with:

python -m spatial_agent.entrypoints.launch_gpu_server --num_gpus 1 --reconstruct_backend pi3

I encountered the following error:

RuntimeError: The NVIDIA driver on your system is too old 
(found version 12050).
Please update your GPU driver...

My server configuration is:

• GPU: NVIDIA A40
• NVIDIA Driver: supporting CUDA 12.5
• Python: 3.11

I suspect that the issue is caused by a mismatch between the installed PyTorch CUDA version and the NVIDIA driver version.

Could you please provide the exact environment configuration used for your experiments, especially:

1. PyTorch version
2. CUDA version (both toolkit/runtime if applicable)
3. NVIDIA driver version
4. The recommended installation command for PyTorch

For example, an environment.yml, requirements lock file, or Docker image configuration would be very helpful.

Thank you very much for your help!

Start with requirements-agent.txt and the launch command in spatial_agent.entrypoints.launch_gpu_server. Confirm the PyTorch, CUDA, and driver versions used for the experiments, then document the recommended installation command or environment configuration. Done means a reproducer can install the stated environment without relying on an unpinned latest torch version.