lasuomela / visual_robot_localization Goto Github PK

Hello,

I am currently working on the visual_robot_localization project and have encountered an issue I'm unable to resolve. Specifically, I'm facing an ImportError related to libcudnn.so.8 when attempting to run the ./do_SfM.sh script. The traceback suggests that the torch import in SfM_from_camera_poses.py is failing due to this.

Here is the error message for reference:

ImportError: libcudnn.so.8: cannot open shared object file: No such file or directory

Steps Taken:

The Docker container is built and runs without errors.
colcon build also completes without errors.
Checked that CUDA and cuDNN libraries are available in the Docker container. However, I was not able to locate them using ls /usr/local/cuda/lib64 and ls /usr/lib/x86_64-linux-gnu/.
Tried explicitly adding library paths in the Python script but to no avail.

Environment:

Ubuntu 20.4
CUDA:
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243
cuDNN [version]:
Python 3.9
Docker [version]: Docker version 24.0.5, build ced0996
Torch [version]: 1.12.1+cu102

I would be extremely grateful if anyone could provide guidance on resolving this issue. Any advice or troubleshooting tips would be much appreciated.

Thank you for your time and assistance.

Best regards,
KJ