As of now, we have different instructions and docker run command variants for different GPU architectures and OS. In order to make things easier for users, we should be able to automatically detect available GPUs and set the required environment variables accordingly. We should prioritize NVIDIA architectures, and still allow users to override the automatic selection process.
That would be ideal, but some users that have Nvidia have problems installing the necessary packages. Also I'm transfering this issue to RoboticsAcademy
As of now, we have different instructions and
docker run
command variants for different GPU architectures and OS. In order to make things easier for users, we should be able to automatically detect available GPUs and set the required environment variables accordingly. We should prioritize NVIDIA architectures, and still allow users to override the automatic selection process.