Notes on installing isaac SDK on Jetson Nano

I’m trying to get the Isaac SDK installed on Jetson Nano (with the Nano acting as the desktop). Here are my notes so far (I’ll try to keep this thread updated.)

To get install_dependencies.sh to work, I had to install:

  • liblapack-dev
  • libffi-dev
  • gfortran
  • llvm-7

Then I had to set

LLVM_CONFIG=/usr/lib/llvm-7/bin/llvm-config

This gets most things installed.

Outstanding issues:

  • Bazel will say it installed correctly, but the script downloads a tarball full of x86_64 binaries, which is not super helpful. (Error message ``` /usr/local/lib/bazel/bin/bazel-real: cannot execute binary file: Exec format error ``` I think I will need to build Bazel from source, since the github releases page only seems to have x86_64 builds.
  • I don't know where to find the nvidia-driver-418 package for the Jetson Nano. I'm unclear if I need a separate package, or if the equivalent drivers are already installed
  • I'm having a similar issue to https://devtalk.nvidia.com/default/topic/1044858/jetson-tx2/tensorflow-integrate-with-baselines-package-in-openai-gym/ : ``` ValueError: invalid version number '1.13.1+nv19.4' ``` I think this is because baselines is trying to copy the tensorflow version number into its own version number, and it doesn't like the +nv19.4 part.

We do not support Jetson as development platform. One main reason are for example an extremely long compilation time on Nano. Cross compilation, deployment and running on Jetson is however extremely easy to use and explained in the docs.

Getting a version of bazel on the nano would be really useful. Then it would be possible to build the latest version of tensorflow on the nano.

@Davidweikersdorfer
Which doc explains cross compilation, deployment and running on Jetson nano? Can you please share.