0
$\begingroup$

Rosanswers logo

jumping from https://gitlab.com/autowarefoundation/autoware.auto/AutowareAuto/-/issues/938

I am trying to start LGSVL with a clean autoware.auto environment I have installed docker, nvidia-docker2, ade, restarted pc, etc according to the documentation (https://autowarefoundation.gitlab.io/autoware.auto/AutowareAuto/lgsvl.html) When I start lgsvl simulator inside ade, a window appears then suddenly disappears. No error message, the program only returns with error code 1.

From the AutowareAuto master/1.0.0 branch I tried the following:

$ ade --rc .aderc-lgsvl start --update --enter

$ /opt/lgsvl/simulator

The documentations mentions mesa-vulkan-drivers may conflict with lgsvl, but removing the driver does not fix the issue.

sourcing /opt/AutowareAuto/setup.bash does not fix the issue either

Full logs here: https://gitlab.com/-/snippets/2095270

Originally posted by vrichard on ROS Answers with karma: 108 on 2021-03-25

Post score: 2

Original comments

Comment by Josh Whitley on 2021-03-25:
What kind of GPU does your host machine have? What OS/kernel is it running? The one thing I didn't see you mention installing is the Nvidia proprietary graphics driver. Can you run nvidia-smi on your host and post the output?

Comment by vrichard on 2021-03-25:
I have the following setup:

  • Ubuntu 18.04
  • kernel 5.4.0-67-generic
  • nvidia driver 450.102.04
  • cuda 11.0
  • gpu is Quadro P4000
  • nvidia-smi gives this output from host: https://gitlab.com/-/snippets/2095278
  • result is similar from ade except no Processes are detected

Comment by Josh Whitley on 2021-03-29:
This one may need help from LG. Can you please create a new snippet with the contents of the file .config/unity3d/LG Silicon Valley Lab/LGSVL Simulator/player.log in your adehome folder? This is the log from the simulator and may help with troubleshooting.

Comment by vrichard on 2021-03-29:
https://gitlab.com/-/snippets/2097364 According to the logs it seems it is just the vulkan drivers that dont work

vulkaninfo also give me an error: WARNING: [Loader Message] Code 0 : loader_icd_scan: Can not find 'ICD' object in ICD JSON file /usr/share/vulkan/icd.d/nvidia_layers.json. Skipping ICD JSON /build/vulkan-tools-136mCR/vulkan-tools-1.1.126.0+dfsg1/vulkaninfo/vulkaninfo.h:399: failed with ERROR_INITIALIZATION_FAILED

The first warning line is explained here: https://vulkan.lunarg.com/issue/home?limit=10;q=;mine=false;org=false;khronos=false;lunarg=false;indie=false;status=new,open Moving the file /usr/share/vulkan/icd.d/nvidia_layers.json away does not fix the issue

As for the ERROR_INITIALIZATION_FAILED error, people usually face this issue when their GPU is too old and cannot support vulkan. According to this page: https://developer.nvidia.com/vulkan-driver#:~:text=Pascal%3A%20TITAN%20Xp%2C%20TITAN%20X,Volta%3A%20TITAN%20V, the Quadro P4000 is supposed to be supported

Comment by vrichard on 2021-03-29:
But after some more digging I realized some people had similar vulkan issues when using X forwarding. Nowadays I am working remotely using chrome-remote-desktop and it seems vulkan does not support such use case: https://askubuntu.com/questions/1231155/vulkan-driver-stops-working-when-running-in-google-remote-desktop

However the comment provides some workaround. I will try it out later

Improve this question
$\endgroup$

2 Answers 2

Reset to default
0
$\begingroup$

Rosanswers logo

This is a somewhat known issue, especially if your CPU also has integrated graphics. Inside the container, run sudo apt purge mesa-vulkan-drivers and the issue should be resolved. See https://gitlab.com/autowarefoundation/autoware.auto/AutowareAuto/-/issues/1140 for details.

Originally posted by Josh Whitley with karma: 1766 on 2021-05-26

This answer was ACCEPTED on the original site

Post score: 0

Original comments

Comment by kk2105 on 2021-06-08:
Thanks for the answer. Every time I enter inside the ade, I need to uninstall the vulkan-drivers. Is there a better way to remove so that it does not ask me everytime I enter ade?

Comment by Josh Whitley on 2021-06-08:
See https://gitlab.com/autowarefoundation/autoware.auto/AutowareAuto/-/issues/1140 for discussion about this issue.

Improve this answer
$\endgroup$
0
$\begingroup$

Rosanswers logo

Hi Why don't you use the SVL simulator instead of LGSVL? This is an old version...also you should use the simulator externally, not inside the docker image (you can use but is not good to develop). Then later you could just connect with ROS/Autoware using the ros2-web-bridge or lgsvlbridge..... But I believe the issue is regarding the need to install the libvulkan. For me at least it worked. Take a look on this link: https://zoomadmin.com/HowToInstall/UbuntuPackage/libvulkan-dev

The cmds in this link should solve your issue. I have also used Autoware.Auto with LGSVL internally and had lot of problems. I needed to work around their .aderc file and customize the image to be built properly.

You can open your .aderc-amd64-foxy-lgsvl file inside adehome/AutowareAuto folder (HINT: Use Ctrl+h to enable visualization of the hidden files in this folder) ERASE all the lines and then copy and paste the lines below:

export ADE_DOCKER_RUN_ARGS="--cap-add=SYS_PTRACE --net=host --privileged --add-host ade:127.0.0.1 -e RMW_IMPLEMENTATION=rmw_cyclonedds_cpp"
export ADE_GITLAB=gitlab.com
export ADE_REGISTRY=registry.gitlab.com
export ADE_IMAGES="
  registry.gitlab.com/autowarefoundation/autoware.auto/autowareauto/amd64/ade-foxy:master
  registry.gitlab.com/autowarefoundation/autoware.auto/autowareauto/amd64/binary-foxy:master
  registry.gitlab.com/autowarefoundation/autoware.auto/ade-lgsvl/foxy:2020.06
"

Afterwards just start and enter the image again. You must be aware that you need to ALWAYS remove the line that enable the Nvidia is set to True (Take a look if you have this line in your .aderc file) and enable to false or remove (as I always do):

export ADE_DOCKER_RUN_ARGS="--cap-add=SYS_PTRACE --net=host --privileged --add-host ade:127.0.0.1 -e RMW_IMPLEMENTATION=rmw_cyclonedds_cpp --runtime=nvidia -ti --rm -e DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix \
-e NVIDIA_VISIBLE_DEVICES=all \
-e NVIDIA_DRIVER_CAPABILITIES=compute,utility,display"
export ADE_GITLAB=gitlab.com
export ADE_REGISTRY=registry.gitlab.com
export ADE_DISABLE_NVIDIA_DOCKER=false
export ADE_IMAGES="
  registry.gitlab.com/autowarefoundation/autoware.auto/autowareauto/amd64/ade-foxy:master   registry.gitlab.com/autowarefoundation/autoware.auto/autowareauto/amd64/binary-foxy:master
  registry.gitlab.com/autowarefoundation/autoware.auto/ade-lgsvl/foxy:2020.06
  nvidia/cuda:11.0-base
"

You can take a look here too for more deeper understanding of Nvidia driver issues: https://gitlab.com/autowarefoundation/autoware.auto/AutowareAuto/-/issues/502

Finally if the lgsvl was still not enabled you should clean your Autoware.Auto system and re-compile, rebuild...It is quicker just follow the "CLEANUP" topic here: https://autowarefoundation.gitlab.io/autoware.auto/AutowareAuto/installation-ade.html Or better follow "Start from clean State" topic here: https://autowarefoundation.gitlab.io/autoware.auto/AutowareAuto/building.html#building-colcon-defaults

I hope you can enable the lgsvl inside docker :)

Originally posted by Vini71 with karma: 266 on 2021-05-25

This answer was NOT ACCEPTED on the original site

Post score: 1

Original comments

Comment by vrichard on 2021-05-26:
Hi, Thank you for the detailed answer. Although it seems to be very practical to run the simulator outside the autoware.auto docker, my initial motivation was to reproduce the demos as described in the documentation. As I described in my latest comment, the issue comes from the fact I am working using remote desktop to access my work computer and unfortunately the vulkan driver does not work in such configuration. I pointed to a link that shows a hacky way to make to work anyway, but I have not been able to test the whole thing in practice yet

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.