Friday, December 2, 2022

ROS experiments

I will be experimenting with the newest ROS2 at the time of this writing; Humble Hawksbill. Image a raspberry PI 4 2Gb with 64-bit rasbian OS on a 32Gb microSD. I am using putty to ssh into the pi and VNCviewer to interface with GUIs. 

Learn docker to install ROS2 on the RPI4

This is my first time using docker. After cloning the docker images repo, I went with the humble ros-base image.

The following are some useful docker commands:

  • docker ps
  • docker images
  • docker container start [CONTAINER-ID]
  • docker exec -ti [CONTAINER-ID] /bin/bash
  • docker container list -f "status=exited"
  • docker commit os_name image_name:version
  • docker container update --help (this might have saved me some time and disk space. I was committing so that I could rerun with new parameters)
  • docker rm container_id_or_name
  • The config file is a json file at the location ‘ /var/lib/docker/containers/[container-id]/config.v2.json
Build the container using the appropriate Dockerfile, then rename it "ROS-TEST" to reference it without having to use the randomly generated container number.
  • $ sudo docker rename [container#] ROS-TEST

Set the timezone in a container

  • sudo rm -rf /etc/localtime
  • sudo ln -s /usr/share/zoneinfo/America/Los_Angeles /etc/localtime
  • date

Moving on to the ROS2: Humble Hawksbill tutorials.

The environment set up without problems. 

  • source /opt/ros/humble/setup.bash
  • echo "source /opt/ros/humble/setup.bash" >> ~/.bashrc
  • printenv | grep -i ROS
  • export ROS_DOMAIN_ID=24
  • echo "export ROS_DOMAIN_ID=24" >> ~/.bashrc

The turtlebot tutorial took a while to download and setup. When running the turtlebot simulator an error occurred:

qt.qpa.xcb: could not connect to display 
qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "" even though it was found.
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Available platform plugins are: eglfs, linuxfb, minimal, minimalegl, offscreen, vnc, xcb.
[ros2run]: Aborted

The sim uses qt and qt can't find a display. Is this a problem with docker, vnc, or a combo of both? 

Make an image of the current container to run with environmental variables set: 

  • --env="Display"
  • --net=host
  • FIX: in RPI terminal "xhost +local:docker". This needs to be done every reboot that you wish to use a GUI. It can also be run while a container is running

"exit" docker. The container is no longer running. "docker container start ROS-TEST", the container is running in the background.

Turtlesim GUI still launches with errors:

libGL error: MESA-LOADER: failed to retrieve device information

MESA-LOADER: failed to retrieve device information

MESA-LOADER: failed to retrieve device information

libGL error: failed to create dri screen

libGL error: failed to load driver: vc4

libGL error: failed to open /dev/dri/card1: No such file or directory

libGL error: failed to load driver: vc4

qt.qpa.xcb: QXcbConnection: XCB error: 148 (Unknown), sequence: 191, resource id: 0, major code: 140 (Unknown), minor code: 20

However, teleop keys work for the movement. I'll look into the errors above if I run into problems. 

  • --volume="$HOME/.Xauthority:/root/.Xauthority:rw"

^ Adding this didn't get rid of the errors. I'm not even sure if that is how to share the RPI's xserver as a volume to the docker container though...

Moving on... rqt install is 1198MB. 

Everything works to finish the CLI tutorials.

The following error came up in Tutorials \Beginner: Client libraries \Using colcon to build packages.

CMake Error at CMakeLists.txt:14 (find_package):
  By not providing "Findexample_interfaces.cmake" in CMAKE_MODULE_PATH this
  project has asked CMake to find a package configuration file provided by
  "example_interfaces", but CMake did not find one.

This is caused by a missing install and remedied by:  

  • sudo apt install ros-humble-example-interfaces

The RPI does not like "colcon build --symlink-install" very much. It went unresponsive for 1h 6m. Plus...more errors. I forgot to start from a clear directory though... When I re ran the build, it maxed out my ram and swap file, interrupting my VNC connection, and became generally unresponsive. 12 hours in, it was still unresponsive. I power cycled the RPI and got back into the container to try again. rm -r the ros2_ws directory and started over... It looks like it is going to have the same problem. Next attempt is to increase the RPI swap file from 100M to 1000M and reduce the amount of RAM the container has access to, it is currently unlimited.

Swap increase:

  1. sudo dphys-swapfile swapoff
  2. sudo nano /etc/dphys-swapfile
  3. CONF_SWAPSIZE=100
  4. CONF_SWAPSIZE=2048 (Increased to 6144 later)
  5. sudo dphys-swapfile setup
  6. sudo dphys-swapfile swapon
  7. sudo reboot
Docker's memory and CPU usage: MEM reference. In step 5, I give docker access to only 3 of the RPI's 4 cpus, 1G of its memory, and 1.5G of swap so that it'll remain responsive and not kill other processes like VNC.
  1. sudo cp /boot/cmdline.txt /boot/cmdline.txt.bak
  2. sudo nano /boot/cmdline.txt
  3. add "cgroup_enable=memory cgroup_memory=1 swapaccount=1" to the end of the line, not on a new line.
  4. verify with "docker info"
  5. docker run -it --name guiROSlimited --memory="1g" --memory-swap="2.5g" --cpus="3" --env="DISPLAY" --net=host [IMAGE NAME]
Now running "colcon build --symlink-install" isn't crashing the pi or making ssh/VNC unsresponsive.
I'm starting to question whether I should have made the swap file 4 or 6G now...  26 minutes in, the RPI swap is at 1.47G and 0/22 modules have been built...45 minutes: the container swap is full :-( ctl+c aborted. increasing swap to 6144.
  • want /var/swap=6144MByte, restricting to config limit: 2048MBytes, checking existing: keeping it
  • uncomment CONF_MAXSWAP=2048 and change to CONF_MAXSWAP=6144
31 minutes later, I forgot to run a new docker container with an increased swap. It maxed out faster. I created a new container for a fresh start.
  • docker run -it --name rosFresh1 --memory="1.25g" --memory-swap="4.5g" --cpus="3" --env="DISPLAY" --net=host ros_docker
  • sudo apt update
  • sudo apt upgrade
I updated chrome and lost a chunk of the blog. colcon on the RPI works better with only one package processed at a time. Although, I have not tried 2 or 3. Use the following tag to use one CPU or the second to use a specific amount:
  • --executor sequential
  • --parallel-workers NUMBER
  • This is only necessary when building multiple packages. It is not necessary yet when using --packages-select
Sometimes I create packages in the root (ros2_ws) directory, instead of 'src'. rm -r the created folder. To remove; "build", "install", and "log".
  • rm -rf build/ install/ log/
  • colcon build --executor sequential
Short of it is that I have restarted from the ros-desktop image. No fundamental errors to speak of as of yet.
Here is a free book on ROS.

No comments:

Post a Comment