Building Containerised Development platform

The containerised development platform is shared by both sim and real hardware. This improve the deployability from sim to real with minimal code change.

Curent release uses Nvidia Isaac Ros 3.2 as the basis for this platform

Prerequisities

• Finish Flasing Hadrware or setting up Sim host.

1) Install Docker + NVIDIA Container Toolkit + Git LFS

Install docker with

• Guide from JetsonHacks

  cd hacks
  git clone https://github.com/jetsonhacks/install-docker.git
  cd install-docker
  bash ./install_nvidia_docker.sh
  bash ./configure_nvidia_docker.sh
  

Or use

• Docker: https://docs.docker.com/engine/install/ubuntu/
• NVIDIA Container Toolkit: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html

Git LFS: https://git-lfs.com/

curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
# Install Git LFS
sudo apt update
sudo apt install git-lfs -y

# Initialize Git LFS for your user
git lfs install

To verify CDI interfaces sudo nvidia-ctk cdi list should output

INFO[0000] Found 2 CDI devices                          
nvidia.com/pva=0
nvidia.com/pva=all

if no CDI devices are found

sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker

2) Setup Workspace

cd $HOME
mkdir workspace && cd workspace
mkdir -p agipix_control/src dds lidar_ws/src raicam-ros/src logging/bags logging/flogs logging/crashreports
cd raicam-ros/src
git clone https://github.com/SasaKuruppuarachchi/isaac_ros_common.git # TODO

echo "export WORKSPACES_DIR="$HOME/workspace"" >> ~/.bashrc
echo "export ISAAC_ROS_DEV_DIR="$HOME/workspace/raicam-ros"" >> ~/.bashrc
echo "alias agidocker='cd $HOME/workspace/raicam-ros/src/isaac_ros_common/scripts/ && ./run_dev.sh --skip-registry-check'" >> ~/.bashrc
echo "alias killagidocker='docker container kill isaac_ros_dev-aarch64-container'" >> ~/.bashrc
source ~/.bashrc

3) Build Docker

In this step we pull source images and build the docker continer. To customise the docker configuration edit

code ~/workspace/raicam-ros/src/isaac_ros_common/scripts/.isaac_ros_common-config
# append the additional docker configuration layers you need built

cd ~/workspace/raicam-ros/src/isaac_ros_common/scripts/ && ./run_dev.sh # this alias builds the container

The build will take more than an hour and if successfull you will see Running isaac_ros_dev-aarch64-container and the container will start running in the background.

Run agidocker to attach to the container interactively.

Verify Pytorch with cuda on dcker

agidocker
# Inside docker
python3 - << 'EOF'
import torch
print("CUDA:", torch.cuda.is_available())
print("Device:", torch.cuda.get_device_name(0))
EOF

4) Post setup (Optional)

Build Opencv with cuda From source

Note! OpenCV =< 4.8 had an incompatability with Cuda 12.x

agidocker
# Inside docker
cd /worckspaces/isaac_ros-dev/src/isaac_ros_common
sudo ./build_ocv_cuda.sh
# This will build and install the OCV for this running container locally 
# but saves the build files on the mounted folder so we can use the Dockerfile.opencv layer later

Build torchvision from source (matches NVIDIA PyTorch)

export FORCE_CUDA=1
git clone -b v0.17.2 https://github.com/pytorch/vision.git
cd vision
sudo -E python setup.py develop  # use 'install' if you don't need editable mode

5) Prune Docker cash (Optional)

Docker builder cash consumes lot of memory. You can clean manualy or schedule a clean as follows

docker system prune -a --volumes

Output will show the size of the volume freed up Total reclaimed space: 60.52GB

Optional not recommended: Prevent buildup with a scheduled cleanup (e.g., every week):

docker system prune -a --filter "until=168h" --volumes

sudo du -h --max-depth=1 /home/<user>/workspace | sort -hr

rm -rf ~/.cache/*

rm -rf ~/.local/share/Trash/*
rm -rf ~/.local/share/pip
rm -rf ~/.local/share/virtualenvs