Isaac ROS Setup on Thor#

Set up the Isaac ROS development environment on the computer connected to the Unitree G1. This provides the isaac-ros CLI, ROS 2 workspace, Isaac ROS source repositories, containerized environment, RealSense support, and packages used by the Unitree G1 real static apple workflow path.

Prerequisites and Version Boundary#

Run these steps on the Thor host. Enter the Isaac ROS container only where the procedure requests activation. Run each command in order and continue only after it succeeds. Review G1 safety before powering on or connecting the robot.

The setup below retains the JetPack 7.1 / Jetson Linux R38.4 and Isaac ROS 4.5 baseline. The Isaac ROS 4.5 requirements identify that platform combination. The source checkouts below use its v4.5-0 release snapshots.

Important

This release-aligned setup does not reconstruct the exact v0.1 qualified image. The repository does not lock all binary package versions or the container image digest. Preserve an existing qualified installation until its replacement is qualified. Repository CI does not test physical hardware behavior.

Source documentation

These references are pinned to Isaac ROS 4.5. They cover setup and the Unitree G1 teleoperation and deployment packages:

Initialize the Compute Platform#

  1. Complete the Jetson AGX Thor Quick Start Guide, selecting JetPack 7.1 / R38.4 for this baseline.

  2. Install the matching JetPack components using the Jetson AGX Thor JetPack setup guide. The latest guide may describe a newer release; keep the selected release consistent.

  3. Verify the JetPack release:

    cat /etc/nv_tegra_release
    

    Confirm the output includes R38 (release), REVISION: 4.0.

  4. Set the Thor power mode to MAXN. See the power mode controls guide.

    sudo /usr/sbin/nvpmodel -m 0
    

Create the Isaac ROS Workspace#

  1. Create the workspace:

    mkdir -p "$HOME/workspaces/isaac_ros-dev/src"
    
  2. Add its environment variable:

    echo "export ISAAC_ROS_WS=\"\${ISAAC_ROS_WS:-$HOME/workspaces/isaac_ros-dev/}\"" >> ~/.bashrc
    
  3. Reload the environment:

    source ~/.bashrc
    
  4. Enter the source directory and clone the Isaac ROS 4.5 snapshots in the same terminal:

    cd "${ISAAC_ROS_WS}/src"
    
    git clone --branch v4.5-0 https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_physical_ai.git
    
    git clone --branch v4.5-0 https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_robots.git
    
    git clone --branch v4.5-0 https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_data_tools.git
    
  5. Verify the checked-out commits with git -C <repository> rev-parse HEAD:

    Repository

    v4.5-0 Commit

    isaac_ros_physical_ai

    4172a10b274116c461780bfe8aa04f5fe24eb4ae

    isaac_ros_robots

    d677275d4dc7a0bc75591251da5ac5f0250bbd5c

    isaac_ros_data_tools

    727510a0a7a6da73f722d0b65ca4709488d95d69

    Stop if a tag resolves differently. These checkouts provide scripts and converter source; they do not pin the binary packages in the Docker image.

Configure CloudXR Firewall Ports#

CloudXR requires network ports to be open. For Quest and PICO headsets using the WebXR client, open the CloudXR runtime and WebSocket Secure proxy ports when ufw is enabled:

  1. Open the minimum ports:

    sudo ufw allow 47998/udp
    
    sudo ufw allow 49100,48322/tcp
    
  2. If you run the WebXR client from source, also open its web-server ports:

    sudo ufw allow 8080,8443/tcp
    

Install and Initialize the Isaac ROS CLI#

These instructions use the Docker workflow. For bare-metal or virtual environment workflows, see the Isaac ROS initialization documentation.

  1. Check the host locale:

    locale
    

    Set it to UTF-8 in the same terminal:

    sudo apt update
    
    sudo apt install locales
    
    sudo locale-gen en_US en_US.UTF-8
    
    sudo update-locale LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
    
    export LANG=en_US.UTF-8
    

    Verify the settings:

    locale
    
  2. Install Apt repository dependencies:

    sudo apt update
    
    sudo apt install curl gnupg
    
    sudo apt install software-properties-common
    
    sudo add-apt-repository universe
    
  3. Configure the public NVIDIA Isaac ROS Apt repository for Jetson AGX Thor. Run these commands in the same host terminal:

    k="/usr/share/keyrings/nvidia-isaac-ros.gpg"
    
    curl -fsSL https://isaac.download.nvidia.com/isaac-ros/repos.key | sudo gpg --dearmor \
        | sudo tee -a $k > /dev/null
    
    f="/etc/apt/sources.list.d/nvidia-isaac-ros.list"
    
    sudo touch $f
    
    s="deb [signed-by=$k] https://isaac.download.nvidia.com/isaac-ros/release-4.5 noble-jetpack main"
    
    grep -qxF "$s" $f || echo "$s" | sudo tee -a $f
    
    sudo apt-get update
    
  4. Install the Isaac ROS CLI from the configured 4.5 repository:

    sudo apt-get install isaac-ros-cli
    
  5. Configure Docker to use the NVIDIA runtime by default:

    sudo nvidia-ctk runtime configure --runtime=docker --set-as-default
    
  6. Review the generated configuration:

    cat /etc/docker/daemon.json
    

    Expected output:

    {
      "runtimes": {
        "nvidia": {
          "args": [],
          "path": "nvidia-container-runtime"
        }
      },
      "default-runtime": "nvidia"
    }
    
  7. Add your user to the Docker group:

    sudo usermod -aG docker $USER
    

    Start a shell with the new group membership:

    newgrp docker
    
  8. Restart Docker:

    sudo systemctl daemon-reload
    
    sudo systemctl restart docker
    
  9. Verify Docker and the NVIDIA runtime. Docker must list its runtimes, hello-world must succeed, and the GPU check must print NVIDIA runtime OK.

    docker info | grep -E "Runtimes|Default Runtime"
    
    docker run --rm hello-world
    
    docker run --rm --gpus all ubuntu:24.04 bash -lc 'echo "NVIDIA runtime OK"'
    

    If nvidia is not the default runtime, correct /etc/docker/daemon.json, restart Docker, and repeat these checks before continuing.

  10. Verify that the container can access the GPU:

docker run --rm --gpus all ubuntu:24.04 nvidia-smi

Successful output identifies NVIDIA Thor and reports no runtime-access error. Complete the Jetson AGX Thor Docker setup before continuing if this check fails.

  1. Select Docker mode:

sudo isaac-ros init docker
  1. Enable real-time scheduling support:

echo '--ulimit rtprio=99' > ~/.isaac_ros_dev-dockerargs

This allows the ROS 2 controller manager to run on a real-time thread.

Configure RealSense Support#

This workflow uses the RealSense camera mounted in the Unitree G1’s head. Connect it directly to Thor when using the Thor backpack; use a USB extension cable when Thor is external. The Ethernet connection between Thor and the Unitree G1 remains direct. See Unitree G1 Introduction and Safety.

  1. Start with the camera unplugged and install the RealSense udev rules:

    wget https://raw.githubusercontent.com/realsenseai/librealsense/v2.56.3/config/99-realsense-libusb.rules
    
    sudo mv 99-realsense-libusb.rules /etc/udev/rules.d/
    
    sudo udevadm control --reload-rules
    
    sudo udevadm trigger
    

    Confirm that each command exits without an error, then connect the camera.

  2. Configure the Isaac ROS Docker image with Physical AI workflow and RealSense support:

    mkdir -p ~/.config/isaac-ros-cli
    
    tee ~/.config/isaac-ros-cli/config.yaml << 'EOF'
    docker:
      image:
        additional_image_keys:
          - gr00t_workflow
          - realsense
    EOF
    

    The realsense layer supplies camera drivers. The gr00t_workflow layer supplies the ROS packages required by this workflow.

  3. Build and activate the environment. Image keys select features; they do not lock a container digest or all package versions:

    isaac-ros activate --build-local
    
  4. Verify camera detection:

    rs-enumerate-devices
    
  5. In a graphical environment, optionally inspect the camera:

    realsense-viewer
    

    If the camera is not detected, recheck its USB connection, host udev rules, container configuration, and device enumeration before continuing. For the rs-enumerate-devices, udev, firmware, and streaming failures this workflow hits, see Troubleshooting.

Next Stage#

The camera enumeration must identify the connected RealSense. Record the source commits, installed package versions, and resulting container image digest with the hardware qualification record. The source repositories are in ${ISAAC_ROS_WS}/src. The gr00t_workflow layer includes:

  • ros-$ROS_DISTRO-isaac-ros-unitree-g1-teleop-bringup

  • ros-$ROS_DISTRO-isaac-ros-unitree-g1-recorder

  • ros-$ROS_DISTRO-isaac-ros-unitree-g1-gr00t

Continue to Unitree G1 Teleoperation.