Skip to main content

Setup Guide — Physical AI Tools

Overview

This guide covers Physical AI Tools-specific setup for OMX.

Before starting this page, complete ROS 2 Setup. It prepares the OpenMANIPULATOR Docker container and the leader/follower serial port settings required for Physical AI Tools.

info

System Requirements

Recommended OSUbuntu
Hardware requirementNVIDIA GPU (CUDA-capable)

Software Setup

Prerequisites

  • Operating System: Any Linux distribution

    • The container runs Ubuntu 24.04 + ROS 2 Jazzy
    • The Host OS version does not need to match.
  • Docker Engine

    • Install using the official Docker guide

    • After installation:

      sudo usermod -aG docker $USER
      sudo systemctl enable docker
      docker run hello-world
  • Git

    sudo apt install git
  • NVIDIA Container Toolkit

Set up Physical AI Tools Docker Container

1. Start the Docker container

Clone the repository along with all required submodules:

USER PC

git clone --recurse-submodules https://github.com/ROBOTIS-GIT/physical_ai_tools.git

Start the Physical AI Tools Docker container with the following command:

cd physical_ai_tools/docker && ./container.sh start

2. Configure camera topics

If you are using more than one camera or want to use a custom camera, list the available camera topics and choose the one you want to use:

Enter the Physical AI Tools Docker container:

USER PC

./container.sh enter

list the available topics to find your camera stream:

USER PC

ros2 topic list

And open the configuration file and update it as described below:

USER PC 🐋 PHYSICAL AI TOOLS

sudo nano ~/ros2_ws/src/physical_ai_tools/physical_ai_server/config/omx_f_config.yaml

Then update the fields outlined in red in the UI to point to your desired camera topic.

Configure camera topic in the UI
info

Note: The topic you set must always end with compressed
(for example, camera1/image_raw/compressed).

🎉 Physical AI Tools Container Setup Complete!

Before recording demonstrations, launch the camera node, Physical AI Server, and ROS 2 teleoperation node from Dataset Preparation.