CONVINCE SIT-AW-AIP

This project allows you to use our selected VLM, `Qwen2.5-VL-7B-Instruct`_, deployed locally, for action classification on robotics tasks. The particularity of this package is the integrability of the pipeline using ROS2. The deployment on a robotic system is straightforward, provided that the used middleware is ROS2.

Contents

• Requirements
  • Install ros2 (humble)
  • Install python modules
• Adapt to your setup
  • Data organisation
  • Conversion script
  • Prompts
  • Add your id in the maps
  • Define parameters
• Test on an example
• Include to your project
  • Send inference request
  • Add launch file to your desired launch instance
  • Build your project
  • Visualize publisher info