Developing a ROS2 infrastructure and control system

Abstract

The increasing complexity of industrial automation demands agile and adaptive control systems capable of dynamic task execution. This thesis addresses these challenges by developing a ROS2-based infrastructure and control system for autonomous robots in a simulated factory environment, aligned with the Agile Robotics for Industrial Automation Competition (ARIAC) 2024. The proposed solution integrates an operation runner for task coordination and Convolutional Neural Network (CNN) for real-time part classification, aiming to optimize adaptability and efficiency in a dynamic manufacturing setting.

The control system leverages ROS2’s communication framework such as topics and services to manage Automated Guided Vehicle (AGV), robotic arms, and competition infrastructure such as orders. The operation runner dynamically coordinate tasks by evaluating preconditions and postconditions of executable operations, enabling scalable control of multiple robots. A CNN, trained on HSV-masked and augmented image data, achieves robust part classification despite variations in orientation. The operation runner demonstrated success in AGV coordination and scalability, while the CNN demonstrates real-time capabilities with classification tasks. However, the integration of the control system and vision components together and into the ARIAC competition framework was not fully realized, mainly due to time constraints and technical challenges.