Machine Learning-Based Visual Recognition Approach For Handling Complex Components In Robotic Fabrication

Construction waste poses significant environmental challenges, motivating research into recycling waste materials to enable a circular economy. However, these materials' diverse shapes, sizes, and characteristics make robot identification and handling difficult. Robotic manipulation is crucial for streamlining manufacturing processes, but grasping objects with complex geometries remains challenging. This paper proposes a machine learning-based visual recognition approach to enhance autonomous robot operation in construction environments. The research aims to improve construction efficiency, accuracy, and customization by integrating computer vision, machine learning, and autonomous manipulation. The experiment setup includes equipping a construction robot with an RGB camera, depth sensor, and customized vacuum gripper to identify and grasp complex components. The results demonstrate the potential of machine learning-based visual recognition in improving construction robots' robots' ability to handle diverse waste materials. This approach bridges the gap between virtual design and real-world implementation, supporting effective waste management and reuse to achieve a circular economy.