This research presents the development and validation of a machine learning-assisted closed-loop control system designed for robotic 3D clay printing, with a particular focus on enhancing print quality and structural stability of complex overhang geometries. Unlike traditional open-loop manufacturing processes, the proposed framework integrates real-time visual feedback captured by dual Raspberry Pi cameras and leverages a multi-objective neural network model to dynamically adjust the robotic printing speed. This enables the system to detect and correct extrusion anomalies such as over-extrusion and under-extrusion, thereby significantly improving the printability and quality of challenging overhang structures.
Two advanced deep learning architectures, ResNet-56 and a DINOv2-based hybrid network, were systematically evaluated to determine their effectiveness in defect detection and prediction of overhang success. The system was implemented using a UR5 robotic arm equipped with a clay extruder, demonstrating practical feasibility within a laboratory environment. Experimental results show that the closed-loop control approach substantially enhances print consistency, reduces structural failures, and maintains geometric accuracy compared to baseline open-loop methods.
This foundational work lays the groundwork for future scaling to construction-scale additive
manufacturing, highlighting the potential to extend this system to other construction materials such as concrete. The study discusses key challenges including material variability, sensor integration, real time control complexity, and robotic motion planning, and provides strategic recommendations for future research aimed at achieving robust, adaptive, and scalable additive manufacturing systems for complex architectural applications.