Lung infections in children remain challenging to diagnose despite the use of bronchoalveolar lavage (BAL), in which sterile saline flushes a lung segment and a sample is tested for pathogens. The yield of BAL is low as less fluid can be inserted since the size of the children's lungs is small, and the site of infection is often far distal to the position of the bronchoscope. In addition, it poses additional risks or complications as the saline is also absorbed into the lung parenchyma. A smaller, robotic device could traverse further along the bronchial tree and guide to the site of infection, increasing the successful identification of the pathogen to decrease complications of the BAL procedure. Likewise, treatment of endovascular disease also often involves catheterization, which requires a highly experienced interventional cardiologist to place guidewires at the target location, typically using fluoroscopic imaging to localize the guidewire during the procedure. However, inaccurate placement of these robotic guidewires can lead to unsuccessful procedural outcomes. Robotic manipulation of guidewires could provide better control of the device. We propose an automated approach for closed-loop control of the modified COaxially Aligned STeerable (COAST) guidewire under fluoroscopic imaging in phantoms. Deep learning architectures are used to determine the workspace, position, and orientation of the guidewire in two dimensions. This information is used for path planning and traversal in a phantom. From the centerline of the guidewire, the parameters for the modified COAST guidewire are determined to generate estimated joint actuations, which are compared to the actual joint actuations to autonomously correct for the error in the guidewire tip position for closed-loop control.