<p>An automated robotic platform with radiation detectors can aid in localizing radiation sources under various scenarios while preventing human exposure to ionizing radiation. This study proposes a novel robot-assisted radiation source localization process for multiple sources. The method is applicable in an environment with an unknown number of radiation sources and is capable of estimating source locations from sparse robot measurements while following a predefined path. A source-count classifier is trained with 59,911 high-fidelity 2D radiation flux simulation samples to derive the estimated number of radiation sources from the sparse measurements. Then, the derived number of sources, flux measurements, and measured locations are fed into physics-informed neural networks to localize multiple radiation sources precisely. The developed method outperforms other related research in terms of estimation error. Additionally, the method has been demonstrated with 3D radiation flux with different geometric configurations by case studies using Unreal Engine 5.</p>