This paper introduces Physics-Informed Machine Learning (PIML), an advanced neural network architecture designed to seamlessly integrate established physical models into machine learning frameworks. By harnessing fundamental physical information, PIML significantly enhances training performance on specific datasets, offering a robust approach to modeling complex systems. Focusing on critical domains within transportation such as car-following models and traffic state estimation, our study provides a comprehensive synthesis of existing research findings and analyzes the strengths and limitations inherent in traditional approaches. Furthermore, we delve into the practical application of PIML within these domains, elucidating its underlying mechanisms, efficacy, and synergies with conventional models.